Related Experiment Video
Updated: Mar 13, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
852
CSAFusion: a convolutional neural network (CNN)-based and Swin Transformer network for multi-modal medical image
Liyuan Zhang1,2, Jiachen Zheng1,2, Xiongfeng Tang3
1School of Computer Science and Technology, Changchun University of Science and Technology, Changchun, China.
Quantitative Imaging in Medicine and Surgery
|March 12, 2026
Summary
CSAFusion, a novel unsupervised framework, significantly improves multi-modal medical image fusion by preserving fine details and spatial fidelity. This advanced technique enhances diagnostic accuracy through superior feature extraction across various imaging modalities.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Single-modality medical imaging offers limited diagnostic information.
- Multi-modal fusion integrates complementary data for enhanced diagnosis.
- Existing fusion methods struggle with spatial fidelity and detail preservation.
Purpose of the Study:
- To develop an advanced multi-modal medical image fusion method.
- To improve the preservation of spatial fidelity and fine structural details.
- To enhance diagnostic information extraction from fused medical images.
Main Methods:
- Proposed an unsupervised fusion framework, CSAFusion, utilizing a U-Net backbone.
- Incorporated adaptive convolutions (ACs) in the encoder and Swin Transformer modules in the decoder.
- Employed a dual fusion architecture and a composite loss function for optimized structural integrity and perceptual quality.
Main Results:
- CSAFusion outperformed existing methods in CT-MRI, MRI-T1-T2, and MRI-SPECT fusion tasks.
- Demonstrated superior preservation of CT bone structures, MRI soft-tissue textures, and SPECT functional information.
- Achieved a QNCIE score of 0.9304 and statistically significant improvements in SSIM and EPI for MRI-SPECT fusion.
Conclusions:
- CSAFusion enhances local and global feature extraction for faithful preservation of diagnostic information.
- The method shows robust cross-modality generalization and superior quantitative performance.
- CSAFusion presents a promising tool for clinical applications in multi-modal medical image fusion.
