Combating Epilepsy with Wearables and Machine Learning

June 11, 2019 by Orlando Hoilett

Biometric-monitoring wearables are common for tracking everyday health—but can they monitor seizures?

Biometric-monitoring wearables are common for tracking everyday health—but can they monitor seizures?

Epilepsy is a chronic condition that affects millions of people worldwide. Epilepsy is most commonly characterized by the occurrence of involuntary seizures. Due to the sometimes debilitating nature of these seizures, epilepsy monitoring is of major interest to the medical community.

A team at healthcare wearables startup Empatica has positioned themselves at the forefront of epilepsy monitoring with their recently FDA-cleared, second-generation, wrist-worn wearable called the Embrace2.


The Empatica Embrace2. Image from Empatica


The Embrace2 is intended for children with epilepsy ages six and up, specifically designed for tonic-clonic seizures, those that cause involuntary convulsions in patients. 

According to a January press release on the FDA's 510(k) clearance of the Embrace device, it is "the first non-EEG based physiology signal seizure monitoring system to be cleared by the FDA" specifically for use with pediatric populations.

A Wearable that Measures Possible Seizures

The Embrace2 is designed to monitor locomotion as well as electrodermal activity or response (EDA). The goal of the device is to make it more likely for caregivers to be present at the time of a seizure to render assistance.

Let's take a look at what sensors are needed to monitor these factors.


Image from Empatica


Using Accelerometers to Measure Locomotion

Of these two physiological quantities, locomotion is probably the more recognizable of the two to most readers.

The Embrace2 tracks locomotion using accelerometers which have become a mainstay in the semiconductor industry. These accelerometers are tiny MEMS (microelectromechanical machines) devices that measure the linear acceleration of an object with respect to gravity. Accelerometers are widely employed in consumer electronics for detecting orientation for switching between landscape and portrait mode in cell phones, free fall for elderly fall alert monitors, and for detecting steps in pedometers and smartwatches to name a few examples.

Since epilepsy results in seizures, Empatica is able to use the accelerometer data to quantify the length and intensity of the convulsion, sending an alert to a caretaker or loved one when a seizure is detected.

Gyroscopes are also reportedly in the device to allow similar measurements regarding orientation.


An exploded view of the Embrace2's hardware. Image from Empatica


Measuring Electrodermal Activity with Electrodes

Between locomotion and EDA, most people are probably less familiar with EDA. EDA (also known as GSR or galvanic skin response) is an indirect measure of the human body’s sympathetic nervous response or “flight-or-fight” response.

The sympathetic nervous system (SNS) is the body’s center for responding to arousal due to external stimuli. The activities of the SNS can be detected by measuring local changes in skin conductance often between two fingers or on the wrist, in the case of a wrist-worn device like the Embrace2. Such changes in skin conductance are elicited due to the activation of sweat glands right below the surface of skin that secrete minute amounts of sweat, a conductive solution, consequently increasing the measured skin conductance.


Image from Empatica


EDA has been used in biometric monitoring wearables over the past few years—and famously used in lie detectors for decades. EDA, however, is not necessarily specific to any particular stimuli (skin conductance can be impacted by a number of emotional responses). For this reason, Embrace2 had a challenging task in using such a technology for detecting seizures. Nevertheless, Empatica has been able to develop a working product that was able to pass the scrutiny of the FDA.

According to Empatica's website, the company is choosing to utilize EDA in this context based on 2015 research that suggests that a seizure that "strongly stimulates the amygdala" could cause large EDA, among other citations of neuroscience journal articles.


Machine Learning and Embedded Systems

Empatica openly admits that the technology isn’t perfect and that false positives and false negatives are a certainty, unfortunately. Nevertheless, the development of technologies to help people with epilepsy continues and Empatica is proud to be a part of this active area of research continually refining their device and algorithms with each new patient.

Empatica’s work demonstrates the need for more sophisticated sensing modalities and artificial intelligence to combat chronic illnesses such as epilepsy. Recent developments in reducing the computation intensity of machine learning algorithms have given rise to technologies such as Intel’s Loihi, Movidius’s Myriad 2, and most recently Ambiq Micro's Apollo3 Blue microcontroller, an ARM Cortex-M4F processor running TensorFlow Lite.


The Apollo3 Blue microcontroller from Ambiq Micro is also available on the SparkFun Edge Board, which can run TensorFlow. Image from Ambique Micro


Hopefully Empatica and other medical device companies continue to improve and refine their technologies turning to the semiconductor industry for each new and more capable chipset.

Future Uses of Biometric Monitoring

Biometric-monitoring wearables have been gaining popularity over time. As the relevant algorithms for interpreting sensor data advance, even component manufacturers are making health wearables.

But Empatica is tackling something beyond monitoring steps, heart rate, and blood oxygen levels.

The promise of a wearable for monitoring such seizures could, dare I say, extend into predicting the occurrence of these seizures. While Empatica makes no claims on seizure prediction, such a feature would make a world of difference to those managing their epilepsy.