Related Experiment Video
Updated: Aug 23, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
483
[Application of a parallel branches network based on Transformer for skin melanoma segmentation]
Sanli Yi1,2, Gang Zhang1,2, Jianfeng He1
1School of Information Engineering and Automation, Kunming University of Scienceand Technology, Kunming 650500, P. R. China.
Summary
This study introduces a novel parallel network for accurate skin lesion segmentation, improving early melanoma diagnosis. The proposed method enhances feature extraction for better segmentation results.
Area of Science:
- Dermatology and Medical Imaging
- Artificial Intelligence in Healthcare
- Computer Vision
Context:
- Cutaneous malignant melanoma diagnosis relies on accurate segmentation of skin lesions.
- Existing segmentation methods may lack robustness and detailed feature extraction capabilities.
- Transformer-based architectures offer potential for improved medical image analysis.
Purpose:
- To propose a novel parallel network architecture for enhanced skin lesion segmentation.
- To integrate a Multiple Residual Frequency Channel Attention Network (MFC) and a Visual Transformer Network (ViT).
- To improve the accuracy and effectiveness of melanoma lesion segmentation.
Summary:
- A parallel network combining MFC and ViT branches was developed for skin lesion segmentation.
- The MFC branch focuses on detailed feature extraction using residual modules and frequency channel attention.
- The ViT branch preserves global image features using multi-head self-attention.
- Parallel fusion of features from both branches enables more effective segmentation.
Impact:
- The proposed algorithm achieved high performance on the ISIC 2018 dataset, with IoU of 90.15% and Dice coefficient of 94.82%.
- Results surpass current state-of-the-art skin melanoma segmentation networks.
- Provides dermatologists with more accurate lesion data for improved diagnosis and treatment planning.
