Wearable Technologies for Autism       


Amar Das photo photo Hillol Sarker photo

Wearable Technologies for Autism - overview

Autism is a complex neurodevelopmental disorder that manifests with atypical communication, limited social interaction, and restricted or repetitive patterns of behavior, interests, and activities. Children with autism can face challenges in undertaking daily activities because of stereotypical behaviors, such as hand flapping, head banging, repetitive movement, and vocal protest. We are investigating whether these behaviors can be accurately detected through sensing technology that can objectively characterize their nature, identify triggers for their occurrence and monitor the effect of therapeutic interventions. We have developed a sensor-based platform, called Tabatha, to assess and monitor these stereotypical behaviors in autism. This robust research platform enables us to conduct research studies on behaviors detected by IoT applications in lab and field settings. The system can securely collect de-identified data in the cloud and develop and evaluate machine learning models. Through the cognitive IoT fabric, we will be able to deliver the artifacts as services to clinical partners. 

In work published at the 2018 American Medical Informatics Association (AMIA) Symposium, we report on a detection model for stereotypical motor behaviors associated with autism using a smartwatch. In a lab-based validation study, we found that we can accurately distinguish hand flapping, head banging, and repetitive dropping movements from play-like behaviors using machine learning approaches, such as Gradient Boosting. Our prior work published at 2018 International Conference on Acoustics, Speech, and Signal Processing (ICASSP) focused on analyzing a speech corpus of stereotypical verbal behaviors in autism, such as yelling, screaming, and crying, along with other daily noises, such as coughing or a television show. An analytics model using an ensemble of Gaussian Mixture Models and Convolution Neural Networks was able to accurately classify a given sound snippet as a verbal behavior associated with autism or as confounding noise.

Technology-based Assistance for Children with Autism