Predicting Superagers by Machine Learning Classification Based on the Functional Brain Connectome Using Resting-State Functional Magnetic Resonance Imaging

Abstract

Abstract Superagers are defined as older adults who have youthful memory performance comparable to that of middle-aged adults. Classifying superagers based on the brain connectome using machine learning modeling can provide important insights on the physiology underlying successful aging. We aimed to investigate the unique patterns of functional brain connectome of superagers and develop predictive models to differentiate superagers from typical agers based on machine learning methods. We obtained resting-state functional magnetic resonance imaging (rsfMRI) data and cognitive measures from 32 superagers and 58 typical agers. The accuracies of three machine learning methods including the linear support vector machine classifier (SV), the random forest classifier (RF), and the logistic regression classifier (LR) in predicting superagers were comparable (SV = 0.944, RF = 0.944, LR = 0.944); however, RF achieved the highest area under the curve (AUC; 0.979). An ensemble learning method combining the three classifiers achieved the highest AUC (0.986). The most discriminative nodes for predicting superagers encompassed areas in the precuneus; posterior cingulate gyrus; insular cortex; and superior, middle, and inferior frontal gyrus, which were located in default, salient, and multiple-demand networks. Thus, rsfMRI data can provide high accuracy for predicting superagers, thereby capturing and describing the unique characteristics of their functional brain connectome.