Breast Cancer Classification with Deep Learning

Deep Learning
Medical Imaging
Classification
Research

A deep learning pipeline for classifying mammographic images from the CBIS dataset, leveraging preprocessing and transfer learning.

Published

December 1, 2023

This project develops a deep learning pipeline for mammogram classification using the CBIS-DDSM dataset, addressing preprocessing, augmentation, and model training challenges in medical imaging.

Techniques include ROI extraction, intensity maps, gradient transformations, and custom data augmentation (ROI overlay, geometric transformations, blended cropping). Pretrained CNNs were fine-tuned through transfer learning, achieving state-of-the-art performance among public baselines.

Key results:
- Preprocessing improved convergence and accuracy
- Transfer learning boosted generalization performance
- Augmentation strategies reduced overfitting
- Demonstrated potential for scalable and reproducible breast cancer diagnostics

Read full-text on ResearchGate


Breast cancer mammogram classification with deep learning models

Deep learning for mammogram classification using the CBIS dataset