Development of an Open-source Seizure Detection and Alert System for Apple Watch Using Machine Learning
Shwetank Singh1, James Dolbow2, Suraj Thyagaraj2, Neel Fotedar2
1Case Western Reserve University, 2University Hospitals Cleveland Medical Center
Objective:
To develop an app for Apple watch and subsequently test its utility as a seizure detection and alert system (SDAS)
Background:
Unattended generalized tonic-clonic seizures carry the highest risk for Sudden Unexpected Death in Epilepsy (SUDEP). A need for automated at-home detection of these seizures has led to the development of many devices. Currently, all the available SDAS use proprietary hardware that reveals the diagnosis of the user and are often too expensive.
Consumer smartwatches, such as the Apple watch, have sensors that can perform activity recognition. Development of an alert system that can be run on consumer smart watches could provide a discreet and cost-effective solution. Therefore, a pipeline is needed to collect data, develop models and test the ability of consumer smart watches to act as SDAS. This pipeline’s open-source and scalable nature is designed for a multi-center data collection effort to develop reproducible and publicly validated machine learning models.
Design/Methods:
We developed a data collection and an alert system app for the Apple watch. We tested a machine learning model for pattern recognition (running v not running) and then incorporated the model into the alert system app.
Results:
The data collection app collects data from the accelerometer, gyroscope, magnetometer, EKG sensor and oxygen saturation and synchronizes it to iCloud at regular intervals. The alert system app continuously monitors the watch sensors and sends the user an alert when running with reliable accuracy.
Conclusions:
Our newly developed SDAS utilizes two Apple watch apps and machine learning to detect user activity and alert the user if detected. Consumer smart watches present an opportunity to develop seizure alert systems that could be discreet and inexpensive, thus increasing utilization. Our work has created open-source apps to facilitate multi-center efforts for development of a seizure alert system using apple watches.