Determination of Saccade Latency Distributions using Video Recordings from Consumer-grade Devices

Quantitative and accurate tracking of neurocognitive decline remains an ongoing challenge. We seek to address this need by focusing on robust and unobtrusive measurement of saccade latency - the time between the presentation of a visual stimulus and the initiation of an eye movement towards the stimulus - which has been shown to be altered in patients with neurocognitive decline or neurodegenerative diseases. Here, we present a novel, deep convolutional-neuralnetwork-based method to measure saccade latency outside of the clinical environment using a smartphone camera without the need for supplemental or special-purpose illumination. We also describe a model-based approach to estimate saccade latency that is less sensitive to noise compared to conventional methods. With this flexible and robust system, we collected over 11,000 saccade-latency measurements from 21 healthy individuals and found distinctive saccade-latency distributions across subjects. When analyzing intra-subject variability across time, we observed noticeable variations in the mean saccade latency and associated standard deviation. We also observed a potential learning effect that should be further characterized and potentially accounted for when interpreting saccade latency measurements.