AI-Supported Mental Health Assessment: A Systematic Review and Integrative Framework

Abstract

Mental health disorders now affect more than one billion people globally, yet the infrastructure for timely, accurate, and equitable assessment remains critically underdeveloped, especially in low- and middle-income countries (LMICs). Artificial intelligence (AI) can be used to fill assessment gaps on a scale with transformational potential. This is a systematic review of the empirical evidence on AIbased mental health assessment instruments, an assessment of their psychometric quality and clinical utility, as well as an integration framework across technical, measurement, clinical, and ethical aspects. A systematic review guided by PRISMA 2020 was carried out in Scopus, Web of Science, PubMed, and PsycINFO (2010-2025). The Population Concept Context (PCC) framework was used to determine the eligibility criteria. Narrative synthesis was employed given the heterogeneity of included studies. Among 3,412 records identified, 68 studies were eligible. Five thematic domains were identified: (1) AI-enhanced screening and diagnostic support; (2) computerised adaptive and algorithmically optimised testing; (3) natural language processing-based assessment; (4) passive digital phenotyping; and (5) ethical, psychometric, and governance considerations. AI models showed encouraging predictive validity (AUC values: 0.74-0.93), although evidence on psychometric strength, cross-cultural validity, and fair implementation was inconsistent. While AI holds substantial potential in mental health assessment, clinical practice integration should be more closely aligned with psychometric science, incorporate explicit human-in-the-loop practices, and be governed by robust ethical oversight. A five-domain integrative framework and minimum reporting checklist are proposed to guide future research and policy.