BACKGROUND
Depression is a major psychiatric disorder threatening people’s health. Presently, existing medical guidelines‘ diagnostic criteria are largely qualitative and potentially contribute to over-diagnosis and misdiagnosis.
OBJECTIVE
This study aims to present a method of depression recognition based on eye movement for clinical use.
METHODS
120 participants performed smooth pursuit eye movement (SPEM) and visual fixation (VF) tasks while their eye movements were recorded using iViewX RED 500. Data from 106 participants (53 in depression group VS 53 control group) passed quality control and were included and extracted by the Matlab program platform with machine learning analysis.
RESULTS
Depressed subjects showed significantly different eye movement from controls. In SPEM task, depressed patients had significantly greater RMSE (P<0.001), higher blink frequency (P<0.001), and greater gain value (P<0.001), but significant lower saccade frequency (P<0.001) than control group. In VF task, depressed patients
had significantly greater RMSE (P<0.01), higher blink frequency (P<0.05), and greater saccade amplitude (P<0.001) than those in control group, but had significant lower fixation frequency (P<0.001) and shorter fixation time (P<0.001). Accuracy of identifing depression reached 88.68% (sensitivity 92.31% , specificity 85.19%).
CONCLUSIONS
Characterized eye-tracking patterns have been identified in depressed subjects via SPEM and VF tasks, as compared to healthy subjects. Recognition accuracy of depression by machine learning reached 88.68%. Future studies should validate these results in larger samples and in clinical populations.
CLINICALTRIAL
the Ethics Committee of Shandong University of Traditional Chinese Medicine.