Facial Recognition Task for the Classification of Mild Cognitive Impairment with Ensemble Sparse Classifier

Area of Science:

• Neuroscience
• Biomarkers
• Machine Learning

Background:

• Mild cognitive impairment (MCI) detection often relies on time-consuming and costly methods like cognitive exams and neuroimaging.
• There is a significant need for objective, cost-effective biomarkers for early MCI detection.

Purpose of the Study:

• To investigate the efficacy of familiar, unfamiliar, and inverted faces as visual stimuli for early MCI detection using electroencephalography (EEG).
• To explore the use of advanced machine learning techniques to analyze entire event-related potentials (ERPs) as biomarkers for MCI.

Main Methods:

• Employed a sequential imaging oddball paradigm with familiar, unfamiliar, and inverted faces.
• Utilized an ensemble of sparse classifier (ESC) machine learning technique to analyze full ERPs, unlike traditional methods focusing on specific deflection points.
• Recorded EEG data from five MCI subjects and eight age-matched controls after MoCA (Montreal Cognitive Assessment) exams.

Main Results:

• Traditional time-domain analysis of averaged ERPs showed no statistical significance between groups.
• The ESC machine learning model achieved 95% classification accuracy in discriminating MCI from controls using ERPs elicited by familiar faces.

Conclusions:

• Advanced machine learning techniques, like ESC, can effectively analyze ERPs for MCI detection.
• ERPs elicited by familiar faces show promise as specific biomarkers for accurate MCI diagnosis.