AI-Driven Prediction of Possible Mild Cognitive Impairment Using the Oculo-Cognitive Addition Test (OCAT)

Area of Science:

• Ophthalmology
• Neurology
• Artificial Intelligence
• Cognitive Science

Background:

• Mild cognitive impairment (MCI) diagnosis is challenging due to insensitive brief exams and time-consuming detailed testing.
• The Oculo-Cognitive Addition Test (OCAT) offers a rapid, objective method using eye tracking during mental tasks.
• Current screening tools lack the speed and accessibility needed for widespread early detection.

Purpose of the Study:

• To develop and validate AI-derived predictive models for early MCI detection using OCAT features.
• To assess the efficacy of eye movement and time-based OCAT data in identifying individuals at risk for MCI.
• To establish OCAT as a practical, scalable cognitive screening tool.

Main Methods:

• 250 patients completed the OCAT with integrated eye tracking.
• Time-related and eye movement features were extracted from gaze data.
• Logistic Regression (LR) and K-nearest neighbors (KNN) models were trained using feature selection (Random Forest, decision trees) to predict Dementia Rating Scale (DRS) outcomes.

Main Results:

• LR models using combined time and eye movement features achieved high performance (accuracy=0.97, recall=0.91, AUPRC=0.95).
• Models using only eye or time features also showed good performance (accuracy=0.93).
• Combined feature models consistently outperformed single-feature set models in predicting DRS outcomes.

Conclusions:

• Machine learning models trained on OCAT data can reliably predict cognitive function (DRS PASS/FAIL).
• OCAT shows significant potential as a practical and scalable tool for early MCI identification.
• The OCAT system is suitable for diverse clinical settings, including remote and resource-limited environments.