Related Experiment Video
Updated: Jun 26, 2026
05:32
Retrospective Cardiac Gating with A Prototype Small-Animal X-ray Computed Tomograph
Published on: February 21, 2025
WDBDM: Wavelet-based dual-branch diffusion model for low-dose CT and PET denoising
Qi Sun1, Tongtong Li1, Guowei Wang1
1School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China; Gansu Provincial Key Laboratory of Wearable Computing, Lanzhou University, Lanzhou 730000, China.
Summary
This study introduces WDBDM, a novel framework using wavelet transforms and diffusion models to denoise low-dose CT and PET scans. WDBDM generates high-quality medical images from low-dose data, reducing radiation exposure risks.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Signal Processing
Background:
- Low-dose computed tomography (CT) and positron emission tomography (PET) imaging are crucial for disease assessment but raise radiation exposure concerns.
- Reconstructing high-quality images from low-dose scans is essential to balance diagnostic performance and patient safety.
- Existing methods struggle to effectively denoise low-dose medical images while preserving diagnostic quality.
Purpose of the Study:
- To present WDBDM, a wavelet-based dual-branch diffusion framework designed for denoising low-dose CT and PET images.
- To enhance image quality to levels comparable to normal-dose imaging, thereby reducing radiation risks.
- To improve the feature representation and denoising performance for medical imaging applications.
Main Methods:
- The WDBDM framework integrates Discrete Wavelet Transform (DWT), a Low-Frequency Diffusion Branch (LFDB), a High-Frequency Diffusion Branch (HFDB), and a Fusion Module.
- A Fusion Spatial-Frequency Convolution Module (FSFCM) was developed to extract joint spatial and frequency domain information, enhancing feature representation.
- A novel recovery network, HLF-MEMNet, was integrated to prevent error propagation and enable bidirectional guidance between frequency components during sampling.
Main Results:
- WDBDM demonstrated superior denoising performance compared to existing methods across four public datasets and two imaging modalities.
- The framework effectively generated normal-dose quality images from low-dose CT and PET data.
- The integration of FSFCM and HLF-MEMNet significantly improved feature extraction and model stability.
Conclusions:
- WDBDM offers a promising solution for generating high-quality medical images from low-dose scans, significantly reducing radiation exposure.
- The proposed framework shows excellent denoising capabilities and generalization ability, outperforming current state-of-the-art methods.
- WDBDM has the potential to improve diagnostic accuracy and patient safety in medical imaging.
