duyphuongcri/CLIMB

CLIMB: Controllable Longitudinal Brain Image Generation via Mamba-based Latent Diffusion Model

Table of Contents

• Installation
• Dataset
• Training
• Inference
• Pretrained models
• Acknowledgements

Installation

Clone the repository and install dependencies:

git clone https://github.com/duyphuongcri/CLIMB.git
cd CLIMB  
pip install -r requirements.txt

Dataset

This model uses the ADNI dataset. You can download it from here.

Training

Step1: Train autoencoder model

python train_autoencoder.py 

Step2: Extract latent features

python extract_latents.py

Step3: Train diffusion model conditioned on variables (age, gender, disease, status, biomarker,...)

python train_diffusion_variables.py

Step4: Train diffusion model with all conditional factors (variables and image features)

python train_diffusion_image_features.py

Step5: Train IRLSTM model (for predicting brain volumes structure and disease status at the projected age)

python train_irlstm.py 

Inference

Evaluating Model

python measure_performance.py 

Inference

python inference.py 

Running the program looks like this:

Pretrained models

Download the pre-trained models for CLIMB:

Acknowledgements

We thank the above repositories for their contributions and resources: BrLP, MONAI and its GenerativeModels extension.