Machine Learning Integration of Eye-Tracking and Cognitive Screening for Detecting Cognitive Impairment.

Area of Science:

• Neuroscience
• Cognitive Science
• Medical Technology

Background:

• Cognitive impairment is a frequent challenge in Post-COVID-19 Condition (PCC).
• Comprehensive neuropsychological testing for PCC is time-consuming and resource-intensive.
• Eye movements can reflect underlying cognitive alterations, suggesting potential for eye-tracking (ET) as a screening tool.

Purpose of the Study:

• To investigate the efficacy of eye-tracking (ET) metrics in predicting neuropsychological test scores in individuals with PCC.
• To evaluate machine and deep learning models for predicting cognitive performance using ET data.
• To assess the added value of integrating ET data with the Montreal Cognitive Assessment (MoCA) for enhanced screening.

Main Methods:

• Collected ET data from 172 participants with PCC during visual tasks assessing eye movements and pupil responses.
• Administered standard neuropsychological tests to evaluate cognitive function.
• Employed machine learning (Random Forest, XGBoost) and deep learning models to predict cognitive performance from ET data and MoCA scores.

Main Results:

• Predicting individual neuropsychological test scores from ET data alone was challenging.
• Combining ET metrics with MoCA scores significantly improved the prediction of a global cognitive composite measure.
• The integrated ET-MoCA model achieved 87% sensitivity and 60% specificity, outperforming MoCA alone.

Conclusions:

• Eye-tracking (ET) demonstrates potential as a rapid, non-invasive screening support tool for cognitive deficits in Post-COVID-19 Condition.
• Integrating ET with existing cognitive screening tools like MoCA can enhance diagnostic accuracy.
• Further research into ET-based cognitive assessment could alleviate the burden of traditional neuropsychological testing.