112. Ai-based prediction and personalized treatment of student depression

Abstract Background Traditional campus psychological screening relies on subjective self-assessment, making it susceptible to missed diagnoses due to the influence of social desirability. With the development of affective computing technology, objective identification using multimodal data has become possible. Existing research largely stops at model prediction, lacking closed-loop validation of “prediction-intervention.” This study constructs a dynamic prediction model of depression risk based on multimodal data and implements personalized behavioral activation therapy based on it. By monitoring objective biological indicators, it verifies the effectiveness of this artificial intelligence (AI)-assisted system in improving the depressive state of high-risk students from a physiological perspective. It is hoped that this will provide a new and precise pathway for campus psychological crisis intervention. Methods Using multimodal data from a smart campus, 200 students at high risk of depression were screened and randomly divided into an AI intervention group (n = 100) and a control group (n = 100). The control group received standard psychological counseling, while the AI group wore a smart device connected to the “AI Mental Health System,” which provided personalized micro-intervention instructions (e.g., sleep adjustments, exercise recommendations) based on real-time physiological and behavioral data for 12 weeks. Objective indicators included: (1) serum brain-derived neurotrophic factor (BDNF) concentration to assess neuroplasticity; (2) deep sleep ratio; and (3) vagal tone index (VTI). The intervention effects were analyzed using a mixed linear model. Results Data analysis showed that at baseline, both groups had low BDNF levels. After intervention, the serum BDNF level in the AI intervention group showed a statistically significant increase. In contrast, the experimental group showed a 40.0% decrease in cortisol, a 41.8% increase in HRV-SDNN, and a 52.5% increase in RMSSD, with all interaction effects p<.001; the control group showed a 7.8% decrease in cortisol, a 3.4% increase in SDNN, and a 3.7% increase in RMSSD, confirming the clear biological validity of the AI-based closed-loop intervention strategy in improving the physiological state of depression. Detailed data are shown in Table 1 below. Discussion The research results confirm that AI-based personalized intervention can significantly improve the biological prognosis of students at high risk of depression. The improvement in BDNF levels indicates that AI-guided behavioral activation not only improves emotional experience but also reverses underlying neurobiochemical changes. This 24/7, low-threshold micro-intervention model can effectively compensate for the temporal and spatial limitations of traditional psychological counseling, breaking the cycle of depression through precise behavioral correction. Future university mental health systems should integrate such digital closed-loop systems to achieve a paradigm shift from passive screening to proactive and precise intervention.