Daily Epileptic Seizure Detection Algorithm Based on Multi-modal Physiological Signals

Epilepsy seizure has brought great harm to patients and their families, and generalized tonic-clonic seizure is one of the most dangerous types of epilepsy. It is important in terms of research significance and practical value to detect epilepsy patients' seizures based on wearable devices outside the hospital. In this paper, a wristband is used to simultaneously collect multi-modal physiological signals of epilepsy patients and healthy people under out-of-hospital scenarios. Aiming at the problem of data imbalance, this paper extracts the three-axis continuous noncross ratio feature to eliminate obvious nonseizure data. Then extract features from seizure data of patients and nonseizure data of healthy people and perform feature dimensionality reduction. Finally, a seizure detection model is constructed based on the random forest algorithm. The seizure detection model achieves an average sensitivity of 90% and a false alarm rate of 1.21 times/24h as performing leave one seizure out cross-validation. An average false alarm rate of 1.74 times/24h can be achieved on patients' nonseizure data testing.