Image-domain material decomposition using data-driven sparsity models for dual-energy CT

Zhipeng Li, Saiprasad Ravishankar, Yong Long, Jeffrey A. Fessler

Abstract: Penalized weighted least squares (PWLS) with better image priors is a promising way to develop improved image-domain dual-energy CT (DECT) methods for achieving high quality basis material images. We propose a new method for DECT that combines conventional PWLS estimation with regularization based on sparsifying transforms (DECT-ST) learned from datasets of numerous CT images. Numerical experiments with phantom and patient data show that the proposed method significantly improves the image quality over direct matrix inversion decomposition and over PWLS decomposition with an edge-preserving hyperbola regularizer (DECT-EP).