Seizure Type Classification Using EEG Based on Gramian Angular Field Transformation and Deep Learning.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate classification of seizure types is critical for epilepsy diagnosis and prognosis.
• Existing research predominantly focuses on seizure detection, with limited work on seizure type classification using deep learning.
• Deep learning (DL) methods show promise for advancing seizure type classification.

Purpose of the Study:

• To propose a novel deep learning (DL) approach for classifying four seizure types and seizure-free states.
• To leverage Convolutional Neural Networks (CNNs) for automatic feature extraction and classification from EEG signals.
• To evaluate the proposed method's performance on a publicly available EEG dataset.

Main Methods:

• Generated 2D images from 1D electroencephalogram (EEG) signals using the Gramian Angular Summation Field (GASF) technique.
• Employed Convolutional Neural Networks (CNNs) for automatic feature extraction and classification of the generated images.
• Utilized the Temple University Hospital (TUH, v1.5.2) EEG dataset for experimental evaluation.

Main Results:

• The proposed DL method achieved high classification accuracies: 96.01% for binary classification.
• Multiclass classification accuracies were 89.91% (3-class), 84.19% (4-class), and 84.20% (5-class).
• The results demonstrate the effectiveness of the GASF technique combined with CNNs for seizure type classification.

Conclusions:

• The proposed deep learning approach shows significant potential for accurate seizure type classification.
• This method can aid in improving the diagnosis and prognosis of patients with epilepsy.
• The integration of GASF and CNNs offers a promising direction for future research in seizure analysis.