Predicting Pure Amnestic Mild Cognitive Impairment Conversion to Alzheimer's Disease Using Joint Modeling of Imaging and Clinical Data

Abstract

Predicting the conversion of amnestic mild cognitive impairment (aMCI) to Alzheimer's disease (AD) is a challenging problem for which machine learning could be of great use. In this work, we aim at assessing the independent and joint value of imaging (structural MRI, resting-state functional MRI (rsfMRI)) and clinical data in classifying stable versus progressive aMCI. Surprisingly, we found no previous studies using rsfMRI to predict conversion of MCI to AD. We use singular value decomposition as a feature extractor before combining modalities. We reach accuracies of up to 82% using rsfMRI, 86% using sMRI and rsfMRI combined, and 77% using a combination of all modalities.