Research On Reconstruction Of Limited-view Endoscopic Photoacoustic Images From Sparse Measurements

Endoscopic photoacoustic tomography(EPAT)is interventional application of photoacoustic tomography(PAT)in the visualization of anatomical features and functional components of biological cavity tissues by combining PAT with endoscopic detection.One of the main technical challenges of EPAT is the incomplete measurements due to the limited detectors or the limited-view scanning of the detector enclosed in the cavity or lumen.In this case,standard image reconstruction algorithms suffer from significantly degraded image quality.The purpose of this thesis is to design and implement algorithms to reconstruct EPAT images with high quality from sparse measurements.The main work includes two aspects.First,a method based on compressed sensing(CS)for sparse-data EPAT image reconstruction is developed.Complete photoacoustic data is recovered from incomplete measurements based on the CS theory.Then,the initial acoustic pressure distribution on a luminal cross-section is reconstructed from the recovered complete data by using time reversal(TR)algorithm.Second,A method based on deep learning for limited-view EPAT image reconstruction is designed.The trained convolutional neural network is utilized to optimize the low quality EPAT images reconstructed from limited view measurements by using the standard TR algorithm.The validity of the two methods was numerically demonstrated with computer-simulated arterial vessel phantoms.The numerical results suggest that the CS method provides a 40% improvement of the structural similarity index over the standard TR reconstruction in reconstructing images from full-view sparse measurements.The deep learning method can effectively restore image distortion caused by limited-view scanning and reduce under-sampling artifacts.