Classifying EEG Signals for Seizure Detection with Non-Linear Vector Decomposed Neural Network (NVDN) Techniques

A seizure is a neurological phenomenon that results from episodic pathological neuronal discharge in the brain, and can be recorded on electroencephalogram (EEG). Seizure detection is a major determinant of seizure management and therefore timely seizures detection is important in managing the problem. This work suggests the use of new NonLinear Vector Decomposed Neural Network analysis methods in the context of the classification of EEG signal to detect seizures. By decomposing a relatively complex structure of signals that are recorded from the scalp through the use of the NVDN framework, the model is able to better decode the constellations of features that may be indicative of seizure activity. To that end, we used an extensive data of EEG signals which was preprocessed by cleaning up noise and artifacts before performing feature extraction based on time frequency analysis. NVDN model was trained and tested against traditional machine learning algorithms to show enhanced accuracy sensitivity and specificity. The results show that with the help of the proposed NVDN technique, one can remarkably enhance seizure detection, which opens up the possibility of creating real-time monitoring systems for clinical practice. The findings of this work advance current knowledge of intelligent seizure detection systems and improve patient outcomes.