Abstract
Low-dose computed tomography (LDCT) reduces radiation exposure but inevitably degrades image quality due to increased noise and loss of fine details. We propose a dual-branch deep learning framework that effectively suppresses noise while preserving high-frequency textures and edges. The proposed method separates low- and high-frequency components and processes them through frequency-specialized branches, enabling complementary noise reduction and detail enhancement. Directional frequency information is exploited to improve structural fidelity, and the outputs of both branches are adaptively fused to generate the final denoised image. Experiments on the NIH-AAPM-Mayo Clinic Low-Dose CT Challenge dataset demonstrate that the proposed approach outperforms state-of-the-art methods in both quantitative metrics and visual quality, achieving superior noise suppression while maintaining fine structural details.
| Original language | English |
|---|---|
| Title of host publication | Medical Imaging 2026 |
| Subtitle of host publication | Image Processing |
| Editors | Jhimli Mitra, Yu Gan |
| Publisher | SPIE |
| ISBN (Electronic) | 9781510697874 |
| DOIs | |
| State | Published - 3 Apr 2026 |
| Event | Medical Imaging 2026: Image Processing - Vancouver, Canada Duration: 15 Feb 2026 → 19 Feb 2026 |
Publication series
| Name | Progress in Biomedical Optics and Imaging - Proceedings of SPIE |
|---|---|
| Volume | 13925 |
| ISSN (Print) | 1605-7422 |
| ISSN (Electronic) | 2410-9045 |
Conference
| Conference | Medical Imaging 2026: Image Processing |
|---|---|
| Country/Territory | Canada |
| City | Vancouver |
| Period | 15/02/26 → 19/02/26 |
Bibliographical note
Publisher Copyright:© COPYRIGHT SPIE. Downloading of the abstract is permitted for personal use only.
Keywords
- Deep learning
- Denoising
- Low-dose CT
- Medical image enhancement
- Texture preservation
Fingerprint
Dive into the research topics of 'Frequency-Aware Inter-Branch Fusion Network for Texture-Preserving Denoising in Low-Dose CT'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver