ADHD detection based on human action recognition

Attention Deficit Hyperactivity Disorder (ADHD) is a highly prevalent human neurobehavioral and neurodevelopmental disorder worldwide. Recently, deep learning-based techniques have been exploited in ADHD detection and diagnosis due to their outstanding performance. However, the majority of these methods relying on fMRI and EEG data suffer from the limitation of requiring expensive equipment and incurring high operational costs. Therefore, inspired by the fact that the symptoms of ADHD may manifest in actions and daily behaviors (as stated in the Medical Statistical Manual of Mental Disorders, Fifth Edition (DSM-V)), we introduce a novel ADHD detection system based on human action recognition. We design a novel hyperactivity test for capturing ADHD features and record a real multimodal ADHD dataset (M-ADHD) for the first time. The proposed system detects ADHD symptoms based on acquired action characters from raw RGB videos from the M-ADHD. Our system outperforms conventional competitors in terms of accuracy and AUC on the real multimodal ADHD dataset. Our proposed method, based on simple, non-wearable sensors, has the advantages of being cost-efficient and easy to operate. It is widely applicable for remote ADHD screening and further applies to understanding, treating, and preventing brain disorders.