Optimized Affective Computing Measures of Social Processes and Negative Valence in Youth Psychopathology

点击翻译按钮获取中文摘要

ABSTRACT Difficulties with emotion expression and social behavior characterize multiple psychiatric conditions and negatively impact child development. However, existing measurement tools for indexing social-emotional function are imprecise and subjective, or require specialized training that is costly and time-intensive, prohibiting widespread implementation. The imprecision of existing tools has a major negative impact not only on research, but on the ability to assess and treat individuals with mental health concerns – especially among underserved and under-resourced populations. Here, we propose to address this problem by quantifying social and emotional behavior using novel biobehavioral markers derived from computer vision (facial expression analysis) and computational linguistics (social/sentiment analysis). Our team has successfully used these markers to predict the presence of autism spectrum disorder (ASD) with 91% accuracy. In this proposal, we determine the extent to which our markers can serve as continuous measures of social behavior and negative emotion to advance clinical phenotyping and interventions. The proposal brings together two high-bandwidth clinical research programs at the Children’s Hospital of Philadelphia and Baylor College of Medicine to collect data on 750 adolescents (ages 12-17 inclusive) with ASD, a primary anxiety or depressive disorder, or without any developmental/psychiatric condition. At a single assessment, all youth will participate in an extensive clinical phenotyping battery consisting of validated clinical interviews and child-/parent-report scales assessing converging and diverging mental health constructs, and three tasks eliciting positive/negative emotion, social stress, and mild frustration. A subsample of 150 adolescents will be reassessed 6-10 weeks later to allow retest/stability analyses. A novel camera apparatus will capture naturalistic synchronized verbal and nonverbal signals from dyads. Our analytic approach combines state-of-the-art machine learning, computational linguistics, and computer vision – including facial emotion recognition methods that rival several commonly used alternatives. The ultimate goal of this proposal is to develop valid and objective measures of the Social and Negative Valence Systems using novel biobehavioral markers in a large transdiagnostic sample of youth. Secondary goals are to develop easy-to-follow methods to widely disseminate our tools and procedures, and to characterize individual variability in these key RDoC metrics by age, gender, race/ethnicity, and diagnosis. The achievement of these goals will provide researchers with sorely needed measures of social and emotional behavior, and provide clinicians with a new set of tools for identifying and tracking youth in need of mental health treatment.