Enhancing Suicide Risk Detection through Computer Vision: a Novel Approach to Tissue Damage Analysis in Emergency Care

SUMMARY Suicide is the second leading cause of death among youth and young adults ages 10-24, with rates having risen over 60% in the past 15 years. Emergency Department (ED) visits for psychiatric concerns among this population have also doubled, often prompted by suicide-related concerns. However, ED clinicians often find it difficult to identify those at the highest suicide risk and to judge whom to transition to higher levels of care when such care is limited. A history of prior self-injury, including both nonsuicidal and suicidal self-injury, has consistently been the strongest predictor of future suicidal behavior. Ample theoretical and empirical evidence suggests that the more severe, or medically lethal, prior self-injurious behaviors are, the greater the risk for future self-injury. However, the field has almost entirely relied on self-report to assess the presence and severity of prior self-injury, despite the fact that self-injury frequently leaves physical markings. Our prior R21 was the first application of computer vision with the goal of augmenting suicide risk detection through the analysis of images of tissue damage. Building on our promising R21 results, this R01 proposal seeks to expand the application of computer vision techniques to the ED to predict prospective suicide attempt (SA) risk more accurately by analyzing images of tissue damage for self-injury presence and severity. This study aims to determine the predictive utility of signals derived from standardized skin images in predicting prospective SA risk among ED patients beyond (1a) participant-report of past self-injury severity indicators and (1b) extant Electronic Health Record (EHR)-based suicide risk algorithms trained at Mass General Brigham (MGB). The performance of these models will be evaluated in racial and ethnic minority groups to mitigate bias in future research. Finally, this study aims to characterize implementation determinants of employing image-taking procedures and computer vision-enabled algorithms to automate EHR documentation of self-injury tissue damage within EDs. Youth and young adults ages 12 to 25 presenting with psychiatric concerns will be recruited at MGB EDs. At baseline, study participants will have standardized images of their arms taken; to assess prospective SAs, participants will complete remote assessments at 1 and 6 months and medical records will be examined. Images will be analyzed using deep learning techniques to detect and classify tissue damage indicators of suicide risk. Successful completion of this study will establish the utility of computer vision at the point of care and provide crucial insights into potential barriers and facilitators of its implementation that can be addressed in future scale-up. This research paves the way for implementing a novel, objective approach to suicide prevention that enhances detection and monitoring of youth and young adult suicide risk. This research aligns with the National Institute of Mental Health and the National Action Alliance for Suicide Prevention’s prioritized research agenda, targeting the development of innovative and effective suicide risk assessment tools in clinical settings. Project Number: 1R01MH140004-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Mental Health (NIMH) | Principal Investigator: Taylor Burke | Institution: MASSACHUSETTS GENERAL HOSPITAL, BOSTON, MA | Award Amount: $839,023 | Activity Code: R01 | Study Section: Clinical Informatics and Digital Health Study Section[CIDH] View on NIH RePORTER: https://reporter.nih.gov/project-details/11226623