Epilepsy is one of the most common neurological conditions with approximately 50 million people affected by the disease worldwide, including all ages and genders (1). If diagnosed and treated correctly, up to 70% of those affected can reach a fully controlled disease (2) and thus live a decent normal life. However, a significant proportion of patients suffer drug-resistant epilepsy with many adverse events related to epilepsy, ranging from physical problems and dangers to psychosocial components. Among the most noticeable preventable physical consequences are the increased risk of premature death related to injuries (3), Sudden Unexpected Death in Epilepsy (SUDEP) (4,5), status epilepticus as well as suicide (5). With regard to psychosocial impact of epilepsy, patients are more prone to low self-esteem, depression and anxiety, often related to the unpredictability of seizure occurrences (6,7).
Developing wearable devices that can alert the patient in case of a seizure, or even predict the occurrence of a seizure, could help prevent some of these negative outcomes (8,9) by calling for help before injury or SUDEP can happen but also by giving the patients better control of their disease. Indeed, one study evaluating the benefits of automatic seizure detection using accelerometry showed that patients who tested automatic seizure detection were globally satisfied. They agreed on saying that, on the one hand, they experienced fewer injuries and, on the other hand, felt safer knowing that a device could alert them (8,10). However, some drawbacks have also been brought forward, especially the issue of false alarm rate (FAR), which is too high (8). Indeed patients wish that the number of FAR would at least not exceed the number of correct alarm rate (10).
From the physician point of view, epilepsy is also a challenging condition. Seizures can occur at any time, without true regularity. To initiate a treatment, it is however necessary to know the type and frequency of seizures. Later, in the treatment follow up, it is important to know how the patient responded to therapy in order to adapt it. So far, these seizure and treatment evaluations are made through a journal in which the patient logs every seizure. But it has not proven to be an effective method as patients fail to report more than half of their seizures (11). One study (12) brings forward seizure unawareness as one of the main reasons. Some studies have more precisely reported focal seizures as being the most problematic ones in that context (12,13).
For all these reasons, it is of high importance to find a way to properly detect seizures. Gold standard today is video-EEG, meaning that the patient is continuously filmed and has a standard 10-20-leads EEG in the form of a hat all around the scalp. It works well for an in-hospital setting but is much more problematic for at-home daily use, as it is too bulky and stigmatizing (10,13).
4
An interesting alternative option that is being widely investigated is the field of wearable seizure detection devices. These wearable devices use signals that are less invasive to acquire. To achieve this, the physiopathology of epilepsy is considered, meaning that both movements and autonomic changes are used. The most commonly used signals are electrocardiography (ECG), accelerometry (ACC), electromyography (EMG), electrodermal activity (EDA) and SpO2. Each of these signals will be explained and linked to epilepsy in the following sections.
In this context, our study aims to explore the feasibility and the performance of a multimodal epileptic seizure detection system using four modalities: ECG, SpO2, EDA and ACC. Ultimately, the goal is to characterise the appropriate and specific signals for epilepsy detection with wearable device; such as a wristband capturing the Heart Rate Variability obtained from PPG sensors, combined with EDA, ACC and eventually SpO2.