Lung Cancer Histopathology Classification

Deep learning classifier for lung cancer histopathology images (LC25000 dataset). Utilizes ResNet50 transfer learning and implements rigorous identity leakage mitigation strategies.

Results

Model	Accuracy	AUC (ACA)	AUC (SCC)
Control (5 batches)	77.0%	0.85	0.98
Final (10 epochs)	99.73%	1.00	1.00

Grad-CAM visualizations show the final model focuses on relevant histopathological features:

Installation

git clone https://github.com/yourusername/lung-cancer-classification.git
cd lung-cancer-classification
pip install -r requirements.txt

Dataset: Download LC25000 to data/lung_colon_image_set/lung_image_sets/

Usage

Train

python scripts/train.py --config config/config.yaml --epochs 10

Evaluate

python scripts/evaluate.py --config config/config.yaml

Grad-CAM visualization

python scripts/visualize.py --config config/config.yaml

Features

• Transfer learning with ResNet50/EfficientNet-B0
• Stratified Sequential Block Split to prevent Identity Leakage
• Cross-Entropy Loss (dataset is perfectly balanced)
• Grad-CAM explainability for model interpretability
• Data augmentation (flips, rotations, color jitter)

Citation

Borkowski, A. A., Bui, M. M., Thomas, L. B., Wilson, C. P., DeLand, L. A., & Mastorides, S. M. (2019). Lung and Colon Cancer Histopathological Image Dataset (LC25000). arXiv preprint arXiv:1912.12142.