An AI-based intelligent diagnosis system for adolescent mental health based on multitask deep learning

Area of Science:

• Mental Health Technology
• Artificial Intelligence in Healthcare
• Adolescent Psychology

Background:

• Adolescent depression and anxiety rates in China are high (20-30%), exacerbated by academic pressure and somatization.
• Traditional screening tools (PHQ-9, GAD-7) face limitations like subjective bias, recall errors, and stigma-related underreporting.

Purpose of the Study:

• To develop and validate an AI-based intelligent diagnosis system (IDS) for non-intrusive screening of comorbid depression and anxiety in Chinese adolescents.
• To leverage multitask deep learning on spontaneous textual expressions for predicting mental health severity.

Main Methods:

• Collected textual responses from ~1,275 adolescents, labeled with clinician-assessed PHQ-9 and GAD-7 scores.
• Employed jieba segmentation and VAE-based data augmentation for preprocessing and addressing class imbalance.
• Utilized a Chinese-optimized BERT encoder with self-attention and dual-feature fusion within a multitask deep learning framework.

Main Results:

• Achieved high performance on the test set: Pearson correlations of 0.706 (PHQ-9) and 0.693 (GAD-7); AUCs of 0.877 (PHQ-9) and 0.902 (GAD-7).
• Binary classification yielded F1-scores of 0.762 (PHQ-9) and 0.863 (GAD-7).
• Multitask learning framework improved F1-scores by 6.2%-7.8% and reduced MSE by 14.2%-18.4% compared to single-task baselines.

Conclusions:

• The AI-based IDS provides a robust, culturally sensitive, and scalable tool for adolescent mental health screening.
• It offers a proactive, privacy-preserving alternative to traditional self-report measures.
• Future work includes longitudinal validation, multimodal integration, and ethical deployment strategies.