A Low-Dose Propagation-Based X-Ray Phase Computed Tomography Algorithm

X-ray absorption imaging is one of the most important and successful non-invasive high-resolution medical imaging techniques,while X-ray phase contrast imaging enables larger image contrast utilizing the phase shift when X-ray interacts with object.X-ray phase computed tomography which combines phase contrast imaging(PCI)and computed tomography,could reconstruct three-dimensional phase distribution of object,is expected to play an important role in biomedical imaging and pre-clinical imaging such as monitoring of bone growth,early screening of breast cancer and liver fibrosis.Based on the in-line x-ray phase computed tomography,also known as the propagation-based phase computed tomography(PB-PCT),we first studied the optimization of imaging parameters to increase the edge enhancement of phase contrast projections and phase retrieval accuracy of Homogenous Transport of Intensity Equation(TIE-Hom),then a low-dose tomographic algorithm based on total generalized variation(TGV,0<p<2)was developed and evaluated,at last,we proposed a tomographic strategy enabling the great improvement of the reconstructed spatial resolution of low-dose PB-PCT based on the former two studies.Phase retrieval algorithm quantifies the phase information from the measured phase contrast projections which combine the absorption information and phase information nonlinearly,is an important step in PB-PCT.To reduce the X-ray dose and measurement time,analytical phase retrieval algorithm using only one projection is the most common used one.Based on TIE-Hom which is the most classical phase retrieval algorithm,we developed a non-ideal phase retrieval model which relates the phase retrieval accuracy in inverse process and edge enhancement in forward process by imaging parameters.In this thesis,edge enhancement to noise ratio(EE/N)was defined to quantify the effects of microfocus source size,the full width of half height of point spread function of detector,geometrical magnification and system noise on edge enhancement,also,~2 was defined to quantify the phase retrieval accuracy of TIE-Hom,results show that phase retrieval accuracy in inverse process is positive corrected with the EE/N in forward process,the relationship was validated by the experiment of numerical phantom and polyethylene air-bubble wrap.To reduce the radiation dose and measurement time as low as reasonably achievable,strategies such as collecting sparse-view projections or reducing exposure time of projections are usually adopted in XCT,while this will unavoidably increase the data inconsistency and projection noise,making tomographic reconstructions challenging.Based on the framework of adaptive steepest descent projection onto convex subsets(ASD-POCS),this thesis reconstructed the object with measured low-dose projections by projection onto convex subsets algorithm,and regularized the reconstructions with total generalized variation(TGV).In this thesis,the simulation projections were produced by a simulated cone beam computed tomography platform based on EGSnrc and reconstructed the image by above strategy,the reconstructed images were evaluated more precisely and systematically by frequency assessments including modulation transfer function,noise power spectrum,and noise equivalent quantum(NEQ).Results show that higher reconstructed spatial resolution can be obtained when smaller value of p in TGV was used,while larger image noise would be reconstructed when p deviates from 1,and the proposed strategy with p equaling 1obtains the overall highest reconstruction quality.The mouse dataset obtained by the digital radiography in our laboratory was reconstructed and evaluated,results validated the above conclusion.Phase retrieval algorithm applying TIE-Hom increases the signal to noise ratio of phase projection at the cost of the reduced spatial resolution because of the essential nature of the algorithm as a low-pass filter,and this would decrease the spatial resolution of the reconstructed three-dimensional phase image.Based on the optimization of imaging parameters,an edge enhancement model was established,then measured phase contrast projections with high edge enhancement and the phase projections retrieved by TIE-Hom were weighted summed and fed into the framework of ASD-POCS to increase the spatial resolution of reconstructed image in low-dose PB-PCT.The median root prior was also introduced in the algorithm to increase the linear model accuracy of the system.In this thesis,the strength of edge enhancement and the median root prior were flexibly adjusted by a pair of hyper-parameters,and thus the reconstructed spatial resolution can be well improved while keeping a high linear model accuracy.Compared with the ASD-POCS algorithm,the reconstructed images of the numerical phantom show an NEQ increase about 2 orders in high frequencies,and the reconstructed images of biomaterial dataset indicate an 67.4%increase of the contrast to noise ratio.