Development of a Seizure Detection Algorithm Based on Heart Rate and Movement Analysis

Epilepsy is the 3rd neurological pathology after migraines and dementia syndromes with a high estimate of nearly 600,000 people affected in France. The disease is characterized by the repetition of epileptic seizures on the one hand, but also by the cognitive, behavioral, psychological and social consequences of this condition, especially when the epileptic disease is not stabilized. Epileptic patients feel a great deal of stress due to the unpredictability of the occurrence of seizures. Seizure detection is of great interest to bioinformatics researchers and to people with epilepsy and their caregivers. Recent advances in physiological sensor technologies and artificial intelligence have opened the possibility of developing systems capable of closely monitoring the frequency of epileptic seizures with a direct impact on therapeutic adaptations. This may eventually allow for seizure prediction and/or "seizure weather" (i.e., seizure forecasting) if there is a particular chronotype of seizure occurrence for a given individual. Currently, few devices have a sufficient level of evidence regarding their effectiveness to be recommended. Those that seem to be the most advanced are those that allow the identification of hypermotor seizures, including tonic-clonic generalized seizures and tonic-clonic secondary generalized focal seizures, mostly occurring at night. The latter represent only a small part of epileptic seizures. The objective of the present study is to build a real life database in order to develop a seizure detection algorithm. The recorded data will be heart rate via ECG and movement data via 9 variables measured on 3 axes x, y, z, with 3 sensors: accelerometer, gyroscope, magnetometer. These data will be collected using a connected patch available on the market (CE marking). At the same time, the patients will benefit from a long term video-EEG examination which will be annotated by the doctors and will be used as a gold standard for the identification of seizures in order to train the algorithm. This more complete base will be used to develop an algorithm previously developed from retrospective data.