longitudinal-neighbourhood-embedding

Self-supervised Longitudinal Neighbourhood Embedding (LNE), Submitted to MICCAI2021. paper

Dependency

conda env create -f requirement.yml

Data Preprocessing

data_preprocessing_ADNI.py and data_preprocessing_LAB.py save images and other information in h5 files.

Self-supervised models

change parameters in config.yml (default setting is training LNE)
run python main.py

Downstream classification / regression

change parameters in config.yml
For setting use_feature: ['z'], data_type: single, run python main_classification_single.py
For setting use_feature: ['z', 'delta_z'], data_type: pair, run python main_classification_pair.py

Visualization

see visualization.ipynb for more details.