Nonlocal low-rank and sparse matrix decomposition for low-dose cerebral perfusion CT image restoration

Abstract

To present a nonlocal low-rank and sparse matrix decomposition (NLSMD) method for low-dose cerebral perfusion CT image restoration. Low-dose cerebral perfusion CT images were first partitioned into a matrix, and the low- rank and sparse matrix decomposition model was constructed to obtain high-quality low-dose cerebral perfusion CT images. The cerebral hemodynamic parameters were calculated from the restored high-quality CT images. In the phantom study, the average structured similarity (SSIM) value of the sequential images obtained by filtered back-projection (FBP) algorithm was 0.9438, which was increased to 0.9765 using the proposed algorithm; the SSIM values of cerebral blood flow (CBF) and cerebral blood volume (CBV) map obtained by FBP algorithm were 0.7005 and 0.6856, respectively, which were increased using the proposed algorithm to 0.7871 and 0.7972, respectively. The proposed method can effectively suppress noises in low-dose cerebral perfusion CT images to obtain accurate cerebral hemodynamic parameters.