Exploring Epileptic Seizure Detection with Commercial Smartwatches Theresa Tobollik Bauhaus-Universität Weimar Weimar, Germany theresa.tobollik@gmail.com Sujay Shalawadi Bauhaus-Universität Weimar Weimar, Germany sujay.shalawadi@uni-weimar.de Christopher Getschmann Bauhaus-Universität Weimar Weimar, Germany christopher.getschmann@uni-weimar.de Florian Echtler Aalborg University Aalborg, Denmark floech@cs.aau.dk Abstract—Some forms of epilepsy can randomly trigger severe seizures that degrade patients’ quality of life and may even lead to death. Specially trained dogs that go through a long learning process can sometimes help to warn patients of an imminent seizure, possibly due to an ability to sense subtle changes in the subject’s blood oxygen level. In this paper, we present our exploratory study of using a commercial smartwatch with an oxygen sensor to continuously capture data and detect changes in blood oxygen saturation during or ahead of a seizure to warn the patient or emergency contacts. Our data shows a possible correlation between reported oxygen level and a seizure incident, but higher-frequency readings will be required in the future to determine whether accurate prediction with smartwatches is indeed possible. Index Terms—smartwatch, epilepsy, seizure, prevention, blood oxygen, sensor I. I NTRODUCTION A chronic disease like epilepsy can have a severe influence on the quality of life of the affected person. During epileptic seizures, the patient may experience loss of perception of their surroundings, short term memory loss, and in severe cases even loss of the body’s sensorimotor functions (e.g. falling and convulsions). In some patients, these seizures occur unpredictably, therefore they cannot drive vehicles and are limited in choosing their occupation and physical activities. The prediction of an imminent seizure helps in reducing harm and improves life quality in general. In some cases, specially selected and trained dogs can predict and warn, but these dogs are hard to find, take years to train, and are very expensive. It is assumed that these dogs’ successful seizure predictions are based on their ability to detect drops in blood oxygenation [14], although the exact mechanism is not yet understood. Modern smartwatches continuously monitor physiological data such as heart rate and blood oxygenation, providing a constant stream of information suitable for ad-hoc evaluation and emergency reporting. In this paper, we explore whether a commercial smartwatch can provide some of the benefits of epilepsy dogs. In particular, we explore and compare sensor data recorded over several days from an epileptic patient and a control subject using a Garmin smartwatch, and correlate this data with manually logged seizure incidents provided by the patient. While the data from our deployment indicates that smart- watches’ sensor readings might be useful for seizure detection, the dataset is too small for statistical analysis. More data needs to be gathered to definitely answer the question of whether seizure detection or even prediction at a reliable accuracy level is possible. Nevertheless, our results show that commercially available smartwatches could be useful for further research concerning epileptic seizures and might aid in seizure detection and prediction. II. BACKGROUND We now introduce some core concepts related to this paper. First, we provide an introduction to epilepsy, followed by a discussion of relevant bio-signals. Finally, we review related work regarding the connection between epilepsy, smart devices and seizure detection/seizure prediction. A. Epilepsy Worldwide 0.5-1% of people suffer from epilepsy [10]. An epileptic can have seizures which vary in their degree of severity and the types of symptoms. During a seizure, there is an excessive amount of neuronal activity in the brain, often referred to as an ’overcharge of the brain’. An epileptic seizure can be triggered by various factors in- cluding flashing lights/patterns (photosensitive seizure), stress, lack of sleep, alcohol, or sudden scares. For roughly 70% of epileptics, seizures can be reduced or eliminated by correct medication or surgery, though there remain patients who never get rid of their seizures [18]. Seizures can be differentiated into Focal Aware Seizures (FAS/auras), Focal Impaired Awareness Seizures (FIAS) and bilateral tonic-clonic seizures [17]. During a Focal Aware Seizure (also called aura), the epileptic is conscious and can later describe the symptoms. An aura lasts only for a few seconds and affects the sensitive and/or vegetative system. These symptoms are only perceived by the epileptic and not the people around them. Such sensations are described as an intense feeling of joy or fear, déjà vus, weird smells, or numb limbs. In contrast to other seizure types, the epileptic can remember having had an aura and how it felt, though during the aura itself the patient cannot always express that they are experiencing an aura. On contrary, during a FIAS, the affected person has reduced consciousness during the seizure and often does not remem-