A functional neuroimaging biomarker of mild cognitive impairment using TD-fNIRS

Abstract

INTRODUCTION: Diagnostic assessments of mild cognitive impairment (MCI) are lengthy and burdensome, highlighting the need for new tools to detect MCI. Time-domain functional near-infrared spectroscopy (TD-fNIRS) can measure brain function in clinical settings and may address this need. METHODS: MCI patients (n=50) and age-matched healthy controls (HC; n=51) underwent TD-fNIRS recordings during cognitive tasks (verbal fluency, N-back). Machine learning models were trained to distinguish MCI from HC using neural activity, cognitive task behavior, and self-reported impairment as input features. RESULTS: Significant group-level differences (MCI vs HC) were demonstrated in self-report, N-back and verbal fluency behavior, and task-related brain activation. Classifier performance was similar when using self-report (AUC=0.76) and self-report plus behavior (AUC=0.79) as input features, but was strongest when neural metrics were included (AUC=0.92). DISCUSSION: This study demonstrates the potential of TD-fNIRS to assess MCI with short brain scans in clinical settings.