Learning Mechanistic Subtypes of Neurodegeneration with a Physics-Informed Variational Autoencoder Mixture Model

Sanduni Pinnawala, Annabelle Hartanto, Ivor J. A. Simpson, Peter A. Wijeratne

This repository provides code, configurations, and environment setup to reproduce experiments from the paper.

Prerequisites

• All Python dependencies are listed in requirements.txt
• Docker is recommended for reproducible environment setup

Setup

1. Clone the repository

git clone https://github.com/sanpinnawala/BrainPhys.git
cd BrainPhys

2. Build Docker image

docker build -t brainphys .

4. Run Docker container

docker run --gpus all -v /absolute/path/to/project:/app -it brainphys

Running Experiments

1. Change to the model directory (e.g., reaction-diffusion-mixture-model)

cd reaction-diffusion-mixture-model

2. Create data, checkpoints, and artefacts directories inside the model directory

mkdir data checkpoints artefacts

Train

python src/train.py --config src/configs/config.yaml

Test

python src/test.py --config src/configs/config.yaml

Citation

@inproceedings{10.1007/978-3-032-05573-6_12,
  title = {Learning Mechanistic Subtypes of Neurodegeneration with a Physics-Informed Variational Autoencoder Mixture Model},
  booktitle = {Simulation and Synthesis in Medical Imaging},
  author = {Pinnawala, Sanduni and Hartanto, Annabelle and Simpson, Ivor J. A. and Wijeratne, Peter A.},
  year = {2026}
}

📬 Contact

For questions or collaborations, feel free to contact m.pinnawala@sussex.ac.uk