AUTHOR=Liu Xing-Chang , Chen Ming , Ji Yu-Jia , Chen Hong-Bei , Lin Yu-Qiao , Xiao Zhen , Guan Qiao-Yan , Ou Wan-Qi , Wang Yue-Ya , Xiao Qiao-Ling , Huang Xin-Cheng-Cheng , Zhang Ji-Fan , Huang Ye-Kai , Yu Qian-Ting , Jiang Mei-Jun 

TITLE=Identifying depression with mixed features: the potential value of eye-tracking features

JOURNAL=Frontiers in Neurology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2025.1555630

DOI=10.3389/fneur.2025.1555630

ISSN=1664-2295

ABSTRACT=ObjectivesTo investigate the utility of eye-tracking features as a neurobiological marker for identifying depression with mixed features (DMF), a psychiatric disorder characterized by the presence of depressive symptoms alongside subsyndromal manic features, thereby complicating both diagnosis and therapeutic intervention.MethodsA total of 93 participants were included, comprising 41 patients with major depressive disorder (MDD), of whom 20 were classified as DMF, and 52 healthy controls (HC). Eye-tracking features were collected using an infrared-based device, and participants were evaluated using clinical scales including the Montgomery-Åsberg Depression Rating Scale (MADRS), Young Mania Rating Scale (YMRS), and Brief Psychiatric Rating Scale (BPRS). Performance of extreme gradient boosting (XGBoost) model based on demographic and clinical characteristics was compared with that of the model created after adding ocular movement data.ResultsSignificant differences were observed in certain eye-tracking features between DMF, MDD, and HC, particularly in orienting saccades and overlapping saccades. Incorporating eye-tracking features into the XGBoost model enhanced the predictive accuracy for DMF, as evidenced by an increase in the area under the curve (AUC) from 0.571 to 0.679 (p < 0.05), representing an 18.9% improvement. This suggests a notable enhancement in the model’s ability to distinguish DMF from other groups. The velocity of overlapping saccades and task completion time during free viewing were identified as significant predictive factors.ConclusionEye-tracking features, especially the velocity of overlapping saccades and free viewing task completion time, hold potential as non-invasive biomarkers for the identification of DMF. The integration of these parameters into the XGBoost machine learning model significantly improved the accuracy of DMF diagnosis, offering a promising approach for enhancing clinical decision-making in psychiatric settings.