Pediatric mental health is a growing concern around the world, with over 7 million children in the United States impacted by mental, emotional, or behavioral disorders. Poor mental health in childhood affects children's social-emotional development and increases the risk of adverse behavioral outcomes later in life. However, diagnosing mental health disorders in early childhood is challenging due to systematic barriers such as lack of access to services and caregivers' inability to identify signs of problematic behavior.
Mobile and wearable technologies present a unique opportunity for the development of scalable assessment tools to screen for pediatric mental health concerns. This thesis posits that multimodal signals (e.g., physiological, video-based, acoustic) recorded in conjunction with brief, naturalistic tasks can be utilized to build highly accurate and easy to use at-home screening tools. I will demonstrate the feasibility of this approach via EarlyScreen - a video-based tool that uses facial expressions and body movements during an emotion regulation game to detect low prefrontal cortex neural activity and psychopathology in preschool-aged children. I will then describe how the accuracy of such screening tools can be improved by combining neurophysiological signals with video features using supervised contrastive learning. Next, I will discuss multi-stakeholder perspectives towards the usability and clinical utility of screening tools such as EarlyScreen. I will conclude by proposing a new class of screening technologies and clinician support tools that are more naturalistic, provide richer context about parent-child interactions, and can better support parents and clinicians.
Co-advisors: Deepak Ganesan and Tauhidur Rahman