A Hybrid Deep Learning Model for Early Detection of Epileptic Seizures

Epilepsy, a condition affecting millions worldwide is characterized by recurring seizures. EEG (Electroencephalography) offers a direct assessment of the brain’s electrical activity. This non-invasive measurement technique, with its high temporal resolution, enables the detection of specific seizure patterns. Early and precise prediction of seizures is crucial in preventing consequences like status epilepticus and sudden unexpected death in epilepsy (SUDEP). The objective of this study is to develop an advanced seizure detection algorithm with high accuracy. In our proposed model band analysis and feature extraction are done using Short Time Fourier Transform (STFT) and these features are passed to a hybrid model of CNN2D (network with 2D input) and Bidirectional Long Short Term Memory (BiLSTM) model. This model is validated across various patient profiles. Results indicate that the CNN2D+BiLSTM model outperforms other existing models in terms of prediction accuracy and speed. Furthermore, adaptive learning mechanisms enhance detection capabilities over time. This study makes use of the CHB MIT Scalp EEG Database and the Siena Scalp EEG Database, which contain EEG recordings from 500 individuals. Key performance measures include: OVLP: Overlap ratio signifying intersection between true and predicted events. EPOCH: Analyzing the alignment of labels and predictions in a time frame, for each EEG channel. ROC Curve: Illustrates the balance between sensitivity and specificity.