A Robust Deep Learning Approach for Automatic Classification of Seizures Against Non-seizures

Abstract

Identifying epileptic seizures through analysis of the electroencephalography (EEG) signal becomes a standard method for the diagnosis of epilepsy. Manual seizure identification on EEG by trained neurologists is time-consuming, labor-intensive and error-prone, and a reliable automatic seizure/non-seizure classification method is needed. One of the challenges in automatic seizure/non-seizure classification is that seizure morphologies exhibit considerable variabilities. In order to capture essential seizure patterns, this paper leverages an attention mechanism and a bidirectional long short-term memory (BiLSTM) to exploit both spatial and temporal discriminating features and overcome seizure variabilities. The attention mechanism is to capture spatial features according to the contributions of different brain regions to seizures. The BiLSTM is to extract discriminating temporal features in the forward and the backward directions. Cross-validation experiments and cross-patient experiments over the noisy data of CHB-MIT are performed to evaluate our proposed approach. The obtained average sensitivity of 87.00%, specificity of 88.60% and precision of 88.63% in cross-validation experiments are higher than using the current state-of-the-art methods, and the standard deviations of our approach are lower. The evaluation results of cross-patient experiments indicate that, our approach has better performance compared with the current state-of-the-art methods and is more robust across patients.

Deep Multi-View Feature Learning for EEG-Based Epileptic Seizure Detection

“…The dataset of Patient 12 (the 12th group) corresponds to a special case that the dataset has been excluded from the experiments in many previous studies [1][2][3][4][5]. In [8,10,12] and our study, among the 24 groups of data used in the experiments, the performance of this 12th group is found to be particularly inferior. This suggests that the EEG signals of the 12th group is highly unstable and irregular which introduces a lot of interference to the learning process of the algorithms, thus leading to poor performance in most cases.…”

“…A major goal of EEG-based epilepsy detection is to make such conversion as fast and accurate as possible. A variety of epilepsy detection algorithms have been proposed in recent years [8,10,12,14,22]. In [22], a method adopting transfer learning and semi-supervised learning is used to classify the status of epilepsy with EEG signals.…”

“…In [14], multi-view learning technology is used for automatic recognition of epileptic EEG signals based on shallow features. In [8,10,12], deep learning techniques are used to automatically classify epilepsy by using deep features. These algorithms apply multi-view learning and deep learning techniques separately and demonstrate promising performance.…”

“…We further analyze the proposed method by making comparison with the feature extraction methods proposed in recent studies [1][2][3][4][5]8,10,12] that exploit deep learning and use the same CHB-MIT dataset for seizure detection. Table II gives the feature extraction methods, experimental settings and the classification performance (in the table, NA meaning "not applicable", indicating that the data is not used in the study).…”

“…To avoid the problem, 25% of the samples are only used for training in [3][4][5]. In [8,10,12], different over-sampling methods have been adopted to increase the number of seizure samples. As a result of data imbalance, the accuracy and sensitivity of most of the algorithms under comparison are low.…”