Enhancement of Neonatal Seizure Detection Based on EEG Signal Using PSO Feature Selection

Abstract

Objective in the newborn critical care unit, seizures are one of the most frequent medical emergencies. They are identified by visually electroencephalography (EEG) results assessment, and neuroscientific professionals are responsible for their care. This method takes a long time, and the results are inconsistent. In order to solve this issue, a system for identifying epileptic seizures from EEG data collected from healthy individuals and epileptic patients is presented in this research. This system is based on a study of EEG data utilizing the discrete wavelet transform (DWT) and both linear and nonlinear classifiers. Methods: Neonatal human EEG recordings are included in the dataset used in our method. Totally, 19-channel EEG equipments were used to record the brainwave activity of 79 term newborns referred to the NICU at the Helsinki University Hospital. Five patients with structural seizure and non seizure annotations were chosen for the pilot study from these databases. Three experts independently identified the presence of seizures and non-seizures in the EEGs using visual interpretation, while statistical features were retrieved using the PSO Algorithm and LS SVM was utilized to identify seizures. Result: To improve the model's performance, statistical features were extracted, PSO's feature selection using LS SVM, and the accuracy of C3, CZ ranges from 97.30 to 99.91. Conclusion: A reliable, autonomous epileptic seizure detection system that can be used in real time to enhance healthcare and life quality is seizure detection based on DWT statistical attributes and LS SVM.