AI-Powered Detection of Eye Rubbing Behavior for Keratoconus Analysis Using Wearable Devices

Eye rubbing is widely believed to significantly contribute to the development and progression of keratoconus, a potentially blinding disorder of the cornea. Keratoconus involves corneal apical thinning and scarring and can result in painful episodes of swelling of the cornea termed corneal hydrops. As a result, keratoconus can necessitate corneal transplantation for visual recovery and can result in irreversible vision loss in patients who are not candidates for corneal transplantation. Accurately quantifying eye rubbing behavior and rigorously assessing its impact on keratoconus progression are critical needs. Currently, eye rubbing assessment in individuals at risk for keratoconus onset or progression relies mainly on self-reports and occasional clinical observations, which are insufficient to quantify the effects of eye rubbing on keratoconus progression. In this paper, we introduce an innovative deep learning-based analysis framework for real-time, automatic detection of eye rubbing using sensor data from wrist-based wearable devices. To develop and validate the proposed system, a dataset of 1,068 non-overlapping recordings from 9 subjects was built. Our results show that the framework effectively identifies eye rubbing behaviors during normal daily activities, achieving an average F1-score of 93.55% and AUC of 96.23%. This system offers a valuable tool for ophthalmologists and researchers by providing rapid, reliable quantification of eye rubbing behavior in patients.