A Hybrid CNN-GRU Model with Attention Mechanism for the Detection of Epileptic Seizures in EEG Signals

Epilepsy, a persistent neurological condition marked by spontaneous seizures due to uncontrolled functioning of the brain. The uncertain nature of seizures has an immense effect on patients mental health. The electroencephalography (EEG) serves as a reliable approach for diagnosing epileptic seizures. Therefore, the effective detection of seizures minimizes the risk and enhances the diagnosis. In this project, a hybrid deep learning model integrated with Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) with attention mechanism is proposed to classify the epileptic EEG data. Firstly, the raw EEG data was filtered using Butterworth bandpass filter and segmented into overlapped 4-s segments, and normalized using Z-score normalization. The hybrid CNN GRU model consists of three stages of CNN layers followed by two stages of GRU layers with an attention mechanism. CNN extracts spatial features, GRU process the obtained features from CNN to capture temporal dependencies and finally, the dot-product attention mechanism selects the significant temporal patterns to improve the model's ability in classification of seizure states. The model is trained using the CHB-MIT EEG database and evaluated with various performance metrics. The presented model achieved Accuracy -93% , sensitivity -92% , specificity -93% , and precision -93% . The resultant outcomes demonstrated the adopted model performs better than most of the existing techniques.