Advancing PTSD Treatment and Understanding with Multimodal AI Framework

Post-traumatic stress disorder (PTSD) affects 3.6% of the global population, with higher rates among women (5.2%) than men (1.8%) and up to 30% in conflict-exposed groups (WHO, 2022). Traditional therapies are effective but limited by the availability of trained professionals. Narrative exposure therapy helps reframe traumatic experiences but faces challenges in delivering personalized, consistent, and interactive care to large populations (Neuner et al., 2018). However, delivering personalized, consistent, and interactive NET while addressing resource limitations and maintaining emotional responsiveness to large populations remains a significant challenge (Neuner et al., 2018). Recent advances in multimodal Large Language Models (LLMs), combined with AI-driven infrastructure, offer transformative potential for augmenting PTSD treatments by the possibility of automating therapeutic workflows, enhancing patient interaction, and ensuring contextual, personalized care (Devlin et al., 2019). This seed project proposes a pilot study to build a multimodal AI-based infrastructure that supports the qualitative analysis in NET. The proposed system will incorporate autonomous AI agents, each specialized in tasks such as narrative analysis and physiological biomarker monitoring, working collaboratively within a secure, privacy-preserving infrastructure to ensure high-quality, personalized therapeutic development (Rieke et al., 2020). By integrating multimodal data inputs—such as text (patient-anonymized narratives) and anonymized physiological biomarker data (heart rate variability)—the project aims to be able (1) to deliver improvements in emotionally responsive, adaptive care by dynamically tailoring interventions to individual patients (Insel, 2017), (2) providing a holistic understanding of patient needs through AI-driven multimodal analysis, and enabling adjustments of therapeutic strategies based on evolving patient responses.