Machine Learning Detects Distinct Subtypes of Minimal Cognitive Impairment

Minimal cognitive impairment (MCI), a potential precursor to Alzheimer’s disease (AD), may be a heterogeneous entity consisting of distinct subtypes. To evaluate this hypothesis, we applied unsupervised machine-learning to a subset of the Alzheimer’s disease Neuroimaging Initiative (ADNI) data set, and detected MCI subtypes with distinct clinical correlates. Our subtype-detection system consists of preprocessing, clustering, validation, and visualization modules. We applied this system to data from MCI subjects in the ADNI-2 cohort. The resulting six subtypes demonstrated different profiles with respect to cognitive and laboratory assessment, potentially indicating differing clinical trajectories and treatment responses.