Predicting Alcohol Dependence

This work was published in the journal of Human Brain Mapping in October of 2020, and can be found here.

This project was to identify neuroimaging biomarkers of alcohol dependence (AD) from structural magnetic resonance imaging. Data were grouped into a training set used for internal validation including 1,652 participants (692‚ÄČAD, 24 sites), and a test set used for external validation with 382 participants (146‚ÄČAD, 3 sites).

This project included an exploratory data analysis, followed by a novel evolutionary search based feature selection procedure, designed to select to highest performing and more generalizable subset of brain measurements.

Data distrubtion
This project sourced a quite varied collection of data from different sites, with very different underlying distributions of case to control.
a. The top 15 features (threshold chosen for readability), as ranked by average weighted feature importance (where 0 indicates a feature appeared in none of the GA final models, and 1 represents a feature appeared in all) are shown. b. The cortical thickness and b. cortical average surface area feature importance scores, above an a priori selected threshold of 0.1, are shown as projected onto the fsaverage surface space.