Related Experiment Video
Updated: May 27, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
349
MAFA-Uformer: Multi-attention and dual-branch feature aggregation U-shaped transformer for sparse-view CT
Xuan Zhang1, Chenyun Fang1, Zhiwei Qiao1
1School of Computer and Information Technology, Shanxi University, Taiyuan, Shanxi, China.
Journal of X-Ray Science and Technology
|February 20, 2025
Summary
This study introduces a novel network combining Convolutional Neural Networks (CNNs) and Transformers to reduce streak artifacts in computed tomography (CT) scans. The developed method effectively suppresses artifacts while preserving image details for improved diagnostic accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Computed tomography (CT) is essential for disease detection but involves health risks from X-ray radiation.
- Reducing projection views in CT minimizes radiation but introduces streak artifacts, hindering image quality.
- Existing methods struggle to balance artifact suppression with detail preservation in sparse-view CT.
Purpose of the Study:
- To develop an advanced artifact suppression method for sparse-view CT by integrating Convolutional Neural Networks (CNNs) and Transformers.
- To leverage the local feature extraction of CNNs and the global information processing of Transformers for enhanced CT image reconstruction.
- To mitigate streak artifacts while preserving crucial image details for accurate medical diagnosis.
Main Methods:
- Proposed a novel Multi-Attention and Dual-Branch Feature Aggregation U-shaped Transformer network (MAFA-Uformer).
- The Transformer branch utilizes a coordinate attention mechanism for global context and structural understanding.
- The CNN branch employs channel spatial attention for extracting critical local image features and enhancing detail recognition.
- A feature fusion module effectively integrates global and local features from both branches.
Main Results:
- The MAFA-Uformer demonstrated superior performance in suppressing artifacts, evidenced by improved peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and root mean square error (RMSE).
- Achieved a 0.76 dB increase in PSNR, a 0.44% improvement in SSIM, and an 8.55% reduction in RMSE compared to the Restormer model.
- Experimental results confirm the method's effectiveness in enhancing image quality for sparse-view CT.
Conclusions:
- The proposed MAFA-Uformer effectively suppresses streak artifacts in CT images.
- The method excels at preserving fine details and features, crucial for reliable diagnosis.
- This approach offers robust support for accurate interpretation of CT scans, particularly in low-dose or sparse-view scenarios.
