Multimodal large language models for women's reproductive mental health.
Review
Overview
abstract
BACKGROUND: Women's risk of mental health conditions fluctuates across the lifespan with hormone-mediated reproductive transitions. Reproductive psychiatry, a relatively new subspecialty, focuses on preventing and treating these conditions throughout various reproductive stages. Multimodal large language models (MLLMs) are advanced artificial intelligence (AI) systems that can process and integrate information across multiple modalities, including text, images, audio, and video. Although MLLMs have shown broad utility in healthcare, their potential in reproductive psychiatry remains largely unexplored. OBJECTIVE: To explore how MLLMs could advance research and clinical care in women's reproductive mental health and to outline opportunities, requirements, and barriers for safe, equitable deployment. METHODS: This perspective synthesizes the literature and domain expertise using a consistent analytical framework applied to each application domain in women's reproductive mental health: (1) define gaps in current clinical knowledge and practice; (2) explain why prevailing AI methods are insufficient; and (3) specify the distinctive advantages of MLLMs, including example data modalities and use cases relevant to reproductive psychiatry. FINDINGS: We identify seven application domains: (1) menstruation, (2) pregnancy, (3) abortion, miscarriage and recurrent pregnancy loss, (4) the postpartum period, (5) menopause, (6) psychiatric comorbidities in infertility, and (7) gynecologic conditions (e.g., endometriosis, polycystic ovary syndrome). Across these domains, MLLMs could enable multimodal risk stratification, longitudinal symptom trajectory modelling, clinical decision support, and patient-tailored education and self-management resources that adapt to evolving reproductive stages. Realizing these benefits requires addressing bias in training corpora; safeguarding privacy and consent for sensitive reproductive data; ensuring consistent, high-quality longitudinal data collection across life stages; and establishing standardized, well-governed multimodal repositories specific to women's health. CONCLUSIONS: MLLMs hold promise to foster more personalized and precise care in reproductive psychiatry. By mapping opportunities and constraints and proposing a structured evaluation lens, this perspective aims to inform clinicians and researchers, stimulate cross-disciplinary dialogue, and guide responsible development and integration of MLLMs in women's mental health.
