Wearable Sensor-Based Physiological and Physical Activity Biomarkers for Use in Laboratory and Naturalistic Environments to Assess Arousal and Repetitive Motor Movements in Individuals with Autism Spectrum Disorder

Friday, May 12, 2017: 4:10 PM
Yerba Buena 8 (Marriott Marquis Hotel)
M. S. Goodwin, Northeastern University, Boston, MA
Background: Despite autism spectrum disorder (ASD) being widely recognized as the fastest growing and most costly neurodevelopmental disorder worldwide, there is no specific biomarker, laboratory test, or behavioral assessment procedure to identify, characterize, or monitor its progression. ASD is defined exclusively by past and present behavior determined from developmental history interviews, parent reporting on current behavior, and structured and semi-structured tasks that involve social interaction between an examiner and a child.

Objectives: Describe wearable sensor-based physiological and physical activity technologies enabling multimodal assessments of autonomic arousal and repetitive motor movements in ASD in both laboratory and naturalistic environments, and review their utility for biobehavioral phenotyping and response to intervention in ASD.

Methods: Ubiquitous and wearable computing is making it possible to capture data in laboratory, clinical, school, and home settings on an unprecedented scale. Coupled with advances in pattern recognition algorithms and large-scale computing, semi-automated measures of physiology and behavior are emerging including wireless sensors for monitoring physiological arousal and wireless 3-axis accelerometers and pattern recognition algorithms that can automate the detection of stereotypical hand flapping and body rocking.

Results:  Results from two lines of research will be reported. The first employs wireless measures of autonomic nervous system (ANS) activity and demonstrates that severely affected children with ASD have different cardiovascular response patterns to various psychological, social, and sensory demands, including substantially higher and less variable heart rate (HR) than a typically developing group of age-sex matched peers. Moreover, in a follow-up replication study repeating the same assessment protocol in 43 severely affected children with ASD, 6 distinguishable subgroups within the ASD sample were identified who display normal, high, and extremely high HR, along with unreactive (‘stabile’) and reactive (‘labile’) response patterns using time series-based cluster analysis. In the second line of research, wireless 3-axis accelerometers and pattern recognition algorithms are employed as automated measures of stereotypical motor movements (SMM) in six severely affected individuals with ASD. Using five time and frequency domain kinematic features and a C4.5 decision tree classifier, average automated hand flapping and body rocking recognition rates of 89.5% in the laboratory and 88.6% in the classroom were achieved. Moreover, a direct replication wherein the same six individuals with ASD were observed three years later in their classrooms while wearing 3, 3-axis accelerometers to determine whether previously trained sensor-based classifiers maintain accuracy over time will be reported. Comparing automated recognition results for two different classifiers – Support Vector Machine and Decision Tree – using our previously established feature set yielded average accuracy across all participants over time ranging from 81.2% to 99.1% for all combinations of classifiers and features.

Conclusions: Identifying more objective and automated physiological and physical activity measures has the potential to transform our ability to better understand the pathophysiology of autism, enhance gold-standard assessments of ASD, aid in subtype identification, individualize treatment protocols, monitor treatment efficacy, and track developmental outcomes.