Predicting the apolipoprotein E 4 allele carrier status based on gray matter volumes and cognitive function

Background: Apolipoprotein E (ApoE) epsilon 4 carriers have a higher risk of developing Alzheimer's disease (AD) and show brain atrophy and cognitive decline even before diagnosis.Objective: To predict ApoE epsilon 4 status using gray matter volume (GMV) obtained from magnetic resonance imaging images and demographic data with machine learning (ML) methods.Methods: We recruited 74 participants (25 probable AD, 24 amnestic mild cognitive impairment, and 25 cognitively normal older people) with known ApoE genotype (22 ApoE epsilon 4 carriers and 52 noncarriers) and scanned them with three-dimensional (3D) T1-weighted (T1W) and 3D double inversion recovery (DIR) sequences. We extracted GMV from regions of interest related to AD pathology and used them as features along with age and mini-mental state examination (MMSE) scores to train different ML models. We performed both receiver operating characteristic curve analysis and the prediction analysis of the ApoE epsilon 4 carrier with different ML models.Results: The best model of ML analyses was a cubic support vector machine (SVM3) that used age, the MMSE score, and DIR GMVs at the amygdala, hippocampus, and precuneus as features (AUC = .88). This model outperformed models using T1W GMV or demographic data alone.Conclusion: Our results suggest that brain atrophy with DIR GMV and cognitive decline with aging can be useful biomarkers for predicting ApoE epsilon 4 status and identifying individuals at risk of AD progression.