SeizureNet-BiLSTM: A Hybrid Deep Learning Framework for Identifying Ictal and Interictal Phases Prabhat Kumar Upadhyay Department of Electrical & Electronics Engineering Birla Institute of Technology, Mesra Ranchi, India pkupadhyay@bitmesra.ac.in Priyaranjan Kumar Department of Electrical & Electronics Engineering Birla Institute of Technology, Mesra Ranchi, India priya7870@gmail.com Abstract—A neurological condition called epilepsy is charac- terized by intermittent, frequent seizures. Traditionally, seizure identification has depended on neurologists’ laborious visual examination of electroencephalogram (EEG) data. This article introduces a novel seizure detection model, SeizureNet-BiLSTM (Bidirectional Long Short-Term Memory), which integrates deep convolutional networks (ResNet-like structures) and bidirectional LSTMs to detect seizures across both ictal and inter-ictal phases. The process begins by extracting ictal and inter-ictal segments from raw multichannel EEG data, followed by feature extraction using Mean, Skewness, Kurtosis, Short-Time Fourier Transform (STFT), and Power-Spectral-Density (PSD). These features are then input into the suggested model for seizure detection. Ex- perimental results validate a test accuracy of 99.66%, sensitivity of 98.49%, specificity of 98.49%, a false positive rate (FPR) of 0.02107, and F1-score of 97.94%. Tested on the public CHB- MIT (Children’s Hospital Boston - Massachusetts Institute of Technology) dataset, the SeizureNet-BiLSTM proves to be a computationally efficient and effective approach for automated seizure detection. Index Terms—Feature Extraction in EEG, Epilepsy, Ictal and Inter-ictal Classification, Seizure Detection, Deep Learning. I. I NTRODUCTION Epilepsy is a neurological condition that originates from the central nervous system, caused by abnormal electrical activity in the brain, which causes recurrent seizures. These often affect behaviors. Epilepsy is among the most common neu- rological disorders; the World Health Organization estimates more than 50 million people around the world are afflicted with epilepsy [1]. The electroencephalogram (EEG) serves as a critical diagnostic tool for clinicians in evaluating epilepsy patients. EEG is particularly valuable in distinguishing various phases of seizure activity, including pre-ictal, ictal, inter-ictal, and post-ictal periods. Fig. 1 demonstrates an EEG wave- form illustrated the different phases associated with Epilepsy seizures. • Pre-ictal: This is the phase leading up to a seizure, often marked by subtle changes in brain activity that may precede the onset of ictal events. Identifying these changes is crucial for predicting seizures in real-time. Fig. 1. EEG signal spectrum after applying short-time Fourier transform. • Ictal: This is the active phase of the seizure, characterized by abnormal, synchronized electrical discharges in the brain. It is during this phase that patients exhibit seizure symptoms, such as convulsions or altered awareness. • Post-ictal: Following the ictal phase, this period involves recovery, where patients may experience confusion, fa- tigue, or other symptoms as the brain returns to its normal state. • Inter-ictal: This refers to the period between seizures, where the brain’s electrical activity is relatively normal. However, subtle disruptions in brain activity can still occur during this time, making it important for continuous monitoring. The accurate detection and prediction of epileptic seizures rely heavily on analyzing and interpreting EEG signals through sophisticated feature extraction techniques. Statistical mea- sures like Mean, Variance, Skewness, and Kurtosis provide initial insights by describing the amplitude and distribution characteristics of EEG signals [2], [3]. Frequency domain features, such as Band Power, and Power Spectral Density (PSD), delve into the power distribution across different EEG frequency bands, highlighting changes during seizures [4]. Time-frequency techniques like the Wavelet Transform and Short-Time Fourier Transform (STFT) offer dynamic, 2024 7th International Conference on Signal Processing and Information Security (ICSPIS) 979-8-3503-6867-3/24/$31.00 ©2024 IEEE 2024 7th International Conference on Signal Processing and Information Security (ICSPIS) | 979-8-3503-6867-3/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICSPIS63676.2024.10812648 Authorized licensed use limited to: Birla Institute of Technology. Downloaded on January 28,2025 at 06:52:53 UTC from IEEE Xplore. Restrictions apply.