Seizure Types Classification by Generating Input Images With in-Depth Features From Decomposed EEG Signals for Deep...

Area of Science:

• Neurology
• Signal Processing
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) based seizure type classification is crucial for epilepsy diagnosis and prognosis.
• Distinguishing between seizure types using EEG signals presents significant challenges due to subtle signal variations.

Purpose of the Study:

• To develop an effective method for classifying different types of epileptic seizures from EEG data.
• To explore underlying EEG signal features through decomposition for improved classification.

Main Methods:

• EEG signals were decomposed using Hilbert vibration decomposition (HVD) to preserve phase information.
• 2D images were generated from high-energy subcomponents using continuous wavelet transform for deep learning (DL) input.
• A hybrid DL model combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) was employed for feature extraction.

Main Results:

• The proposed method achieved a classification accuracy of 99% and an F1-score of 99% on the Temple University EEG dataset (TUH v1.5.2).
• The HVD-based decomposition effectively extracted in-depth features for accurate seizure classification.
• The hybrid CNN-LSTM model demonstrated superior performance in classifying five seizure types and seizure-free data.

Conclusions:

• The developed HVD and hybrid DL approach offers a highly accurate and efficient solution for EEG-based seizure type classification.
• This method significantly advances the potential for improved diagnosis and prognosis in patients with epilepsy.
• The study highlights the effectiveness of signal decomposition techniques combined with advanced DL architectures in analyzing complex biomedical signals.