Towards Affordable Smartphone Eye Tracking for Nystagmus Analysis and Monitoring

Area of Science:

• Ophthalmology
• Neurology
• Computer Science

Background:

• Nystagmus is a key indicator for neurological and vestibular disorders.
• Current clinical eye tracking is costly and requires specialized training, limiting widespread use.

Purpose of the Study:

• To develop and evaluate a smartphone-based eye tracking pipeline for nystagmus analysis.
• To assess the efficacy of deep learning models for pupil tracking in nystagmus detection.

Main Methods:

• Utilized two U-Net deep learning models (3-class and 4-class) for pupil segmentation on the MOBIUS dataset.
• Applied and evaluated the models on 12 optokinetic nystagmus (OKN) videos.
• Compared smartphone tracking performance against the ICS Impulse system.

Main Results:

• The 3-class U-Net model outperformed the 4-class model in segmentation metrics (cross-entropy loss, IoU, DICE score).
• Both models demonstrated reasonable accuracy in capturing nystagmus movements (avg R² of 0.755 and 0.768).
• No significant statistical difference was found between the 3-class and 4-class models for tracking performance (p-value 0.80).

Conclusions:

• Smartphone-based eye tracking presents a viable, cost-effective tool for objective nystagmus assessment.
• The proposed pipeline can enhance diagnostic accuracy and accessibility for neurological and vestibular disorders.
• Further refinements are necessary to optimize tracking performance for clinical applications.