Reconstruction Algorithms For Low-dose Computed Tomography And Its Application To Dual-energy Breast Material Decomposition And Detection

Computed tomography(CT)is an integral component of contemporary medicine that offers tremendous benefits in the diagnosis of human disease in the whole body like the brain,chest,bone,abdomen and even soft tissues like the breast.Dedicated breast CT has shown a higher development potential than conventional CT,because the radiation dose of dedicated breast CT is lower than that of conventional CT,and dedicated breast CT has more potential value in the early diagnosis of breast diseases.Low-dose imaging and dual-energy decomposition algorithms are the key technologies.The thesis focuses on the imaging problems of dedicated breast CT and the research on low-dose reconstruction,dual-energy material decomposition and noise reduction algorithms based on sparse sampling mode.The characteristics and innovations of the paper are as follows:To achieve LDCT reconstruction,smoothed l0-norm constraint dictionary learning(SL0-DL)method was proposed.The SL0-DL introduced smoothed l0-norm regularization to the formulated objective function at each iteration and the algorithm performance was assessed with both simulated phantom and real CT datasets of sheep and mouse.The SL0-DL has the mean value of peak signal-to-noise ratio(PSNR),structural similarity index(SSIM),and normalized mean absolute deviation(NMAD)of 44.2,0.9920 and 0.7652 for the head phantom respectively.The results showed that the proposed SL0-DL is effective for the reconstruction of low-dose acquisition and under-sampled noisy data and also outperformed other competing algorithms like conventional dictionary learning(DL).gradient projection Barzilai Borwein,and Total-variation algorithm in terms of artifacts suppression and texture details preservation.More so,the SL0-DL dictionary-based method was further improved by incorporating simultaneous algebraic reconstruction technique(SART)into the optimization problem and replacing the dictionary patches with a group of non-local similar patches to form the hybrid optimization method(HOM)which was adopted for the DECT material decomposition.DECT breast datasets were simulated to assess the proposed HOM algorithm.The HOM algorithm had estimated mean of 49.0,0.9968,and 0.0028 for the PSNR,SSIM,and root mean square error(RMSE)in the simulated low-dose dataset respectively,while the signal-to-noise ratio(SNR)and normalized absolute error(NAE)of decomposed normal breast tissue were 22.0 and 0.0547 respectively.The results revealed that the method is capable of separating microcalcifications from healthy breast tissue via the DECT technique without noise amplification on a material basis compared to other competing algorithms.Finally,a comparison of the diagnostic and reconstruction accuracy of cone-beam breast computed tomography(CBBCT)and digital breast tomosynthesis(DBT)for adequate breast cancer diagnosis was carried out using experimental and meta-analysis approaches.The result was tested on both scanned breast phantom and real breast datasets.The DBT had 31.6,0.9353 and 0.1273,while CBBCT had 29.7,0.9136 and 0.1273 for PSNR,SSIM,and NAE respectively.The experimental result revealed that the reconstruction through DBT geometry shows higher image quality assessment than CBBCT reconstructed images for all competing algorithms and capable of revealing more breast lesions.Likewise,the pooled sensitivity,specificity,and area under the receiver's operating curve(AUC)are 86%,91%,and 0.9371 for DBT,while 89%,62%,and 0.8634 were estimated for the CBBCT at 95%confidence interval.The meta-analysis results further supported previous experimental results showing higher diagnostic performance over that of the CBBCT.In summary,based on the above experimental analysis,the proposed new algorithm and its improved algorithm show the effectiveness of LDCT and DECT image noise reduction.At the same time,based on the current limited clinical data set,the higher diagnostic performance of DBT on CBBCT requires deeper clinical trials and verification.