Abstract
Early detection and monitoring of Parkinson's disease (PD) remain challenging, highlighting the need for accessible, cost-effective tools. Saccadic eye movement abnormalities are promising noninvasive biomarkers for PD screening and monitoring. Here, we present an iPad-based system that uses a deep learning algorithm to extract saccade metrics and validate these metrics against the clinical-grade EyeLink 1000 Plus. Twenty-five participants (10 with PD, 15 controls) completed pro-saccade, anti-saccade, memory-guided-saccade, and self-generated-saccade tasks. Relative to the EyeLink, the iPad system achieved average subject-level errors of 2 ms for latency and 0.7(∘) for amplitude in pro-, anti-, and memory-guided saccades, and 0.003 s(-1) for inter-saccadic rate and 1.6(∘) for amplitude in self-generated saccades. A review of 22 studies on PD-related saccadic impairments established benchmarks for clinically meaningful effects. The iPad-based system meets or exceeds these benchmarks, supporting its use as a scalable and cost-effective tool for screening and monitoring PD.