Related Experiment Video
Updated: Mar 13, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.4K
A deep learning framework for breast cancer diagnosis using Swin Transformer and Dual-Attention Multi-scale Fusion
Murdhy A Aldawsari1, Saad Jamhan Aldosari1, Atef Ismail2
1Department of Mathematics, College of Sciences and Humanities, Prince Sattam bin Abdulaziz University, 11942, Al-Kharj, Saudi Arabia.
Scientific Reports
|March 12, 2026
Summary
A new hybrid AI model, Swin-DAMFN, enhances breast cancer detection from mammograms by combining global and local feature analysis. This approach improves accuracy and sensitivity for early cancer identification.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Machine Learning for Breast Cancer Detection
Background:
- Breast cancer is a leading cause of death in women globally, making early detection via mammography crucial.
- Convolutional Neural Networks (CNNs) excel at local feature detection in mammograms but struggle with long-range dependencies.
- Transformer models capture global context but demand extensive data and computational resources, limiting their use in medical imaging.
Purpose of the Study:
- To introduce Swin-DAMFN, a novel dual-branch hybrid architecture for improved breast cancer classification in mammograms.
- To leverage the strengths of both CNNs and Transformers to overcome individual limitations in mammogram analysis.
- To enhance model generalization and address dataset limitations through advanced data augmentation techniques.
Main Methods:
- Developed a hybrid architecture with a Swin Transformer branch for global dependencies and a CNN-based Dual-Attention Multi-scale Fusion Network (DAMFN) branch for local features.
- Incorporated custom Multi Separable Attention (MSA) and Tri-Shuffle Convolution Attention (TSCA) modules for multi-scale feature extraction within the CNN branch.
- Utilized an attention-guided fusion mechanism to integrate global and local features.
- Employed Generative Adversarial Networks (GANs) and photometric augmentation for advanced data augmentation to mitigate class imbalance and improve generalization.
Main Results:
- Swin-DAMFN achieved high performance on the MIAS and CBIS-DDSM datasets.
- Achieved 99.30% accuracy, 99.14% sensitivity, and 99.15% F1-score on the CBIS-DDSM dataset.
- Attained 98.75% accuracy, 98.37% sensitivity, and 98.42% F1-score on the MIAS dataset.
Conclusions:
- The proposed Swin-DAMFN architecture effectively integrates global and local feature learning for accurate breast cancer classification.
- Advanced augmentation strategies significantly enhance model generalization and address data scarcity in medical imaging.
- Swin-DAMFN demonstrates superior diagnostic accuracy and efficiency for mammogram analysis, offering a promising tool for early breast cancer detection.
