Deep learning enhanced seizure monitoring from wearable sensors Over 1 million patients in the United States have uncontrolled epilepsy despite ongoing medical therapy. When seizures are prolonged or violent, there is significant risk of injury or even Sudden Unexpected Death of Epilepsy (SUDEP) which occurs in approximately 1 in 500 patients per year. Novel seizure monitoring could help alleviate this burden. Our research team has created a software application, EpiWatch, to capture different sensor measurements related to seizure activity such as convulsions (accelerometers), heart rate increases (photo- plethysmography-PPG), and unresponsiveness to behavioral prompting (interactive user interface). Our hope is to offer accurate seizure detection with improved false positive performance to encourage usage. Our team proposes to develop multi-modal sensor analysis driven by deep learning technology to enhance seizure monitoring. We are uniquely positioned to accelerate development by leveraging our team?s prior EpiWatch IRB approved study which generated over 6,000 hours of sensor data. In Phase I, we will teach EpiWatch how to read time series sensor data and how to discriminate seizure activity. EpiWatch will employ a convolutional neural network, a technique rooted in deep learning, to self-characterize seizure features from labeled sensor data. In order to infer additional information from vast amounts of unlabeled sensor data from US epilepsy patients, EpiWatch will incorporate a deconvolutional neural network technique to increase predictive performance. A pilot study of EpiWatch will test its ability to identify the presence of seizures in a prospective new cohort of epilepsy patients. If we are successful, we envision a Phase II proposal which is focused on clinical translation of the technology and assessment of its impact. Our goal is to combine recent advances in deep learning and scalable parallel computing to create EpiWatch. In the long term, we hope this monitoring technology will aid epilepsy patient management and improve outcomes.
Recurring seizures are disabling, dangerous, and often limit independence. We are developing novel seizure detection using a consumer friendly device with wearable sensors (EpiWatch) to enable monitoring and emergency alerting for seizures that occur without warning (~50% of all seizures) and without witnesses, especially when they are prolonged (> 5 min) or accompanied by cardiac arrhythmias responsible for SUDEP (Sudden Unexpected Death with Epilepsy), a 1 in 500 annual risk for patients with uncontrolled seizures. If emergency care can be summoned under these circumstances, patients can live more safely and independently, in turn encouraging app usage. When integrated with current disease monitoring activities, EpiWatch will have the long range impact of providing a unique platform for individualized epilepsy care.
