EEG Absence Seizure Detection with Autocorrelation Function and Recurrent Neural Network

Epilepsy patients experience challenges in daily life, and epilepsy seizures might cause injuries or endanger the life of the patients or others. The electroencephalogram (EEG) signals, recorded by a machine or device, are often used to analyze the brain electrical activity, which is noninvasive. Locating the seizure period in EEG recordings is usually difficult and time consuming for doctors. Therefore, automatic detection of seizures is necessary. In this paper, we use the autocorrelation function to extract the EEG features, and propose a method based on Recurrent Neural Network to detect the seizure period of the EEG signal, which combines the gated recurrent unit and a 1-D convolutional embedding head. We use the clinical EEG recording of 15 patients to simulate the results of our proposed method. The experimental results demonstrate that our method achieves an excellent performance with 99.6% detection accuracy for Absence Seizure, which can greatly reduce the workload of doctors in clinical diagnosis.