Image Reconstruction Algorithms For Photoacoustic Imaging

Photoacoustic tomography(PAT),which is also referred to as thermoacoustic tomography,is a novel technique of noninvasive medical imaging.Combining the high contrast of optical imaging and the good resolution of ultrasound imaging,PAT is suitable for tumor detection,vessel imaging,brain structural and functional imaging, and flow speed measurement,etc.In PAT,an electromagnetic pulse(normally laser or microwave) is used to irradiate the biological tissue.The tissue absorbs the energy and then radiates photoacoustic waves(ultrasound).Detected by the ultrasonic transducer,photoacoustic waves can be utilized to calculate the optical absorption distribution of the tissue based on a certain image reconstruction algorithm.The reconstruction algorithm is a key problem of PAT.Until now many reconstruction algorithms have been proposed.However,most of these algorithms require some ideal conditions and are therefore restricted in practicability.These algorithms need to detect photoacoustic signals from a full view,and suppose the acoustic speed in the tissue to be constant.However,the full-view scanning is sometimes impossible in many applications,and the acoustic speed variation can be as large as 10%.This dissertation focuses on these two problems and conducts theory and simulation studies on reconstruction algorithms of PAT.The major work of this dissertation is described as follows.A computer simulation platform is set up as an important analyzing and verifying tool for the PAT algorithm study.In the simulation,the parameter model of the tissue is firstly built.Then the Monte Carlo method is employed to simulate the optical absorption distribution of the tissue,and the finite-difference time-domain(FDTD) method is used to simulate the generated photoacoustic signal.Here the FDTD method is modified for the PAT simulation,and its precision is better than the popular time-of-flight method.The deconvolution reconstruction(DR) algorithm is proposed for both the full-view and the limited-view PAT.The DR algorithm models the imaging process as a linear shift-invariant system,whose input is the optical absorption distribution of the tissue and output is a new function constructed from photoacoustic signals.Thus the reconstructed image can be calculated by the Fourier-based deconvolution.Computer simulations are carried out to validate this algorithm.In the full-view case,the DR algorithm has good precision and strong noise robustness,nearly equivalent to the popular time-domain reconstruction(TDR) algorithm and filtered back-projection (FBP) algorithm.However,the DR algorithm is much faster than TDR and FBP.In the limited-view case,the reconstructed image of the DR algorithm is superior to those of TDR and FBP with clearer boundaries,fewer artifacts and stronger noise robustness.A correlation-based method is proposed to estimate the acoustic flight time in the tissue.Then the traditional TDR algorithm and our DR algorithm are modified, respectively,to compensate for the acoustic speed heterogeneity using the estimated acoustic flight time.These two modified algorithms do not require the prior knowledge of acoustic speed distribution and can quickly reconstruct the image as inverse algorithms.It is shown by simulation results that these two algorithms are quite precise and are not sensitive to random noises when the acoustic speed variation is lower than 10%.The modified TDR algorithm is a bit more precise than the modified DR algorithm while the latter is faster.Since the acoustic speed variation of biological tissues is normally within 10%,these two algorithms are suitable for practical applications and outperform other model-based iterative algorithms which require the prior knowledge of acoustic speed distribution.Three major PAT modes,the forward mode,the backward mode and the sideward mode,are analyzed and compared.First,the basic suitable applications of these three modes are analyzed from perspective of physical configuration.Then the simulation on mouse chest is carried out to compare the performance of these three modes.Three popular algorithms are chosen for the image reconstruction in each of three modes,respectively,and the precision and noise characteristics of reconstructed images are systematically analyzed.It is shown that the forward and backward modes are suitable for the fast imaging of plane human parts,and the optimum imaging region is at an appropriate distance from the scanning line.The sideward mode is suitable for the precise imaging of protuberant human parts,and the optimum imaging region is around the center of the scanning circle.If an image with good precision is required,the sideward mode is preferably considered.If the full-view scanning is infeasible in the sideward mode due to physical limitation,the forward mode, backward mode or limited-view sideward mode can be used for the imaging.