Interictal Epileptiform Discharge Detection Using Dual-Domain Features and GAN

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Interictal Epileptiform Discharges (IEDs) are crucial for epilepsy diagnosis but challenging to detect due to EEG signal complexity.
• Manual EEG interpretation is time-consuming and subjective, necessitating automated solutions.
• Existing deep learning methods like CNNs struggle with long-term dependencies in EEG time-series data.

Purpose of the Study:

• To develop an automated system for reliable Interictal Epileptiform Discharge (IED) detection using advanced deep learning.
• To overcome the limitations of CNNs in capturing long-term dependencies in EEG signals.
• To address the challenge of limited IED data through data augmentation.

Main Methods:

• Proposed an IED Detector (IEDD) combining convolutional layers for local feature extraction and a Transformer for modeling long-term dependencies.
• Utilized a 3D tensor representation of EEG data with channel topology for spatial feature extraction via CNN and temporal modeling with LSTM.
• Developed a novel Transformer-based Generative Adversarial Network (GAN) for augmenting scarce IED datasets.

Main Results:

• Achieved high accuracy in binary classification: 96.11% on augmented Dataset 1 and 95.25% on Dataset 2.
• Demonstrated strong performance in multi-label classification with an average sensitivity of 87.26% and precision of 89.96%.
• The integrated CNN-Transformer architecture effectively captures both local and long-term dependencies in EEG data.

Conclusions:

• The proposed IEDD model shows significant promise for automated and accurate IED detection.
• Transformer-based GANs offer an effective solution for augmenting limited neurological datasets.
• This work advances deep learning applications in epilepsy diagnosis and EEG analysis.