Automating the analysis of eye movement for different neurodegenerative disorders

Abstract

AbstractThe clinical observation and assessment of extra-ocular movements is common practice in assessing neurological disorders but remains observer-dependent and subjective. In the present study, we propose an algorithm that can automatically identify saccades, fixation, smooth pursuit, and blinks using a non-invasive eye-tracker and, subsequently, elicit response-to-stimuli-derived interpretable features that objectively and quantitatively assess patient behaviors. The cohort analysis encompasses persons with mild cognitive impairment (MCI) and Alzheimer’s disease (AD), Parkinson’s disease (PD), Parkinson’s disease mimics (PDM), and controls (CTRL). Overall, results suggested that the AD/MCI and PD groups exhibited significantly different saccade and pursuit characteristics compared to CTRL when the target moved faster or covered a larger visual angle during smooth pursuit. When reading a text passage silently, more fixations were an AD/MCI-specific feature. During visual exploration, people with PD demonstrated a more variable saccade duration than other groups. In the prosaccade task, the PD group showed a significantly smaller average hypometria gain and accuracy, with the most statistical significance and highest AUROC scores of features studied. The minimum saccade gain was a PD-specific feature distinguishing PD from CTRL and PDM. Furthermore, the PD and AD/MCI groups displayed more omitted antisaccades and longer average antisaccade latency than CTRL. These features, as oculographic biomarkers, can be potentially leveraged in distinguishing different types of NDs in their early stages, yielding more objective and precise protocols to monitor disease progression.