Automated Seizure Detection Tailored Towards Portable Patient Care Using A Novel, Computationally Efficient Method of Analysis

Abstract

Epilepsy is a devastating neurological disorder that affects approximately 1-3% population worldwide. Wearables have recently gained more popularity as having a promising future in epilepsy management including seizure alert and close-loop therapy for severe forms of seizures. Due to the random and low frequency of seizures, seizure evaluation requires continuous, long-term EEG monitoring, which produces a large volume of data. The future treatment systems will rely on algorithms that can detect seizures with high precision and low computational cost. Previous implemented algorithms have used various computationally expensive methods of data transformation and feature extraction. In the present study, we developed a new computationally efficient seizure detection algorithm based on analysis of the broad, global shape EEG data of tonic-clonic seizures from a mouse model of temporal lobe epilepsy (TLE). To perform this algorithm, EEG data was normalized and processed through a rolling mean function, producing smoothed, simplified EEG clips that represent the global shape of the clip. These signals were then directly inputted for SVM training and testing. This novel method of seizure analysis only requires a small fraction of EEG data points, yet achieved an accuracy rate of approximately 98.51%. Our study provides a proof of principle that this simpler method could have an advantage in low-power platform such as wearables. Recently, the FDA approved a seizure detection app called embrace2 by Empatica. However, their algorithm detects seizures indirectly by monitoring heart rate and muscle contractions. Our novel algorithm detects seizures directly through analysis of brain activity. Thus, our algorithm may be better suited for future wearables in epilepsy management.