Parallel Ictal-Net, a Parallel CNN Architecture with Efficient Channel Attention for Seizure Detection

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Epilepsy affects 70 million globally, causing unpredictable seizures due to abnormal neural activity.
• Early and reliable diagnostic tools are crucial for managing epilepsy's impact on patients and families.
• Existing research has not fully explored specific EEG data subsets (D and E) from the Bonn University dataset.

Purpose of the Study:

• To introduce a novel neural network architecture, Parallel Ictal-Net (PIN), for high-accuracy EEG signal classification.
• To evaluate the PIN model's effectiveness in distinguishing between ictal (seizure) and interictal (non-seizure) states using specific EEG data subsets.
• To provide a reliable diagnostic aid for early epilepsy detection.

Main Methods:

• Utilized scalograms derived from continuous wavelet transform of EEG signals.
• Developed and implemented the Parallel Ictal-Net (PIN) neural network architecture.
• Focused analysis on EEG subsets D and E from the Bonn University dataset, corresponding to the epileptogenic zone during ictal and interictal events.

Main Results:

• The PIN model achieved high-accuracy classification of EEG signals into ictal or interictal states.
• Performance metrics including accuracy, precision, recall, and F1 scores consistently reached approximately 99%.
• The model demonstrated superior performance compared to previous approaches in the literature.

Conclusions:

• The Parallel Ictal-Net (PIN) model is highly effective for distinguishing ictal from interictal EEG events.
• The proposed method offers a reliable and accurate approach for the early detection of epilepsy.
• This advancement has the potential to significantly alleviate the social and emotional distress experienced by epilepsy patients.