Unsupervised Learning of Eye State Prototypes for Semantically Rich Blinking Detection

Area of Science:

• Ophthalmology and Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Eye blinking is crucial for ocular health and offers diagnostic potential for neurological and muscle disorders.
• Current blink detection methods (open/closed states) lack detail on closure speed, duration, and percentage, limiting medical applications.
• Accurate detection of blink intervals in high-temporal resolution recordings is needed for advanced analysis.

Purpose of the Study:

• To develop a reliable method for detecting eye blink events and intervals using data-driven analysis.
• To establish an unsupervised eye state prototype for blink detection and inter-eye synchronicity measurement.
• To compare the efficacy of unsupervised versus manually defined prototypes for blink analysis.

Main Methods:

• Utilized data-driven analysis of the eye aspect ratio to detect blinking events.
• Developed an unsupervised eye state prototype to identify blink intervals.
• Measured inter-eye synchronicity at peak eye closure moments.
• Compared results from unsupervised and manually defined prototypes.

Main Results:

• Successfully demonstrated reliable detection of blinking events and intervals.
• Achieved precise measurement of inter-eye synchronicity, with results up to 4.16 milliseconds.
• Showed that manually defined prototypes yield comparable results to unsupervised methods.
• Validated the potential for high-temporal resolution blink analysis.

Conclusions:

• The developed data-driven method reliably detects eye blink intervals and synchronicity.
• Unsupervised and manual prototype methods offer comparable results for blink analysis.
• Precise blink metrics can be extracted, offering potential for novel diagnostic tools.
• Future applications include defining disease-specific blink prototypes for medical professionals.