Taking fNIRS home: enabling cognitive biomarker discovery using naturalistic neuroimaging

Abstract

Attention-deficit hyperactive disorder (ADHD) affects nearly half of all adolescent children and has significant long-term ramifications for adult wellbeing and productivity. Current ADHD screening techniques struggle from heavy reliance on subjective surveys, variability in clinician interpretation, and the lack of quantitative biomarkers. Functional neuroimaging using near-infrared spectroscopy (fNIRS) is poised to enable quantitative diagnosis of ADHD by inferring cognitive patterns from brain hemodynamics of the prefrontal cortex (PFC). Unfortunately, functional neuroimaging studies require more longitudinal data to overcome inter- and intra-subject variability. Additionally, lab environments are unnatural, and data collected in such settings lack ecological validity. The large volume of data required alongside the challenges of ecological validity require that studies engage research subjects in naturalistic environments such as homes. To address this need, new tools are needed to affordably perform neuroimaging in real-world settings when no researchers are present in-person. In this dissertation, we develop and validate a solution for naturalistic imaging: 1) a low-cost (<$1000), comfortable, lightweight (150g), and high density (30-channel) fNIRS headband and 2) an accompanying augmented reality (AR) tool to align the headband using a “selfie camera” found ubiquitously. The fNIRS headband was built using a flexible printed circuit board (PCB) embedded in a silicone-fabric enclosure and controlled using a credit-card sized controller. Data from the fNIRS headband is transmitted wirelessly to an accompanying Android tablet application. The headband’s functionality was successfully validated using breath holding tests and arithmetic tests. The AR app was built using open-source computer vision libraries such as OpenCV and Dlib and validated using human subject testing. The tests compared the alignment performance and setup durations using the AR app versus placement by a researcher (2nd-person) or the research subject themselves (self). The AR app provided lower alignment error than 2nd-person or self-placement scenarios and provided more than 3x the reduction in setup duration.