Multimodal Response Analysis for Behavioral Intervention of Children with ASD
Zheng, Zhaobo
0000-0002-1277-2476
:
2021-11-12
Abstract
This dissertation focuses on multimodal response analysis and behavioral intervention in children with autism spectrum disorders (ASD). ASD is a neurodevelopmental disorder, characterized by repetitive behaviors, fixed interests as well as social and motor skill differences. The prevalence of ASD has increased significantly in the past two decades with huge cost to autistic individuals, their families and the society. This dissertation explores three main directions to improve the overall quality of life for individuals with ASD: 1) Early diagnosis of ASD through understanding of behavioral pattern using machine learning; 2) Prediction of Problem behavior (PB) that many autistic individuals experience through machine learning based predictive models; and 3) the design of an Augmented Reality (AR) intervention system to support learning of personal hygiene skills. For our first goal, we designed a multisensory stimulus delivery system, including a novel tactile stimulator to impart affective touch and to collect multimodal behavioral responses of toddlers including peripheral physiological, eye gaze and electroencephalogram (EEG) data. The machine learning model built on the pilot study data demonstrates the potential for early prediction of ASD developmental risk from multimodal behavioral responses. For the PB prediction, we designed customized wearable sensing system to collect motion, physiological, audio, social orientation and facial expression data from autistic children. The trained machine learning models were able to detect PB at an average offline accuracy of 98.51%. Later, we applied these machine learning models to predict PB in real-time. The comparison between the real-time machine learning predictions and the human expert observations showed the potential of an automatic, consistent and low-cost approach to predict PB to parents and caregivers. For the third goal, we designed an AR-based intervention system for toothbrushing skills in autistic children. It provided a closed-loop and immersive learning environment for children with ASD. We also designed a mechatronic toothbrush to measure their brushing movements pre and post to the coaching sessions. From the experiments, we observed statistically significant improvements on their brushing skills after the coaching sessions. The results of this dissertation demonstrates the potential of technology-based autism intervention.