Exploration and Analysis of Seizure Spikes Through Spectral Domain Transformation

Seizure detection is the most crucial area of investigation when it comes to understanding brain disorders. This proposed research study embarked on an automated model for epileptic seizure diagnosis by means of different kinds of Spectral transformation using EEG inputs from seizure sufferers and healthy subjects. This automated model accommodates non-invasive brain electrical activity monitoring. This method aims to facilitate the analysis and identification of epileptic seizure states since, monitoring and diagnosing such brain electrical activity is a complex task due to its numerous divisions and underlying features. The primary objective of this research study is to distinguish between EEG-based seizures and healthy individuals. To achieve this goal, a combination of spectral transformation and EEG analysis techniques is utilized. These techniques include examining the frequency spectrum, magnitude spectrum, correlation, and T-Distributed Stochastic Neighboring Embedding (T-SNE) analysis. This analysis yields valuable insights from EEG data, refining the input data and making it more suitable for prediction and identification. The model's performance is evaluated using two distinct datasets: real-time EEG data from individuals experiencing epileptic seizures and EEG data from healthy subjects. These datasets are sourced from the Bangalore EEG Epilepsy Dataset (BEED), India and the BONN epilepsy dataset from the UCI repository. In a comparative study of spectral transformation methods, including Complex Fast Fourier Transform (CFFT) and Real-Valued Fast Fourier Transform (RFFT), it is discovered that reducing the data dimension by using feature extraction is not the optimal approach. This simplification leads to the loss of valuable information. Therefore, preserving the full spectrum of EEG characteristics is crucial for gaining valuable insights into brain neuronal functions, ultimately enabling more accurate seizure prediction.