Prostate Cancer Classification

Multi-task learning pipeline for prostate cancer classification using the SICAPv2 dataset.

Features

• Multi-label classification for mixed Gleason patterns
• Multi-resolution processing (224x224 + 384x384)
• Transformer-based feature fusion
• Specialized cribriform detection
• Stain normalization

Data Structure

The SICAPv2 dataset is located in data/SICAPv2/ with the following structure:

data/SICAPv2/
├── images/           # 18,783 histology patch images (512x512, 10X magnification)
├── masks/            # Corresponding annotation masks
├── partition/
│   ├── Test/        # Train/test split files
│   └── Validation/  # Cross-validation folds (Val1-Val4)
├── wsi_labels.xlsx  # WSI-level Gleason scores
└── readme.txt       # Dataset documentation

Quick Start

1. Setup virtual environment:

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Test the setup:

python test_imports.py

3. Run training:

python main.py

Data Information

• Total samples: 18,783 patches (512×512 pixels)
• Magnification: 10X
• Labels: Multi-label classification (NC, G3, G4, G5, G4C)
• Cribriform detection: Specialized detection for G4C patterns
• Cross-validation: 4-fold patient-based splits available

Results

Results and model checkpoints will be saved to SICAPv2_results/ directory.