Predicting Visuo-Motor Diseases From Eye Tracking Data.

Area of Science:

• Ophthalmology and Neuroscience
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Eye tracking (oculography) offers insights into visual attention and motor behavior.
• Advances in camera technology and machine learning enable high-resolution gaze tracking on smartphones.
• A gap exists in scalable oculography for diagnosing visuo-motor dysfunction, specifically in task design, diagnostic algorithms, and dataset size.

Purpose of the Study:

• To develop and validate novel visual task paradigms and machine learning algorithms for diagnosing visuo-motor dysfunction using eye tracking.
• To assess the predictive power of oculography data for disease states in neurological conditions.
• To determine the minimum recording duration required for reliable disease detection.

Main Methods:

• Utilized a 500 Hz infrared oculography dataset from healthy controls and patients with neurological diseases.
• Developed novel visuo-motor tasks, including rapid reading of single-digit numbers.
• Created and applied a machine learning algorithm to predict disease state from oculography data.

Main Results:

• Oculography during a rapid reading task predicted visuo-motor dysfunction with high accuracy (ROC-AUC = 0.973).
• Short recordings (approx. 2.5 seconds) were sufficient for disease detection (ROC-AUC = 0.831).
• The system could discriminate between different disease categories (eye movement disorders, vision loss) and healthy controls with 81% accuracy.

Conclusions:

• Smartphone-based eye tracking with novel visual tasks and machine learning is a powerful tool for diagnosing visuo-motor dysfunction.
• Brief eye tracking recordings are sufficient for effective disease detection.
• The developed paradigms can differentiate between various neurological conditions affecting vision and eye movement.