BrainIAC: A foundation model for generalized Brain MRI analysis

Overview

BrainIAC (Brain Imaging Adaptive Core) is vision based foundation model for generalized structural Brain MRI analysis. This repository provides the BrainIAC and downstream model checkpoints, with training/inference pipeline across all downstream tasks. Checkout the Paper

Installation

Prerequisites

• Python 3.9+
• CUDA 11.0+ (for GPU training)
• python >= 3.9

Setup Environment

# Clone the repository
git clone https://github.com/YourUsername/BrainIAC_V2.git
cd BrainIAC

# Create conda environment

# Create conda environment
conda create -n brainiac python=3.9
conda activate brainiac
pip install -r requirements.txt

Model Checkpoints

Download the BrainIAC weights and dowstream model checkpoints and place it in ./src/checkpoints/:

Download Link: Model Checkpoints

Quick Start

See quickstart.ipynb to get started on how to preprocess data, load BrainIAC to extract features, generate and visualize saliency maps. We provide data samples from publicly available UPENN-GBM License (with no modifications to the provided preprocessed images) and the Pixar License dataset in the sample_data directory.

Train and Infer Downstream Models

• Brain Age Prediction
• IDH Mutation Classification
• Mild Cognitive Impairment Classification
• Diffuse Glioma Overall Survival Prediction
• MR Sequence Classification
• Time to Stroke Prediction
• Tumor Segmentation

Application Demos

Test BrainIAC's downstream models on your own data with our interactive demos, hosted on hugging face spaces!

• Brain Age Prediction: Brain age prediction from structural T1 MRI scans

  • Upload T1w MRI scan to get the brain age

• Mild Cognitive Impairment (MCI) Classification: Mild cognitive impairment risk prediction from structural T1 MRI scans

  • Upload T1w MRI scan to get the MCI risk score

interactive demos coming soon for other downstream tasks!

Docker

For local deployment, run BrainIAC downstream model application using Docker:

IDH app

docker run -p 7860:7860 --name brainiac-idh-app divyanshutak1/brainiac-idh:latest

MCI app

docker run -p 7860:7860 divyanshutak1/brainiac-mci:latest

Citation

@article{tak2024brainiac,
    title={BrainIAC: A Foundation Model for Generalized Brain MRI Analysis},
    author={Tak, Divyanshu and others},
    journal={medRxiv},
    year={2024},
    doi={10.1101/2024.12.02.24317992}
}

License

BrainIAC is released under the CC BY-NC License. See LICENSE for details.