Evaluating the predictive validity of computational markers of self-injury severity among high-risk youth

Suicide is the second leading cause of death among those ages 15-24. A history of self-injury, including both nonsuicidal self-injury and suicidal self-injury, is the strongest predictor of future suicidal behavior, with evidence suggesting that the more severe such behaviors are, the greater the risk for future self-injury. Importantly, however, our current means of assessing severity of prior self-injury is almost entirely reliant on self-report, despite the fact that self-injury frequently leaves tangible physical markings. Although applications of machine learning in medical image analysis are growing exponentially, none have attempted to augment suicide risk detection through automated analysis of self-directed tissue damage. Leveraging computer vision to automatically assess images of tissue damage has the potential to obviate complete reliance on subjective patient report of self-injury severity characteristics. The objective of this proposal is to extend ongoing research supported by the National Institute of Mental Health aimed at utilizing computer vision techniques to automate the assessment of self-injury visual severity indicators and to determine the utility of these visual signals in predicting suicide risk. Psychiatrically hospitalized adolescents with a history of self-injury will be recruited if they have currently visible physical marking(s) secondary to self-injury. Adolescents will securely upload images of markings secondary to intentional self-injury and will provide information about these images as well as their history of self-injury. One month after psychiatric hospitalization discharge, adolescents will be assessed for prospective engagement in suicidal behavior. Deep convolutional neural networks will be applied to the images to detect severity indices of self-injury and to examine their accuracy in predicting prospective suicide risk in this high-risk clinical sample. This study will help assess the feasibility of pursuing our long-term goal of integrating this technology into psychiatric care entry-points to assess whether it can improve suicide risk assessment models and in turn, serve as a clinical decision-support tool.