abdullahtahir/resnet50-mammography-birads

ResNet50 Mammography Classification

Binary classification of mammographic images into Normal, Benign, and Malignant categories.

Model Details

• Architecture: ResNet50 (pretrained on ImageNet, fine-tuned)
• Task: 3-class classification (Normal / Benign / Malignant)
• Dataset: King Abdulaziz University Mammogram Dataset (5,662 images)
• Classes: Normal (BI-RADS 1), Benign (BI-RADS 2-3), Malignant (BI-RADS 4-5)

Performance (5-Fold Cross Validation)

Metric	Mean ± Std
Accuracy	84.53% ± 1.37%
AUC	0.9665 ± 0.0038
Recall	83.58% ± 0.87%

Best Fold (Fold 4)

• Accuracy: 86.32%
• AUC: 0.9734
• Recall: 84.74%