Classification and staging of Parkinson's disease using video-based eye tracking

Area of Science:

• Ophthalmology
• Neurology
• Biomarkers

Background:

• Parkinson's Disease (PD) often progresses through stages of cognitive impairment, including mild cognitive impairment (PD-MCI) and dementia (PDD).
• Current diagnostic methods may not fully capture the spectrum of PD-related cognitive dysfunction.
• Sensitive and specific biomarkers are needed for early and accurate diagnosis.

Purpose of the Study:

• To develop a simple, non-invasive eye-tracking test using machine learning.
• To assess the test's sensitivity and specificity in detecting PD and its cognitive stages (PD-MCI, PDD).
• To establish eye-tracking biomarkers for PD diagnosis and cognitive assessment.

Main Methods:

• Collected video-based eye-tracking data from 121 PD patients and 106 healthy controls during a pro/anti-saccade task.
• Extracted features related to saccade, pupil, and blink behavior.
• Trained a machine learning classifier to predict diagnostic confidence scores for PD, PD-MCI, and PDD.

Main Results:

• The classifier achieved an overall ROC-AUC of 0.88.
• AUC was progressively higher for cognitive dysfunction subgroups, reaching 0.95 for PDD.
• The classifier demonstrated 83% sensitivity and 78% specificity, predicting PD motor and cognitive performance.

Conclusions:

• Video-based eye-tracking biomarkers can effectively identify the spectrum of PD cognitive and motor dysfunction.
• This quick, non-invasive eye-tracking task offers a promising supplementary screening tool for clinical use.
• The developed model provides a framework for utilizing eye-tracking in PD diagnostics.