Search

83% of those diagnosed with Parkinson's Disease (PD) eventually progress to PD with mild cognitive impairment (PD-MCI) followed by dementia (PDD) - suggesting a complex spectrum of pathology concomitant with aging. Biomarkers sensitive and specific to this spectrum are required if useful diagnostics are to be developed that may supplement current clinical testing procedures. We used video-based eye tracking and machine learning to develop a simple, non-invasive test sensitive to PD and the stages of cognitive dysfunction. From 121 PD (45 Cognitively Normal/45 MCI/20 Dementia/11 Other) and 106 healthy controls, we collected video-based eye tracking data on an interleaved pro/anti-saccade task. Features of saccade, pupil, and blink behavior were used to train a classifier to predict confidence scores for PD/PD-MCI/PDD diagnosis. The Receiver Operator Characteristic Area Under the Curve (ROC-AUC) of the classifier was 0.88, with the cognitive-dysfunction subgroups showing progressively increased AUC, and the AUC of PDD being 0.95. The classifier reached a sensitivity of 83% and a specificity of 78%. The confidence scores predicted PD motor and cognitive performance scores. Biomarkers of saccade, pupil, and blink were extracted from video-based eye tracking to create a classifier with high sensitivity to the landscape of PD cognitive and motor dysfunction. A complex landscape of PD is revealed through a quick, non-invasive eye tracking task and our model provides a framework for such a task to be used as a supplementary screening tool in the clinic. • Structured eye-tracking produces biomarkers sensitive to PD cognitive subtypes. • Saccade, pupil, and blink reveal a complex landscape of PD spectrum. • Functional data analysis can automatically suggest features for complex signals. • ROC-AUC of 0.88 for PD classification can be achieved through machine learning. [ABSTRACT FROM AUTHOR]