MCA-Net: A Lightweight Multi-order Context Aggregation Network for Low Dose CT Denoising

Low-dose computed tomography (LDCT) is widely utilized to reduce patient radiation exposure but often results in elevated noise levels. While deep learning models for LDCT denoising have made significant strides, they continue to struggle with effectively capturing discriminative features from high-resolution feature maps. This work proposes a lightweight network based on multi-order context aggregation for LDCT denoising, called MCA-Net. The proposed method enables an enhanced discriminative feature representation through stacking multiple multi-order context aggregation blocks, which are consisted of effective convolutions and gated aggregation. Furthermore, MCA-Net mitigates the loss of spatial detail information by incorporating prior edge-enhanced information and preserving high-resolution feature maps. Experiments conducted on two public datasets demonstrate superior performance compared to state-of-the-art models. Additionally, our method exhibits high competitiveness concerning parameter counts and computational efficiency. The code and pre-trained models will be released on https://code.ihep.ac.cn/lijf/MCANet .