Attention-Based LSTM System for Epileptic Seizure Detection from EEG Signals

The electroencephalogram (EEG) has become one of the most important tools for clinicians to detect seizures and other neurological irregularities of the human brain over the past few decades. An accurate diagnosis of epilepsy is essential due to its unique characteristics and the adverse consequences of epileptic seizures on individuals. An urgent need exists for an automated epilepsy detection system utilizing electroencephalography (EEG) for clinical use. This paper employs the Discrete Wavelet Transform (DWT) to decompose EEG signals into multiple subbands. Various features used to discriminate spike events and extracted from each subband signal of an EEG trial. The attention mechanism augments the network's capacity to concentrate on discriminative features and temporal steps within the feature sequence, thereby enhancing interpretability and detection precision. The weighted number of features effectively distinguishes the underlying characteristics of EEG signals indicative of seizure and non-seizure events. The attention mechanism utilizing Long Short-Term Memory (LSTM) is employed to classify seizure and non-seizure EEG signals. The effectiveness of the proposed method is assessed through multiple experiments utilizing a public dataset acquired from the University of Bonn. The experimental findings indicate that the proposed seizure detection method obtains a classification accuracy of 99.65%, outperforming the performance of current techniques. The efficacy of the LSTM with attention model is compared with support vector machine classifiers, which exhibit a classification accuracy of 98.52%. Thus, the proposed method is validated as a potential indicator for EEG-based seizure detection