Eye-Rubbing Detection Using a Smartwatch: A Feasibility Study Demonstrated High Accuracy With Machine Learning.

Area of Science:

• Ophthalmology
• Machine Learning
• Wearable Technology

Background:

• Eye rubbing can worsen ectatic disorders like keratoconus, potentially leading to blindness.
• Accurate detection and prevention of eye rubbing are crucial in ophthalmology.
• Smartwatches offer a potential platform for monitoring hand-face interactions.

Purpose of the Study:

• To develop and evaluate a machine learning method for detecting eye rubbing using smartwatch sensor data.
• To leverage transformer neural networks for accurate identification of eye-rubbing behavior.
• To establish a foundation for monitoring hand-face interactions via wearables.

Main Methods:

• Utilized a transformer neural network architecture, known for its success in natural language processing.
• Collected a new dataset of smartwatch sensor data capturing various hand-face interactions.
• Evaluated the proposed method against established baseline algorithms.

Main Results:

• Achieved over 80% accuracy in eye-rubbing detection.
• Demonstrated improved accuracy up to 97% with moderate user-specific fine-tuning (3 hours).
• Showcased minimal fine-tuning (20 minutes) yielding significant detection capabilities.

Conclusions:

• Eye rubbing is detectable and distinguishable from other hand gestures using only a wrist-worn device.
• This research provides a proof-of-concept for smartwatch-based eye-rubbing detection.
• The findings support further studies in hand-face interaction monitoring and keratoconus management.