Developing an AI-Enhanced Individualized Prediction Tool for Psychopathological Symptoms in Vietnam: A Study Protocol

Artificial intelligence (AI) is increasingly leveraged in mental healthcare for early detection, monitoring, and personalized intervention. However, most existing AI applications are based on categorical diagnostic systems like DSM-5 or ICD-11, which often lead to comorbidity issues, ambiguous diagnoses, and insufficient personalization. These tools typically target specific disorders (e.g., depression or anxiety), neglecting the broader, interconnected nature of psychopathological symptoms. Addressing these limitations, recent innovations in psychopathology emphasize transdiagnostic and network-based approaches, such as the Hierarchical Taxonomy of Psychopathology (HiTOP), which conceptualize mental disorders as dimensional and interconnected constructs. This study proposes an AI-powered tool that integrates data-driven principles from both the HiTOP and symptom network models to generate individualized risk profiles for internalizing mental disorders (e.g., depression, anxiety, bipolar disorders). Our solution aims to assess individual's current psychopathological traits and symptom components, providing a comprehensive, nuanced profile supporting clinical diagnoses and monitoring. The study unfolds in three phases: (1) model ideation; (2) model implementation in a large-scale Vietnamese sample; and (3) deployment in clinical and psychological practice settings in Vietnam. Central to our method is the development of a Risk-aware Taxonomy-enhanced Symptom Encoder (RiTaSE), which encodes symptom data and their severities into rich representations processed via a Transformer-based model. The model is trained using high-quality, validated datasets mapped to the HiTOP framework. This project is among the first to employ AI for personalized psychopathological profiling in Vietnam,, as well as other low- and middle-income countries. Expected outcomes include an advanced diagnostic-support tool for clinical use, improved cross-cultural insights into symptom comorbidity, and practical utility in mental health monitoring and intervention evaluation. Future extensions aim to broaden the scope across all HiTOP dimensions and predict transitions to clinical states through longitudinal and multi-modal data integration.