Related Experiment Video
Updated: Jun 10, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
LAM-CATNet: lambda-aware multi-scale cross-attention swin transformer network for mammogram classification
Gautam Ankoji1, N Thirupathi Rao2, C H V V Ramana3
1Department of Computer Science and Engineering, Vignan's Institute of Information Technology (A), Visakhapatnam, Andhra Pradesh, 530049, India.
Scientific Reports
|June 8, 2026
Summary
A new deep learning model, LAM-CATNet, enhances automated breast lesion detection in mammograms. This advanced segmentation and classification tool improves early cancer identification, offering better diagnostic performance and interpretability for clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Accurate segmentation and classification of breast lesions in mammograms are crucial for early breast cancer detection.
- Small or difficult-to-identify lesions pose significant challenges in mammogram analysis.
Purpose of the Study:
- To introduce LAM-CATNet (Lambda-Aware Multi-scale Cross-Attention Swin Transformer Network), a novel deep learning model for automated breast lesion detection.
- To improve the accuracy and consistency of breast lesion segmentation and classification in mammograms.
Main Methods:
- Developed LAM-CATNet, a transformer fusion model integrating statistical intensity modeling with deep learning.
- Employed a Lambda Distribution-Guided Transformer-Fused Hybrid Segmentation Network approach.
- Utilized statistical analysis of variability to enhance model performance.
Main Results:
- LAM-CATNet achieved superior segmentation performance with Dice coefficients of 0.786 (CBIS-DDSM) and 0.869 (MIAS).
- The model attained 94.7% accuracy and 98.2% area under the ROC curve for lesion classification.
- Attention map visualizations demonstrated increased model interpretability by focusing on relevant lesion areas.
Conclusions:
- LAM-CATNet effectively improves automated breast lesion segmentation and classification, outperforming existing models.
- The model's high performance, interpretability, and statistical reasoning show potential as a supportive tool for computer-aided diagnosis in mammography.
- Further clinical validation is recommended for its application in breast cancer screening.
