Epileptic Seizure Detection Based on EEG Signals and CNN

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Epilepsy affects 50 million people globally, necessitating accurate and timely diagnosis for effective treatment.
• Electroencephalography (EEG) is crucial for epilepsy monitoring and diagnosis, but manual analysis is time-consuming.
• Automated seizure detection can reduce diagnostic delays and improve patient outcomes.

Purpose of the Study:

• To evaluate the performance of a convolutional neural network (CNN) for epileptic seizure detection using raw EEG signals.
• To compare the efficacy of time and frequency domain EEG signals for classifying ictal, preictal, and interictal states.
• To assess the potential of frequency domain signals over time domain signals in CNN-based epilepsy diagnosis.

Main Methods:

• A CNN model was developed to analyze raw EEG signals without manual feature extraction.
• Experiments were conducted using intracranial (Freiburg) and scalp (CHB-MIT) EEG databases.
• Three classification tasks were performed: interictal vs. preictal, interictal vs. ictal, and interictal vs. preictal vs. ictal.

Main Results:

• Frequency domain signals achieved high average accuracies (92.3-97.5%) across both databases and all classification tasks.
• Time domain signals showed significantly lower accuracies (47.9-91.1%), with effective identification only for some patients.
• CNNs demonstrated superior performance with frequency domain EEG signals compared to time domain signals for epilepsy detection.

Conclusions:

• Frequency domain EEG signals are highly effective for automated detection of epileptic seizure states (ictal, preictal, interictal) using CNNs.
• Time domain signals are less reliable for CNN-based epilepsy detection, particularly in scalp EEG data.
• Frequency domain analysis offers greater potential for advancing CNN applications in epilepsy diagnosis and monitoring.