Can brain connections help us understand Alzheimer’s disease? Leakage-Controlled Multimodal Connectome Stacking for Alzheimer's Disease-Related Clinical Status and Cognitive Outcomes

Abstract

This study investigates whether brain connectivity patterns can distinguish healthy aging from mild cognitive impairment (MCI) and dementia and predict cognitive performance. Using data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), we compared three representations of brain connectivity: a regional structural map containing 68 brain regions, a detailed structural map containing 5,124 locations, and a detailed functional map containing 5,124 locations. Predictive models integrated these connectivity measures with brain-volume and participant information and were evaluated using participants held out from model training. All three connectivity representations distinguished dementia from healthy aging relatively well, with the detailed functional map producing the strongest result. In contrast, none of the map types reliably distinguished MCI from healthy aging, indicating that earlier-stage cognitive decline remains more difficult to detect. Prediction of cognitive test scores was modest overall, although ADAS13 was the most predictive measure. Combining information from multiple model branches improved performance in eight of nine cognitive-score analyses. These findings demonstrate that MRI-derived connectivity maps contain meaningful signals related to dementia while also highlighting important limitations in detecting MCI and predicting individual cognitive outcomes.