A Preliminary Study to discriminate aMCI and dMCI with Multiple Clinical Neuroimaging Characteristics Using Random Forests Classifier

Abstract

In this study, a classification scheme, using the features from resting-state functional MRI (rs-fMRI) and voxel-based morphometry (VBM), was proposed to discriminate two subtypes of mild cognitive impairment (MCI): amnestic MCI (aMCI) subtypes and dys executive MCI (dMCI) subtypes. More specifically, this scheme employed random forests (RF) algorithm to classify three study groups i.e., healthy controls (NC), aMCI, and dMCI. With the hybrid framework, the classification accuracy achieves 77.42% (AUC=0.8101) between aMCI and NC, and 82.14% (AUC=0.8473) between dMCI and NC. If comparing two MCI subtypes against each other, the accuracy can reach 79.57% (AUC=0.8410). The preliminary results suggest that pattern matching using the features from multiple modalities can achieve a clinically relevant accuracy for the a priori diagnosis in MCI subtypes.