Classification of epileptic seizures in EEG data based on iterative gated graph convolution network.

Area of Science:

• Neuroscience
• Machine Learning
• Medical Diagnostics

Background:

• Accurate epilepsy type classification using electroencephalogram (EEG) data is crucial for patient diagnosis.
• Traditional Graph Convolutional Neural Networks (GCN) face challenges with predefined graph topologies and capturing long-term temporal dependencies in EEG signals.

Purpose of the Study:

• To develop an advanced epileptic seizure classification model that overcomes limitations of existing GCN approaches.
• To improve the precision and efficiency of epilepsy diagnosis through automated EEG analysis.

Main Methods:

• Proposed an Iterative Gated Graph Convolutional Network (IGGCN) model for epileptic seizure classification.
• Implemented iterative graph optimization with multi-head attention and Gated Graph Neural Networks (GGNN) to capture complex brain region correlations and long-term EEG dependencies.
• Utilized Focal Loss to address data imbalance issues common in epileptic EEG datasets.

Main Results:

• Achieved outstanding performance on the Temple University Hospital EEG Seizure Corpus (TUSZ) for classifying four epileptic seizure types.
• Attained an average F1 score of 91.5% and an average Recall of 91.8%, significantly outperforming current state-of-the-art models.
• Ablation studies confirmed the effectiveness of iterative graph optimization, gated graph convolutions, and Focal Loss.

Conclusions:

• The IGGCN model demonstrates superior capability in analyzing complex EEG data for epilepsy classification.
• Dynamic graph structure optimization and enhanced temporal feature extraction are key to improving diagnostic accuracy.
• The proposed method offers a significant advancement in the automated diagnosis of epilepsy.