Eye-Rubbing Detection Tool Using Artificial Intelligence on a Smartwatch in the Management of Keratoconus

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Eye rubbing is a significant risk factor for the progression of keratoconus and post-refractive surgery corneal ectasia.
• Current methods for assessing eye rubbing are subjective and lack objective quantification.
• There is a need for a non-invasive, objective tool to monitor and manage eye rubbing behavior.

Purpose of the Study:

• To develop and validate an innovative solution for objectively quantifying and preventing eye rubbing.
• To introduce a deep-learning artificial intelligence (AI) algorithm deployed on a smartwatch for eye rubbing detection.
• To provide a familiar and accessible device for patients to monitor their eye rubbing habits.

Main Methods:

• Motion data, including gyroscope and accelerometer readings, were collected from a Samsung Galaxy Watch 4 during eye rubbing and daily activities.
• Two deep-learning algorithms (LSTM, GRU) and four machine learning algorithms (Random Forest, KNN, SVM, XGBoost) were trained to detect eye rubbing.
• The algorithms were evaluated for their accuracy in recognizing and quantifying eye rubbing events.

Main Results:

• The developed AI model achieved a high accuracy of 94% in detecting eye rubbing.
• The application successfully recognized, counted, and displayed the number of eye rubbing instances.
• The Gated Recurrent Unit (GRU) and XGBoost algorithms demonstrated particularly promising performance.

Conclusions:

• Automated eye rubbing detection using deep-learning AI is feasible and accurate.
• This technology offers a potential breakthrough in managing patients with keratoconus and those who have undergone refractive surgery.
• The tool can quantify eye rubbing and alert patients, aiding in the reduction of this detrimental behavior.