ReMiND: Recovery of Missing Neuroimaging using Diffusion Models with Application to Alzheimer's Disease

Evnironment

conda create -n remind
conda activate remind
pip install -r requirements.txt

Dataset

Data Collection and Preparation

Please collect your MRIs from ADNI. ReMiND use the ANTs Longitudinal-SST for MRI preprocessing.

Train/Val/Test splits

ReMiND takes train/val-list.txt for model training and test-list.txt for generation. Please organize your splits into the following format:

<past-visit-mri-path> <current-visit-mri-path> <future-visit-mri-path> <past-visit-id> <current-visit-id> <future-visit-id> <past-visit-stage> <current-visit-stage> <future-visit-stage>
# stage could be NL, MCI, or AD
...

Model Training

Please refer to the following script for ReMiND-PF training:

CUDA_VISIBLE_DEVICES=0 python main.py --config configs/adni_remind_pf.yml --exp remind-pf --ni --data_path <path-to-your-data-root>

Training hyperparameters can be tuned by modifying ./configs/adni_remind_pf.yml:

• data.data_root: path to your dataset
• data.train_list_path: path to the training split file
• data.val_list_path: path to the validation split file
• data.test_list_path: path to the test split file
• data.channels: number of segments
• data.cont_local: number of slices in a local clip

Pretrained Model

The pretrained ReMiND-PF is provided in here.

Model Inference

Please refer to the following script for missing MRI interpolation:

python generation.py \
--test-listpath <path-to-the-test-list> \
--checkpoint-path <path-to-your-checkpoint> \ # the config.yml should be in the same folder as the checkpoint
--save-as-nii 

The interpolated MRI will be saved at ./interpolated-<current-visit-name>.nii

Acknowledgement

This codebase is built on top of mcvd-pytorch. Thanks for their excellent contribution.

Reference

Please cite our paper as

@article{yuan2023remind,
  title={Remind: Recovery of missing neuroimaging using diffusion models with application to alzheimer’s disease},
  author={Yuan, Chenxi and Duan, Jinhao and Tustison, Nicholas J and Xu, Kaidi and Hubbard, Rebecca A and Linn, Kristin A},
  journal={medRxiv},
  year={2023},
  publisher={Cold Spring Harbor Laboratory Preprints}
}