Multiscale penalized weighted least-squares image-domain decomposition for dual-energy CT

Abstract

Image-domain dual-energy CT (DECT) is practical and critical for medical diagnosis and treatment evaluation. The bottleneck of DECT imaging is the significantly magnified noise after a direct inversion decomposition. A good balance is not readily achievable between noise suppression and spatial resolution maintenance. Noise suppression is conventionally included in DECT decomposition implicitly or explicitly and results in the spatial resolution loss. To tackle this problem, we propose a multiscale framework for DECT imaging. High- and low-energy CT images are decomposed into the scale space to fully explore the flexibility of signal processing in each scale. The scale images are then decomposed into two material-specific images using penalized weighted least-squares (PWLS) algorithm with adjustable penalty parameters in different scales using the statistical information in the DECT decomposition process. The final material-specific images are generated by accumulating the decomposed images throughout all scales. The proposed algorithm is evaluated on two physical phantoms: Catphan©600 evaluation phantom and an anthropomorphic head phantom. Noise standard deviation (STD) is reduced by 93% and 98% using the proposed strategy for bone and soft-tissue images as compared with the one using direct inversion for Catphan©600 evaluation phantom, and by 96% and 98% for the anthropomorphic head phantom. Compared with its single-scale counterpart, the proposed multiscale PWLS-based method increases the spatial resolution of soft-tissue images by 28% using a lower scaling factor (k = 20), and by 51% with a higher scaling factor (k = 80). The noise STDs of the decomposed material images remain at the similar level using single- and multi-scale schemes. The proposed multiscale PWLS-based decomposition method is thus attractive for low-dose DECT imaging and clinical practice.