Final Dissertation Project: Advanced Computational Pathology Methods

This repository contains a comprehensive collection of state-of-the-art computational pathology methods for whole slide image (WSI) analysis, implemented as part of a final dissertation project. The project encompasses multiple deep learning approaches including attention-based multiple instance learning, graph neural networks, and foundation models for pathological image analysis.

🔬 Project Overview

This project explores various computational pathology approaches for analyzing whole slide images (WSIs) in cancer diagnosis and prognosis. The repository integrates several cutting-edge methods:

• CLAM: Clustering-constrained Attention Multiple Instance Learning for weakly supervised WSI classification
• Patch-GCN: Graph Convolutional Networks for context-aware survival prediction
• TRIDENT: Large-scale WSI processing toolkit with foundation model support
• SUGAR: Subgraph Neural Network with Reinforcement Learning pooling
• GigaPath: Whole slide foundation model for pathology

📁 Repository Structure

final_dissertation/
├── CLAM/                    # CLAM implementation for WSI classification
├── Patch-GCN/               # Graph-based survival prediction
├── TRIDENT/                 # WSI processing toolkit
├── SUGAR/                   # Subgraph neural networks with RL
├── prov-gigapath/           # GigaPath foundation model
├── TCGA-WSI/                # TCGA dataset processing utilities
├── TCGA-CLAM-graph/         # Processed TCGA data for graph methods
├── TCGA-patches/            # Extracted patches from TCGA
├── trident_processed/       # Processed WSI data using TRIDENT
└── test/                    # Test data and validation files

🚀 Quick Start

Prerequisites

• Python 3.8+
• CUDA-compatible GPU (recommended)
• Minimum 16GB RAM
• Large storage capacity for WSI data

Installation

1. Clone the repository

   git clone <your-repository-url>
   cd final_dissertation
   

2. Set up individual environments

   Each method has its own environment requirements. Choose the method you want to use:

   For CLAM:

   cd CLAM
   conda env create -f env.yml
   conda activate clam
   

   For TRIDENT:

   cd TRIDENT
   conda create -n trident python=3.10
   conda activate trident
   pip install -e .
   

   For SUGAR:

   cd SUGAR
   pip install -r requirements.txt
   

   For GigaPath:

   cd prov-gigapath
   conda env create -f environment.yaml
   conda activate gigapath
   

🔧 Usage

1. CLAM - Weakly Supervised WSI Classification

CLAM enables slide-level classification without patch-level annotations.

cd CLAM

# Segmentation and patching
python create_patches_fp.py --source DATA_DIRECTORY --save_dir RESULTS_DIRECTORY --patch_size 256 --seg --patch --stitch

# Feature extraction
python extract_features_fp.py --data_h5_dir RESULTS_DIRECTORY --data_slide_dir DATA_DIRECTORY --csv_path dataset_csv/tumor_vs_normal_dummy_clean.csv --feat_dir FEATURES_DIRECTORY

# Training
python main.py --drop_out --early_stopping --lr 2e-4 --k 10 --label_frac 1.0 --exp_code tumor_vs_normal_CLAM_50 --weighted_sample --bag_loss ce --inst_loss svm --task tumor_vs_normal --model_type clam_sb --log_data --data_root_dir FEATURES_DIRECTORY

2. Patch-GCN - Graph-based Survival Prediction

Formulates WSIs as graphs for context-aware analysis.

cd Patch-GCN

# WSI-Graph Construction
python "WSI-Graph Construction.py"

# Training
python main.py --task tcga_kidney_cv --split_dir tcga_kidney_100 --gc 32

3. TRIDENT - WSI Processing Toolkit

High-performance WSI processing with foundation model support.

cd TRIDENT

# Process single slide
python run_single_slide.py --slide_path ./wsis/slide.svs --job_dir ./processed --patch_encoder uni_v1 --mag 20 --patch_size 256

# Batch processing
python run_batch_of_slides.py --task all --wsi_dir ./wsis --job_dir ./processed --patch_encoder uni_v1 --mag 20 --patch_size 256

4. SUGAR - Subgraph Neural Networks

Graph neural networks with reinforcement learning pooling.

cd SUGAR
python train.py

📊 Datasets

The project works with several major pathology datasets:

• TCGA (The Cancer Genome Atlas): Multi-cancer WSI dataset

  • BLCA (Bladder Cancer)
  • BRCA (Breast Cancer)
  • LUAD (Lung Adenocarcinoma)
  • UCEC (Uterine Corpus Endometrial Carcinoma)

• Graph Datasets: MUTAG, DD, NCI1, NCI109, PTC_MR, ENZYMES, PROTEINS

Data Organization

TCGA-WSI/
├── BLCA/           # Bladder cancer WSIs
├── BRCA/           # Breast cancer WSIs  
├── LUAD/           # Lung cancer WSIs
└── UCEC/           # Uterine cancer WSIs

TCGA-patches/       # Extracted tissue patches
TCGA-CLAM-graph/    # Graph representations

🏗️ Key Features

CLAM Features

• Weakly supervised learning using only slide-level labels
• Attention-based multiple instance learning
• Clustering-constrained feature learning
• Heatmap visualization for interpretability

Patch-GCN Features

• Graph representation of WSI spatial relationships
• k-NN connectivity based on patch coordinates
• Context-aware survival prediction
• Message passing for spatial feature learning

TRIDENT Features

• Support for 25+ foundation models (UNI, CONCH, TITAN, etc.)
• Efficient tissue segmentation and patch extraction
• Scalable processing pipeline
• Multiple encoder options

SUGAR Features

• Reinforcement learning-based graph pooling
• Self-supervised mutual information mechanism
• Subgraph neural network architecture

📈 Results and Evaluation

The project includes comprehensive evaluation scripts and benchmarks:

• Classification accuracy metrics
• Survival prediction C-index
• Attention visualization and interpretation
• Cross-validation results
• Statistical significance testing

📚 Documentation

Detailed documentation for each method can be found in their respective directories:

• CLAM/docs/ - CLAM documentation and tutorials
• TRIDENT/docs/ - TRIDENT user guide
• Patch-GCN/docs/ - Patch-GCN implementation details

🤝 Contributing

This is a research project developed for academic purposes. For questions or collaboration opportunities, please feel free to reach out.

📄 License

Each component retains its original license:

• CLAM: GPLv3 License
• TRIDENT: MIT License
• Patch-GCN: Academic use
• SUGAR: Academic use

🔗 References

If you use this work, please cite the relevant papers:

CLAM

@article{lu2021data,
  title={Data-efficient and weakly supervised computational pathology on whole-slide images},
  author={Lu, Ming Y and Williamson, Drew FK and Chen, Tiffany Y and Chen, Richard J and Barbieri, Matteo and Mahmood, Faisal},
  journal={Nature biomedical engineering},
  volume={5},
  number={6},
  pages={555--570},
  year={2021},
  publisher={Nature Publishing Group}
}

Patch-GCN

@incollection{chen2021whole,
  title={Whole Slide Images are 2D Point Clouds: Context-Aware Survival Prediction using Patch-based Graph Convolutional Networks},
  author={Chen, Richard J and Lu, Ming Y and Shaban, Muhammad and Chen, Chengkuan and Chen, Tiffany Y and Williamson, Drew FK and Mahmood, Faisal},
  booktitle={Medical Image Computing and Computer Assisted Intervention--MICCAI 2021},
  pages={339--349},
  year={2021},
  publisher={Springer}
}

TRIDENT

@article{chen2025trident,
  title={Trident: A foundation model for digital pathology},
  author={Chen, Richard J and others},
  journal={arXiv preprint arXiv:2502.06750},
  year={2025}
}

---

Developed as part of UCL Final Dissertation Project
Advancing Computational Pathology through Deep Learning and Graph Neural Networks