Symposium
Artificial Intelligence and Technology-based Interventions
Tomas Meaney, B.S., M.S., Ph.D. (he/him/his)
post-Doctorate
University of New South Wales
Sydney, New South Wales, Australia
Isaac Galatzer-Levy, PhD, MSC, BA
Associate Professor
New York University
New York City, New York, United States
Vijay Yadav, PhD, MSC, BA
Adjunct Professor
New York University
New York City, New York, United States
Richard Bryant, B.A., Ph.D., PsyM
Scientia Professor, head of Traumatic Stress Clinic
University of New South Wales
Sydney, New South Wales, Australia
PTSD is a well-established psychological condition that has a 6% global prevalence rate and highly deleterious impacts on individual wellbeing, but has a tailored CBT treatment approach that has been demonstrated to be mostly efficacious. To date, research on the assessment and measurement of PTSD has been primarily limited by a reliance on self-report measures and structured interviews, which each have their own specific limitations in facilitating timely, accessible and accurate assessment of PTSD. The present study advances on past PTSD assessment research by examining the capacity of facial, vocal and linguistic features extracted from recordings of participants responses to contribute to the measurement of PTSD in war veterans. War veterans (N = 157) were asked to describe traumatic events they had experienced whilst being audio and visually recorded for 1-minute. These recorded responses were analyzed using OpenWillis software, which is able to simultaneously extract facial (e.g., facial expressivity), acoustic (e.g., pitch, volume), and linguistic (e.g., sentiment) features from a short participant recording. Features shown to be relevant to PTSD from past research (e.g. facial expressivity of fear and negative language sentiment) were fed into several different Large Language Models, which were each asked to identify, based on participant scores on the relevant facial, vocal and language measures, those who were scoring above the threshold for possible PTSD on the PCL-5. Findings will be described in terms of how PTSD can be more accurately and accessibly measured and assessed using scalable digital phenotyping software that facilitates the analysis of multiple emotional channels simultaneously and large language models that can process and use this data to identify those at greater risk of having PTSD. This research also aims to provide an improved understanding of the emotive and expressive mechanisms that are relevant to the presentation of PTSD.