Tabatha: A Mobile Health Platform to Evaluate Stereotypical Behaviors in Autism

Background: Children with Autism Spectrum Disorder (ASD) face challenges in undertaking daily activities if they have stereotypical behaviors, such as hand flapping, head banging, repetitive movement, and vocal protest. Accurate detection of these behaviors through wearable devices and other sensing technologies can allow researchers to characterize their nature objectively and identify triggers for their occurrence. Sensing technologies can also be used to monitor therapeutic interventions and assess their efficacy in an unbiased manner.

Objectives: Our primary objective is to create an end-to-end mobile health (mHealth) platform (called TABATHA) to support research on challenging behaviors in ASD. Researchers can use TABATHA to collect and store sensor readings, validate physiological parameters, and analyze the data to discover novel biomarkers and triggers for vocal protest and motor stereotypy. Our secondary objective is to ensure that the data collection approach complies with human subject regulations, such as subject privacy.

Methods: The TABATHA platform consists of a commercially available smartwatch (Motorola Moto 360), Android tablet, Hyperledger Blockchain, and IBM Cloud. The smartwatch collects raw sensor data sampled at high frequency: accelerometer (96 Hz), gyroscope (48 Hz), photoplethysmogram (24 Hz), Heart rate (1 Hz), and the acoustics (44.1 kHz). To minimize the re-identification risk, we have developed software within the smartwatch to anonymize the sensed samples right at the source. We obfuscate the identifiable traits captured by acoustic sensing by computing and storing time and frequency domain speech features. The anonymized streaming data is sent to the tablet on which researchers note the occurrence of stereotypical behaviors; both the anonymized data and identified behavioral occurrences are sent to the IBM Cloud, where we can apply machine learning algorithms. TABATHA also uses Blockchain to support privacy and security. Prior to enrollment, consent information, collected from subjects, are securely stored on the immutable ledger. Study data ownership, access rights, and secondary data usage are enforced by smart contracts written into the Blockchain.

Results: The design of TABATHA can facilitate six central roles within an ASD research environment: subject (uses a wearable); parent (who provide consent); primary investigator (monitors, manages, and approves the ongoing activities), study coordinator (recruits subjects, acquires informed consent, and administers study sessions), auditor (evaluate the protocol compliance), and third-party collaborator (access anonymized study data).

We have validated TABATHA by conducting a feasibility study to detect stereotypical motor behaviors in the presence of other confounding child-type playing activities, such as card matching and drawing. We obtained a nearly 93% accuracy in a cross-subject evaluation. Moreover, we hand-curated a speech dataset on vocal protest (defined as sensory overload induces crying or screaming among children with autism) and developed neural network-based detection models with accuracy over 93%.

Conclusions: We have developed a novel mHealth platform to collect research data on stereotypical behaviors of children with ASD. The developed platform facilitates consent management, annotation, anonymized datasets, and summaries of the occurrence of behaviors. Future work will involve using TABATHA to conduct a field study on children with ASD and identify triggers for challenging behaviors.