Related Experiment Videos
Breast cancer detection and classification via a robust deep learning approach
Magy Makram1,2, Alber S Aziz3, Mary Monir Saeid4
1Computer Science Department, Faculty of Computers and Artificial Intelligence, Fayoum University, Fayoum, Egypt. mm5600@fayoum.edu.eg.
Scientific Reports
|July 17, 2026
Summary
A new deep learning framework accurately classifies breast cancer from mammograms. This leakage-controlled approach achieved 97.12% accuracy, improving early detection capabilities.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate breast cancer classification from mammography is crucial for early detection and treatment.
- Deep learning models show promise but require robust frameworks to handle image variations and ensure reliable performance.
Purpose of the Study:
- To develop and evaluate a leakage-controlled deep learning framework for enhanced breast cancer classification using the CBIS-DDSM dataset.
- To introduce an Inter-View Attention Fusion (IVAF) module for adaptive fusion of mammographic views.
Main Methods:
- A patient-level data partitioning strategy was employed before data augmentation.
- A two-stage transfer learning approach utilizing ResNet50 was implemented.
- The Inter-View Attention Fusion (IVAF) module was developed as a lightweight convolutional gating strategy for feature map fusion.
Main Results:
- The framework achieved a high accuracy of 97.12% on the CBIS-DDSM dataset.
- Sensitivity reached 96.44%, specificity 97.68%, and AUC-ROC was 0.9876.
- Consistent performance was observed with an average accuracy of 97.11% ± 0.18% across 100 random seeds.
Conclusions:
- The proposed leakage-controlled deep learning framework demonstrates high efficacy in breast cancer classification.
- The IVAF module effectively fuses multi-view mammographic information, enhancing classification performance.
- This framework offers a promising tool for improving automated breast cancer detection in clinical settings.