Related Experiment Video
Updated: Jul 17, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
441
A Novel Deep Learning Model for Medical Image Segmentation with Convolutional Neural Network and Transformer
Zhuo Zhang1, Hongbing Wu2, Huan Zhao1
1Tianjin Key Laboratory of Optoelectronic Detection Technology and Systems, School of Electronic and Information Engineering, Tiangong University, Tianjin, 300387, China.
Interdisciplinary Sciences, Computational Life Sciences
|September 4, 2023
Summary
A new deep learning model, MRC-TransUNet, enhances medical image segmentation accuracy by integrating transformer and UNet architectures. This approach improves contextual information capture, outperforming existing methods for breast, brain, and lung imaging.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate medical image segmentation is crucial for clinical decisions.
- Deep learning models, particularly those combining transformers and U-Nets, show promise but face limitations in capturing contextual information due to traditional skip connections.
Purpose of the Study:
- To introduce a novel deep learning architecture, the coordinated mobile and residual transformer UNet (MRC-TransUNet), designed to overcome limitations in medical image segmentation.
- To enhance the capture of both contextual and detailed information in medical images.
Main Methods:
- Proposed the MRC-TransUNet, utilizing a lightweight Mobile and Residual Vision Transformer (MR-ViT) to bridge the semantic gap and a reciprocal attention (RPA) module to preserve details.
- Modified the U-Net architecture by limiting skip connections to the first layer and incorporating MR-ViT and RPA in downsampling layers to better leverage long-range contextual information.
Main Results:
- Evaluated the MRC-TransUNet on breast, brain, and lung medical image segmentation datasets.
- The proposed method demonstrated superior performance compared to state-of-the-art techniques, evidenced by improved Dice coefficient and Hausdorff distance metrics.
Conclusions:
- The MRC-TransUNet significantly enhances the accuracy of medical image segmentation.
- The model shows strong potential for practical clinical applications due to its improved performance.
