Deep learning project for automated detection of breast abnormalities in mammogram images using the VinDr-Mammo dataset.

• Faster R-CNN and RetinaNet architecture for multi-class object detection
• Balanced dataset creation through stratified sampling based on:
  • BI-RADS classifications (1-5)
  • Breast density categories (A-D)
  • Finding categories (Mass, Calcification, Asymmetry, etc.)
• Data pruning to address class imbalance:
  • Reduced overrepresented "No Finding" class from 18k to 2.5k samples
  • Preserved rare finding combinations across train/test splits

git clone https://github.com/yourusername/vindr-mammo.git

pip install -r requirements.txt

TODO: Add results as a notebook