Low-Dose CT with Deep Learning Regularization via Proximal Forward Backward Splitting

TL;DR

This paper introduces deep learning regularization within a proximal forward-backward splitting framework for low-dose CT reconstruction, combining analytical and iterative methods to improve image quality.

Contribution

It develops a novel unrolled proximal splitting method that fuses analytical and iterative reconstruction with deep learning regularization for LDCT.

Findings

DL-regularized methods outperform conventional AR or IR.

PFBS-AIR outperforms PFBS-IR due to AIR.

Deep learning enhances low-dose CT image reconstruction quality.

Abstract

Low dose X-ray computed tomography (LDCT) is desirable for reduced patient dose. This work develops image reconstruction methods with deep learning (DL) regularization for LDCT. Our methods are based on unrolling of proximal forward-backward splitting (PFBS) framework with data-driven image regularization via deep neural networks. In contrast with PFBS-IR that utilizes standard data fidelity updates via iterative reconstruction (IR) method, PFBS-AIR involves preconditioned data fidelity updates that fuse analytical reconstruction (AR) method and IR in a synergistic way, I.e. fused analytical and iterative reconstruction (AIR). The results suggest that DL-regularized methods (PFBS-IR and PFBS-AIR) provided better reconstruction quality from conventional wisdoms (AR or IR), and DL-based postprocessing method (FBPConvNet). In addition, owing to AIR, PFBS-AIR noticeably outperformed PFBS-IR.