Random Neural Network Based Epileptic Seizure Episode Detection Exploiting Electroencephalogram Signals

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Epileptic seizures stem from abnormal brain electrical activity, impacting awareness and requiring precise identification for effective management.
• A significant portion of patients exhibit drug-resistant epilepsy, underscoring the need for continuous monitoring and accurate detection of seizure events.
• Current monitoring involves neurophysiological signals (EEG, EMG), cardiac activity (ECG), and physical/environmental data, with EEG showing promise for distinguishing epileptic from normal brain activity.

Purpose of the Study:

• To leverage advanced deep learning algorithms for classifying various epileptic seizure types from non-seizure events using electroencephalogram (EEG) data.
• To evaluate the performance of Random Neural Networks (RNN), Convolutional Neural Networks (CNN), Extremely Random Trees (ERT), and Residual Neural Networks (ResNet) on a recent EEG dataset.

Main Methods:

• Utilized a state-of-the-art EEG dataset for training and testing deep learning models.
• Applied multiple deep learning architectures: Random Neural Network (RNN), Convolutional Neural Network (CNN), Extremely Random Tree (ERT), and Residual Neural Network (ResNet).
• Classified epileptic seizure events against non-seizure baseline activity.

Main Results:

• The Random Neural Network (RNN) model demonstrated superior performance compared to other evaluated algorithms.
• An overall accuracy of 97% was achieved by the RNN model in classifying epileptic seizures.
• Cross-validation slightly enhanced the accuracy of the RNN model, indicating robust performance.

Conclusions:

• Deep learning, particularly RNNs, offers a highly accurate method for classifying epileptic seizures from EEG data.
• The findings support the potential of advanced AI techniques for improving the diagnosis and management of epilepsy, especially in treatment-resistant cases.
• Accurate and automated seizure detection through EEG analysis can significantly aid clinicians in patient care and treatment strategies.