Multiclass Breast Cancer Classification using CNN
Deep learning for histopathology-based cancer diagnosis
Project Overview
This project developed a convolutional neural network (CNN) to classify breast cancer histopathology images into eight subclasses (benign and malignant). The pipeline included dataset resampling, augmentation, and CNN-based feature extraction, achieving high classification accuracy compared to baseline methods
Key Features
🧠 Deep CNN Architecture: Multi-layer convolution, pooling, and dropout for robust learning
🩺 Eight Subclasses: Benign (Adenosis, Fibroadenoma, Tubular Adenoma, Phyllodes Tumor) and Malignant (Ductal, Lobular, Mucinous, Papillary Carcinomas)
📊 Resampling & Augmentation: Balanced class distribution with rotations, flips, shifts, and brightness variations
⚡ High Accuracy: Training accuracy of 93.8% and validation accuracy of 88.8%
Research Contributions
Designed and trained a CNN with class weighting and augmentation for imbalanced datasets
Compared performance against baseline models, achieving a 15% improvement in validation accuracy
Demonstrated applicability of deep learning in medical imaging and cancer diagnostics
Technical Achievements
Built a custom CNN model in Python with Keras/TensorFlow
Achieved 88.8% validation accuracy, outperforming benchmark models (73.5%)
Validated results through cross-validation and extensive dataset resampling
Applications
Computer-Aided Diagnosis (CAD): Supporting oncologists in breast cancer detection
Medical Imaging Analysis: Automated classification of histopathology slides
AI in Healthcare: Enhancing diagnostic speed and reducing error rates
Impact and Recognition
This project highlighted the potential of deep learning in biomedical applications, demonstrating how CNNs can improve diagnostic accuracy and efficiency in cancer detection