Related Experiment Video
Updated: Jan 15, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
735
Transformer-assisted broad learning for hybrid intelligence-based skin cancer segmentation
Yisa Cai1,2, Lizheng Cai1,3, Ao Song1,2
1Translational Institute for Cancer Pain, Chongming Hospital, Chongming Branch), Shanghai University of Health & Medicine Sciences (Xinhua Hospital, Shanghai, 202150, China.
Scientific Reports
|October 10, 2025
Summary
This study introduces VIWDNet, a novel hybrid deep learning model combining Broad Learning Systems and Transformers for enhanced skin lesion segmentation. VIWDNet improves diagnostic accuracy in early skin cancer detection.
Area of Science:
- Artificial Intelligence
- Medical Image Analysis
- Deep Learning
Background:
- Deep learning models, particularly Transformers, are increasingly used for medical image analysis, outperforming traditional Convolutional Neural Networks.
- Broad Learning Systems (BLS) offer efficient training and scalability but have limitations in feature design and representation.
- Transformer architectures address BLS limitations through self-attention and hierarchical structures.
Purpose of the Study:
- To propose VIWDNet, a hybrid network integrating Transformer and BLS advantages for skin lesion segmentation.
- To leverage Transformer's feature extraction and BLS's efficiency for improved medical image analysis.
- To enhance early skin cancer screening and diagnosis through accurate segmentation.
Main Methods:
- Developed VIWDNet, a hybrid network combining Transformer and Broad Learning System architectures.
- Utilized self-attention mechanisms and hierarchical structures from Transformers.
- Integrated efficient training and scalability features from BLS.
- Evaluated performance on four public skin lesion datasets (ISIC 2016, 2017, 2018, PH2).
Main Results:
- VIWDNet demonstrated superior performance in skin lesion segmentation tasks.
- Achieved improvements of 1.48-2.23% in Dice and mIoU metrics compared to state-of-the-art models.
- Validated effectiveness across multiple public medical image datasets.
Conclusions:
- VIWDNet offers a powerful and efficient solution for skin lesion segmentation in medical imaging.
- The hybrid approach enhances feature abstraction and model performance.
- Provides reliable technical support for computer-aided dermatological diagnosis and early skin cancer detection.
