Epilepsy Detection of Generalized Tonic-Clonic Seizures based on Wrist Signals

Epilepsy is a common chronic brain disease, mainly caused by abnormal brain discharges. There are many types, among which generalized tonic-clonic seizures are the most dangerous. In this context, this paper proposes a method based on a combination of Boruta algorithm, support vector machine (SVM), and particle swarm optimization (PSO) to detect systemic tonic-clonic seizures from wrist signals. First, a wearable device is used to collect signals from four channels of heart rate(HRT), number of wrist movements(NOWM), acceleration(ACC), and skin conductance response(SCR), and the corresponding features of the four channels are extracted respectively. The Boruta algorithm is introduced for feature dimensionality reduction to eliminate noise and highly correlated features. Then the particle swarm optimization algorithm is used to assist SVM in learning. The results show that after dimensionality reduction, there are only 29 features, and SVM can detect all generalized tonic-clonic seizures, with a false recognition rate of 0.083/d. Compared with previous methods, the epilepsy detection achieved through the Boruta algorithm, SVM and particle swarm optimization algorithm has lower model complexity and is easier to solve the problem of wearable device memory on the premise that the false recognition rate and accurate recognition rate are similar.