Clinical availability of eye movement during reading

Abstract

Eyes provide valuable information for neurological diagnosis. So far, the use of diagnostic devices to analyze eye movement is limited. We explored whether the analysis of eye movements can be efficacious. Patients with Parkinson's disease (PD) (n = 29), spinocerebellar degeneration (SCD) (21), progressive supranuclear palsy (PSP) (19), and control individuals (19) participated in this study. The patients read aloud two sets of sentences displayed on a monitor: one was displayed horizontally, and the other vertically. Parameters such as eye movement speed, travel distance, and fixation/saccade ratio were extracted, and comparisons between groups were performed. Maneuvers of eye movements were also subjected to image classification using deep learning. Reading velocity and fixation/saccade ratio were altered in the PD group, and the SCD group exhibited ineffective eye movements due to dysmetria and nystagmus. Vertical gaze parameters showed aberrant values in the PSP group. Vertical written sentences were more sensitive than horizontal ones in detecting these abnormalities. In the regression analysis, vertical reading indicated a high accuracy in identifying each group. The machine learning analysis showed more than 90 % accuracy in distinguishing between the control and SCD groups and between the SCD and PSP groups. Analyzing eye movements is useful and easily applicable.

Keywords: Deep learning; Eye tracking; Horizontal and vertical reading; Parkinsonian and cerebellar syndromes.