1 PHOTOACOUSTIC TECHNOLOGIES AND ITS APPLICATION IN THE STUDY OF EPILEPTIC S EIZURES IN RATS By BO WANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FO R THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2014
2 Â© 2014 Bo Wang
3 To my super visor, Dr. Jiang; M y dad and mom; And all who have been supportive to me in my life
4 ACKNOWLEDGMENTS This wor k is the culmination of my work and thoughts in the f ive more years in UF. However, I could never have finished it without the support of many people, whom I d like to share the joy with and express my appreciation to . First of all , I want to thank my sup ervisor , Professor Jiang. He gave me the opportunity and financial support to continue my training in biomedical imaging and education towards Ph . D . in this beautiful country far across the sea from China . Over the years, h e provided immeasurable support t o me through seemingly endless dedication and attention. H e cultivated my interest in research and my scientific personality , and he always encouraged me to be creative in work, which will certainly benefit my career . Then , I need to thank all other membe rs in my academic committee: Dr. Mingzhou Ding, Dr. David Gilland, and Dr. Henry Zemuda . The y provide invaluable comments and suggestions to my study and research. I also appreciate the valuable time they spen d reading my dissertation and attending my defe nse . I also need to thank the guys who have already left the lab in recent years. When I just got into the lab, Dr. Yao Sun and Dr. Qizhi Zhang guided me into the new field of photoacoustics and taught me the experimental skills I needed. Dr. Zhen Yuan tau ght me the knowledge in finite element based photoacoustic reconstruction algorithm and computation . They also gave me valuable support in accommodating the life here in the beginning. Dr. Jianjun Yang and Dr. Liangzhong Xiang helped me to build the multi channel system . I really enjoy the collaboration with them. Finally, I would like to thank my family and friends for their support and encouragement through the years. To my current lab mates and friends here in UF , I can t forget the time we spent having parties , celebrating festivals , and riding for view of sights together. It s you making
5 these four more years here seem so short, wonderful and unforgettable . To my families in China, You provide limitless support through not only this PhD, but my entire l ife. I really miss you all .
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 8 LIST OF FIGURES ................................ ................................ ................................ ......................... 9 LIST OF ABBREVIATIONS ................................ ................................ ................................ ........ 11 ABSTRACT ................................ ................................ ................................ ................................ ... 13 CHAPTER 1 BACKGROUND AND SIGNIFICANCE ................................ ................................ .................. 15 1.1 Photoacoustics ................................ ................................ ................................ .................. 15 1.2 Epilepsy ................................ ................................ ................................ ............................ 19 1.3 Motivation and Objectives ................................ ................................ ................................ 22 2 MULTI CHANNEL BASED REAL TIME PAT SYSTEM ................................ ..................... 25 2.1 System Overview ................................ ................................ ................................ .............. 25 2.2 Multi Channel Data Acquisition System ................................ ................................ .......... 26 2.3 System Calibration and Performance ................................ ................................ ............... 27 2.4 I n vivo Experiments ................................ ................................ ................................ .......... 29 2.5 Conclusion ................................ ................................ ................................ ........................ 30 3 FOUR DIMENSIONAL PAT SYSTEM ................................ ................................ ................... 38 3.1 Pre Investigation Works ................................ ................................ ................................ ... 38 3.2 4D PAT System Based on a Sparse Spherical Transducer Array ................................ .... 40 3.3 System Performance ................................ ................................ ................................ ......... 41 3.3.1 Syst em Characterization ................................ ................................ ........................... 42 3.3.2 Static Phantom Experiments ................................ ................................ .................... 43 3.3.3 Dynamic 3D Phantom Results ................................ ................................ ................. 43 3.4 Real Time imaging of epileptic foci ................................ ................................ ................. 43 3.5 GPU Accelerated 3D Delay&sum Algorithm ................................ ................................ .. 45 3. 6 Conclusion ................................ ................................ ................................ ........................ 48 4 ICTAL NEUROVASCULAR COUPLING STUDY BY 4D PAT SYSTE M ........................... 55 4.1 Background ................................ ................................ ................................ ....................... 55 4.2 Experimental S etup and M ethods ................................ ................................ ..................... 57 4.3 Data Processing for Neurovascular Coupling Study ................................ ........................ 59 4.4 Res ults ................................ ................................ ................................ ............................... 60
7 4.4.1 Seizure Generation ................................ ................................ ................................ ... 60 4.4.2 PAT brain vascular images ................................ ................................ ...................... 60 4.4.3 Hemodynamic Signals in Selected ROIs ................................ ................................ . 61 4.4.4 Correlation Study ................................ ................................ ................................ ..... 63 4.4.5 Granger Causality Analysis ................................ ................................ ..................... 64 4.5 Discussion ................................ ................................ ................................ ......................... 66 4.5.1 Significance of This Work ................................ ................................ ....................... 66 4. 5.2 Data Processing Method ................................ ................................ .......................... 68 4.5.3 Other Factors in Neurovascular Coupling ................................ ................................ 69 4.5.4 Limitations ................................ ................................ ................................ ............... 70 4.6 Conclusion ................................ ................................ ................................ ........................ 70 5 SIMULTANEOUS PAT/EEG RECORDING OF FREE MOVING RATS .............................. 82 5.1 Backgrou nd and S ignificance ................................ ................................ ........................... 82 5.2 Material and methods ................................ ................................ ................................ ....... 84 5.2.1 System Overview ................................ ................................ ................................ ..... 84 5.2.2 Photoacoustic Sensor Design ................................ ................................ ................... 85 5.2.3 Surgery ................................ ................................ ................................ ..................... 85 5.2.4 Data Collection and Processing ................................ ................................ ............... 86 5.2.5 Rat Experiments ................................ ................................ ................................ ....... 87 5.3 Results ................................ ................................ ................................ ............................... 87 5.3.1 Photoacoustic Signal V erification ................................ ................................ ............ 87 5.3.2 Dose and Behavior ................................ ................................ ................................ ... 88 5.3.3 Simultaneous EEG and Hemodynamic Response Study ................................ ......... 88 5.4 Discussion ................................ ................................ ................................ ......................... 90 5.4.1 Photoacoustic Sensor Design ................................ ................................ ................... 90 5.4.2 Dual Modality Study with Video Monitoring ................................ .......................... 90 5.5 Conclusion ................................ ................................ ................................ ........................ 91 6 CONCLUSION AND FUTURE WORK ................................ ................................ ................... 97 6.1 Conclusion and Significance ................................ ................................ ............................ 97 6.2 Limitations and Future Works ................................ ................................ .......................... 99 LIST OF REFERENCES ................................ ................................ ................................ ............. 102 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 114
8 LIST OF TABLES Table page 5 1 The rat e xperiment design. ................................ ................................ ................................ . 96
9 LIST OF FIGURES Figure page 2 1 Block diagram of the multi channel based real time 2D PAT system .............................. 31 2 2 Phot ograph of the real time 2D PAT system ................................ ................................ ..... 31 2 3 Diagram of the multi channel data acquisition system ................................ ...................... 32 2 4 Signal flow chart of th e multi channel data acquisition syste m ................................ ........ 33 2 5 Front control panel of the Labview data acquisition program. ................................ .......... 34 2 6 The calibrati on of the 2D transducer array ................................ ................................ ........ 34 2 7 Reconstructed phantom images before and a fter the calibration ................................ ....... 35 2 8 Photoacoustic experi me nts for the system resolution ................................ ........................ 36 2 9 Selective photoacoustic images of the ink flow in a thin tube. ................................ .......... 36 2 10 Cross sectional PAT images of simulated bloo d vasculature at various depths . ............... 37 2 11 In vivo experiments for comparison of transducer array and single transducer based system of different transducer frequency ................................ ................................ ........... 37 3 1 Comparison of spherical scan and cylin drical scan with a point target ............................. 49 3 2 Influence of the number of detectors to the rec onstructed photoacoustic images. ............ 50 3 3 The spherical transducer arra y ................................ ................................ ........................... 50 3 4 Reconstructed result of the point source for calib ration ................................ .................... 51 3 5 Reconstructed static 3D phantom results ................................ ................................ ........... 51 3 6 Reconstructed results of ink flowing through a tube with an inner d iameter of 0.3mm .... 52 3 7 Real time imaging of epileptic foci in rat brain ................................ ................................ . 52 3 8 Block diagram of CUDA. ................................ ................................ ................................ .. 53 3 9 Data flow chart in CUDA. ................................ ................................ ................................ . 54 3 10 Reconstructed images of the three hair phantom by GPU A) and CPU B) respectively. ................................ ................................ ................................ ....................... 54 4 1 Experimental setup of the combined system ................................ ................................ ..... 72
10 4 2 Electrophysiology of PTZ induced seizure ................................ ................................ ........ 73 4 3 Representative time frames of 2D PAT images of the rat brain, compared with an anatomical image ................................ ................................ ................................ ............... 74 4 4 The reconstruction of 3D PAT images ................................ ................................ .............. 74 4 5 The HbO 2 changes of the SSS, hippocampus and cortex ................................ .................. 75 4 6 The Hb R changes of the SSS, hippocampus and cortex ................................ .................... 76 4 7 The average HbO 2 and Hb R changes of the SSS, hippocampus and cortex across all the rats ................................ ................................ ................................ ................................ 77 4 8 Correlation study of the HbO 2 signals ................................ ................................ ............... 78 4 9 Negative correlation map of HbO 2 in different rats ................................ ........................... 79 4 10 Granger C ausality analysis of the HbO 2 and Hb R signals between different regions of the brain and the ref erence signals ................................ ................................ ................ 80 4 11 Granger C ausality graphs of HbO 2 A) and Hb R B) between different regions of the brain ................................ ................................ ................................ ................................ ... 81 4 12 White noise assessment of the reconstructed images ................................ ........................ 81 5 1 Schematic of experimental system ................................ ................................ ..................... 92 5 2 The design of the photoacoustic sensor ................................ ................................ ............. 92 5 3 Photograph of a rat skull, with the positions of optical cannula and EEG electrodes as indicated by arrows. ................................ ................................ ................................ ........... 93 5 4 EEG data de noising in frequency domain ................................ ................................ ........ 93 5 5 Phantom experiment for validating photoacoustic signal of SSS ................................ ...... 94 5 6 Awake and freely mov ing rat in acrylic cage for experiments. ................................ ......... 94 5 7 Distribution of Racine score (in %) in freely moving rat seizure experiments for each PTZ dose. ................................ ................................ ................................ ........................... 94 5 8 EEG (blue) and HbO2 (red) signals for rat experiments of different models .................... 95
11 LIST OF ABBREVIATIONS 2 D Two dimensional 3D Three dimensional 4D Four dimensional A/D Analog to digital AIC Akaike Inf ormation Criterion BMI Bicuculline methiodide BOLD Blood oxygen level dependent CBF Cerebral blood flow CBV Cerebral blood volume CUDA Compute Unified Device Architecture DOT Diffuse optical tomography EEG Electroencephalography fMRI Functional mag netic resonance imaging FWHM Full width half maximum GABA Gamma aminobutyric acid GPU Graphic Processing Unit HbO 2 Oxy hemoglobin HbR Deoxy hemoglobin HWHM Half width of half maximum MAP Maximum amplitude projection MEMS Microelectromechanical syst ems MRI Magnetic resonance imaging NIR Near infrared NIRS Near infrared spectroscopy
12 OCT Optical coherence tomography ORIS Optical recording of intrinsic signals PACT Photoacoustic computed tomography PAM Photoacoustic microscopy PAT Photoacoustic tomography PCB Printed circuit board PET Positron emission tomography PTZ Pentylenetetrazol ROI Regions of interest SD Standard deviation SNR Signal to noise ratio SPECT Single photon emission computed tomography SS Straight sinus SSS Superior sag ittal sinus TPM Two photon microscopy TS Transverse sinuses
13 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PHOTOAC OUSTIC TECHNOLOGIES AND ITS APPLICATION IN THE STUDY OF EPILEPTIC S EIZURES IN RATS By Bo Wang August 2014 Chair: Huabei Jiang Major: Biomedical Engineering Epilepsy is a most common serious brain disease, of which, however, the understanding is limited . In epileptic seizures, intensive hemodynamic changes are induced as a result of the synchronized electrical discharges from groups of neurons in the brain, and there is a growing trend to detect these hemodynamic changes for epilepsy diagnosis. Compared to other existing hemodynamic detection methods, photoacoustic s features the vascular imaging and hemodyamic signal tracking in small animal brain s , thus is a promising tool for epilepsy research on small animals. We here present a real time three dimensio nal ( 3D ) photoacoustic system for the study of epilepsy in small animals. The system is based on a spherical array containing 192 transducers with a 5MHz central frequency , offering an isotropic resolution about 0.2mm . With the 64 channel data acquisition system and the 10Hz laser, i t can record a complete set of 3D data in 0.3s . Phantom experiments and in vivo experiments were conducted to demonstrate the high imaging quality and real time imaging ability of the system. N eurovascular coupling in generaliz ed seizures was studied using the proposed system with simultaneous electroencephalograph y ( EEG ) . Two groups of rats were imaged with two different wavelengths for the signals of oxy hemoglobin (HbO 2 ) and deoxy hemoglobin (Hb R )
14 respectively. Hemodynamic ch anges in different regions of the brain are extracted and analyzed with correlation and Granger C ausality methods . Study shows that the change of Hb R is less significant than that of the HbO 2 during the seizure , and hyperoxygenation were concentrated aroun d the hippocampus following the seizure onset. It is also noticed that t he hemodynamic connections between different brain regions were found to be closer as the seizure evolved. To study the n euro hemodynamic activities in seizure in freely moving rat s , we built a photoacoustic sensor with EEG electrodes that can be attached on the rat head . Experiments showed that both the neural and vascular responses to seizure in freely moving rats have characteristics, which are observed to be different and more dive rse from that of anesthetized rats, and this calls for more detailed study in future. This technology also promises for other hemodynamic related research study in freely moving small animals.
15 CHAPTER 1 BACKGROUND AND SIGNIFICANCE Epilep sy is the most co mmon brain disease that affecting 2 % of the world population 1 2 . In epileptic seizures, the synchronic electrical charge s of neurons result in prominent hemodynamic c hange in the blood vessels 3 5 , which can be detected by photoacoustic techniques 6 7 . As a hy brid technology that based on the use of laser generated ultrasound , photoacoustic s combines the merits of the high optical contrast from optical imaging as well as the high penetration depth and spatial resolution from the ultrasound in a single modality , featuring comprehensive label free detection of the blood vessels deep in the tissue, and has been investigated in epilepsy study 8 10 . The goal of this thesis resear ch is to further explore the potential of photoacoustic techn iques into the diagnosis of epilepsy, which is embodied in the work in chapter 2 to 6. This chapter firstly gives a n introduction to the photoacoustic techn iques (which emphasis on the applicatio n of photoacoustics in vascular detection in small animal brain) and epilepsy , followed by the motivation and significance, and then the context and o utline of this dissertation . 1.1 P hotoacoustic s In photoacoustic effect , a pulsed laser is employed for t he illumination of an object. When the pulsed energy is absorbed and converted to heat, wide band pulsed ultrasound is generated due to transient thermoelastic expansion which is inherent to the absorption distribution in the object 6 7 . Photoacoustic effect was firstly reported by Alexander Graham Bell in 1880 . Only until the mid 1990s, however, was photoacoustics developed as an imaging technology named as photoacous tic tomography (PAT) , and began to be investigated for biomedical imaging . T he first truly compelling in vivo PAT image s were obtained in 2003 11 . From this point onwards, th is
16 field has witnessed major growth in terms of the instrumentation development 12 13 , image reconstruction algorithms 14 , vascular and functional imaging 15 16 , and the in vivo application of this technology in clinical medicine and basic biological research 17 . PAT in biomedicine c ombines the merits of optical and acoustical methods: sensitive optical absorption contrast and low acoustic scattering in soft tissue. Using safe illumination sources, the photoacoustic effect can be applied for functional and molecular imaging in vivo bi ological tissues . In particular, t he optical wavelengths in the visible and near infrared (NIR) part of the spectrum are most commonly used in PAT, which are mainly absorbed by hemoglobin in the blood vessels, the main carrier of oxygen in the tissue 15 , 18 20 . By utilizing the optical absorption of hemoglobin as contrast, PAT features the label free vascular imaging with a h igh signal to noise ratio (SNR). Using multi wavelength method, multiple functional parameters regarding the hemodynamic changes such as oxy saturation, total blood volume can be achieved 21 23 . Besides, it uses less scattered ultrasound as its probing signal, thus overcome the light diffraction limit 6 7 . Thus it o ffers a great pen etration depth extending to several centimeters , and able to depict subsurface information of the tissue more accurately than other optical imaging methods. The spatial resolution of PTA ranges from um to mm varying with the designs of the PAT setup 13 , which is superior to that of the traditional clinical functional imaging modalities such as functional magnetic resonance imaging (fMRI) 24 26 , positron emission tomography (PET) 27 , and single photon emission computed tomography (SPECT) 28 . With all these merits, PAT has show its excellent p erformance in vascular structural imaging and heamodynamic signal changes tracking in small animal brain 29 32 , which is one of the major directions of the PAT .
17 There are different approaches to PAT imaging , including different illumination schemes, detection schemes, and reconstruction techniques. The early PAT system mainly works in two dimensional ( 2D ) , and can be categorized in two modes: photoacoustic microscopy (P AM ) 33 37 and photoacoustic computed tomography (PACT) 38 42 . In PAM, a focuse d transducer or excitation laser is employed to get the photoacoustic signal along the axial of the focal line; then the transducer or laser is mechanically scanned on the measurement surface, and the acquired signals are directly mapped according to the t ime of flight in the signals to produce image s . In PACT signals from multiple directions around the target are collected by unfocused transducers, and a reconstruction algorithm is applied to reconstruct the image 43 45 . While simplified 2D PAT detection geometries and reconstruction algorithms have been used in most of the previous studies, w ith the fast growing of PAT in recent years, high quality real time three dime nsional ( 3D ) PAT is strongly demanded for monitor ing the hemodynamic changes in small animal brains. Similar to 2D, 3D PAT can be achieved with PAM 46 or PACT 9 , or some hybrid configurations of these two modes 47 48 . Although PAM has been employed to provide excellent 3D vascular image of small animal brain s, the imaging speed is hig hly restricted by the repetition rate of the lasers. Some groups have demonstrated real time 3D PAM systems using lasers with a repetition rate up to more than 10 0 kHz , leading to a 3D frame rate of several frames per second 49 50 , but for most of the PAM system s the need of scanning the focused transducer or excitation laser usually takes hours , which makes it impossible for real time imaging. Compared to PAM, PACT can significantly reduce the imaging time by using various transducer arrays. However, most of the transducer arrays forms in a curve or line and can only image 2D planner objects. In this case, mechanical scanning in z direction is still required for 3D imag ing pu rpose , and this leads to a non optimal elevat ional resolution in z
18 direction . R eal time 3D PAT imaging has also been realized with some hybrid setups 47 48 , but no convincing result has been published using th ese kind of scheme on small animal brain vascular imaging yet. Therefore, 2D array such as planar or spherical array which consist of hundreds of transducers along with parallel data acquisition system elimi nate the need of mechanical scanning, thus are most ideal for real time 3D PAT imaging . The only problem of this kind of system is that the complicacy and expensive price of the parallel data acquisition system limits the number of parallel acquisition cha nnels. However, if the number of parallel acquisition channels is close to the number of transducer elements in the array, only several laser shots are needed to produce a complete 3D image. 2D planar array of transducers has recently been employed for 3D PAT imaging, but due to the limited aperture of 2D planar array transducers, features with high aspect ratio or with orientations oblique to the transducer surface suffer from distortion, and the andazimuthal resolution is reduced. Compar atively , a spheric al array can offer more complete angular views of the object, providing both high resolution and accurate feature definition regardless of shape or location of the object. The use of sparse spherical arrays for 3D PAT imaging has been recently reported, bu t due to the insufficient number of transducers used, the image quality w ere not satisfying and they can not be used for small animal brain imaging 51 . There s another trend of PAT in developing fiber based endo scope or hand held reflection mode miniature PAT probes, which have small rotation motor or Microelectromechanical systems ( MEMS ) mirror integrated for light scanning purpose 52 53 . This offers great flexibility in b iomedical application practice, and they have been widely used for intravascular imaging 54 , skin disorder diagnosis 55 56 , and tumor detection 57 61 . However, c urrent PAT systems for small animal brain imaging are relatively complicated , in which the
19 excitation laser is d elivered in free space. Therefore, only anesthetized or immobilized animals can be used for PAT neurovascular imaging. As the research of neurovascular imaging advances towards free moving animals, there s a growing interest of developing fiber based PAT s ystems 62 63 . 1.2 Epilepsy Epilepsy is a common, chronic, neurological disorder characterized by epileptic seizures. T wo percent of people worldwide will be diagnose d with epilepsy at some time in their lives 1 2 . In most cases, seizures are controlled, although not cured, with anticonvulsant medication , but 20 to 30% of the pati ents are refractory to all forms of medical treatment 64 . Seizure types are classified according to whether the source of the seizure is localized (partial or focal onset seizures) or d istributed (generalized se izure s ) 65 66 . Localization related epilepsies arise from an epileptic focus. A partial seizure may spread within the brain, a process known as secondary generalizatio n. Generalized epilepsies, in contrast, arise from many independent foci (multi focal epilepsies) or from epileptic circuits that involve the whole brain. In epilepsies of unknown localization, it remains unclear as to whether they arise from a portion of the brain or from more widespread circuits. For those patients with medically intractable focal epilepsy, the best treatment option is resective brain surgery 67 69 . A lthough several factors can impact the success of epilepsy surgery, the primary reason of failure is the incomplete mapping of the local epilepsy network which results in incomplete resection of epileptogenic foci 70 71 . Much of the attention on epilepsy surgery has been directed at identifying single neuronal populations 72 74 . Thi s approach has, in many cases, led to failed surgical outcomes, because seizures typically involve groups of neurons interacting both locally and across several cortical and subcortical brain regions 75 78 . A better understanding of the regional interactions occurring at the site of seizure onset and spread may
20 provide important insights about the pathophysiology of seizures and aid with accurate brain mapping and resec tion of the epileptic focus. In theory, removing the focus should result in a patient's seizures being cured. However, there is much evidence to suggest that the focus is more of a region of seizure onset with a number of sites that can act independently to initiate seizures 79 . Seizures in animal models and in people often have a multifocal or broadly synchronized onset. The best evidence for multifocality within the seizure onset zone comes from surgical experi ence with intracranial monitoring. The challenge of mapping the epileptic focus stems from the observation that the pathology associated with focal epilepsy is often distributed across a number of brain sites 80 and that current diagnostics methods frequently fall short of identifying such sites. Animal studies indicate that the neurons involved in the epileptic circuitry have enhanced excitability throughout 81 83 . The implication of these observations is that each of the sites could act independently to initiate a seizure or, potentially, to drive another site into a seizure. Thus, in focal epilepsy, one may view a cortical r egion as a broad seizure onset zone, with the potential that multiple foci can act as a seizure focus for any given seizure. Much of our understanding about focal seizure circuitry comes from electrophysiological recording methods 84 87 . Although EEG recording is mapping the epileptic focus, it is often inadequate to define the boundary of the epilepsy circuitry due to spatial sampling limitations and volume conduction. What we truly desire is a brain mapping modality, which could give a high resolution real time dynamics of cortical processing and seizures. It has been well established that n euron al activity and these hemodynamic related changes in the brain are closely coupled in time, spatial, and amplitude , which is termed as
21 n eurovascular coupling 88 89 . In epileptic seizures, the synchronous excessive neuron disch arges cause enormous increase in the metabolic rate of oxygen, which will place supra normal regulatory mechanisms, thus hemodynamic parameters such as blood volume and blood oxygenation can be greatly changed. Over the recent years, there s a growing interest employ ing various neuroimaging modalities for the study of epilepsy, which detect the hemodynamic changes in the brain to indirectly infer the neural activities in epilepsy based on the neurovascular coupling mechanism 90 91 . Among these technologies, traditional clinical functional imaging technologies such as fMRI, SPECT, PET , have been employed in the diagnosis and therapy assessment of epilepsy 92 95 . However, their spatiotemporal resolution is low . More importantly, because fMRI detects blood oxygen level dependent (BOLD) signal, and is only sensitive to dexoy hemoglobin (HbR) 96 ; Similarly, PET and SPECT rely on the injection of radioactive extraneous agents, and are only sensitive to blood flo w 92 . T hey can only detect focused seizures, and the interpreting of the ir acquired results for generalized seizures is complex and critically dependent on a comprehensive understanding of the neurovascular coupl ing. In contrast, recent developed optical imaging methods such as optical recording of intrinsic signals ( O RI S ) 97 98 , near infrared spectroscopy (NIRS) 99 , and optical coherence tomography ( OCT ) 100 have enough high spatial and temporal resolution and are well used in the neurovascular coupling study of epilepsy, helping the patho logical study of this disease . However, due to the strong scattering of light, they can only see superficial changes in the cortex , while epilepsy involves wild spread networks deep in the brain, thus help less for clinical use. Other optical imaging metho ds such as diffuse optical tomography (DOT) can access to greater depth information, but the spatial resolution is poor 101 .
22 PAT combines both high contrast and spectroscopic specificity based on the optical abs orption of both oxy hemoglobin (HbO 2 ) and HbR with high ultrasonic spatial resolution 18 , and has been demonstrated in focused seizure experiments . Relative to other optical imaging modalities, PAT has the advantage of mitigating both scalp and skull light scattering by a factor of about 1000 102 . The end result is that PAT allows for high spatial resolution imaging of brain at a depth considerably beyond the soft depth limit of conventional optical imaging techniq ues such as confocal microscopy 103 104 , two photon microscopy (TPM) 105 , and OCT . The strong preferential optical absorpti on of hemoglobin makes photoacoustic imaging to have a better imaging contrast than ultrasound , which is difficult to visualize the microvessels with pulse echo ultrasound owing to the weak echogenicity 106 . 1.3 Motivation and O bjectives Based on the review of literature in previous sections, PAT is more suitable for the study of epilepsy for its unique compromise between the high spatiotemporal resolution and deep penetration , as well as the high endogenous opt ical contrast from the blood vessels. The first 2D PAT for epilepsy study was demonstrated for the first time by our lab imaging seizure focus on an acute seizure rat model 8 , but the observation of hemodynamic changes during seizure onset was hindered by t he long time scanning of a single transducer. Besides, light scattering and pressure wave propagation in tissue is inherently 3 D, and 2D approximation to a real 3 D problem (2 D reconstruction model with 2 D detecting geo metry surrounding 3 D targets) will ine vitably bring errors, blurring and missing structures in the reconstructed 2D images 107 . In another hand , all the current epilepsy studies are performed on anesthetized animals , in which both the neural and vasc ular response to seizure may have been altered. This research was supported in part by a grant from the US Department of Defense Congressi onally Directed Medical Program , the B.J. and Eve Wilder endowment fund, and the
23 , which ai m s to realize the PAT monitoring of epilepsy in free moving awaken rats for the prediction, diagnosis, and therapy assessment of epilepsy. The entirety of this work describes our further effort to towards this goal, which can be separated into four related projects. The early work sough to realize a multi channel based real time 2D PAT system, which is a promotion of our previous work. This system has a collection of 120 ultrasound transducers forming a 2D circular array, and a 192 channel parallel data ac quisition system insisting of 8 8 channel high speed analog to digital ( A/D ) boards in a industrial computer with up to 80dB on board preamplifiers and 50MHz sampling rate , as well as a home made 192 channel printed circuit board ( PCB ) preamplifier system (26dB gain ) for the amplifying of the signals from the transducers . T his enables a frame rate of 3.33f/s . Chapter 2 outlines the formation, calibration , and characterization of this system through a series of phantom and in vivo experiments. F rom there, a spherical 192 transducer array based real time 3 D PAT system is built, with the pre mentioned 192 channel data acquisition system . In Chapter 3, we tested the performance of the system, and demonstrated the dynamic tracking of a seizure focus in the rat br ain. This is the first real time 3D PAT system for rat brain imaging. Some pre investigation work was also introduced. Due to the dramatically increased computation cost for the image reconstruction in 3D PAT compared to that in 2D, a G raphic P rocessing U n it (GPU) 108 accelerated 3D delay&sum algorithm 109 110 is applied to reduce the computation time cost. Further more, we i ntegrated an EEG system with our real time 3 D PAT system, and generalized seizure experiments were carried out with two groups of rats using different wavelengths. The dynamic coupling during seizure ictal process among HbO 2 , HbR , and the neural activity a re revealed . Regions with hyperfusion were visualized with cross correlation
24 methods, and the directional influences between hemodynamic changes in different regions in rat brain were assessed using Granger C ausality analysis 111 113 . This part is introduced in Chapter 4. The focus of Chapter 5 is aimed towards the applic ation of photoacoustic technologies in epilepsy experiments in free moving awaken rats. We fabricate d a small probe that insists of a photoacoustic sensor and two EEG electrodes , and mounted it on the skull of mature rats to study the neurovascular coupling of epilepsy. We also recorded video for the assessments of the severity of seizure. The difference s in neurovascular coupling between anesthetized and awaken rats were revealed. Chapter 6 gives a summ ary of our current work, and points out the limitations. Our findings suggest that PAT represents a powerful tool for noninvasively mapping seizure onset and p ropagation patterns, and the functional connectivity within epileptic brain networks was revealed using our real time 3D PAT system. However, the performance of our system is still not satisfying due to the limited number of transducers, and small blo od vessels and many important features that can be detected with some other reported systems were missing. Besides, m ore detailed data processing is still demanded. W hile for our free moving rat experiments there s only a single transducer for detecting th e HbO 2 signals in one blood vessel , and no concrete conclusion was derived due to the limited cases studied. Hemodynamic changes at multiple locations and study of more cases should be done in the future work.
25 CHAPTER 2 MULTI CHANNEL BASED REAL TIME PAT S YSTEM Real time PAT enables the noninvasive recording of whole surface and depth hemodynamic changes in small animal brain with high spatial and temporal resolution. In this chapter, we built a novel PAT system that has three orders of magnitude higher tem poral resolution and four fold higher spatial resolution relative to our previous PAT prototype 8 . With 64 channel parallel data acquisition and 3:1 electronic multiplexing, the imaging speed is about 3 .33 f/s . Using a 120 element full ring ultrasound trans ducer array with 5 MHz central frequency, the system resolution can reach 0.16 mm. We verified the imaging stability of the system with p hantom and in vivo experiments, which proved that this system was suit able for real time monitoring of hemdynamics in s mall animal brain and was a powerful tool for e pileptic s eizure focus localization and making diagnosis 114 . 2. 1 S ystem O verview Fig ure 2 1 depicts the block diagram of the real time 2D PAT system , which is compo sed of three key components: a tunable laser, ultrasound transducer array and the acquisition system with integrated contro ller . A tunable Ti:Sapphire laser pumped with a Q switched Nd:YAG laser was used to provide 690 to 950nm laser pulses with a full wi dth half maximum (FWHM) value of 8 25ns pulse width with a repetition rate of 10Hz for generating photoacoustic signals. A 5 MHz full ring transducer array with 120 transducers was used for recording the photoacoustic sign als. 16 home made preamplifier b oards ( each supporting 12 input channels and 4 output channels , and sealed in four separate metal boxes ) amplify the signals from the transducer elements wi th a fixed gain of 26dB , and transmitted the amplified signals to 8 8 channel A/D boards , with 3:1 m ultiplexing. T he multiplexed signal is subsequently amplified with the AD604 amplifier s on
26 the A/D boards with a programmable gain of 0 to 80dB, and then digitized , and then stor ed in the industrial computer for offline processing. Figure 2 2 shows the ph otography of the entire real time 2D PAT system. In theory, the parallel data acquisition system that consists of the 192 /64 channel home made preamplifier and multiplexer module , and the 8 8 channel A/D boards allows an input of 192 signals and an output of 64 signals. However, b ecause only 120 transducers are employed in the full ring configuration, the amplifie d signals with the preamplifier boards were multiplexed into three groups ( 40 signals in a group ) and subsequently transmitted to 5 8 channel A/D boards for further amplifying and digitalization. With a laser repetition rate of 10Hz, three laser pulses are needed for one complete set of 2D PAT data. Detailed information of this parallel data acquisition system will be presented thoroughly in the fo llowing par t . 2. 2 M ulti C hannel D ata A cquisition S ystem As shown in Figure 2 3, each of the 16 preamplifier board s has 12 input channels, 4 output channels, and 2 digital signal control inputs. More specific ally, within each board, 12 dedicated operational amplifier (AD8099) modules individually amplify the input signals with a fixed 26dB gain, and then the amplified signals were multiplexed into the 4 output channel by 4 multiplexer chips (MAX 4051), which were controlled by a USB IO digital module ( USB 10 24 LS, as seen in Figure 2 4) through the 2 digital inputs. The power supply to the preamplifier board is 5V DC. Because the multiplexer is easily damaged if signals are present before the 5V DC power su pply, we also ad de d diodes on the power supply port on the boards for protection, which leaves the low switch resistance and low leakage characteristics of the devices un affect ed . However, the difference between V+ and V should not exceed 17V. The 16 prea mplifier board s were evenly sealed in 4 metal boxes, thus each box includes 4 boards with 4 12 AD8 099 chips and 4 4 MAX4051 chips , and the whole system has 192 AD8099 and 64
27 MAX4051. The mapping of the 1 96 transducers to the 64 output channels on the 16 preamplifier boards ensure s that for each output channel there were 3 transducers, so that three laser pulses were needed for a complete data capture. The 8 PCIAD850 data acquisition boards are plugged in to an industrial computer . Each PCIAD850 has 8 channels with a sample clock up to 50MHz, 10bit A/D converter. In our current settings, the first board se rves as the master board, and the remaining 7 boards are set to slave mode . A ll channels on the slave b oards will start taking data upon receiving a trigger signal from the master board. The jitter between channels is less than 2 nanoseconds. The whole system is synchronize d with the laser Q switch sync signal through t he master board . The industrial comput er sends out the synchronize signal to the USB 1024 LS to control the 16 preamplifier boards. The signal flow chart is shown in Figure 2 4. Data acquisition software based on Labview program sets up the acquisition parameters, reads and parses the acquisit ion memory, and creates acquisition files ( s ee the front control panel and in Figure 2 5 ). The software was written to allow maximum flexibility for vario us configurations of the system . At each trigger event, all the channels will simultaneously convert t he analog signals to digital data with user selected post trigger delay and waveform length. Other programmable parameters include sampling rate, source of trigger , gain, DC off set, low and high pass filters . Data is later processed using the delay&sum alg orithm to produce one image frame . 2. 3 System C alibration and P erformance As shown in Figure 2 6 (A and B), the transducers in the 2D array interface were mounted with a small screw, and the distance from the surface of the transducer to the center of the array may be different for different transducers. This distance difference must be calibrated before image reconstruction.
28 For the calibration of the system, a point source placed in the center of the array was employed . Then the photoacoustic signals fr om the transducers were acquired (Figure 2 6 (C), blue lines), in which we can see that the time delay of the point source (seen as the pulsed signal between 8 to 10 s ) is different for different transducers. W e calibrated the transducers in the radius dir ection by applying different time delay to different transducer signals . T he photoacoustic signals for the point source were calibrated to 8 s (Figure 2 6 (C), red lines) . This is equivalent to generating a virtual transducer array with a radius of 7 6 . 1 mm. Figure 2 7 (A) and (B) show the reconstructed PAT images of the point source before and after the calibration, in which we see that the calibration greatly improved the spatial resolution and signal to noise ratio (SNR) of the reconstructed images. This i s further confirmed by a phantom experiment, in which a phantom contains two target of different seizes (4mm and 2mm respectively) was employed. The optical contrast between the target and the background is 2:1. T he calibrated PA T image (Figure 2 7 (D)) sh ows a good correlation with the target sizes and the shape , whose image quality is significantly than that without calibration (Figure 2 7 (C)) . It demonstrate s that the calibration is validated for the image reconstruction. To evaluate the system resolu tion, a phantom with two 100 inserted around the center was imaged, as shown in its reconstructed photoacoustic image Figure 2 8 (A). Figure 2 8 (B) shows the enlarged photoacoustic image of the two targets, as indicated with the dotted rectangle in (A). Figure 2 8 (C) i s the line profile of the reconstructed image around y=10mm that across the center of the two targets . AB and CD are the half width of half maximum (HWHM) of the two targets respectively, and the spatial resolution of the system is defined as AB+C D 2r, where r is the radius of the targets 115 . The measured spatial resolution of
29 the imaging system is estimated to be 0.16 m m, which is close to the theoretical resolution of 0.15 m m according to the central fre quency of the transducers 116 . With a capture rate of 3 .33 f /s , the real time tomographic imaging ability was demonstrated with phantom experiments . Figure 2 9 depicts some selected frames from a 10 second sequenc e of dynamic ink flow through a 1.6 mm diameter polyethylene tube using a manual syringe push. The images can clearly track the flow through the tub e with high spatial and temporal resolution. The imaging speed is 900 times faster than the previous single transducer based system (270s) . Thus it is suit for real time monitoring of brain hemodynamics in vivo during e pileptic s eizure onset. Figure 2 10 presents the sectional images of simulated blood v essels over a depth range of 4.6cm. ICG solution of 500 M c oncentration was sealed into four short plastic tubes with 0.3mm inner diameter and embedded on the surface of a phantom, on which chicken breast tissue with different thickness were laid on. This illustrate d the capability of the system to provide high pe netration imaging depth with a high resolution and good image contrast up to 3.5cm. The photographs of the phantom and the chicken breast tissue (3.5cm) laid on the phantom are shown in Figure 2 10 (A) and (B) respectively. 2. 4 I n v ivo E xperiments Male Spr ague Dawley rats (Harlan Labs, Indianapolis, IN) weighing 35 45 g on arrival were allowed one week to acclimate to the 12 hour light/dark cycle and given food and water. Animals were anesthetized intraperitoneally with 1g/kg of body weight dose of urethane. Then the rats were have hair on the top head removed, fixed on a home made rat holder, elevated from the bottom of the water tank for imaging. In this experiment, 532nm light from a Q switched Nd:YAG laser was employed, and the incident power was about 12 mJ/cm 2 on the rat head . We also carried out single transducer experiments with a 5MHz transducer (the same transducer in
30 the array) and a 3MHz transducer for comparison. All procedures were approved by the University of Florida Animal Care and Use Committe e and conducted in accordance with the National Institutes of Health Guide for the Care and Use of Experimental Animals. Figure 2 11 (A) presents the reconstruct im age for the rat brain using the transducer array, in which the signals were averaged only on ce , and the c orresponding open skull photograph is shown in Figure 2 11 (B). The resulting images for single 3 MHz and 5 MHz transducers were shown in Figure 2 11 (C) and (D) respectively, which are slightly different from that in (A) probably because they u sed different rats and the position and the orientation of the rat head were different. We can see that the main blood vessels that seen in the open skull image can be clearly reconstructed in the three PAT images. The quality of the image using the 5MHz t ransducer array is similar to that using a single 5MHz transducer, while the spatial resolution of the image using a 3MHz transducer seems a little worse due to a lower central frequency. 2. 5 Conclusion A real time PAT system using a full ring ultrasound array of 120 discrete transducers has been developed. Using 64 channel parallel data acquisition and 3:1 electronic multiplexing, the imaging speed reaches up to 3 .33f/s and is 900 times faster than a single transducer based system. The spatial resolution is about 0.16mm. Phantom and in vivo experiments verified the stability of the imaging system , which proves that it is suitable for the seizure experiments in the next step on rats .
31 Figure 2 1 . Block diagram of the multi ch annel based real time 2D PAT system . Figure 2 2 . Photograph of the real time 2D PAT system .
32 Figure 2 3 . Diagram of the multi channel data acquisition system .
33 Figure 2 4 . Signal flow chart of the multi channel data acquisition system .
34 Figure 2 5 . Front control panel of the Labview data acquisition program. Figure 2 6 . The calibration of the 2D transducer array . A) Photography of the array, B) d iagram of the transducer mounting, C) t he calibration of the array.
35 Figu re 2 7 . Reconstructed phantom images before and after the calibration. A) and C) before the calibration, B) and D) after the calibration.
36 Figure 2 8 . Photoacoustic experiments for the system resolution. A) P hotoacoustic image of the two metal wires, B) enlarged image of the rectangle in A), C) line profile of Y=10mm as indicated in B) with dotted line. Figure 2 9 . Selective photoacoustic images of the ink flow in a thin tube.
37 Figure 2 10 . Cross sectional PAT images of si mulated blood vasculature at various depths . ICG tubes of 500 M were embedded on the surface of a phantom with chicken breast tissue of different thickness laid on . A) P hotograph of the phantom with no chicken breast tissue, B) photograph of the phantom with 4.6cm thick chicken breast tissue. Figure 2 11 . In vivo experiments for comparison of transducer array and single transducer based system of different transducer frequency . (A) Recovered photoacoustic image with the real time 2D PAT system with 5MHz circular array . (B) Photograph of the rat cerebral with skull opened . C) Recovered photoacoustic image with one 3MHz transducer, D) recovered photoacoustic image with one5MHz transducer.
38 CHAPTER 3 FOUR DIMENSIONAL PAT SYSTEM A method to produce real time 3D photoacoustic images could drastically influence the landsc ape of photoacoustic imaging applications. The entirety of this chapter describes our approach to design and test a four dimensional ( 4 D ) PA T system (3D PAT system with real time imaging ability) . The system is based on a 2D spherical array of 19 2 transduc ers , and coupled with the 192/64 channel parallel data acquisition system we introduced in Chapter 2 . It can record a complete set of 3D data at a frame rate of 3.3 3 f/s, and the spatial resolution is about 0.2mm. We demonstrate this system using static/dy namic phantom and in vivo animal experiments. S ome pre investigation work is also presented . To reduce the significantly increased reconstruction time in 3D PAT, we also developed a 3D delay&sum algorithm that is accelerated with GPU parallelization techni ques. 3.1 Pre I nvestigation W orks Pre investigation works were carried out for the design of the 2D spherical array, which includes phantom experiments for 3D scanning scheme investigation, and numerical simulations to assess the influence of the number o f transducers of the array . For the 3D scanning scheme investigation, a point source (a small piece of graphite particle, less than 0.1mm, embedded in a phantom) was imaged with two different 3D scanning schemes: spherical scan and cylinder scan. The form er scanning scheme offers more complete views from different directions in 3D, while the latter one is easy to implement . The scanning scheme s of the spherical scan and cylindrical scan w ere shown in Figure 3 1 (A) and (B) respectively . For the spherical scan, we scanned 50 steps in the x z plane, covering an angle of 240 degree. In the x y plane, 18 steps were scanned, covering an angle of 96 degree.
39 For the cylindrical scan, we scanned 71 steps in the x y plane, and scanned 11 steps in the z direction, w ith an elevation step of 3mm. Figure 3 1( C ) shows the reconstructed results with the spherical scan, compared with the cylindrical scan in ( D ). It s evidently that the resolution in x y plane is almost the same for the two scanning schemes, while the reso lution in z direction for the spherical scan is much better than the cylindrical scan, which is in accordance with many published theories and reported experimental results. Furthermore, numerical simulations were carried out to assess the reconstruction results with different number of detectors in the spherical scan . As shown in Figure 3 2, the target is a cube of 5mm in the center of the image domain, which is set to be a 10mm cube. The resulted photoacoustic signals were generated with the Green functi on: (3 1) where in which is the distance from the target to the detector, is the del ta function, and is the velocity of ultrasound in the media (here we set =1.5mm/ s ). The reconstruction results were acquired with the delay&sum algorithm in 3D, and the voxel value below zero is set to be zero. In this simulation, three different detector distributions (with 163, 298, 622 detectors respectively) were employed. We ca n see that the target can all be clearly reconstructed under the three detector patterns, and the quality of the reconstructed image is better with more detectors . We thus determine that a detector number of higher than 163 is enough for high
40 quality image s in spherical scan. In the next step, we proceed to build a spherical array based real time 3D PAT system with 192 5MHz transducers and the 192/64 channel data acquisition system that employed in the circular array in Chapter 2. 3.2 4D PAT S ystem B ased on a S parse S pherical T ransducer A rray I n this system, a Ti:sapphire laser optically pumped with a Q switched Nd:YAG laser was employed to sen d 8 12ns pulses at 10 Hz with a wavelength tunable from 690 to 1015nm to the spherical array . The beam was delivered with an optical fiber bundle through an opening on the top of the spherical array and produced an approximately uniform illumination in a 2cm diameter area onto the sample. Figure 3 3 ( B ) depicts the design and detector distributions of the spherical tran sducer array , and the photograph of it is shown in (A) . T he array consisted of 192 transducers placed along a custom fabricated white ABS spherical interface containing 610 through holes with counter bores . More holes were drilled so that the selection of the transducer positions on the ball can be flexible. The interface has an outer diameter of 160mm and an inner diameter of 140mm, and the diameter of the holes in the ball is 5.7mm, which fitted well with the transducer (5.5mm outer diameter). Each transd ucer (Custom designed from Blatek, Inc.) has a central frequency of 5 MHz with a reception bandwidth of greater than 80%. The active area of the transducer is 3mm in diameter and the angular acceptance is about 15 degree. The transducers were glued onto th e interface with epoxy which can be removed to allow the position change of the transducers. As previously introduced, t here were 16 preamplifier boards separately sealed in 4 metal boxes, and each board had 12 input channels, 4 output channels, and 2 dig ital signal control inputs. Within each board, 12 dedicated operational amplifier modules (AD8099) individually amplify the input signals with a fixed gain of 26dB, and then the amplified signals were multiplexed into the 4 output channels by 4 multiplexer chips (MAX 4051), which were
41 controlled by a USB IO digital module (USB 1024LS, Measurement Computing) through the 2 digital inputs. The system was featured in the selection of transducer positions on the spherical interface. The total 610 holes formed 1 1 evenly spaced layers on the ball, and the 196 transducer positions were indicated with three different colors in Figure 3 3 (B) . Three different colors were used because of the 3 to 1 multiplexing, which grouped the 196 transducers into 3 groups. As seen in the image, the transducers were only distributed on the top 7 layers. The 64 holes in the 7th layer just in the middle of the array were all filled with transducers, and formed a single group. Another group of 64 transducers were distributed on the 5th and 6th layer, and the last group occupied the top 4 layers. In this way, just one laser pulse is needed for the recording of the 64 transducers in each group, so that this system can also be used as a real time 2D system operating at 10f/s with our Ti: S apphire laser using the 64 transducers in the 7th layer. In in interface through a pipe fixed on the tank bottom, whose top was about 15mm beneath the interface center, and a tr ansparent plastic warp was used to cover the pipe top to keep off water and keep the rat brain acoustically coupled to the transducer array. For phantom experiments, a homemade silicone holder is used to hold the phantoms. 3. 3 S ystem P erformance Three pha ntoms were used: one containing a point source for system calibration, two phantoms containing hairs with different spatial distribution and orientation for static imaging, one phantom with ink flowing through a thin tube for real time 3D imaging. The poi nt object used for calibration was a small spherical graphite particle (0.1mm in diameter) located at the center of the spherical array and ensured an isotropic acoustic emission profile for all directions. We calibrated the system by measur ing and compens at ing the delay of
42 time for all the 192 channels, and reconstructed this point source after the calibration. Two phantoms containing hairs were used to demonstrate the high imaging quality of our system. The first phantom contained five hairs placed in a h orizontal plane, and the second phantom contained three tilted hairs with different orientations. Finally, we imaged a tube filled with flowing ink to prove the real time imaging ability of our system. The tube had a 0.3mm inner diameter and it was horizon tally placed in the phantom. No averaging of signals was performed for the phantom experiments except for the point object experiment where 10 times averaging was applied. 3 . 3 .1 System C haracterization We calibrated the system by recording the emission pr ofiles of a small point source of all the 192 channels, and then measured and compensated the delay of each channel. After that, we reconstructed the point source. Figures 3 4 ( A ) and ( C ) present the reconstructed x y and z x cross section images of the p oint object located at the array center. The quality of these images is determined by both the distribution and the characteristics of the transducers. The profiles of the two reconstructed images were also extracted in x and z directions, as shown in Fig u re 3 4 ( B ) and ( D ) , respectively. The full width at half maximum (FWHM) of the profiles was measured to be 0.19mm (x direction) for Fig ure 3 4 ( B ) , and 0.27mm (z direction) for Fig ure 3 4 ( D ) , compared to the theoretical value of 0.16mm for the 5MHz centra l frequency transducer with an estimated cut off frequency of 7MHz. It is noted that the profile in Fig ure 3 4 ( D ) is noisier than that in Fig ure 3 4 ( B ) . This along with a larger FWHM from Fig ure 3 4 ( D ) was due to the asymmetric distribution of the trans ducers. For targets located away from the center of the array, the radial resolution will stay nearly the same as that for a centrally located target, while the lateral resolution will be linearly reduced with increased distance away from the array center. In our
43 system, the lateral resolution will be reduced by 0.1mm when the target is located 5mm off the array center. 3 . 3 . 2 Static P hantom E xperiments Fig ure 3 5 ( A ) is the photograph of the first phantom containing five hairs placed in a plane and Fig ure 3 5 ( B ) and ( C ) shows the reconstructed 3D images from two different views. The reconstruct ion domain is 14 10 5mm with a 0.1mm voxel size. The reconstructed shape of the hairs matched the photograph very well. The results for the three tilted hairs in the second phantom are shown in Fig ure 3 5 ( D ) ( F ). The reconstruc tion domain is 10 10 10mm with a 0.1mm voxel size. The spatial distribution and tails for all the three hairs were clearly re vealed. The two phantom results indicate that our system is capable of three dimensionally imaging small objects of different spatial distribution and orientation in high quality. 3.3 .3 Dynamic 3D P hantom R esults The rec onstructed 3D images of ink flowing through a thin tube are shown in Fig ure 3 6 . The image domain is 15 5 10mm with a 0.1mm voxel size. Fig ure 3 6 ( A ) is the photograph of the phantom containing a tube fi lled with ink, and Fig ure 3 6 ( B ) ( J ) are the reconstructed 3D images at different time points. The time interval between two consecutive images was 0.3s. These 3D images clearly tracked the flow through the tube over the course of 2.4 s with high spatial a nd temporal resolution, and the flowing speed of the ink was measured to be 6 0.9mm/s from the reconstructed results with a time interval of 0.3s. 3. 4 Real T ime imaging of epileptic foci In Epilep tic seizures, the intensive hemody namic changes provide high endogenous contrast for PAT imaging due to the strong absorption of blood at visible and NIR wavelengths. Although 2D PAT has been demonstrated in a focused seizure model in rats, but the observation
44 of hemodynamic change was hin dered by the long time scanning of a single transducer. Here we test our real time 3D system using the same model to reveal the hemodynamic changes, and show s the 3D vascular structures in the rat brain. Two small Sprague Dawley rats (~40g) were imaged wi th intact skull and skin but hairs on the head were removed. The rats were anaesthetized and mounted on the homemade plastic methiodide (BMI) into the neocortex of one rat, while saline solution was injected into the brain of another rat as control. In each experiment, the rat was elevated to the transducer array center and kept alive under the water tank through the whole experiment. The incident energy of the 730nm li ght was maintained at 8mJ/cm 2 below the safety standard. Seizure process was recorded for 50 minutes, and the measurement from the control rat was recorded for 3s. All animal procedures were performed in accordance with the approved University of Florida I ACUC protocols. Figure 3 7 ( B ) presents the 3D images for the rat with BMI injection during the seizure onset at 6 time points, compared with that for the control rat in Figure 3 7 ( D ) . The corresponding photographs of the two rats with scalp removed right after the experiments are shown in Figure 3 7 ( A ) and ( C ) , respectively. The reconstructed domain for the images shown in both Figure 3 7 ( B ) and ( D ) is 20 20 4.5mm, with a 0.1mm vo xel size. T hree large blood vessels on the rat brain are clearly revealed for both cases, as indicated by the white arrows, and for the rat with BMI injection a seizure focus can be clearly seen right at the BMI injection position (the white circle in Figure 3 7 ( B ) ), where str ong absorption is observed during seizure onset. The rapid changes of the absorption both in the main blood vessel and in the seizure focus were observed from Figure 3 7 ( B ) , while no such changes were noted for the control rat ( Figure
45 3 7 ( D ) ). The seizur e focus had a diameter of ~3mm, which is consistent with the results published before 8 . This experiment demonstrates that our system can be used to investigate the hemodynamic changes in small animal brain both spatially and temporally during seizure onset , although the complex microvasculature cannot be resolved due to the limited number of transducers and the simple reconstruction method used here. The microvasculature can be revealed if sophisticated reconstruction methods such as the finite element base d algorithms coupled with the total variation minimization scheme are used 117 119 . 3.5 GPU A ccelerated 3D D elay&sum A lgorithm An effective reconstruction algorithm is a critical component of PAT. While different reconstruction algorithms have been proposed for more rigid and accurate reconstruction of absorbed energy distribution, delay&sum algorithm remains the most common method for its simplicity , easy adaptation an d effectiveness . Delay&sum is beam forming algorithm that is commonly used in radar signal processing . This is based on the back projection method according to the flight time of the received photoacoustic signals. The image expression in the case of near field for photoacoustic imaging can be stated as ( 3 2 ) where is the image output at a particular focus point f , is the time signal from the th receiver, is the delay applied to this signal, and is an amplitude weighting factor, which is used to enhance the beam shape, to reduce sidelobe effects, or to minimize the noise level. The summed signal is typically normalized to make the output independent of the
46 actual set of transducers. This delay and sum algorithm has been tested and evaluated using considerable phantom and in vivo experiments 9 , 15 . The computation cost of delay&sum algorithm is in proportional to , where , , are the reconstructed voxel numbers in the x, y, z directions respectively , and are the number of detectors. In 2D PAT, is set to be 1. However, the computation cost can be increased by hundred to thousan d times in 3D PAT. GPU is a specialized electronic circuit designed to help the processing of image s in a computer . Because of the ir highly parallel data processing structure , modern GPUs begin to be commonly used for processing of large blo cks of data in parallel , in which their efficiency is much higher than the general purpose CPUs 120 122 . Compute Unified Device Architecture ( CUDA ) is the programming platform intro duced for parallel computing with GPUs, and CUDA along with GPU provides a low cost, powerful tool for parallel calculation with a personal computer. In CUDA, the functions parallel executed on the GPU in an application are called kernels. Each kernel is executed at a time by a collection of independent threads , with implicit barrier synchronization between kernels . All threads that execute the same kernel function are called a grid, which is divided into many thread blocks. T he blocks are totally independ ent with each other, while the threads in a block can synchronize and cooperate through the S hared M emory. Each block in a grid and each thread in a block have their unique ID to compute memory address and make control decisions. The dimension of a grid an d block are defined with two dim3 vectors : grid(num_block_x, num_block_y, num_block_z) and threads (num_thread_x, num_ thread _y, num_ thread _z) respectively.
47 In the aspect of hardware , GPUs have a number of multiprocessors, each of which executes in para llel with the others. Each multiprocessor has 8 or 16 stream processors (which can also be referred as cores) , depending on the card model . Each core can execute a sequential thread, and all cores in the same multiprocessor execute the same instruction at the same time . Threads in a same block in CUDA are executed in the same multiprocessor, and the threads in the same block are actually executed in groups of 32 threads, called a warp. If each multiprocessor has 8 cores, then a warp can be finished in four clock cycles. The S hared M emory shared among threads in the same block is a small data cache attached to each multipr ocessor . It is a low latency, high bandwidth, software manage able memory which runs essentially at register speeds. However, large data are stored in the main memory of the GPU card with high latency , called Device M emory (seen in Figure 3 8) . In delay&sum algorithm, the calculation of each voxel value is highly independent, thus can be easily parallelized with GPU. The flow chart of the par allelization is shown in Figure 3 9 . Because almost all the calculation s are done in the GPU, t he raw PAT data along with the coordinates of the transducers, the calibration data, and other data necessary for the calculation i s firstly transferred from the Host M emory (the main memory of the computer) to the Device Memory . Then a memory is allocated and initialized in the Device Memory for saving the calculation results. The gird and block sizes in our algorithm are set to be grid( , ,1) and threads( , 1,1) respectively. After that, the CPU assign s the calculation of a voxel value to a thread according to the block ID and thread ID. Each thread sequentially calculates the delay of time of different transducers, and does the summing of PAT signals according to Equation 3 2 . All the memory access including reading and saving on the GPU card are in the Device Memory ,
48 and the Shared Memory is not used. Finally, t he calculation result of a 3D frame is copied to the host memory for later processing and display. To verify the performance of the proposed parallelization algorithm using GPU , the reconstruction results of the three hair phantom experiment (seen in 3.3.2) with the GPU based me thod and that with the CPU based method are compared, as seen in Figure 3 10 (A) and (B). The reconstruction s were carried out on a desk computer with 2 AMD Athlon 64 CPU, 4 GB RAM and NVIDIATM GTX 650 graphics card of an 1 GB on board memory. The serial c omputation on CPU and parallel computation on GPU are both programmed by C++ programming with the same data structure and calculating process. The reconstruction results for the two images are found to be almost identical in further inspection, which prove s that there is little influence of the parallel calculation to the reconstruction accuracy. However, it s noted that the calculation is greatly accelerated by the parallel computation with GPU. While it takes about 130s for the calculating of a complete s et of 3D set images (as in Figure 3 10 ( B )) using the CPU based method, it is only about 1.1s with the proposed method . 3. 6 Conclusion We have presented a 4D PAT system for real time 3D imaging of small animal brain. The system is based on a spherical arra y containing 192 discrete transducers. With the 64 channel data acquisition system coupled with 3:1 multiplexing, it can achieve a frame rate of 3.3 3 f/s and a spatial resolution of 0.2mm. The 3D imaging performance of the system was demonstrated by both st atic and dynamic phantom experiments. We have also tested our system using an acute epilepsy rat model and obtained 3D images showing the hemodynamic changes during seizure onset. In addition, a GPU base method is proposed and implemented for the paralleli zation of 3D delay&sum algorithm, which ha s greatly accelerated the 3D PAT reconstruction.
49 Figure 3 1 . Comparison of spherical scan and cylindrical scan with a point target . A) and B) are the scanning schemes of the spherical scan and cylindrical scan respectively. C) and D) shows reconstructed images of the point target in x y plane and x z plane.
50 Figure 3 2 . I nfluence of the number of detectors to the reconstructed photoacoustic image s . Figure 3 3 . The spherical transducer array . A) T he design a nd detector distributions of the spherical transducer array B) Photograph of the spherical transducer array.
51 Figure 3 4 . Reconstructed result of the point source for calibration. A ) and B ), the x y and x z cross section images through t he center of the point source. C ) and D ), profile extra cted in x and z direction from A ) and B ) respectively. Units are in mm . Figure 3 5 . Reconstruc ted static 3D phantom results. A ), P hotog raph of five hairs in a plane. B ) C ), 3D reconstructed results of the five hairs in two different views. D ), photography of the phantom containing three tiled hairs; E ) F ), corresponding 3D reconstruction results for D ) .
52 Figure 3 6 . Reconstructed results of ink flowing through a tube with an inner diameter of 0.3mm. A ), photograph of the phantom containing the tube. B ) J ), reconstructed 3D images at different times. The time interval is 0.3s. Figure 3 7 . Real time imaging of epileptic foci in rat brain. A ) and C ): photograph of the rat with BMI injection and the control rat after scalp removed. B ) and D ): reconstructed 3D images for A ) and C) at 6 different time points respectively. The reconstructed three main blood vessels are indicated by the white arrows, and the seizure focus is indicated in A ) and B ). The time interval betwee n two successive images is 0.3s. Scale bar represents 10mm.
53 Figure 3 8 . Block diagram of CUDA .
54 Figure 3 9 . Data flow chart in CUDA . Figure 3 10 . Reconstructed images of the three hair phantom by GPU A) and CPU B) respectively .
55 CHAPTER 4 ICTAL NEU ROVASCULAR COUPLING STUDY BY 4D PAT SYSTE M C urrent neuroimaging techniques such as fMRI and ORIS for epileptic study are l imitated in spatiotemporal resolution and imaging depth respectively 90 91 . Here we for the first time present the detailed study of ge neralized ictal neurovascular coupling in rats with 4 D PAT and simultaneous EEG recording , which is a natural extension of the work in Chapter 3 . Two groups of rats (n=16) were intraperitoneally injected with pentylenetetrazol (PTZ) 123 , and investigated under two different wavelengths for HbO 2 and Hb R respectively. Hemodynamic changes in different regions of the brain are extracted and analyzed. Study shows that the overall change of Hb R is less significant than that of the HbO 2 during the seizure, and it also lagged the prominent increase of HbO 2 in hippocampus. Since fMRI is only sensitive to Hb R , this may indicates that PAT is probably more suitable than fMRI for epileptic study. Correlation study shows that hy perfusion and hyperoxygenation were more concentrated in the hippocampus and its surroundings than other brain regions following the seizure onset. The hemodynamic connections between different brain regions calculated with granger causality method were fo und to be closer as the seizure evolved. Discussions are presented in aspects of the significance of the findings, data processing method, other factors in neurovascular coupling, and the limitations in our work. 4.1 Background As mentioned in Chapter 1, PAT inherits the high optical contrast from optical imaging, and breaks through the diffraction limits of the light by the use of the ultrasound as the detection method , and has proved to be a promising imaging modality in biomedical applications. With di fferent configurations , it bridge s the resolution and penetration gaps between clinical modalities at the organ level and optical microscopy technologies at the cellular level. It has been
56 widely used in vivo to measure epidermal melanin 124 , detect breast tumor 125 and osteoarthritis 21 , 107 , estimate oxygenation levels 21 , and track molecular probes 126 127 . In particular, as a major direction for PAT application , the small rat brain imaging with PAT c an greatly facilitate the in vivo research of epilepsy, helping the understanding of neuronal activity and energetic s d uring epileptic seizures, making it a potential substitution of current neuroimaging modalities in epilepsy study. The requirement of rea l time 3D imaging deep into the rat brain impose significant challenges in terms of experimental paradigm and instrument design for the study of epilepsy using PAT . Most PAT systems require mechanical scanning using step motor for 3D imaging purpose, leadi ng to a low frame rate. With the application of high repetition laser, multi channel DAQ system, and various transducer arrays, real time 3D PAT of different configurations begin to emerge in recent years. However, most of these systems can not be used for brain vascular imaging, and detailed research using PAT in the study of epilepsy is still rare. Although our group has reported 2D PAT with concurrent EEG for epilepsy study 10 , but their conclusions were in con flict to existing knowledge maybe due to the wrong interpretation of the data. Recently we have reported a 4D PAT system that is based on a sparse spherical 192 transducer array and a 64 channel DAQ system operating at a frame rate of 3.3 3 f/s for epileptic focus imaging in rat brain, but without concurrent EEG recording. In this work, we incorporated EEG into that system, and further explored its potential in epilepsy study through the neurovascular coupling investigation of PTZ induced seizure in rats with dual wavelength. Neurovascular coupling study can help to reveal the mechanical of epileptic seizure, and is the base for possible seizure prediction and localization 128 129 . By comparing the hemodynamic signals from different structures in the rat brain with high temporal resolution, we are trying to
57 address many important disputed issues in neurovascular coupling of epilepsy. Studies include the timecourse of hemod yamic changes in different brain regions during seizure, the correlation study of PAT signals to find important regions that are more sensitive to epileptic seizure and possible seizure localization, and the neural network analysis using granger causality method. This study can be of great importance to the current clinical practice in epilepsy diagnosis with fMRI and other neuroimaging modalities, but the ultimate goal is to use PAT as a better alternative to these technologies. 4.2 Experimental S etup and M ethods The PAT system was previously reported in other publication, and the main feature of this system is the sparse spherical array which is consisted of 192 discrete transducers, as shown in Fig ure 4 1. Rats were elevated to the center of the spherica l interface through a chamber fixed at the tank bottom, whose top was about 15 mm beneath the interface center, and a transparent plastic wrap was used to cover the chamber top. The spherical transducer array was placed in the water bath, acoustically coup led to the rat head with ultrasound gel. Laser beam was delivered through a concave lens resulting a homogenously illumination on the rat head. In this study, two wavelengths of 1064nm and 710nm were used. The 1064nm light was from a Q switched Nd:YAG lase r, and the 710nm light was from a tunable Ti:sapphire laser. The light intensity on the rat head was about 50mJ/cm 2 for 1064nm, and 4mJ/cm 2 for 710nm, all below the ANSI exposure limits (ANSI Z136). Both the two lasers operate at 10Hz, so that one complete set of 3D PAT data can be recorded in 0.3s. Two syringe needles were inserted about 2mm into the front cortex of rat brain served as EEG electrodes. One electrode was chosen as reference and ground, while the other one was used as signal channel. The EEG signal was amplified (RA16PA, TDtucker Davis Tech.) and recorded (RZ5 Bioamp Processor, TDtucker Davis Tech.) with a sampling rate about 50kHz.
58 Both the two electrodes were firmly glued on the rat head and the EEG cable was tightly tied on the animal hold er to avoid noise induced by any possible unnecessary movement of the operators. The PAT system gave synchronize signal to the EEG system for synchronization. Two groups of rat experiments (n=16 rats, about 35~45g, with 8 rats for 1064nm and the other 8 ra ts for 710nm) were carried out. The rats were firstly anaesthetized with about 0.25ml Urethane (0.25g/mL) and the hair on the top head was shaven. A syringe containing 0.04g/ml PTZ with a scalp needle was inserted in the rat abdomen and fixed with glue. Th is was to easy the intraperitoneal injection later in the experiment, which has already been established 123 . The rat was immobilized on a custom made plastic holder with ear bar using glue tapes. Two EEG electrodes were inserted, and then the rat was eleva ted into the center of the spherical transducer array. The seizure process was recorded for 25min, and PTZ was administrated with a dose of 0.5mL at 5min. The injection of the drug lasted about 30s. All rats kept alive through the whole data recording proc ess. All the animal procedures performed were in accordance with the approved University of Florida IACUC protocols. With a frame rate of 3.3 3 f/s, there were 5000 complete sets of PAT data for the 25min experimental time. Before doing 3D reconstruction and processing, 2D time series images were reconstructed with the signals from the 64 transducers in the lowest layer of the array, which were indicated with blue in Fig ure 4 1. Then all 2D time series were first screened for movement artifacts using a movie function, although rats were completely paralyzed during experiments. In rats for 710nm were quantified. Then 3D images were reconstructed using 3D delay&su m algorithm, which was accelerated by GPU parallel technique. The resulted 3D images had a voxel number of 201 201 101 representing a 10 10 5mm volume, and the negative voxel
59 values were set to zero. The reconstruction speed is about 0.97s for one frame. Amira (from Visage Imaging, Inc.) was used to render the 3D images. There were two sources of noise in the EEG signal, a strong sharp 10Hz noise from the Q switch of the laser, and the 60Hz alternating current noise. For de noising, the data deteriorated by the Q switch was replaced with adjacent data, and the 60Hz alternating current noise is removed by filtering out 58 62Hz in frequency domain. Then the EEG data was applied with a low pass filter of 100Hz, and transformed into a 300000 point signal for the 25min EEG recording, corresponding to sampling rate of 200Hz. 4. 3 Data Processing for Neurovascular C oupling S tudy Because the 1064nm and 710nm light w ere mainly absorbed by HbO 2 and Hb R in the tissue respectively, with reasonable approximations the PAT signals from them can be taken as the total volume change of those two absorbers 18 . For neurovascular coupling study, averaged signals from specific anat omical regions of interest (ROI) are compared to that from the superior sagittal sinus (SSS) as well as the EEG signal. The PAT signals were expressed as the percentage changes from the baseline, which is defined to be the average value over first 4min of the recording. Next, we further investigated the correlation relationship between these signals. We also calculated the correlation coefficient of every voxel with the SSS, and got the negative correlation map following the seizure onset to identify import ant regions in the brain. We choose has the highest SNR. In this step, the reconstructed 3D image was applied with a moving window of 9 voxel wide cube for avera ging to improve the low SNR due to the limited number of transducers. Finally, the granger causality networks between the SSS, hippocampus, and cortex at different times were calculated. All the acquired results were exami ned by experts in the neurology .
60 4 . 4 Results 4 . 4 . 1 Seizure Generation EEG signals were recorded to confirm the onset, duration, and morphology of the seizure, which is shown in Fig ure 4 2. The EEG signal was stable before the drug injection. Ictal discharges were induced shortly after the injection of PTZ, which were characterized by the low amplitude fast activity with high amplitude (Fig ure 4 2 ( A ) ). The epileptic pattern progressively increased in amplitude in the following 2 to 3 minutes. As the seizure evolves, it gradually turned into clonic paroxysmal discharges of lower frequency. 2D plot of the time frequency power map of the EEG signal shows that a mixture of lower si gnal frequency particularly in 1 3 Hz was seen throughout the seizure, and higher frequency power components (up to 4 6 Hz) were stronger early in the seizure and gradually decreased in power (Fig ure 4 2 (B) ) . This is close to results in other publications. Seizures were induced successfully in all the 16 rats, and they were similar in the shape and amplitude. The injec tion time is easily distinguished as isolated high amplitude spikes around 5min in the EEG signal. 4 . 4 . 2 PAT brain vascular images Representative time frames of 2D images through the experiment are displayed in Fig ure 4 3 ( A ) with a time interval of 1min, compared with a rat brain photo taken after the experiment (Fig ure 4 3 ( B )). Although the image quality is not excellent due to the limited transducer positions (only 64 transducers are averaged once), the SSS and two transverse sinuses (TS) are clearly re constructed, and some other small blood vessels are also revealed. In Fig ure 4 3 ( A ) it is observed that there is no palpable shift of the rat brain which is a prerequisite for the following data processing. In our experiments, 3 out of 8 rats with 1064nm moved, but all the rats in the eyes from these 2D images, so quantitative analyze is needed.
61 Representative 3D images for the HbO 2 and Hb R are displayed in F ig ure 4 4 ( A ) and ( B ) . Similar to 2D, HbO 2 the Ostia end of the straight sinus (SS) which is below the SSS in z direction can also be seen. The SNR of the 3D image for Hbr is not as good as that for HbO 2 partly due to relatively low laser intensity, and the other reason lies in the fact that the concentration of Hb R is much lower than that of HbO 2 in the blood vessels. 4 . 4 . 3 Hemodynamic Signals in Selected ROIs To investigate the hymodynamic changes during the seizure initiation and evolution, ROI s are selected for the SSS, as well as other regions near the hippocampus and the cortex in the brain that are reported to have high gamma aminobutyric acid (GABA) receptor concentrati ons (Fig ure 4 5 and Figure 4 6 ) 130 131 . ROI was also chosen in the upper posterior region in the reconstruction domain, which contained no tissue of the rat served as reference. With a high frame rate of 3.3f/s, the rapid hemodynamic changes in these regions can be the maximum amplitude projection (MAP) PAT vascular struct ural images precisely onto the high resolution anatomical magnetic resonance imaging ( MRI ) slices of the rat brain obtained from online database 132 . Fig ure 4 5 shows the averaged HbO2 signals of the ROIs during the seizure, with corresponding EEG signal presented to confirm the onset, duration, and morphology of the seizure. The HbO 2 signals are converted into the relative percentage change to the baseline. Because the SSS has the highest SNR in the reconstructe d 3D images, the standard deviation ( SD ) for SSS is much higher than the other three signals. The HbO 2 signals and the EEG signal are mutually well correlated. All the HbO 2 signals of the SSS, hippocampus and cortex remained quite still at first. We can la rgely divide the change of these three signals after the injection of
62 PTZ into three for stages, which is indicated in Fig ure 4 5 (B) . We can see that there was an HbO 2 was s hown in the hippocampus at the same time which was probably induced by the dilation of the arteries. the cortex and the SSS are positive correlated. In the t hird stage, both the HbO 2 and Hb R signals in the SSS and cortex dropped, while the increase of HbO 2 in the hippocampus sustained through the whole experiment. Comparably, the reference signal remains the same all although the experiment. Fig ure 4 6 present s result in another rat showing how the changes of Hb R in the SSS, hippocampus, cortex and the EEG signal are mutually correlated. The sizes and the relative positions of the three ROIs are defined to be the same as those in Fig ure 4 5 (A) . The EEG signal here looked almost the same in pattern with that in Fig ure 4 5 ( B ) , indicating that our PTZ induced generalized seizure modal was quite stable and the drug administration was well controlled, so that results in Fig ure 4 5 and Fig ure 4 6 were comparable. We also divided the timecourse of the Hb R changes into the three stages with the exact timing as that in Fig ure 4 5 (B) for easy comparison. Although with a relatively low SNR, we clearly see that the decrease of Hb R ure 4 5 ( B ) . In addition, there was a slightly decrease of Hb R in the hippocampus followed by an increase in the first and second stages, but these changes were not as notable as the increase of HbO2 in the hippocampus at the sa me time in Fig ure 4 5 ( B ) . In general, the Hb R signals for the hippocampus and cortex were generally the same in shape, which were negatively correlated with the Hb R of SSS. Fig ure 4 7 shows the statistical results of the HbO 2 and Hb R changes across all th e rat experiments, which are similar to those in Fig ure 4 5 and Figure 4 6 . The average signal in each
63 rat is calculated every 30 seconds, and then further averaged between different rats, and the corresponding SD is calculated and shown. 4 . 4 . 4 Correlation Study Based on the above results, the hippocampus showed strong hyperfusion and hyperoxgenation during the seizure, and the HbO 2 change of it was inversely related with that of the SSS. Fig ure 4 8 ( A ) shows the time change of the correlation coefficient b etween the HbO 2 (Fig ure 4 8 ( B ) ), but it turned negative for most of the time after the seizure onset (Fig ure 4 8 ( C ) ). All the correlation coefficients in this paper were calculated in a short time window of 1min. For further investigation, we calculated the 3D HbO 2 negative correlation coefficient map between every voxel and the SSS (Fig ure 4 8 ( D ) and ( E ) ). In this step, the 3D HbO 2 images were firstly smo othed with a moving window of a 9 voxel wide cube to improve the SNR. Fig ure 4 8 ( D ) presents the 3D negative correlation coefficient map at different times, as indicated in Fig ure 4 8 ( A ) with vertical red lines. Fig ure 4 8 ( E ) shows the same result as Fi gure 4 8 ( D ) (e), but from different views, along with the MAP images of it in the x y and x z planes coregistered with the corresponding MAP HbO 2 vascular structural images, and the MRI anatomical images. We found that voxel with high negative coefficient (with a threshold of 0.7) were hardly found before the drug injection, but they began to show and concentrated in regions around the hippocampus after the seizure onset, although asymmetry is present in these images (Fig ure 4 8 ( E ) ) 133 . Especially, when the seizure just began later in the first stage of the seizure (around 6 7min), the HbO 2 signal in the hippocampus was correlated with that of the SSS with a large negative correlation coefficient, so that the 3D negative correlation map is preventative at this time (Fig ure 4 8 ( D ) ). We also calculated the negative correlation maps of HbO 2 in other rat experiments, and similar results were obtained (Fig ure 4 9 ). In summery, voxel with high
64 negative correlation coe fficient began to show after the seizure initiation in the hippocampus and its surroundings, which indicates that the hippocampus seemed to be more sensitive in this intraperitoneal PTZ seizure model 134 135 . This also implies that this method may be of great help for seizure localization especially in generalized seizure cases where fMRI is of less help. 4 . 4 . 5 Granger Causality Analysis In Fig ure 4 8 ( B ) and ( C ) , we se e that the hemodynamic coupling of the high frequency fluctuations between different brain regions is different with different times. This implies that the that c an not be inferred from the overall trend. We applied short time window spectral pairwise G ranger C ausality method to identify the functional interaction among the SSS, hippocampus and cortex during the seizure process 111 113 . Granger C ausality allowed for assessment of the magnitude and direction of temporal relationships among multiple signals during overlapping time series windows, holding the promise of revealing p aths of information flow within the nervous system. This method has emerged as the leading statistical quantities to furnish directional information from multivariate neural data. To make use of this formulation, the PAT signals in each short time window is firstly detrend with a 3rd order and then normalized by dividing the standard deviation. In this way, the PAT signals can be approximately regarded as a stationary stochastic process with locally stationary segments, and the collection of the PAT signal s from different rat experiments can be treated as an ensemble of realizations. In this paper, the time window is set to be 1min, which is 200 points for the frame rate of 3.33Hz. This is a promise between preserving the time variability and maintaining th e smoothness of the estimated spectral quantities. The model order is determined with the Akaike I nformation Criterion (AIC ) 136 . Because the AIC generally decreases monotonically with increasing order for all th e windows, we set the model order to be 19
65 selected as a tradeoff between sufficient spectral resolution and overparameterization, and the model order is validated with white noise test. The directional causality was calculated as the average value in the 0 0.3Hz range which makes up about 84% of the total power in the 1.6Hz frequency band. Granger C ausality spectral analysis of both the HbO 2 and Hb R data were performed for all pairs between the SSS, hippocampus, cortex, and the reference signals which is u sed for significance assessments. Fig ure 4 10 ( A ) and ( B ) show the time varied G ranger C ausality of HbO 2 between pairs among the SSS, hippocampus and cortex. Fig ure 4 10 ( C ) and ( D ) are the G ranger C ausality of HbO 2 between these three brain regions with t he reference, which is less varied with time, and the average value of it can be taken as a noise background or baseline, as indicated in Fig ure 4 10 ( A ) and ( B ) with dotted black horizontal lines. Similarly Fig ure 4 10 ( E ) and ( F ) show the G ranger C ausali ty of Hb R for these three brain regions. The results of HbO 2 shows that strong influence of the hippocampus to the SSS is present in the first and second stages of the seizure; strong influence of the cortex to the SSS is seen in the second and early third stages; and as the seizure evolves, significant causality for all pairs between these three regions can be seen in both directions late in the third stage. Comparatively, the causality of the Hb R for pairs between these three brain regions is overall less significant, which may be due to the poor SNR of the data. However, some time relevant features can still be obtained through the calculated results. The G ranger C ausality in different times can be better presented in graphs, as shown in Fig ure 4 1 1 . In Fig ure 4 11 , t he arrow presents the direction of the influence, and the visibility presents the magnitude of the causality. The granger causality of less significance is not shown, and the thresholds for HbO 2 and Hb R are chosen to be the average value of t he causality between the reference and the reference, or rather the dotted lines in Fig ure 4 10 ( A ) ( or ( B) ) and ( E ) (or
66 ( F ) ) respectively. The value of the causality can be read from the colorbars. We see that as the seizure evolved, the connectivity betw een these three brain regions tended to be more and much closer. 4.5 Discussion 4.5.1 Significance of T his W ork Epilepsy affects about 1% of the population world wide, and about 20 to 30% of these patients are refractory to all forms of medical treatment 64 . For those patients the best treatment option is resective brain surgery, which however critically depends on the complete mapping of the local epileptic seizures, and the well understanding of the possible epileptic circuit. Much of our understanding ab out seizure comes from electrophysiological recording methods, and intracranial EEG is commonly used for presurgery seizure localization. However, its defects of invasiveness, low spatial sampling and volume conduction may leads to incomplete seizure mappi ng. As a result, various neuroimaging technologies have been employed to give a high As the most common neuroimaging method, fMRI is well applied for the diagnosi s and presurgery localization of focused seizure, where the hemodynamic changes mainly occurred within the seizure focus and can be well distinguished in fMRI 137 139 . However, fMRI can not localize generalized seizure because in this case the hemodynamic changes can be found wide spread in the brain, and it remains unclear whether they arise from a portion of the brain or from more widespread circuits. This predicamen t mainly owes to two reasons. One reason is that the spatial and temporal resolution of fMRI is poor, so the initial site of seizure is missed. Most importantly, fMRI is only sensitive to Hb R 96 . It relies on the drop of Hb R in the vein due to the speedy of the blood flow to localize the activation sites of the neurons. However, the change of Hb R and the blood flow is
67 two counter parts during the seizure, so that the changes of Hb R in the tissue may not be as nota ble as that of HbO 2 , which mainly results from the dilation of the artery and the blood flow 140 141 . In generalized seizure, although current diagnostic methods inclu ding intracranial EEG that all these regions are the origin of seizure or equally in function. Primary regions or seizure network may be revealed if more i nformation is given. With the real time PAT operated at two different wavelengths, relative focused regions that were negatively correlated with the SSS were found in the hippocampus and its surroundings following the onset of the seizure using the HbO 2 da ta, and the change of HbO 2 in the hippocampus was found to be ahead of the Hb R in the PTZ induced generalized seizure model (Fig ure 4 5 , Figure 4 6 , and Figure 4 7 ). Since the SSS is decreased in HbO 2 when the seizure began, the above regions with high neg ative correlation coefficient represent regions with hyperoxygenation and hyperfusion, which is usually the marker of the seizure location. PTZ is a GABA antagonist, and this result is in coincidence with the findings hippocampus is high in GABA receptor c oncentration. All these imply that PAT may serve better for seizure localization with appropriate laser wavelength settings and data processing methods. Besides, generalized seizure occurs in multiple regions all over the brain that form a seizure network The spatial and temporal resolutions of our realtime PAT system are both about 10 times higher than that of fMRI, thus more subtle hymodynamic changes related to the seizure could be identified. We applied short time window spectral pairwise granger causality method to both the HbO 2 and Hb R data, and the time resolved interaction between different regions of the brain was
68 visualized, resulting in a better understanding of epil eptic networks at high spatiotemporal resolution. 4.5.2 Data P rocessing M ethod PAT has already been employed in the neurovascular coupling study with various animal models, which however are mainly limited in the monitoring of hemodynamic signals in large blood vessels. Actually the changes in capillaries are more related to the metabolic status of the cells. Yet the direct monitoring of the capillary is difficult because the resolution of PAT under the epidermis is usually not enough due to the diffusion of the light. In this work, the signals are averaged in the surrounding region to increase the SNR, and converted into percentage changes regarding to the baseline for comparison. With this method, the small changes in deep tissue that is submerged in the signal of larger blood vessels can be revealed. For example, Fig ure 4 1 2 shows a 2 minute signal summed within a 9 voxel wide cube around the hippocampus with 1064nm (the black line in the upper right corner), expressed as ( 4 1 ) compared with and (lines in the lower right corner in red and blue respectively, which are the summed signals with the two halves of the voxel in the cube), where is the value of a voxel in the cube. The difference of and can almost be neglected, and calculation shows that the SD of is just 0.04 of the SD of . This means the white noise can almost be neglected. Actually, the high frequency fluctuations in the PAT data here are more related to the hemodynamic change in the tissue, which can be inferred from the correlation between the hippocamp us and the SSS. The HbO 2 signals in the hippocampus and SSS are positively correlated before seizure, but negatively correlated with a high correlation coefficient
69 shortly after the seizure onset. This can be totally explained by the fluctuation of the las er or the influence of the reconstruction algorithm. 4.5.3 Other F actors in N eurovascular C oupling Contradictory data exists for neurovascular coupling 142 145 . One reason for this lies in the model difference and the individual difference, as well as the systematic influence of the animals , and the other reason may caused by the different principles of various technologies leading to different conclusion in case of incomplete acquisition of the information. Neurovascular coupling refers to all the aspects regarding the neural metabolic demands and the energy source supply by the vascular system, including the local filed potential, the oxygen consumption rate, oxyge n partial pressure, HbO 2 and Hb R volume, blood flow, and so on. More parameters we can measure, the more correct understanding of the neurovascular coupling we can get for the diagnosis of epilepsy. In this work, through the measuring of HbO 2 and Hb R , oth er factors such as the dilation of the arteries and blood flow, which works as means of power supply to the neurons from the vascular system, can also be indirectly inferred. However, signal analysis can still be complicated sometimes. We only get informat ion of HbO 2 and Hb R , making it difficult to infer the status of the neurons. For example, we observed that the HbO 2 is decreasing while the Hb R is increasing at the second stage in hippocampus, but since the overall blood supply is increase in the rat brai n, we are not sure whether the temporal decreasing of the HbO 2 in the hippocampus which factor dominates. Furthermore, the oxygen partial pressure in the ti ssue which is more related to the neural activity can be uncoupled with the HbO 2 change 146 . This also impedes our understanding of neurovascular coupling.
70 The preceding of the hemodynamic signal to the EEG sign al in ictal seizures has also been reported with various kinds of technologies 147 150 . However, in other cases and in our studies, this phenomenon is not seen. Becaus e this is of quite importance in aspect of seizure prediction and manual intervention, more ictal neurovascular coupling studies should be done with different seizure models to find out the condition and the mechanism of this phenomenon. 4.5.4 Limitation s There are limits in our work. First of all, the image contrast and resolution are greatly hindered by the limited transducer numbers with our system. Besides, although the morphology of seizures exhibit quite similarity for all the rats in our experiments, images with the two different wavelengths are not acquired on the same animal. Although this will be not likely to refute the main conclusions we got, valuable information that need simultaneous dual wavelength recording to be unearthed is missed. Finally and most important of all, generalized seizure involves wide spread network all over the brain, in which the electrical and hemodynamic signal of different regions are closely related. In our work, we only analyzed the granger causality of the hemodynamic change in SSS, hippocampus, and cortex. Detailed study needs to be done to dig out the interaction of more different regions. 4.6 Conclusion In this work, we monitored the hemodynamic changes of PTZ induced seizure in rats with our proposed 4 D PAT system and simultaneous EEG recording. Time series of 3D images were acquired at two different wavelengths for the HbO 2 and Hb R respectively. Compared with PAT imaging using single wavelength for single parameter recording, which is more similar to the case in f MRI, more information regarding the activities of neurons can be acquired. We observed a more focal increasing of HbO 2 around the hippocampus at seizure onset. We also analyzed the granger causality among the SSS, hippocampus and cortex during the seizure. Since
71 fMRI is not sensitive in localizing generalized seizure, our findings here could be of great significance to di agnosis of generalized seizures .
72 Figure 4 1 . Experimental setup of the combined system. The anaesthetized rat was elevated to the cen ter of the spherical transducer array for simultaneous PAT and EEG recording. The 192 channel signals were collected and multiplexed with 3:1 into a 64 channel data acquistion system, and with a 10Hz laser, a frame rate of 3.3 3 f/s is achieved. Then the col lected data is stored for later GPU accelerated reconstruction using 3D delay&sum code .
73 Figure 4 2 . Electrophysiology of PTZ induced seizure. A ) An example of EEG traces. The arrow highlights the time of PTZ injection. The seizure always starts with s ynchronized high frequency oscillations that increase in amplitude, progressing into clonic paroxysmal discharges of lower frequency afterwards. B ) Time frequency analysis for the same EEG signal above. Analysis was performed using a short time Fourier tra nsform. Peak power was normalized to 1. The dominant ictal signal frequency was 1 3 Hz. Higher frequency components up to 6 Hz occurred earlier in seizure and dim inished in power gradually .
74 Figure 4 3 . Representative time frames of 2D PAT images of the rat brain, compared with an anatomical image. A ) Representative time frames of 2D PAT images of the rat brain with 1064nm. The time interval is 1min, and each image covers 10Ã—10mm2 with 201Ã—201 pixels. B ) Photo of rat brain with hair and scalp being remove d. The SSS, TS, and the reconstruction area are indicated. Scale bar is 5mm through the figure . Figure 4 4 . The re construction of 3D PAT images. A) and B) show the representative 3D images for HbO2 and Hbr in different views respectively. The image doma in is 10Ã—10Ã—5mm containing 201Ã—201Ã—101 pixels. Thresholds and colorbars are presented for the 3D images. The SSS and TS are well distinguished in both the HbO 2 and Hb R images, and the Ostia end of the inferior sagittal sinus can also be seen in the HbO 2 im age.
75 Figure 4 5 . The HbO 2 changes of the SSS, hippocampus and cortex. A) 3D vascular images of the rat brain (a1) and the MAP vascular structural images (a2 and a3 in the x y and x z plane respectively) overlaid on the MRI anatomical images using 1064nm . The ROIs for the SSS, hippocampus, cortex, and the reference are indicated. Thresholds, colorbars, and scale bar are presented. B) The HbO 2 changes of the SSS, hippocampus, cortex and the reference as well as the corresponding EEG signal. The changes of the HbO 2 signals are expressed in percentage relative to the baseline, which is defined as the first four minutes of the recording. The changes of these HbO 2 signals after the PTZ injection are divided into three stages, as indicated .
76 Figure 4 6 . The Hb R changes of the SSS, hippocampus and cortex. A) 3D vascular images of the rat brain (a1) and the MAP vascular structural images (a2 and a3 in the x y and x z plane respectively) overlaid on the MRI anatomical images using 710nm. The ROIs for the SSS, hipp ocampus, cortex, and the reference are indicated. Thresholds, colorbars, and scale bar are presented. B) The Hb R changes of the SSS, hippocampus, cortex and the reference as well as the corresponding EEG signal. The changes of the Hbr signals are expressed in percentage relative to the baseline, which is defined as the first four minutes of the recording. The changes of these Hb R signals after the PTZ injection are also divided into three stages with the same timing as that in Fig ure 4 5 B) , as indicated.
77 Figure 4 7 . The average HbO 2 and Hb R changes of the SSS, hippocampus and cortex across all the rats. The signals are averaged every 30 seconds, and corresponding SD between different rats are also calculated and shown (in red). A) and B), HbO 2 and Hb R ch anges of the SSS; C) and D) HbO 2 and Hb R changes of the hippocampus; E) and F) HbO 2 and Hb R changes of the cortex.
78 Figure 4 8 . Correlation study of the HbO 2 signals. A) The time change of the correlation coefficient be tween the hippocampus and SSS. B) a nd C) the HbO 2 signal changes of the SSS, hippocampus and cortex at two different time window as indicated in A with blue rectangles. D) The 3D negative correlation coefficient map at the time of the green vertical bar in A, shown in different views, and t he MAP images of it in the x y and x z plane imposed on the corresponding MAP HbO 2 vascular images and the MRI anatomical images. E) The 3D negative correlation coefficient map at different times as indicated in A with vertical bars. The threshold for the negative correlation correlation voxel with the SSS for HbO 2 were concentrated in the hippocampus and its surroundin gs after the seizu re initiation .
79 Figure 4 9 . Negative correlation map of HbO 2 in different rats. The time window for generating the correlation map is chosen to be from 6min to 7min. The results are presented in 2D maps, which are imposed on the corresponding MAP HbO 2 va scular images and the MRI anatomical images in x y and x z planes, and also presented in 3D images, which are coregistered with the 3D vascular images. The threshold for the negative correlation map is chosen to be 0.6 in the 2D images, and 0.4 in the 3D i mages. A) D) represent the results for different rats.
80 Figure 4 10 . Granger C ausality analysis of the HbO 2 and Hb R signals between different regions of the brain and the reference signals. A) and B) the granger causality of HbO 2 between th e SSS, hippoc ampus and cortex; C) and D) the G ranger C ausality of HbO 2 between the above three br ain regions and the reference; E) and F) the G ranger C ausality of Hb R between the SSS, hippocampus and cortex. The horizontal dotted lines in A) and B) are the average valu e of the G ranger C ausality of HbO 2 between these three brain regions and the reference, or rather the average value of C) and D). Similarly, the hori zontal dotted lines in E) and F) are the average value of the G ranger C ausality of Hb R between these three brain regions with the reference. SSS, superior sagittal sinus; Hipp, hippocampus; Cort, cortex.
81 Figure 4 11 . Granger C ausality graphs of HbO 2 A ) and Hb R B) between different regions of the brain. The granger causality values are coded by the visibility of the lines between these brain regions, and the arrowheads indicate the direction of G ranger C ausal influence. Lines between site pairs not reaching significance in the coherence and Granger C ausality measures are not shown. SSS, superior sagittal sinus ; Hipp, hippocampus; Cort, cortex. Figure 4 12 . White noise assessment of the reconstructed images. Upper right corner shows the summed signal of a 9 voxel wide cube, compared with the signals summed with the two halves of the total voxel in the cube (t he red and blue lines in lower right corner).
82 CHAPTER 5 SIMULTANEOUS PAT/EEG RECORDING OF FREE MOVING RATS Study of neuro hemodynamic activities in freely moving small animals (rather than anesthetized) provides more realistic and accurate information abo ut diseases, and hence better understanding. While it has been proved that hemodynamic changes are closely related to epileptic seizures, methods for detection in freely moving (small) animals are limited. In this chapter , we integrated photoacoustic senso r and EEG system into a small device that can be attached on the head of awaken rats, and for the first time it shows the realization of long time and simultaneous monitoring of photoacoustic and EEG signals in PTZ induced seizure on freely moving small ra ts. Results showed that both the neural and vascular responses to seizure in freely moving rats have characteristics, which are observed to be different and more diverse from that of anesthetized rats, and this calls for more detailed study in future. This technology also promises for other hemodynamic related research study in freely moving sm all animals . 5 .1 Background and S ignificance Epilepsy is the most common brain disease that affects 2 % of the world population 1 2 . Epileptic seizures are caused by synchronous, rhythmic firing of a population of neurons in the brain that involves enormous increase in metabolic rate of oxygen and thus hemodynamic parameters such as cerebral blood flow (CBF), cerebral blood volume (CBV) and oxy saturation are greatly changed 3 4 . A better understanding of coupling between the neural and vascular systems is crucial to diagnosis, prediction and treatment for this disease. Currently ava ilable neural imaging modalities such as fMRI, PET, SPECT have already been employed in the study of neurovascular coupling in epilepsy both in human and animal models. However, one remaining problem for these technologies is that they are expensive and bu lk in size, so the objects must be anesthetized or keep still during the entire recording process.
83 For recently developed optical mapping methods such as ORIS, OCT and voltage sensitive dye, pre operation surgeries like skull opening are generally required due to their limited imaging depth and thus they permit only anesthetized animals. Problem associated with anesthetized animals is: the anesthetization suppresses the activities of neurons and cardiovascular system, so the neural and vascular responses t o seizure are altered. As a consequence, awakening and freely moving animals with vascular response uninterrupted by extrinsic factors is ideal and preferred model for epilepsy study. Conventional EEG or telemetric EEG system with video monitoring is the g old standard for seizure detection and quantification, and has already been widely used on freely moving animals 151 152 . Yet the compact devices, which can be firmly fixed on the animals and hence can reliably detect hemodynamic changes, are still missing and no studies have been reported. Photoacoustic s have been used for the hemodynamic monitoring of epilepsy . However, most of the photoacoustic techniques focus on im aging that permits only the anesthetized animals. Besides, the imaging speed of photoacoustic systems is often limited by the laser repetition rate. On other hand, its potential for real time monitoring as a sensor in simple and compact design that can be used on freely moving animals is not fully explored. The purpose of this work is to develop an interface that consists of an EEG system and a photoacoustic sensor, and to study the interaction between the EEG signal and the hemodynamic signal of SSS in PT Z induced seizure on freely moving rats. PTZ has been employed to induce generalized seizure in rats to study seizure phenomenon, where hemodynamic changes have reportedly been found all over the brain. As one of the most distinguishable as well as main bl ood vessels in brain, SSS allows blood to drain from lateral aspects of anterior cerebral hemisphere to confluence of sinuses and return to venous circulation, covering large region of
84 brain. Thus the hemodynamic signal of SSS is closely related to the sta tus of the PTZ induced generalized seizure. Through this dual modality, monitoring epilepsy in freely moving animals, one can get an in depth knowledge of responses from vascular and neural systems, and hence better understanding of mechanisms of epilepsy . This technology may have greater utility in applications such as drug development and therapy assessment in epilepsy, and other hemodynamic related studies. 5 .2 Material and methods 5.2.1 System O verview The entire system can be divided into four main pa rts as: photoacoustic system, EEG system, video camera, and small interface mounted on rat head, as shown in Fig ure 5 1 . Pulse light of wavelength, 1064nm from a Q switch Nd:YAG laser operating at 10Hz was coupled into a 0.4mm core multimode fiber patch ca ble using a convex lens and delivered to the rat brain. The transient photoacoustic signal generated by the short laser pulse was detected by a pair of transducers with central frequency 5MHz. The signal was amplified using a 26dB PCB preamplifier and a 17 dB commercial preamplifier (AH 1100, ONDA Co.), and finally digitalized by a PCI data acquisition card (NI PCI 5154) with sampling rate 50MHz. The intracranial EEG signal was collected through a pair of small screws drilled in the skull of rat. The depth o f the screws through the skull was about 1mm. The EEG signal was amplified (RA16PA, TDtucker Davis Tech.), and collected (RZ5 Bioamp Processor, TDtucker Davis Tech.) and stored into a computer for processing later. The entire system was synchronized throug h triggering with Nd:YAG laser. Continuous video recording was achieved with a video camera (we used iPhone 4S). The transducers, multimode fiber and EEG connecting wires are detachable from the rat
85 head when the experiment is over. All the data including the photoacoustic signal, EEG signal and video are jointly analyzed later for study of epilepsy. 5.2.2 Photoacoustic S ensor D esign Fig ure 5 2 ( A ) depicts the schematic of photoacoustic sensor being mounted on a rat head. Multimode fiber patch cable (0.4mm core diameter) for delivery of light was connected to an optical cannula (CFMC14L02, Thorlabs Inc.) with a ceramic mating sleeve (see Fig ure 5 2( B )). The cannula was implanted right above SSS, with a tip (2mm) reaching into brain through a small hole in s kull. In rat experiment, the optical energy coming out of the multimode fiber was about 6mJ/pulse and that from the connula tip into the brain was about 4mJ/pulse. The cannula was fixed with dental cement, and then the skull surface was sealed with a thin layer of super glue. The two transducers were tilted and placed with their axes pointing to SSS. They were supported by a 3D printed VeroWhitePlus interface and fastened with small screws and thermal glue. The space between the skull and the transducers wa s filled with ultrasound gel for proper acoustic coupling. Fig ure 5 2 ( C ) shows a photograph of the above mentioned design being mounted on rat head (after the surgery). 5.2.3 Surgery Four Sprague Dawley rats, whose weights were around 400g, were employed in the experiments. SSS was about 2mm below the skull. Fig ure 5 3 shows the top view of a rat skull during surgery. The positions of the optical cannula and the two EEG electrodes are noted by arrows. The surgery was carried out as follows. Firstly, rat w as anesthetized with isoflurane and hair on the head was removed, and subsequently, the scalp was opened. Next, a hole of diameter 0.5mm was carefully drilled in the skull, and cannula was inserted into the hole and fixed with dental cement. The surface of the skull was cleaned and applied with antibiotic, and then sealed with a thin layer of super glue to prevent from infection. After that, two small screws were
86 drilled at forehead and they served as two EEG electrodes. Finally, 3D Printed VeroWhitePlus in terface was fixed onto the rat skull with dental cement, and the two EEG screws were also embedded in dental cement, leaving two wires tied on the screws stretching out for later connecting to recording leads. After four days (for complete recovery), rat w as adopted for (epilepsy) experiment. All the procedures were carried out in accordance with an approved University of Florida IACUC protocol. 5.2.4 Data C ollection and P rocessing Phantom experiments were carried out to validate the photoacoustic signal f rom SSS. A tube with inner diameter 0.2mm was filled with ink, and embedded in a phantom (we employed agar phantom) at a depth of 2mm. Then VeroWhitePlus interface along with optical cannula and transducers was placed onto the phantom to record the photoac oustic signal from the tube. By comparing the resulting photoacoustic signal with that of rat experiments, the photoacoustic signal of SSS can be verified (amplitude vs. time was calculated). Because the optical absorption at wavelength 1064nm is dominantl y sensitive to HbO 2 as compared to that of HbR , the photoacoustic signal at this wavelength can be taken as the volume change of HbO 2 in SS S . For each experiment, the signals from the two transducers were compared, and the better one was selected for recor ding. There were two strong sources of noise in the collected EEG data. One is from the alternating current supply at 60Hz and other is from Q switch of the laser at 10Hz (Fig ure 5 4 ( B ), blue line). The EEG data was firstly Fourier transformed and the alt ernating current noise of 60Hz was eliminated by filtering out the frequency band in the range 58Hz to 62Hz, and the Q switch noise was removed by filtering out the frequency band of 10 0.5Hz, 20 0.5Hz, 3 0 ure 5 4 ( A )). Then the filtered EEG data was inverse Fourier transferred back to obtain the time resolved data (red line as shown in Fig ure 5 4 ( B )).
87 5.2. 5 Rat Experiments All the rat experiments were car ried out between 2:00pm and 5:00pm. The rats were first placed in an acrylic cage and acclimatized for at least one hour before the experiment. Then the rats were anaesthetized, and the EEG leads, the multimode fiber, and the transducers were connected to the head of rats for recording. For free moving rat, the seizure experiment was initiated around 10min after the awake of rat, in which seizures were generated by injecting PTZ of 30 75mg/Kg into abdomen 123 . Two kinds of control tests were performed. In first case, sterile saline of same volume was injected, and in other case, rats were continuously anesthetized with a PTZ dose of 120mg/Kg. The concentration for PTZ was fixed to 50mg/mL. The duration of recording is 30mins for each experiment, and the inj ection of drug was induced around 5mins. Behavioral seizures of all rats were scored usi ng 5 graded Racine Score system 153 , and the score was based on the severest behavior of rats. A total number of 22 experime nts (n=22) were carried out using 4 rats, as shown in Table 5 1. The time interval between two experiments for the same rat was at least 3 days. All the results were confirmed by experts in neurology. 5 .3 Results 5.3.1 Photoacoustic S ignal V erification Pho tograph of a phantom (employed in experiment) is shown in Fig ure 5 5 ( A ) and photoacoustic signals for phantom experiment (blue line) and rat experiment (red line) are shown in Fig ure 5 5 ( B ). Both the curves show strong and sharp bipolar signals at the in stant around 3us with clear background, which is equivalent to about 5.3mm in tissue. We conclude that the photoacoustic signal from SSS is about 3 s in rat experiment.
88 5.3.2 Dose and B ehavior In all unanesthetized rat experiments, the rat was put into an uncovered acrylic cage, as shown in Fig ure 5 6 . The dose of PTZ was carefully administrated ranging from 30~75mg/Kg. One rat died shortly after a dose of 75mg/Kg, while all the rats injected with 50mg/Kg or under survived. During the seizure process, the rats showed behavior of facial clonus, head nodding, forelimb clonus, rearing and falling (loss of postur al control ). V ideo was recorded about 0.5 m in front of the cage with iPhone 4S. In some experiments, the rats in the cage were well trapped. In this case, the rats only turned around the two ends of the cage, and then the operator needed to turn the cage in opposite direction to prevent the multi mode fiber from twisting. However, in other experiments, the rats jumped out of the box, or torn off the optical fiber and/or transducers. Thus, only 30% (3 out of 10) of the freely moving rat experiments with more than 50mg/Kg PTZ can be fully recorded al l through the experiment (Table 5 1). No seizure was observed in the rats injected with sterile saline. In general, the severity of seizure tended to be higher with a higher dose of PTZ injection (see Fig ure 5 7). For anesthetized rat experiments, the rats were continuously anesthetized with isoflurane breathing circuit, and the body temperature of the rats was maintained at with a thermo pad. 5.3.3 Simultaneous EEG and H emodynamic R esponse S tudy Fig ure 5 8 shows the experiment results of three different rat models: anaesthetized seizure model (Fig ure 5 8 ( A )), control model (Fig ure 5 8 ( C )), and freely moving seizure model (Fig ure 5 8 ( B ) using rat2 and Fig ure 5 8 ( D ) using rat3) respectively. All the injections were administrat ed at the instant 5min. By comparing these curves, we noted that these three different rat models showed distinguishable features for both HbO 2 (red) and EEG (blue) signals. For anesthetized rats (with 120mg/Kg PTZ injection, Fig ure 5 8 ( A )), there are fe w sharply isolated waves in the EEG signals before the administration of PTZ, and at the same time
89 some fluctuations in HbO 2 are observed. The PTZ injection causes some high amplitude spikes in the EEG signal at 5min. After seizure begins, the EEG signal s lowly increases in amplitude and frequency, and in the mean while, the HbO 2 signal first shows a sudden drop, and then gradually increases. Sooner after, some hypoxia gaps appear in the HbO 2 signal, which becomes ed that two anesthetized rats were killed due to overdose of PTZ when they got awake after taking isoflurane breathing circuit away with completion experiments. For freely moving rats (with 50mg/Kg PTZ injection) in Fig ure 5 8 ( B ), the EEG signal is activ e before the PTZ injection, while HbO 2 level is relatively stable. When the rats were caught and taken out for injection at 5min, the EEG signal activity becomes low, and the HbO 2 signal increases remarkably for a short interval of time. The seizure onset is about 3min after the injection. At the same time, the HbO 2 level increases significantly and then gradually gets lower, and the EEG signal displays continuous high amplitude that shows bursting activities. The result in Fig ure 5 8 ( D ) (with 0.5mg/Kg PTZ injection) is much different from that in Fig ure 5 8 ( B ). The fluctuation of HbO 2 signal in Fig ure 5 8 ( D ) is much larger (all through the experiments) than that in Fig ure 5 8 ( B ). The overall activity of EEG signal is prominently increased due to seizure onset in Fig ure 5 8 ( D ), but not in Fig ure 5 8 ( B ). Apparently, the results for freely moving rat experiments seemed more individual and case dependent. For the test experiment under control ( Fig ure 5 8 ( C )), in which the rats were injected with 0.4mL of 0.9% sterile saline, no dramatic change is seen both in HbO 2 and EEG signals HbO 2 signal or EEG signal as show n in Fig ure
90 5 8 ( C ), which is much different from the results for freely moving rat experiments in Fig ure 5 8 ( B ) and Fig ure 5 8 ( D ). This may be due to individual difference of rats. 5 .4 Discussion 5.4.1 P hotoacoustic S ensor D esign Photoacoustic technol ogies have gained much attention for their diverse applications in photoacoustic sensor and EEG system in a small device for the dual modality study of epileps y in free moving rats. By utilizing ultrasound as means of detection, the penetration depth of PAT can be much deeper than that of pure optical imaging while keeping high spatial resolution. Thus non invasive detection of blood vessels in small rat brain c an be achieved. However, in this work, penetrate scalp and skull of mature rat. Small rat can not be employed here because fiber and transducers can not be fir mly attached to the loose scalp of small rats. There are other limitations in this work as well. For example, we only used one wavelength, but it can be easily adapted to dual wavelength measurement with the same Q switch Nd:YAG laser; and multiple illumin ation and detection sites in the brain can also be realized too. All these can be improved in future work. 5.4.2 Dual M odality S tudy with V ideo M onitoring When rat was anesthetized, the fiber, transducers, and EEG leads were swayed but no notable fluctuat ions was observed in both the photoacoustic and the EEG signals, proving the reliability of the system. Video monitoring was used to confirm the presence of a motor component associated with an EEG seizure and to score the severity of motor seizures. We no ticed that there are some artifacts in both the photoacoustic signal and EEG signal before the injection of drug, which are absolutely not related to seizure. These artifacts could not be discriminated without video monitoring.
91 For simultaneous monitoring of EEG and photoacoustic signals, the results for anesthetized rats are generally similar in which hypoxia gaps can be easily observed in the HbO 2 signal in all the cases. In contrast, results in freely moving rats are more case dependent and complicated than that of anaesthetized rats. In some experiments with freely moving rats, we also observed the proceeds of hemodynamic signal to the EEG signal; and EEG signals of different frequencies are decoupled sometimes. However, in this study, no statistic base d analysis is derived due to limited number of cases. The EEG and hemodynamic signal recordings after seizure (which we missed to carry out in this work) are also important and interesting issues for epilepsy study. As a result, more detailed study is requ ired (in future) with more cases and longer observation time. But it is beyond the scope of this communication. 5. 5 Conclusion This study demonstrates the practicality of photoacoustic detection combined with EEG and video monitoring of epilepsy in freely moving rats. Rats with different seizure model were tested, and we found that the EEG and hemodynamic signals during seizure in freely moving rats are more complicated than that of anesthetized rats, which requires more detailed research in future. With th e detachable design of fiber and transducers in this type of recording system, one rat can be used many times, allowing long term studies of epilepsy. In our experiments, rats survived more than one month until they were killed. With appropriate modificati ons, this technology can easily find other applications related to hemodynamic study than epilepsy, in which freely moving rat model is preferred.
92 Figure 5 1 . Schematic of experimental system. The integration of photoacoustic system, EEG system, video camera, and small interface mounted on the rat head is depicted. Figure 5 2 . The design of the photoacoustic sensor. A ) Schematic of the photoacoustic sensor in 3D. Laser is delivered to SSS through an optical cannula, which is implanted on rat skull. T hen a pair of transducers receives the generated ultrasound signals. The diameter of the receiving transducers is approximately 5.5mm, and they are supported by a 3D printed VeroWhitePlus interface. B ) Photograph of optical cannula, mating sleeve and multi mode fiber patch cable. The optical cannula has a diameter of 2.5mm,
93 with a piece of multimode fiber (0.4mm core diameter) in its center. C ) Photograph of rat head, with all detectors being connected, after surgery. The EEG electrodes are implanted in fore head of the rat, as indicated. Figure 5 3 . Photograph of a rat skull, with the positions of optical cannula and EEG el ectrodes as indicated by arrows . Figure 5 4 . EEG data de noising in frequency domain. A ) Frequency power chart of EEG signal (before de noise (red) and after de noise (blue)). B ) EEG signal before (blue) and after (red) de noise. The noises are indicated with arrow and green encircles: A (the Q switch noise (10Hz) from the laser) and B (the alterna ting power supply noise (60Hz)) .
94 Fig ure 5 5 . Phantom experiment for validating photoacoustic signal of SSS. A ) Photograph of phantom. B ) Photoacoustic signals in phantom experiment (blue), and in rat experiment (red). The signal for SSS is at instant about 3.5 s. Figure 5 6 . Awake and freely moving rat in acrylic cage for experiments. Figure 5 7 . Distribution of Racine score (in %) in freely moving rat seizur e experiments for each PTZ dose .
95 Figure 5 8 . EEG (blue) and HbO2 (re d) signals for rat experiments of different models. A ) Anaesthetized rat with 120mg/Kg PTZ injection. C ) Freely moving rat with 0.4mL injection of 0.9% sterile saline. B ) and D ) Freely moving rats with 50mg/Kg PTZ injection. .
96 Table 5 1. The rat e xperiment design. C ase number R at1 Rat2 Rat3 Rat4 Anaesthetized 1 * 1 * 1 0 30mg/kg PTZ 2 (2) 1 (1) 0 1 (1) 50mg/kg PTZ 2 4 (1) 3 (2) 0 75mg/kg PTZ 0 0 0 1 Control 2 1 2 0 * T he rat died after experiment. Note: T he number s of rats that were successfull y vid eo recorded through 30mins are indicated in the brackets .
97 CHAPTER 7 CONCLUSION AND FUTURE WORK 6.1 C onclusion and S ignificance In t his dissertation , we are trying to fully explore the potential of photoacoustic technologies in the study of epilepsy thro ugh different kinds of rat experiments. First of all, it s proved that epilepsy involves hemodynamic changes over the whole brain, so that a 4 D PAT system with both high spatial and temporal resolution is required. Chapter 2 and 3 describe our work for dev eloping such a system based on multi acquisition channel for small rat brain imaging. In chapter 2, we have setup a 192/64 channel data acquisition system including the transducers, pre amplifier boards (26dB gain), and 8 8 channel high speed data acquisit ion boards, and an USB IO control, based on which a real time 2D circular 120 transducer array PAT system was built and calibrated. With the 10Hz repetition rate from a tunable Ti:Sap p h i re laser, the system can achieve a frame rate of 3.33f/s, and a spatia l resolution about 0.2mm. Intensive phantom experiments were done for the system characterization and performance. This work is a further elevation of the single transducer based system, and also a pre investigation work for Chapter 3, since they are based on the same data acquisiton system and the calibration of these two systems are similar. In Chapter 3, we remodeled the real time 2D P AT system into 3D with a spherical transducer interface. The new system was capable of 3D imaging with the same frame rat e of 3.33f/s, and also retained the 2D imaging ability with a frame rate of 10f/s using the 64 transducers at the bottom layer of the interface. The system has a near isotropic resolution about 0.2mm, and its real time imaging ability was tested using ink flow experiments. Finally, t he feasibility of the system for in vivo real time hemodynamic imaging in the whole rat brain was demonstrated sing a focused seizure model.
98 Based on that, generalized seizure experiments were carried out with this system, empl oying two groups of rats for the detection of HbO 2 and HbR signals , which are described in Chapter 4 . Compared to the focused seizure model in Chapter 3, g eneralized seizure model are more complicated and more representative, where hemodynamic changes happ ens in the whole brain. EEG signals were also recorded to confirm the initiation, evolution and propagation of the epileptic seizures. We extracted the hemodynamic signals from different regions in the rat brain, and largely divided the seizure process int o three stages. We found that the increase of blood supply is in adequate for the consumption of oxygen in the beginning stage of seizure. At the same time, the HbO 2 signal in the hippocampus increased up to 200%, indicating the high neural activities there . The HbO 2 and HbR signals in the SS S are well negative correlated, and regions of hyper oxygenation were found to be coagulated around the hippocampus after the seizure initiation with correlation study. We also analyzed the Granger C ausality influences a mong different regions, and results show that the interaction s between different regions tend to be stronger. This also indicates that the fluctuations in the hemodynamic signal we obtained may convey more useful information that needs to be dinged out. On the other hand, current stud ies on epilep sy with neuroimaging modalities are mostly performed on anesthetized animals. However, it s proved that the anesthetization suppress the neural activity, so that the results acquired in these experiments may not be as genuine as that in free moving animal experiments. In Chapter 5, we have built a small photoacoustic sensor integrated with EEG electrodes that can be mounted on the rat brain, to collect the hemodynamic signal and EEG signal simultaneously in free mov ing mature rats. Video recording was also present to identify seizure related changes in the photoacoustic signal and EEG signal. We performed rat experiments with different drug level, and did control experiments. One thirds of
99 free moving rat experiments can be fully recorded, whose results however were quite diverse and different from those in anesthetized rat experiments. This calls for more cases for statistical study in the future. 6. 2 Limitations and Future W ork s Limitations exist in our current work . This work consists of researches from two different aspects. The first one is to develop high er performance photoacoustic setups, and the other is about experiment design and data processing. In our work, a 4 D PAT system was developed. However, the capa bilities of this system need improving. The biggest problem with this system is the limited transducer numbers, due to which some important features of the objects can be missing. Although the static phantom experiments showed acceptable results, there are artifacts in the ink flow experiments, and only three main blood vessels can be clearly distinguished in the rat brain. A direct solution to this problem is to increase the number of detectors in a 3D scan. This can be achieved by increasing the number of transducers and data acquisition channels. It s also a common practice to rotate the transducer array to increase the number of detectors in a 3D scanning, although at a cost of lowering the frame rate of the system. In addition, the in homogeneity of the transducer distribution on the array results in artifacts in the reconstruction results. Thus, numerical simulations and should be done for the optimization of the transducer numbers and distribution. P hantom tests should also be done with dif ferent transd ucer distributions, and the transducer mounting must be flexible to allow the changing of transducer positions. In our previous studies, only the influence of the transducer numbers on the image quality was investigated , and other works remain to be done. We have proved that the hemodynamic changes in free moving rats can be detected photoacoustically. However, only the HbO 2 signals in the SSS was revealed, and this system can
100 be easily adapted into dual wavelengths with current Nd:YAG laser. Besides the 10 64nm we employed, the laser gives the second harmonic wavelength 532nm, which can be approximately regarded as a n iso sbestic wavelength for measuring the total hemoglobin in the blood vessel. Thus the change s of HbR and the oxy saturation can also be obtai ned. This will provide other complementary information that is valuable for the diagnosis of epilepsy other than HbO 2 . In addition, we only monitored the hemodynamic changes at one site, but detection of multiple sites can also be realized. More detailed d ata processing and better experimental design n eed to be carried out in further study. In the generalized seizure experiments, we only extracte d the signals in three regions due to the presenting of the low SNR and artifacts in the reconstructed images. If the system performance can be improved, we could have examined more regions related to epilepsy in deeper regions in the brain, such as thalamu s , amygdala , and brainstem . Besides, different regions in the cortex should have distinguished hemodynamic chang es, but we only selected one brain cortex region to investigate in our early study. Then more reliable and complete correlation and Granger causality analysis can be done with h e modynamic changes from more regions. Post ictal state should also be examined for the study of seizure termination, and it s critical for the epilepsy therapy assessment , which may be carried out in further study. M o st importantly, a lthough many conclusions have been derived, the significance of these statistical f in ding s was hinder ed by the limited number of cases. In our generalized seizure experiments, only eight rats were employed in each experiment group, in which only five rat experiments were successful for the HbO 2 experiment , and the PAT data from most of the eight rats has a low SNR for the HbR experiment . The number of successful cases in the free moving rat experiments was also limited, thus no statistical results were obtained. If more cases can be
101 done, we not only get more convincible conclusions, but also more possible new findings, which may help the better understanding of epilepsy.
102 LIST OF REFERENCES 1 B. Meldrum , Concept of activity induced cell death in epilepsy: historical and contemporary perspectives , Prog Brain Res, 135 , 3 11 (2002) . 2 B. Siesjo, T. Wieloch , Epileptic brain damage: pathophysiol ogy and neurochemical pathology , Adv . Neurol . 44 , 813 847 (1986) . 3 M . Zhao, H . Ma, M . Suh, and T. Schwartz , Spatiotemporal dynamics of perfusion and oximetry during ictal discharges in the rat neocortex , The Journal o f Neuroscience 29 (9) , 2814 2823 (2009) . 4 V . Junior, L . Ana, R . Silva, D . Abud, S . Rosset, R . Pereira, and J . Leite , Neurovascular coupling and functional neuroimaging in epilepsy , J. epilepsy clin. Neurophysiol 15 (1) , 30 36 ( 2009 ). 5 R . Kuzniecky, R . Knowl ton , Neuroimaging of epilepsy , In Seminars in neurology 22 (3) , 279 288 ( 2002 ). 6 P . Beard , B iomedical photoacoustic imaging , 1 , 602 631 (2011) . 7 C . Li, and L . V . Wang , Photoacoustic tomography and sensing in biomedicine , Phys. Med. Biol. 54 , R59 R97 ( 2009 ) . 8 Q. Zhang, Z. Liu, P. R. Carney, Z. Yuan, H. Chen, S. N. Roper, and H . invasive imaging of 53 , 1921 1931 (2008). 9 B . Wang, L . Xiang, M . S. Jiang, J . Yang, Q . Zhang, P. R. Carney, and H . Jiang , photoacoustic tomography system for non invasive real time three dimensional imaging of epilepsy , Bio . Opt . Express 3 (6) , 1427 1432 ( 2012 ). 10 V. Tsytsarev, B. Rao, K. Maslov K, L. Li, and L.V. Wang , Photoacou stic and optical coherence tomography of epilepsy with high temporal and spatial resolu tion and dual optical contrasts , J . Neurosci . Methods 216 (2) , 142 14 5 ( 2013 ). 11 X . Wang , Y . Pang, G . Ku , X . Xie, G . Stoica , and L. V. Wang , Noninvasive laser induced pho toacoustic tomography for structural and functional in vivo imaging of the brain , Nature Biotechnology 21 , 803 806 (2003) . 12 L. V. Wang, computed tomography , N ature P hotonics 3 , 503 509 (2009). 13 L. V. Wang, and S. Hu, t omography: In v ivo i maging from o rganelles to o rgans , Science 335 , 1458 1462 (2012). 14 B . Cox, J . Laufer, S . Arridge, and P . Beard , Quantitative photoacoustic imaging: a review , J. Biomed. Opt. 17 (6), 061202 ( 2012 ) .
103 15 S. Hu, and L. V. Wa ng, Neurovascular photoacoustic tomography , Frontiers in Neuroenergetics 2 , 10 (2010). 16 H . Zhang, K . Maslov, G . Stoica, and L . V . Wang , Functional photoacoustic microscopy for high resolution and noninvasive in vivo imaging , ature Biotechnology 24 (7) , 848 851 ( 2006 ). 17 L. V. Wang, Prospects of photoacoustic tomography , Med. Phys. 35 ( 12 ), 5758 5767 (2008). 18 W. G . Zijlstra , and A . Buursma , Spectrophotometry of Hemoglobin: Absorption Spectra of Bovine Oxyhemoglobin, Deoxyhemoglobin, Carboxyhemoglobin, and Methemoglobin , Comparative Biochemistry and Physiology Part B: Biochemistry and Molecular Biology 118 ( 4 ) , 743 749 ( 1997 ). 19 F . A. MartÃn, D . Rojas DÃaz, M. L . Luis GarcÃa, J . L. GonzÃ¡lez Mora, and M . A. Castellano , oxyhemoglobin and deoxyhemoglobin from small areas of the rat brain by in vivo visible spectroscopy and a least square approach , Journal of Neuroscience Methods 140 (1 2) , 75 80 (2004). 20 W. G. Zijlstra, A. Buursma, H. E. Falke, and J. F. Catsburg , Spectrop hotometry of hemoglobin: absorption spectra of rat oxyhemoglobin, deoxyhemoglobin, carboxyhemoglobin, and methemoglobin , Comparative Biochemistry and Physiology P art B: Comparative Biochemistry 107 ( 1 ) , 161 166 (1994). 21 J Xiao, and J . He , Multispectral qua ntitative photoacoustic imaging of osteoarthritis in finger joints , Applied Optics 49 (30) , 5721 5727 (2010) . 22 B. T. Cox, S. R. Arridge, and P. C. Beard , Estimating chromophore distributions from multiwavelength photoacoustic images , JOSA A 26 (2), 443 455 (2009) . 23 X . Wang , X . Xie , G . Ku , and L. V. Wang , concentration and oxygenation in the rat brain using high resolution photoacoustic tomography , Journal of Biomedical Optics 11 (2), 024015 (2006). 24 N . K. Logothetis, J . Pau ls, M . Augath, T . Trinath , and A . Oeltermann , Neurophysiological investigation of the basis of the fMRI signal , Nature 412, 150 157 (2001). 25 K. J. Friston, A. P. Holmes, C. J. Price, C. BÃ¼chel , and K. J. Worsley , Multisubject fMRI Studies and Conjunction Analyses , NeuroImage 10 (4) , 385 396 (1999). 26 K. J. Friston, P. Fletcher, O. Josephs, A. Holmes , M. D. Rugg , and R. Turner , Event Related fMRI: Characterizing Differential Responses , NeuroImage 7 (1) , 3 0 40 (1998). 27 J. M. Ollinger, and J. A. Fessler, "Posi tron Emission Tomography , " IEEE Signal Processing Magazine 14 (1): 43 55 (1997) .
104 28 D. L. Bailey, B. F. Hutton, and P. J. Walker , Improved SPECT using simultaneous emission and transmission tomography , Journal of nuclear medicine: official publication, Socie ty of Nuclear Medicine 28 ( 5 ), 844 851 , 1987. 29 L. Liao , M. Li , H. Lai , Y. Shih , Y. Lo, S. Tsang , P. Chao , C. Lin , F. Jaw , and Y. Chen , Imaging brain hemodynamic changes during rat forepaw electrical stimulation using functional photoacoustic microscopy , Ne uroimage 52 (2) , 562 70 (2010). 30 X . Yang , X . Cai , K . Maslov , L . V. Wang , and Q . Lu o , High resolution photoacoustic microscope for rat brain imaging in vivo , Chin. Opt. Lett. 8 (6) , 609 611 ( 2010 ). 31 G . Ku, X . Wang, X . Xie, G . Stoica, and L . V. Wang , Imaging of tumor angiogenesis in rat brains in vivo by photoacoustic tomography , Applied Optics 44 (5) , 770 775 (2005) . 32 Y . Wang, X . Xie, X . Wang, G. Ku, K. L. Gill , D. P . O'Neal , G. Stoica , and L . V. Wang , Photoacoustic t omography of a n anoshell c ontrast a gent in the in v ivo r at b rain , Nano Letters 4 (9), 1689 1692 ( 2004 ). 33 E . M. Strohm, E . Berndl, and M . C. Kolios , High frequency label free photoacoustic microscopy of single cells , Photoacoustics 1 ( 3 ), 49 53 (2013). 34 G. Ku, K. Maslov, L. Li, and L.V. Wang , acoustic microscopy with 2 micron transverse resolution , J . Biomed . Opt . 15 , 021302 (2010) . 35 S. Yang, F. Ye, and D. Xing , free gold nanorods imaging with photoacoustic microscopy , Opt . Express 20 , 10370 10375 (2012) . 36 J . Jo , and X . Yan g , imaging to observe regional brain activation induced by cocaine hydrochloride , J. Biomed. Opt. 16 (9) 090506 (2011). 37 S . Hu, K . Maslov, and L . V. Wang , Noninvasive label free imaging of microhemodynamics by optical resolution p hotoacoustic microscopy , Opt . Express 17 ( 9 ) , 7688 7693 (2009) . 38 J . Xia, C . Huang, K . Maslov, M . A. Anastasio, and L . V. Wang , Enhancement of photoacoustic tomography by ultrasonic computed tomography based on optical excitation of elements of a full ring transducer array , Optics Letters 38 (16) , 3140 3143 (2013) . 39 J. Gamelin, A. Aguirre, A. Maurudis, F. Huang, D. Castillo, L. V. Wang, and Q. Zhu, "Curved array photoacoustic tomographic system for small animal imaging," J. Biomed Opt 13 (2), 024007 (2008). 40 X. Yang, A. Maurudis, J. Gamelin, A. Aguirre, Q. Zhu, and L. V. Wang , Photoacoustic tomography of small animal brain with a curved array transducer , J . Biomed . Opt. 14 (5) , 054007 ( 2009 ).
105 41 A . Buehler, E . Herzog, D . Razansky, and V . Ntziachristos , Real time imaging of cardiovascular dynamics and circulating gold nanorods with multispectral optoacoustic tomography , Opt . Express 18 (19), 19592 19602 (2010) . 42 A. Buehler , E. Herzog , D. Razansky , and V. Ntziachristos , Video rate optoacoustic tomography of mouse ki dney perfusion , Opt . Letters 35 , 2475 2477 (2010) . 43 R . J. Zemp , Quantitative photoacoustic tomography with multiple optical sources , Appl . Optics 49 (18) , 3566 3572 (2010) . 44 P . Shao, B . Cox, and R . J. Zemp , Estimating optical absorption, scattering, and G rueneisen distributions with multiple illumination photoacoustic tomography , Appl . Optics 50 (19) , 3145 3154 (201 1 ) . 45 H . Jiang, Z . Yuan, and X . Gu , Spatially varying optical and acoustic property reconstruction using finite element based photoacoustic tomo graphy , JOSA A, 23 (4) , 878 888 (2006) . 46 E. Zhang, J. Laufer, and P. Beard , Backward mode multiwavelength photoacoustic scanner using a planar Fabry Perot polymer film ultrasound sensor for high resolution three dimensional imaging of biological tissues , Appl . Opt . 47 , 561 577 (2008). 47 S. Hu, K. Maslov, K. K. Shung , array based real time photoacoustic microscopy of human pulsatile dynamics in vivo , J . Biomed . Opt. 15 (2) , 021303 ( 2010 ). 48 J. Gateau, M. A. Caballero, A. Dima A, and V. Ntziachristos , dimensional optoacoustic tomography using a conventional ultrasound linear detector array: whole body tomographic system for small animals , Med . Phys. 40 (1) , 013302 ( 2013 ). 49 D. A. Nedosekin, M. Sarimollaoglu, E. V. Shashkov, E. I. Galanzha, and V. P. Zharov , Ultra fast photoacoustic flow cytometry with a 0.5 MHz pulse repetition rate nanosecond laser , Opt . Express 18 (8) , 8605 86 20 ( 2010 ). 50 S . Hu, K . Maslov, and L . V. Wang , Second generation optical resolution photoacoustic microsc opy with improved sensitivity and speed , Optics Letters, 36 (7) , 1134 1136 (2011) . 51 P. Ephrat, L. Keenliside, A. Seabrook, F. S. Prato, J. J. Carso n, Three dimensional photoacoustic imaging by sparse array detection and iterative image reconstruction , J . Biomed . Opt. 1 3 ( 5 ) , 054052 ( 20 08). 52 L . Xi, C . Duan, H . Xie, and H . Jiang , Miniature probe combining optical resolution photoacoustic microscopy and optical coherence tomography for in vivo microcirculation study , Appl . Optics 52 (9) 1928 1931 (2013) . 53 L . Ze ng, G . Liu, D . Yang , and X . Ji , Portable optical resolution photoacoustic microscopy with a pulsed laser diode excitation , Appl. Phys. Lett. 102, 053704 (2013) .
106 54 D . Yeager , Y . Chen , Si . Litovsky , and S . Emelianov , 637 i ntravascular u ltrasound and p ho toacoustic i maging of p lasmonic n anoparticles for c ombined d iagnosis and i mage g uided t herapy of a therosclerotic p laques , J . Am . Coll . Cardiol. 62 (18_S1) , B194 B194 ( 2013 ). 55 E . Z. Zhang, B . Povazay, J . Laufer, A . Alex, B . Hofer, B . Pedley, C . Glittenberg, B . Treeby, B . Cox, P . Beard, and W . Drexler , Multimodal photoacoustic and optical coherence tomography scanner using an all optical detection scheme for 3D morphological skin imaging , Bio. Opt. Express 2 ( 8 ) , 2202 2215 (2011) . 56 E . Z . Zhang, J . G . Laufer, R . B . Pedley , and P . C . Beard , In vivo high resolution 3D photoacoustic imaging of superficial vascular anatomy , Phys. Med. Biol. 54 , 1035 1046 (2010). 57 M . Li , J . C . Wang , J . A. Schwartz , K . L. Gill Sharp , G . Stoica , and L. V. Wang , vivo photoacoustic microscopy of nanoshell extravasation from solid tumor vasculature , J. Biomed. Opt. 1 4 ( 1 ) 0 10 50 7 (20 09 ). 58 J . Jose, S . Manohar, R . Kolkman, W. Steenbergen, T . Leeuwen , vasculature using Twente photoacoustic systems , Journal of Biophotoni cs 2 (12) , 701 717 (2009). 59 C . Kim, T . N. Erpelding, L . Jankovic, M . D. Pashley, and L. V. Wan g, in vivo photoacoustic imaging using a clinical ultrasound array system , Bio . Opt . Express 1 ( 1 ) , 27 8 284 ( 201 0). 60 R . A . Kruger, R . B. Lam, D . R . Reinecke, S . P. Del Rio, and R . P. Doyle , Photoacoustic angiography of the breast , Med. Phys. 37 ( 1 1), 6096 6100 (2010). 61 S. Yang, G. Yin, and D. Xing, m onitoring of i ndocyanine g reen for t umor d etection u sing p hotoacoustic i maging , C hi n . phys . lett . 27 (9) , 094302 (2010) . 62 P. C. Beard, F. PÃ©rennÃ¨s, E. Draguioti, and T. N. Mills , photothermal probe , Opt. Lett. 23 (15), 1235 1237 (1998) . 63 J . Yang, K . Maslov, H . Yang, Q . Zhou, K. K . Shung, and L . V. Wang , Photoa coustic endoscopy, Opt . Lett . 34 (10), 1591 1593 (2009) . 64 W. A. Hauser, J. F. Annegers, and W. A. c ontributions of population based studies from Rochester , Mayo Clinic Proceedings 71 (6), 576 586 (1996). 65 K. A. J ellinger, Reading EEGs: A Practical Approach , European Journal of Neurology, 17 34 (2010) . 66 M . M. Frucht, M . Quigg, C . Schwaner, and N . B. Fountain , Distribution of Seizure Precipitants Among Epilepsy Syndromes , Epilepsia, 41 (12) , 1534 1539 ( 2000 ).
107 67 G. L . Birbeck, R. D. Hays, X. Cui, and B. G. lifeimprovements in people with epilepsy , 43 (5) , 535 538 (2002). 68 A. T. Berg, J. T. Langfitt , S. S. Spencer, and B. G. afte r epilepsy surgery: a survey of U.S. epilepsy center neurologists , Behav. 10 (2) , 219 222 (2007). 69 J. S. Duncan, J. W. Sander, S. M. Sisodiya, and M. C. , Lancet 367 (9516) , 1087 1100 (2006). 70 J. T he goal of epilepsy therapy: no seizures, no side effects, as soon as possible , CNS Spectr. 9 (1) , 95 97 ( 2004 ) . 71 L. E. Jeha, I. Najm, W. Bingaman, D. Dinner, P. Widdess Walsh, and H. outcome and prognostic factors of frontal lobe epilepsy surgery , Brain 130 , 574 584 ( 2007 ) . 72 D . A. Marks, J. Kim, D. D. Spencer, and S . S. Spencer, Seizure localization and pathology following head injury in patients with uncontrolled epilepsy , Neurology 45 (11), 2051 2057 (1995). 73 A. M. Murro, Y. D. Park, D. W. King, B . B. Gallagher, J. R. Smith, and W. Littleton , Use of scalp sphenoidal EEG for seizure localization in temporal lobe epilepsy , J . Clin . N europhysiol . 11 (2) , 216 21 9 ( 1994 ). 74 C. C. Rowe, S. F. Berkovic, S. T. Sia, M. Austin, W. J. McKay, E. M. Kalnins, an d P. F. Bladin, Localization of epileptic foci with postictal single photon emission computed tomography , Ann . Neurol. 26 , 660 668 (1989) . 75 E . H . Bertram, D . X . Zhang, P . Mangan, N . Fountain, and D . Rempe , Functional anatomy of limbic epilepsy: a proposa l for central synchronization of a d iffusely hyperexcitable network , Epilepsy Res. 32 , 194 205 ( 1998 ). 76 W . Liao, Z . Zhang, D . Mantini, Q . Xu, Z . Wang, G . Chen, Q . Jiao, Y . Zang, and G . Lu , Relationship Between Large Scale Functional and Structural Covaria nce Networks in Idiopathic Generalized Epilepsy , 3 (3) , 240 254 ( 2013 ). 77 E . Guevara, P . Pouliot, D . K . Nguyen, and F . Lesage , Optical imaging of acute epileptic networks in mice , J. Biomed. Opt. 18 (7), 076021 ( 2013 ). 78 K. Lehnertz, S. Bia lonski, M. T. Horstmann, D. Krug, A. Rothkegel, M. Staniek, and T. Wagner , Synchronization phenomena in human epileptic brain networks , J. Neurosci. Methods 183 (1) , 42 48 (2009). 79 M. Thom, G. W. Mathem, J. H. Cross, and E . H. Bertram , be epilepsy: 68 , 424 434 ( 2010 ). 80 E. H. Behav. 14 , 32 37 (2009) .
108 81 E. H. Bertram, D. X. Zhang, P. Mangan, N. Fountain, and D. Rem of limbic epilepsy: a proposal for central synchronization of a diffusely hyperexcitable network , Epilepsy Res. 32 , 194 205 ( 1998 ) . 82 N. B. Fountain, J. Bear, E. H. Bertram, and E. W. cortex are epileptiform in an electrogenic rat model of chronic temporal lobe epilepsy , Neurophysiol. 80 , 230 240 ( 1998 ) . 83 A. J. Cadotte, T. B. DeMarse, T. H . Mareci, M. B. Parekh, S. S. Talathi, D. U. Hwang, W . L. Ditto, M. Ding, and P. R. lity relationships between local field potentials in an animal model of temporal lobe epilepsy , 189 , 121 129 (2010) . 84 E . V. Diessen , W . M. Otte , K . Braun, C . J. Stam, and F. E. Jansen , Improved d iagnosis in c hildren with p artial e pile psy u sing a m ultivariable p rediction m odel b ased on EEG n etwork c haracteristics , Plos ONE ONE 8 (4 ), e59764 (2013). 85 S . J . M . Smith , EEG in the diagnosis, classification, and management of patients with epilepsy , J . Neurol . Neurosurg . Psychiatry 76 , ii2 i i7 ( 2005 ). 86 S . Shinnar, H . Kang, A . T. Berg, E . S. Goldensohn, W. A . Hauser, and S . L. MoshÃ© , Abnormalities in Children with a First Unprovoked Seizure , Epilepsia 3 5 (3) , 471 476 (1994). 87 A . M. Murro, D . W. King, J . R. Smith, B . B. Gallagher, H . F. Fla nigin, and K . Meador , Computerized seizure detection of complex partial seizures , Electroencephalography and Clinical Neurophysiology 79 (4), 330 333 (1991). 88 T . H. Schwartz , Neurovascular coupling and epilepsy: hemodynamic markers for localizi ng and pred icting seizure onset , Ep ilepsy Currents 7 (4) , 91 94 ( 2007 ). 89 H . Girouard , and C . Iadecola , Neurovascular coupling in the normal brain and in hypertension, stroke, and Alzheimer disease , J . Appl . Physiol . 100 (1), 328 335 (2006). 90 N. Voges, S. Blanchard, F . Wendling, O. David, H. Benali, T. Papadopoulo, M. Clerc, and C. BÃ©nar , , 25 (2) , 136 1 56 (2012). 91 M . Suh, H . Ma, M . Zhao, S . Sharif, and T . H. Schwartz , Neurovascular coupling and oximetry during epileptic events , Molecular Neurobiology 33 (3), 181 197 (2006). 92 W. V . Paesschen, P. Dupont, S. Sunaert, K. Goffin, and K. V . Laere , The use of SPECT and PET in routin e clinical practice in epilepsy , 20 (2), 194 202 ( 2007 ). 93 C. L. FougÃ¨re, A. Rominger, S. FÃ¶rster, J. Geisler, P. Bartenstein , PET and SPECT in epilepsy: a critical review , Epilepsy Behav. 15 (1) , 50 5 5 ( 2009 ).
109 94 P. A. Bandettini , , 61 , A 3 A11 ( 2012 ). 95 J. A. Detre , fMRI: applications in epilepsy , Epilepsia 45 (4), 26 31 (2004). 96 S. Ogawa, T. M. Lee, A. S. Nayak, and P. sensitive contrast in magnetic resonance image of rodent brain at high magnetic fields , Med. 14 , 68 78 ( 1990 ) . 97 T . H. Schwartz, and T . Bonhoeffer , In vivo optical mapping of epileptic foci and surround inhibition in ferret cerebral cortex , Nature Medicine 7 (9) , 1063 1067 ( 2001 ). 98 M . M. Haglund , Optical imaging of visual cortex epilepti c foci and propagation pathways , Epilepsia 53 (s1) , 87 97 (2012) . 99 D. K. Sokol , O. N. Markand , E. C. Daly , T. G. Luers sen , and M. D. Malkoff , Near infrared spectroscopy (NIRS) distinguishes seizure types , Seizure 9 , 323 327 ( 2000 ) . 100 M . M. Eberle, C . L. Reynolds, J . I. Szu, Y . Wang, A . M. Hansen, M . S. Hsu, M. S . Islam, D . K. Binder, and B. H . Park , seizure activity with optical coherence tomography , Bio . Opt . Express 3 ( 11 ) , 2700 2706 ( 2012 ). 101 H. Jian g, K. D. Paulsen, U. L. Osterberg, B. W. Pogue, and M. S. image reconstruction using frequency domain data: simulations and experiments , J. Opt. Soc. Am. A 13 , 253 266 ( 1996 ) . 102 J . Yao, and L . V. Wang , Photoacoustic microscopy , Laser P hoton . Rev . 7 (5), 758 778 (2013). 103 D. Sipkins, X. Wei, J. Wu, J . M. Runnels, D. Cote, T. K. Means, A. D. Luster, D. T. Scadden, and C. P. Lin, microdomains for tumour engraftment , Nature 435 , 969 973 (2005) . 104 X . Chen, C . Sheng, and X . Zheng , neurons with diaminoanthraquinone and confocal microscopy , Cell Biology International 25 (7), 593 598 ( 2001 ). 105 W. Denk, J. H. Strickler, and W. W. Webb, Two phot on laser scanning fluorescence microscopy , Science 248 , 73 76 ( 1990 ) . 106 E. Mace, G. Montaldo, I. Cohen, M. Baulac, M. Fink, and M. Tanter, Functional ultrasound imaging of the brain , Nat. Meth ods 8 , 662 685 ( 2011 ) . 107 J . Xiao, L . Yao, Y . Sun, E . S. Sobel, J . He, and H . Jiang , Quantitative two dimensional photoacoustic tomography of osteoarthritis in the finger joints , Opt . Express 18 (14), 14359 14365 (201 0 ) .
110 108 J. D. Owens, M. Houston, D. Luebke, S. Green, J. E. Stone, and J. C. GPU computing, Proc eedings of the IEEE 96 (5), 879 899 (2008). 109 C. G. A. Hoelen, F. F. M. de Mul, R. Pongers, and A. Dekker , dimensional photoacoustic imaging of blood vessels in tissue , Optics Letters 23 ( 8 ), 648 650 ( 1998 ). 110 C. G. A. Hoelen , reconstruction for photoacoustic scanning of 39 (31), 5872 5883 (2000). 111 A . Brovelli, M . Ding, A . Ledberg, Y . Chen, R . Nakamura, and S . L. Bressler , Beta oscillations in a large scale sensorimotor cortical network: Directiona l influences revealed by Granger causality , PNAS, 101 (26) , 9849 9854 ( 2004 ). 112 X . Wang, Y . Chen, S . L. Bressler, and M . Ding , Granger causality between multiple interdependent neurobiological time series: blockwise versus pairwise methods , International J ournal of Neural Systems, 17 (2) , 71 7 8 ( 2007 ). 113 M . Ding, S . L. Bressler, W . Yang, and H . Liang , window spectral analysis of cortical event related potentials by adaptive multivariate autoregressive modeling: data preprocessing, model validation, and variability assessment , 83 , 35 45 ( 2000 ). 114 L . Xiang, L . Ji, T . Zhang, B . Wang, J . Yang, Q . Zhang, M . S. Jiang, J. Zhou, P . R. Carney, and H . Jiang , Noninvasive real time tomographic imaging of epileptic foci and networks , NeuroImage 66 , 240 248 (2013) . 115 D . Yang, D . Xing , H . Gu, Y . Tan, and L . Zeng , controlled photoacoustic imaging based on limited field filtered back projection algorithm , Appl. Phys. Lett. 87 , 194101 ( 2005 ). 116 M . Xu , and L . V. Wang , on of spatial resolution related to bandwidth and detector aperture size in thermoacoustic or photoacoustic reconstruction , P hysical Review E 67 , 056605 (2003). 117 L. Yao , detector and limited angle 2 (9), 2649 2654 (2011). 118 13. L. Yao , based photoacoustic tomography using 50 (25), 5031 5041 (2011). 119 Z. Yuan, and H. Jiang, Three dimensional finit e element based photoacoustic tomography: Reconstruction algorithm and simulations , Med. Phys. 34 (2) , 538 546 (2007) . 120 J . D. Owens , D . Luebke, N . Govindaraju, M . Harris, J . KrÃ¼ger, A . E. Lefohn, and T . J. Purcell , A s urvey of g eneral p urpose c omputation on g raphics h ardware , Computer Graphics Forum 26 (1) , 80 1 13 ( 2007 ). 121 M. J. Harvey, G. Giupponi , and G. Fabritiis , b iomolecular d ynamics in the m icrosecond t ime s cale , J. Chem. Theory Comput. 5 (6), 1632 1639 ( 2009 ).
111 122 G . Pratx , and L . Xi ng , GPU computing in medical physics: A review , Med. Phys. 38 , 2685 2693 (2011) . 123 P. L. De Riu , V. P etruzzi , C. T esta , M. M ulas , F. M elis , M. A. C aria , and O. M ameli, Propofol anticonvulsant activity in experimental epilepstic status , 9(2) , 177 181 (1992). 124 J. A. Viator, J. Komadina , L. O. Svaasand , G. Aguilar , B. Choi , and J. S . Nelson , A comparative study of photoacoustic and reflectance methods for determinati on of epidermal melanin content , J . Invest . Dermatol. 122 (6) , 1432 143 9 ( 2 004 ). 125 M. Heijblom, D. Piras, W. Xia, J. C. G. van Hespen, J. M. Klaase, F. M. van den Engh, T. G. van Leeuwen, W. Steenbergen, and S. Manohar , Visualizing breast cancer using the Twente photoacoustic mammoscope: What do we learn from twelve new patient me asurements? Optics Express 20 (11), 11582 11597 (2012) . 126 X . Liang , Z . Deng, L . Jing, X . Li, Z . Dai, C . Li , and M . Huang , Prussian blue nanoparticles operate as a contrast agent for enhanced photoacoustic imaging , Chem. Commun. 49 , 1 1029 11031 ( 2013 ). 127 X . X ia, M . Yang, L. K . Oetjen, Y . Zhang, Q . Li, J . Chen , and Y . Xia , An enzyme sensitive probe for photoacoustic imaging and fluorescence detection of protease activity , Nanoscale 3 , 950 953 (2011). 128 L. Tyvaert, P. LeVan, F. Dubeau, and J. e dynamic imaging of seizures in epileptic patients , . Brain Mapp. 30 , 3993 4011 (2009) . 129 H . Ma, M . Zhao, and T . H. Schwartz , during ictal onset, propagation, and termination revealed by simultaneous in vi vo optical imaging of neural activity and local blood volume , 23 (4) , 885 899 (2013) . 130 T. Reese, B. Bjelke, R. Porszasz, D. Baumann, D. Bochelen, A. Sauter, and M. Rudin , tic resonance imaging. Time resolved assessment of bicuculline induced changes in local cerebral blood volume using an intravascular contrast agent , 13 , 43 49 (2000) . 131 T. Mueggler, D. Baumann, M. Rausch, and Bicuculline induced rain activation in mice detected by functional magnetic resonance imaging , 46 , 292 298 (2001) . 132 B. Nie , K. Chen , S. Zhao , J. Liu , X. Gu , Q. Yao , J. Hui , Z. Zhang , G. Teng , C. Zhao , and B. Shan , A rat brain MRI template with digital s tereotaxic atlas of fine anatomical delineations in paxinos space and its automated application in voxel wise analysis , Hum Brain Mapp. 34 (6) , 1306 13 18 ( 2013 ). 133 N . Camp, R . Hooge, M . Verhoye, R . R. Peeters, P . Deyn, and A . Linden , Simultaneous electroencephalogra phic recording and functional magnetic resonance imaging during pentylenet etrazol induced seizures in rat , 19 , 627 636 ( 2003 ).
112 134 U . Schridde, M . Khubchandani, J . E. Motelow, B . G. Sanganahalli, F . Hyder, and H . Blumenfeld , ge increases in neuronal activity , 18 , 1814 1827 ( 2008 ). 135 B. P. Keogh, D. Cordes, L. Stanberry, B. D. Figlera, C. A. Robbins, B. L. Tempel, C. G. Green, A. Emmi, K. M. Maravilla, and P. A. Schwartzkroin. BOLD fMRI of PTZ induced seizures i n rats , Epilepsy Research 66 , 75 90 ( 2005 ). 136 D . Posada , and T . R. Buckl ey , Model s election and m odel a veraging in p hylogenetics: a dvantages of Akaike Information Criterion and b ayesian a pproaches o ver l ikelihood r atio t ests , Syst . Biol . 53 (5) , 793 808 (2 004) . 137 M . Seeck, F . Lazeyras, C. M . Michel, O . Blanke, C. A . Gericke, J . Ives, J . Delavelle, X . Golay, C. A . Haenggeli, N . Tribolet , and T . Landis , invasive epileptic focus localization using EEG triggered functional MRI and electromagnetic tomography , Electroencephalogr . Clin. Neurophysiol. 106 (6) , 508 512 ( 1998 ). 138 F . Pittau, F . Dubeau, and J . Gotman , epileptic focus , 78 (19) , 1479 1487 (2012). 139 P . Le v an, L . Tyvaert, F . Moeller, and J . Gotman , dynamic ictal BOLD responses in EEG fMRI data from focal epilepsy patients , NeuroImage 49 (1) , 366 378 (2010) . 140 G . Carmignoto, and M . GÃ³mez Gonzalo , neurovascular coupling , 63 , 138 148 (2010) . 141 M. Zonta, M. C. Angulo, S. Gobbo, B. Rosengarten, K. A. Hossmann, and T. Pozzan , to astrocyte signaling is central to the dynamic control of brain microcirculation , Nat . Neurosci . 6 (1) , 43 50 (2003) . 142 B. S. Meldrum , induced cell death in epilepsy: historical and contemporary perspectives , . Brain Res . 135 , 3 11 (2002) . 143 B. K. Siesjo , and T. Wieloch , pathology , . Neur ol . 44 , 813 847 (1986) . 144 S . Bahar, M . Suh, M . Zhao, and T . H. Schwartz , , 17 (5) , 499 503 (2006) . 145 H. Nersesyan, F. Hydeer, D. L. Rothman, and H. Blumenfeld, MRI and EEG recording during Spike Wave seizures and generalized tonic clonic seizures in WAG/Rij rats , Blood Flow Metab. 24 , 589 599 ( 2004 ) . 146 M . Zhao, H . Ma, M . Suh, and T . H. Schwartz , Spatiotemporal dynamics of perfusion and oximetry during i ctal discharges in the rat neocortex , The Journal of Neuroscience 29 (9) , 2814 2823 (2009) .
113 147 T . H. Schwartz, S . Hong, A . P. Bagshaw, P . Chauveld, and C . BÃ©nar , cerebral haemodynamics: Review of findings and insights from intracerebral E EG , Epilepsy Research 97 , 252 266 (2011) . 148 M. Zhao, M. Suh, H. Ma, C. Perry, A. Geneslaw, and T. H. Schwartz , perfusion and decreases in hemoglobin oxygenation precede seizure onset in spontaneous human epilepsy , 48 , 2059 206 7 (2007) . 149 T. J. O'Brien, E. L. So, B. P. Mullan, G. D. Cascino, M. F. Hauser, B. H. Brinkmann, F. W. Sharbrough, and F. B. Meyer , ictal SPECT is predictive of extratemporal epilepsy surgery outcome , 55 , 1668 1677 (2000) . 150 P. Fed erico, D. F. Abbott, R. S. Briellmann, A. S. Harvey, and G. D. Jackson , MRI of the pre ictal state , 128 , 1811 1817 (2005) . 151 P . Williams, A . White, D . Ferraro, S . Clark, K . Staley, and F. E . Dude , radiotelemetry to evaluate ele ctrographic seizures in rats with kainate induced epilepsy , Neuro. Meths. 155 (1) , 39 48 ( 2006 ) . 152 D . Lapray, J . Bergeler, E . Dupont, O . Thews, and H . J. Luhmann , telemetric system for recording EEG activity in freely moving rats , euro. Meths. 168 (1) , 119 1 26 ( 2008 ) . 153 R. J. Racine , Modification of seizure activity by electrical stimulation: II. Motor seizure , Electroencephalogr . Clin. Neurophysiol. 32 (3) , 281 294 ( 1972 ) .
114 BIOGRAPHICAL SKETCH Bo Wang received his Bachelor of Science in the Department of Physics in Tsinghua University in 2006. He worked in Dr. Xue s lab as a recommended postgraduate candidate in the last year of his undergraduate study for the graduation project , and began to get access to the field of biomedical imagi ng. He continue d work ing in Dr. Xue s lab since 2006 focused on OCT imaging, and received his Master of Science in Physics from Tsinghua University in 2009. At the same year, he was admitted to Dr. Jiang s lab in the Biomedical Department in University of Florida for his PhD education , and received his Ph.D . degree in the summer of 2014. His current research mainly focuse d on PAT system development, and the application of PAT in the study of epilepsy.
Photoacoustic tomography system for noninvasive real -time three-dimensional imaging of epilepsy Bo Wang ,1 Liangzhong Xiang ,1 Max S. Jiang,1 Jianjun Yang ,1 Qizhi Zhang ,1 Paul R. Carney ,1,2,3,4,5 and Huabei Jiang1,* 1J. Crayton Pruitt Family Department of Biomedical Engineering, University of Florida, Gainesville, FL 32611, USA 2Department of Pediatrics, University of Florida, Gainesville, FL 32611, USA 3Neurology and Neuroscience, University of Florida, Gainesville, FL 32611, USA 4Wilder Center of Excellence for Epilepsy Research, University of Florida, Gainesville, FL 32611, USA 5McKnight Brain Institute, University of Florida, Gainesville, FL 32611, USA *email@example.com Abstract: A real time three dimensional (3D) photoacoustic imaging system was devel oped for epilepsy imaging in small animals. The system is based on a spherical array containing 192 transducers with a 5 MHz central frequency. The signals from the 192 transducers are amplified by 16 homemade preamplifier boards with 26 dB and multiplexed into a 64 channel data acquisition system. It can record a complete set of 3D data at a frame rate of 3.3 f/s, and the spatial resolution is about 0.2 mm. Phantom experiments were conducted to demonstrate the high imaging quality and real time imaging abi lity of the system. Finally, we tested the system on an acute epilepsy rat model, and the induced seizure focus was successfully detected using this system. 2012 Optical Society of America OCIS codes: (110.5120) Photoacoustic imaging; (170.6920) Time res olved imaging. References and links 1. R. A. Kruger , R. B. Lam , D. R. Reinecke , S. P. Del Rio, and R. P. Doyle , â€œPhotoacoustic angiography of the breast,â€ Med. Phys. 37 ( 11 ), 6096â€“ 6100 ( 2010 ). 2. A. A. Karabutov, E. Savateeva , and A. Oraevsky , â€œImaging of l ayered structures in biological tissues with opto acoustic front surface transducer,â€ Proc. SPIE 3601 , 284 â€“ 295 ( 1999 ). 3. J. Xiao , L. Yao , Y. Sun , E. S. Sobel , J. He , and H. Jiang , â€œ Quantitative two dimensional photoacoustic tomography of osteoarthritis in the finger joints ,â€ Opt. Express 18( 14), 14359 â€“ 14365 ( 2010 ). 4. X. Wang , Y. Pang , G. Ku , G. Stoica , and L. V. Wang , â€œ Three dimensional laser induced photoacoustic tomography of mouse brain with the skin and skull intact ,â€ Opt. Lett. 28 ( 19 ), 1739â€“ 1741 ( 200 3 ). 5. Q. Zhang , Z. Liu , P. R. Carney , Z. Yuan , H. Chen , S. N. Roper , and H. Jiang , â€œ Non invasive imaging of epileptic seizures in vivo using photoacoustic tomography ,â€ Phys. Med. Biol. 53 (7 ), 1921 â€“ 1931 ( 2008 ). 6. J. Gamelin , A. Maurudis , A. Aguirre , F. Hu ang , P. Guo , L. V. Wang , and Q. Zhu , â€œ A real time photoacoustic tomography system for small animals ,â€ Opt. Express 17 ( 13 ), 10489 â€“ 10498 ( 2009 ). 7. D. W. Yang , D. Xing , S. H. Yang , and L. Z. Xiang , â€œFast full view photoacoustic imaging by combined scanning w ith a linear transducer array ,â€ Opt. Express 15( 23), 15566â€“ 15575 ( 2007 ). 8. S. Manohar , A. Kharine , J. C. G. van Hespen , W. Steenbergen , and T. G. van Leeuwen , â€œ Photoacoustic mammography laboratory prototype: imaging of breast tissue phantoms ,â€ J. Biomed. Opt. 9 (6 ), 1172â€“ 1181 ( 2004 ). 9. T. N. Erpelding , Y. Wang , L. Jankovic , Z. Guo , J. Robert, G. David, C. Kim , and L. V. Wang , â€œ Three dimensional photoacoustic imaging with a clinical two dimensional matrix ultrasound transducer ,â€ Proc. SPIE 7899 , 78990A , 789 90A -6 ( 2011 ). 10 . M. B. R ou meliotis , I. Kosik , and J. J. L. Carson , â€œ 3D photoacoustic imaging using staring, sparse array with 60 transducers ,â€ Proc. SPIE 8223, 82233F , 82233F -6 ( 2012 ). 11 . C. G. A. Hoelen and F. F. M. de Mul , â€œ Image reconstruction for photoacoustic scanning of tissue structures ,â€ Appl. Opt. 39( 31), 5872 â€“ 5883 ( 2000 ). 12 . L. Yao and H. Jiang , â€œPhotoacoustic image reconstruction from fewdetector and limited angle data ,â€ Biomed. Opt. Express 2 (9 ), 2649â€“ 2654 ( 2011 ). 13 . L. Yao and H. Jiang , â€œEnhancing finite element based photoacoustic tomography using total variation minimization ,â€ Appl. Opt. 50 ( 25 ), 5031â€“ 5041 ( 2011 ).
1. Introduction Photoacoustic tomography (PAT) is a hybrid method that is capable of imaging optical absorption of tissue through the detection of ultrasound waves generated by a short laser pulse due to transient thermoelastic expansion. It has an imaging resolution that is superior to pure optical imaging at centimeter scale depths. To date PAT has been applied to the detection of breast cancer, skin cancer and osteoarthritis in humans [ 1â€“3 ], and functional brain imaging in small animals [ 4,5 ]. For small animal brain imaging, most prior PAT studies were mostly based on a single transducer scanning system or a circular or linear array of transducers for only 2D imaging [5â€“7 ], and the scanning in z direction is needed for 3D imaging purpose, which leads to a non optimal elevational resolution in the z direction. 2D planar array of transducers has recently been employed for 3D PAT imaging [ 8,9 ]. However, due to the limited aperture of 2D planar array transducers, features with high aspect ratio or with orientations oblique to the transducer surface suffer from distortion, and the azimuthal resolution is reduced. Thus, 2D planar array of transducers is not suitable for small animal brain imaging. Compared to a 2D planar array, a spherical array can offer more complete angular views of the object, providing both high resolution and accurate feature definition regardless of shape or location of the object. The use of sparse spherical arrays for 3D PAT imaging has been recently reported [ 1,10], but due to the insufficient number of transducers used, these arrays were not designed for small animal brain imaging. The goal of this work is to present a sparse spherical array based PAT system that is specifically designed for real time 3D imaging of small animal brains. The work is a natural extension/improvement of our previous work which reported for the first time 2D PAT imaging of epileptic f ocus in small animals [ 5 ]. The current PAT system is based on a 2D spherical array of 196 transducers coupled with parallel data acquisition, offering a temporal resolution of 0.33 s for data acquisition. We demonstrate this system using static/dynamic pha ntom and in vivo animal experiments. To the best of our knowledge, this is the first work reporting 3D PAT imaging of epileptic focus in small animals. 2. System descriptions Figure 1 depicts the block diagram of our real time 3D PAT system. A Ti:sapphire laser optically pumped with a Q switched Nd:YAG laser sent 8 â€“ 12 ns pulses at 10 Hz with a wavelength tunable from 690 to 1015 nm. The beam was delivered with an optical fiber Fig. 1. Block diagram of our real time 3D PAT system. The inset is a photograp h of the close up view of the chamber holding the rat head.
through an opening on the top of the transducer array and produced an approximately uniform illumination in a 2cm diameter area onto the sample. The transducer array consisted of 192 transducers placed along a custom fabricated white ABS spherical interface containing 610 through holes with counter bores, as shown in Fig. 2(a) . More holes were drilled so that the selection of the transducer positions on the ball can be flexible. The interface has an outer diameter of 160 mm and an inner diameter of 140 mm, and the diameter of the holes in the ball is 5.7 mm, which fitted well with the transducer (5.5 mm outer diameter). Each transducer (Custom designed from Blatek, Inc.) has a central frequency of 5 MHz with a reception bandwidth of greater than 80%. The active area of the transducer is 3 mm in diameter and the angular acceptance is about 15 degree. The transducers were glued onto the interface with epoxy which can be removed to allow the position ch ange of the transducers. There were 16 preamplifier boards separately sealed in 4 metal boxes, and each board had 12 input channels, 4 output channels, and 2 digital signal control inputs. Within each board, 12 dedicated operational amplifier modules (AD8099) individually amplify the input signals with a fixed gain of 26 dB, and then the amplified signals were multiplexed into the 4 output channels by 4 multiplexer chips (MAX 4051), which were controlled by a USB IO digital module (USB 1024LS, Measurement C omputing) through the 2 digital inputs. The 64 channel parallel data acquisition system consisted of eight 8 channel PCI cards (PCIAD850, US Ultratek) in an industria l computer. For each channel, 3 000 sampling points were collected at 50 MHz sampling rate in 10 bits, and stored in a 32k on board memory before they were transferred to the host machine. Amplifiers with a programmable gain of 0 to 80 dB along with a 16 MHz low band pass filter were built into the data acquisition system. A Labview program cont rolled the data acquisition, and the acquired data was stored on hard disk for further image processing. Images were reconstructed with a delay and sum algorithm [ 11]. For 3D image display, the reconstructed results were normalized to 0~255 after setting t he negative values to be zero. Then 3D images were rendered with Amira (from Visage Imaging, Inc.) with different thresholds as indicated in the colorbar shown in each image. Fig. 2. The spherical transducer array. (a) Photograph of the transducer array. (b) 3D schematic of the transducer distribution on the interface. The system allows the selection of transducer positions on the spherical interface. The total 610 holes formed 11 evenly spaced layers along the vertical direction of the ball, and the 192 transducer positions were indicated with three different colors in Fig. 2b. Three different colors were used to indicate the 3 to 1 multiplexing from the 192 transducers to the 64channel data acquisition. This system can also be used as a real time 2D sys tem operating at 10 f/s using the 64 transducers arranged in the vertical center layer (blue). For in vivo experiments, the rat head was elevated to the center of the spherical interface through a chamber fixed at the tank bottom, whose top was about 15 mm beneath the interface center, and a transparent plastic wrap was used to cover the chamber top. For phantom experiments, a homemade silicone holder was used to hold the phantom.
3. Phantom experiments Three different types of phantoms were used: One containing a point object for system calibration, one phantom containing three hairs tilted along different orientations for static imaging, and one phantom with ink flowing through a thin tube embedded in a phantom for real time 3D imaging. The point object us ed for calibration was a small spherical graphite particle (0.1 mm in diameter) located at the center of the spherical array and ensured an isotropic acoustic emission profile for all directions. We measured and compensated the delay of time for all the 192 channels in the radical direction, and reconstructed this point object after calibration. The hairs containing phantom was used to demonstrate the high imaging quality of our system. Finally, we imaged an embedded tube filled with flowing ink to show the real time imaging ability of our system. The tube had a 0.3 mm inner diameter and was horizontally placed in a phantom. No averaging of signals was performed for the phantom experiments except for the point object experiment where 10 times averaging was applied. 3.1 . System characterization We calibrated the system by recording the emission profiles from the spherical graphite particle for all the 192 channels, and then measured and compensated the time delay of each channel. We then evaluated the system resolution by reconstructing the image of the point object. Fig. 3. (a) and (c): x y and z x cross section images through the center plan of the point object. (b) and (d): the profile extracted in x and z directions from (a) and (c), respectively. Units are in mm. Figures 3a and 3c present the reconstructed x y and z x cross section images of the point object located at the array center. The quality of these images is determined by both the distribution and the characteristics of the transducers. The profiles of the two reconstructed images were also extracted in x and z directions, as shown in Figs. 3b and 3d, respectively. The full width at half maximum (FWHM) of the profiles was measured to be 0.19 mm (x direction) for Fig. 3b , and 0.27 mm (z direction) f or Fig. 3d , compared to the theoretical value of 0.16 mm for the 5 MHz central frequency transducer with an estimated cut off frequency of 7 MHz. It is noted that the profile in Fig. 3d is noisier than that in Fig. 3b. This along with a larger FWHM from Fi g. 3d was due to the asymmetric distribution of the transducers. For targets located away from the center of the array, the radial resolution will stay nearly the same as that for a centrally located target, while the lateral resolution will be linearly reduced with increased distance away from the array center. In our system, the lateral resolution will be reduced by 0.1 mm when the target is located 5 mm off the array center. 3.2 . Static phantom experiments Figure 4a is the photograph of the phantom contai ning three hairs tilted along different orientations, and Figs. 4b and 4c show the reconstructed 3D images from two different views. The reconstructed volume is 10 x 10 x 10 mm with a 0.1 mm voxel size. The spatial distribution and tails for all the three hairs were clearly revealed. This result indicates that our system is capable of threedimensionally imaging small objects of different spatial distribution and orientation in high quality.
Fig. 4. (a): photograph of the phantom containing three tiled ha irs; (b)(c): reconstructed 3D images of the three hairs in two different views. Scale bar represents 5 mm. 3.3 . Dynamic phantom experiments The reconstructed 3D images of ink flowing through a thin tube are shown in Fig. 5 . The image domain is 15 x 5 x 10 mm with a 0.1 mm voxel size. Figure 5a is the photograph of the phantom containing a tube filled with ink, and Figs. 5bâ€“ 5j are the reconstructed 3D images at different time points. The time interval between two consecutive images was 0.3 s. These 3D images clearly tracked the flow through the tube over the course of 2.4 seconds with high spatial and temporal resolution, and the flowing speed of the ink was measured to be 6 0.9 mm/s from the reconstructed results with a time interval o f 0.3 s. Fig. 5. Reconstructed 3D images of ink flowing through a 0.3 mm tube embedded in a background phantom. (a): photograph of the phantom containing the tube. (b) (j): reconstructed 3D images at different time points. The time interval is 0.3 s. Sca le bar represents 5 mm. 4. Rat epilepsy experiment Epilepsy is a serious brain disorder involving intensive hemodynamic changes, which provides high endogenous contrast for PAT imaging due to the strong absorption of blood at visible and NIR wavelengths. Compared with current existing neuron imaging methods (such as MRI, CT, PET, and SPECT), PAT provides not only high ultrasound resolution and high optical contrast, but also unprecedented advantage of high temporal resolution over these methods, which is cr itical for capturing seizure dynamics. 2D PAT of seizure focus on an acute seizure rat model was demonstrated for the first time by our lab [ 5 ], but the observation of hemodynamic changes during seizure onset was hindered by the long time scanning of a sin gle transducer. Here we test our real time 3D system using the same animal model to show the hemodynamic changes and reveal the 3D structures in the rat brain. Two small rats (~40g) were imaged with intact skull and skin but hairs on the head were removed. The rats were anaesthetized and mounted on the homemade plastic chamber/holder. Focal seizure was induced by microinjection of 10 (BMI) into the neocortex of one rat, while saline solution was injected into the brain of another rat as control. In each experiment, the rat was elevated to the transducer array center and kept alive under the water tan k through the whole experiment. The incident energy of the 730 nm light was maintained at 8mJ/cm2 below the safety standard. Seizure process was recorded for 50 minutes, and the measurement from the control rat was recorded for 3 s. All
Fig. 6. (a) and (c): photograph of the rat with BMI injection and the control rat after scalp removed. (b) a nd (d): 3D PAT images at 6 time points for (a) and (c) respectively. The three main blood vessels are indicated by the white arrows, and the seizure focus is indicated by the circle in (b). The time interval between two successive images is 0.3 s. Scale ba r represents 10 mm. animal procedures were performed in accordance with the approved University of Florida IACUC protocols. Figure 6b presents the 3D images for the rat with BMI injection during the seizure onset at 6 time points, compared with that for the control rat in Fig. 6b . The corresponding photographs of the two rats with scalp removed right after the experiments are shown in Figs. 6a and 6c , respectively. T he reconstructed domain for the images shown in both Figs. 6b and 6d is 20 x 20 x 4.5 mm, with a 0.1 mm pixel size. The three main blood vessels on the rat brain are clearly revealed for both cases, as indicated by the white arrows, and for the rat with BM I injection a seizure focus can be clearly seen right at the BMI injection position (the white circle in Fig. 6b ), where strong absorption is observed during seizure onset. The rapid changes of the absorption both in the main blood vessel and in the seizur e focus were observed from Fig. 6b, while no such changes were noted for the control rat (Fig. 6c ). The seizure focus had a diameter of ~3 mm, which is consistent with the results published before [ 5 ]. This experiment demonstrates that our system can be us ed to investigate the hemodynamic changes in small animal brain both spatially and temporally during seizure onset, although the complex microvasculature cannot be resolved due to the limited number of transducers and the simple backprojection reconstruction method used here. The microvasculature can be revealed if sophisticated reconstruction methods such as the finite element based algorithms coupled with the total variation minimization scheme are used [ 12,13 ]. 5 Conclusions We have presented a real time photoacoustic system for three dimensionally imaging focal cortical seizures in a rat. The system is based on a spherical array containing 192 discrete transducers. With the 64 channel data acquisition system coupled with 3:1 multiplexing, it can achieve a frame rate of 3.3 f/s with a spatial resolution of 0.2 mm. The 3D imaging performance of the system was demonstrated by both static and dynamic phantom experiments. We have also tested our system using an acute epilepsy rat model and obtained 3D images s howing the hemodynamic changes during seizure onset. Acknowledgments This research was supported in part by a grant from the U.S. Department of Defense Congressionally Directed Medical Program , the B.J. and Eve Wilder endowment fund, and the Childrenâ€™s Miracle Network .