Depression Recognition Using Machine Learning Algorithms With Eye Tracking, Visual Evoked Potentials, and Auditory P300 Among Chinese Medical Students
Open Access
- 1 January 2025
- journal article
- Published by Wiley in Depression and Anxiety
- Vol. 2025 (1) , 8637398
Abstract
Background: Current assessment of depression primarily relies on psychological scales. Although the use of machine learning in depression has grown, limited reports are available on multiple neurophysiological measurements. We employed machine learning algorithms incorporating eye tracking, visual evoked potentials (VEPs), and auditory P300 to classify depression among Chinese medical students. Methods: A total of 66 students with depression and 72 matched controls were recruited; eye tracking, VEPs, and auditory P300 data were collected. Descriptive analyses and group comparisons were performed between the depression and control groups. Then, multivariate logistic regression (LR) analysis was conducted to evaluate the relationship between eye tracking, VEPs, and auditory P300 features and Patient Health Questionnaire‐9 (PHQ‐9) scores. Furthermore, the study employed six classifiers to differentiate between depression and nondepression. Five‐fold cross‐validation was employed. Model performance was assessed using receiver operating characteristic (ROC) curves, area under the curve (AUC), precision, accuracy, recall, and F1 score. We applied SHapley Additive exPlanations (SHAP) values to explain the model. Results: Depression group was characterized by lower response search scores, higher D values, and prolonged P100 latencies in both eyes. No significant differences were observed in auditory P300 features. Random forest (RF) classifier demonstrated superior classification performance relative to the other five machine learning algorithms. Models utilizing combined features showed enhanced performance compared with those based solely on eye tracking or VEP features. Utilizing the SHAP method, we identified that P100 latency in the right eye was the most significant feature across all machine learning models. Conclusions: Chinese medical students with depression exhibited reduced responsive search scores and extended P100 latencies, suggesting impairments in attention and visual information processing associated with depression. The combined eye tracking and VEPs proved to be more effective than single features for distinguishing depression and nondepression. P100 latency in the right eye may be the most significant predictor of depression.Keywords
Funding Information
- National Natural Science Foundation of China (82301689)
This publication has 51 references indexed in Scilit:
- Higher depression scores in cognitively unimpaired individuals affect event‐related potentials similar to those of MCIAlzheimer's & Dementia, 2023
- Explanation of machine learning models using shapley additive explanation and application for real data in hospitalComputer Methods and Programs in Biomedicine, 2022
- Validity, Reliability, and Optimal Cut-off Scores of the WHO-5, PHQ-9, and PHQ-2 to Screen Depression Among University Students in IranInternational Journal of Mental Health and Addiction, 2021
- Cognitive neurophysiology: Event-related potentials2019
- The P300 component decreases in a bimodal oddball task in individuals with depression: An event-related potentials studyClinical Neurophysiology, 2018
- Prevalence of Depression, Depressive Symptoms, and Suicidal Ideation Among Medical StudentsJAMA, 2016
- The P300 component wave reveals differences in subclinical anxious-depressive states during bimodal oddball tasks: An effect of stimulus congruenceClinical Neurophysiology, 2015
- Depression and event-related potentials: emotional disengagement and reward insensitivityCurrent Opinion in Psychology, 2015
- Eye tracking of attention in the affective disorders: A meta-analytic review and synthesisClinical Psychology Review, 2012
- The effect of 5-HT1A receptor ligands in a chronic mild stress model of depressionNeuropharmacology, 1995