By Catherine Chan
The MIT Startup Empatica, developed the first FDA cleared multimodal smartwatch for epilepsy patients, named Embrace, was approved as a medical device in 2018. It’s a seizure-alerting smartwatch sensing the physiological signals of ongoing Generalized Tonic Clonic seizure(GTCS-like events), then send alert to caregivers, which can be a life-saving device for patients who are under high risk of sudden unexpected death in epilepsy(SUDEP), in Generalized Tonic Clonic Seizure (GTCS). Traditionally GTCS can only be detected by video with EEG in hospitals and clinics. The company made this breakthrough by making seizure detection viable in daily life at the convenience of wearing a stylish smartwatch.
The smartwatch Embrace was equipped with accelerometer and EDA sensor, with machine learning algorithm, to detect convulsive seizure-like motion and EDA signals on the skin.
It is Rosalind Picard, founder of the company Empatica and director of Affective Computing Research Lab at MIT, who pioneers the field of research on the use of Electrodermal activity (EDA signals) on the skin for seizure detection (specifically GTCS). Rosalind Picard found out that during onset of a seizure, the patient’s EDA signals goes extraordinary high resulting from arousal of sympathetic nervous system. This results in changes in electrical conductance on the skin. Moreover, EDA signal is able to detect brain activity involved in seizure and anxiety which sometimes even EEG cannot detect if the seizure involves areas that are deep inside the brain.
The product shows promising results in multimodal sensing algorithm compared to unimodal in seizure detection. There have been other products in the market that uses only motion detection (accelerometry alone) for seizure detection, but has lower sensitivity then this Embrace smartwatch. According to the research funded by Dutch Epilepsy Fund/National Science Foundation. sensitivity of seizure detection is only 88% if using acceleromtry alone, compared to 94% sensitivity which uses both accelerometry and electrodermal activity. This proves that multimodal may work better than unimodal sensing.
Whilst EDA signals and motion detection has a higher sensitivity but it’s not foolproof. . According to a few user review, it has false positives if the person is having intense physical activities such as running, as the smartwatch is incredibly sensitive to motion. These users recommended that the product may only be suitable for someone having a less physically active lifestyle, as well as having multiple, higher debilitating seizure a day. It makes sense that Empatica makes the smartwatch as a prescription only device as it’s best to consult the doctor before deciding whether it’s a right device for the epilepsy patient or not.
In fact, some activities and bodily movements in daily life may seem like convulsive seizure to the smartwatch, resulting in false alarms. According to Empatica, the motion of convulsive seizure takes precedence in Embrace smartwatch’s algorithm. EDA also takes part in the smartwatch’s algorithm, but since EDA is more prone to fluctuate, relying too heavily on EDA signals would result in even more false alarms. It’s good that in latest review article in the product it shows there has been improvement on the algorithm owing to growing data availability and more users contributed immediate feedback on the false alarm problem, thereby facilitating machine learning algorithm adjustment to better distinguish normal physical activities from GTCS. The false alarm rate has been lowered from the initial ~2 to 0.2-1 false alarms per day.
It’s hard to achieve 100% accuracy. So false-positive is still a challenge for seizure detection device. One reason to explain this is that seizures events may be different for each individual and seizure patterns fluctuate because of external environment. More data would need to be collected from wide variety of real life situation from users in order to facilitate continuous improvement of the algorithm.
In fact, Dutch Epilepsy Fund and Science Foundation has done research on how to improve accuracy of the devices. They tested on combiningmultiple physiological sensors for seizure detection in a few studies (accelerometers, electromyography(EMG), heart rate, and oximetry, Electrodermal activity(EDA)). It was found that multiple sensors increased sensitivity, with false alarms decreased in a study, but increased in another study. Why is it so?
In fact, while combining different sensors may improve accuracy and lower false alarms in some circumstances, it is not always the case . Often when a device becomes too sensitive, the number of false positives may rise. It’s also possible that with too many sensors, signals may interfere with each other.
The study has shown promising sensitivity for multimodal devices, but minimizing the number of false positives is still challenging. According to Frans S.S. Leijten of the Dutch TeleEpilepsy Consortium , false alarm is still a major problem and is the major impediment for clinical devices to be used at home.
What is encouraging is that the false detections seem to occur only in a minority of patients (around 16%-30% of patients).
In fact, the false detections occurring only in minority of patients opens up the prospect of using generic algorithm in a device for most patients, and personalization of the algorithm for the few patients who turns out to have many false detections. To market these devices in the consumer market, this should be part of the instructions manual and user information before purchase.
It is understandable that some epilepsy patients may expect a smartwatch which can act as a mini caretaker, a device which can monitor their nervous system, and warn them of an impending seizure. Today, this is still limited by the lack of biomarkers that can predict/diagnose Epilepsy. The patient won’t know he/she is a Epilepsy patient until he/she undergoes the cumbersome process of a video-EEG(Electroencephalography) in the hospital/clinic after he/she has the first and second seizure. The Epilepsy Foundation is still looking for potential biomarkers which can diagnose a epileptic seizure before the patient has the first one. I hope there will be a breakthrough in this aspect which would be good news to the patients.