Schizophrenia recognition via eye movement features in video paradigm.

Schizophrenia is a severe mental disorder characterized by abnormal eye movements. However, existing methods for detecting these abnormalities rely primarily on static stimuli that lack ecological validity. This study explores the potential of video paradigms for capturing eye movement patterns specific to schizophrenia. Forty patients with schizophrenia (SZ) and forty healthy controls (HCs) completed eye movement tests and cognitive assessments. The participants watched two types of videos: an animated video with low cognitive load and a documentary with high social cognitive load, from which ten eye movement features were extracted. Group differences were compared using analysis of covariance (ANCOVA). Support vector machine (SVM) and random forest (RF) algorithms constructed the classification models, employing leave-one-out cross-validation to optimize performance. Compared to HCs, the SZ group showed significant differences in eye movement metrics, including longer fixation durations, smaller saccade amplitudes, and a reduced pupil size ratio. Furthermore, the SZ group exhibited greater variability in fixation distribution, particularly during the documentary video. When combining eye movement features from both videos, the SVM algorithm achieved a classification accuracy of 84%, while the RF algorithm reached 80%. The video-based eye movement paradigm effectively detected abnormalities in the SZ group, highlighting its potential for objective schizophrenia detection.