Multi-scale kernel and electrode attention network for EEG-based epileptic seizure detection.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Epilepsy is a common neurological disorder requiring efficient diagnostic tools.
• Automated seizure detection using electroencephalography (EEG) is crucial for clinical practice.
• Current deep learning methods for EEG analysis face limitations like information loss or high computational costs.

Purpose of the Study:

• To develop a novel deep learning network for accurate and efficient epileptic seizure detection.
• To overcome limitations of existing methods by processing full-channel EEG signals without reduction or transformation.
• To enhance feature representation and discriminative capability for improved seizure detection.

Main Methods:

• Proposed the Multi-scale Kernel and Electrode Attention Network (MKEANet), an end-to-end network operating on raw, full-channel EEG data.
• Utilized multi-scale convolutional structures to capture diverse electrode channel combinations and adapt to temporal waveform variations.
• Introduced an Electrode Attention Module (EAM) to adaptively weight channels and enhance spatial modeling.

Main Results:

• MKEANet achieved state-of-the-art performance on two public datasets (CHB-MIT and Siena Scalp).
• Achieved 99.29% accuracy, 98.29% sensitivity, and 99.59% specificity on the CHB-MIT dataset.
• Achieved 99.42% accuracy, 98.69% sensitivity, and 99.54% specificity on the Siena Scalp dataset.

Conclusions:

• MKEANet effectively detects epileptic seizures using raw, full-channel EEG signals without information loss.
• The network's design, incorporating multi-scale kernels and electrode attention, enhances spatial and temporal feature extraction.
• Demonstrated superior performance in resource-constrained settings, offering a promising tool for epilepsy diagnosis.