Efficient low-dose CT image enhancement using MobileMamba-UNet with wavelet-enhanced long-range modeling.
Authors
Affiliations (5)
Affiliations (5)
- School of Information Engineering, Changsha Medical University, Changsha, China.
- Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences, Guangzhou, China.
- Hunan Provincial Key Laboratory of the Traditional Chinese Medicine Agricultural Biogenomics, Hunan Provincial University Key Laboratory of the Fundamental and Clinical Research on Functional Nucleic Acid, Changsha Medical University, Changsha, China.
- Hunan Provincial Maternal and Child Health Care Hospital, Changsha, China.
- China Spallation Neutron Source, Dongguan, China.
Abstract
Deep learning has become a dominant paradigm for low-dose computed tomography (LDCT) image reconstruction. Nevertheless, existing approaches still struggle to simultaneously achieve accurate structural detail preservation and computational efficiency, particularly when handling long-range contextual dependencies. To design a lightweight yet effective LDCT reconstruction framework that captures both global contextual information and fine-grained local details while maintaining low memory consumption and fast inference speed. We propose MobileMamba-UNet, a hybrid neural network that integrates a MobileMamba backbone with a multi-scale U-Net architecture. The model incorporates a Wavelet Transform Enhanced Mamba mechanism to emphasize high frequency and diagnostically relevant structures, together with a multi-receptive field feature interaction module that jointly models local textures and long-range dependencies. All components are constructed with linear computational complexity to ensure efficiency in large-scale LDCT reconstruction tasks. Extensive experiments conducted on the Mayo-2016 and Mayo-2020 LDCT datasets demonstrate that MobileMamba-UNet consistently outperforms existing CNN- and Transformer-based methods. The proposed approach achieves superior image quality while significantly reducing memory usage and inference latency. MobileMamba-UNet represents a promising approach for LDCT image reconstruction, balancing reconstruction performance with computational efficiency and practical applicability.