Research On Reconstruction Algorithms And System Optimization For Photon Counting Spectral CT

As one of the essential methods in modern medicine,CT is necessary for the diagnosis of many kinds of diseases.With the development of photon counting detectors,photon counting spectral CT has been of more and more research interest in recent years.It provides the ability of material discrimination which is lacked in conventional CT,and it has great meaning in contrast enhanced and soft tissue imaging.Meanwhile,spectral CT has lower noise than dual energy CT,which can avoid the significant increase on patient dose.With the aim at reduction of image noises,the dissertation studied spectral CT on both reconstruction algorithms and detector system optimizations.On reconstruction algorithms,a projection decomposition algorithm based on weighted subsets was proposed specified for the redundancy in the acquired information.We split the energy bins into various combinations and for each combination polynomials were used for material decomposition.Then calibration data was used to calculate weighting factors and weighted sum the decomposition results from subsets,in order to minimize the decomposition noise.Simulation and experimental results demonstrated that the proposed method achieved noise level which was near to the optimal noise.During reconstruction of the decomposed image,based on the structural similarity between material decomposed image and the attenuation images,a weighted non-local TV minimized optimization algorithm was proposed.The algorithm calculated weights as the pixel similarity in intensity images,constructed the weighted non-local TV object function and solved the problem via the ASD-POCS algorithm.We thoroughly studied the influence of algorithm parameters and proposed a corresponding adaptive parameter choice strategy.Simulation and experimental results indicated that the proposed methods had better edge-preservation capability compared to some existing methods.The key point in detector optimization is the optimization of threshold.To achieve the equivalent spectrum required in threshold optimization,we first proposed a physical model for the energy response function of photon counting detectors.A hybrid Monte Carlo model was proposed.GEANT4 was used to simulate the process that X-ray photons enter and interact with the detector;then simplified numerical model was utilized to describe the charge sharing effect and statistical fluctuation.Anti-charge sharing logics were also embedded in the detector model.X-ray fluorescence spectrums were used for solving the key parameters in the model.The energy response function model was further verified with acquired spectrum of the X-ray source.During thresholds optimization,Cramer-Rao lower bound(CRLB)weighted by the probability density function of projections was proposed as the object function.The thresholds were optimized to minimize the weighted CRLB by differential evolution algorithm,and the minimization of reconstruction noises could be achieved in the meantime.Simulations and experiments were carried out to verify the method and its robustness against parameter variation.Based on the proposed method,we further studied the detector's key parameters' influence on reconstruction noises and optimal thresholds.