Low-dose CT image processing using artifact suppressed dictionary learning

With low-dose scanning protocol, CT images are often severely corrupted by quantum noise and artifacts. Artifacts often take prominent directional features and are rather hard to be suppressed without blurring tissue structures. In this paper, we propose to improve low-dose CT (LDCT) images using a two-step scheme called “artifact suppressed dictionary learning algorithm” (ASDL). In the first step, artifacts are significantly reduced by a discriminative sparse representation (DSR) operation, in which scale and orientation information of artifacts are exploited to build discriminative dictionaries for artifact suppression. Then, a general dictionary learning (DL) processing is performed to suppress the residual artifacts and noise. Experiments on both abdominal and thoracic data validate the good performance of the proposed method.