Total tonic clonic seizure recognition of wrist signals

It is extremely dangerous for epilepsy patients to become sick when no one is accompanying them in their daily life. The alarm for epilepsy patients can be timely notified to their families to take measures. In this context, a scheme for the identification of general tonic-clonic seizures (GTCs) based on wrist signals is proposed. Firstly, features were extracted from wrist acceleration(ACC), skin conductance response(SCR), number of wrist movements(NOWM) and heart rate(HR) signals. Secondly, in order to reduce the interference of unnecessary features on classification, feature dimensions were reduced by random forest algorithm. Finally, the number of normal data samples is much larger than the number of diseased data samples, and the training model is adopted to sacrifice the accuracy of identifying diseased data and improve the accuracy of identifying normal data. The detection and recognition effects of SVM (Support vector machine), AdaBoost and XGBoost machine learning models are compared. The results showed that the SVM algorithm could recognize all GTCs episodes (median 39.5s, range 5-69s) in the 10 data with a false recognition rate (FRR) of 0.08/d when the continuous predicted onset time reached 9s. When the predicted onset time reaches 19s, the three algorithm models can effectively reduce FRR, but at the same time, more underreporting will be generated. GTCs seizures can be detected through wrist signals, and it has good recognition effect and low FRR, which is conducive to the development of wearable epilepsy recognition devices.