This application compares various models for the classification of stable vs. progressive MCI. Progressive MCI is defined as conversion to dementia within 24 months. Data was sourced from the ADNI dataset, specifically the ADNIMERGE preprocessed and pre-curated table.
Run create_kfolds.py
to divide data into 10x10 nested stratified k-folds.
univariate_analyses.ipynb
performs some traditional statistical analyses to measure the influence of site in each feature.
model_comparison.ipynb
compares a conventional neural network with domain adversarial and mixed effects models.
model_comparison_probe_features.ipynb
adds synthetic confounded features to the dataset and compares each model's ability to separate truly informative from known confounded features, in terms of feature importance.