Eye blink is one of the most common artifacts in electroencephalogram (EEG) and significantly affects the performance of the EEG related applications, such as epilepsy recognition, spike detection, encephalitis diagnosis, etc. To achieve an accurate and efficient eye blink detection, a novel unsupervised learning algorithm based on a hybrid thresholding followed with a Gaussian mixture model (GMM) is presented in this paper. The EEG signal is priliminarily screened by a cascaded thresholding method built on the distributions of signal amplitude, amplitude displacement, as well as the cross channel correlation. Then, the channel correlation of the two frontal electrodes (FP1, FP2), the fractal dimension, and the mean of amplitude difference between FP1 and FP2, are extracted to characterize the filtered EEGs. The GMM trained on these features is applied for the eye blink detection. The performance of the proposed algorithm is studied on two EEG datasets collected by the Temple University Hospital (TUH) and the Children's Hospital, Zhejiang University School of Medicine (CHZU), where the datasets are recorded from epilepsy and encephalitis patients, and contain a lot of eye blink artifacts. Experimental results show that the proposed algorithm can achieve the highest detection precision and F1 score over the state-of-the-art methods.

Benign childhood epilepsy with centro-temporal spikes (BECT) is one of the most common epilepsy syndromes in childhood which is typically characterized by localized discharges in the central and temporal regions. Traditionally, the recognition of spikes requires visual assessment of long-term EEG recordings which is time consuming and subjective because it depends on the knowledge and experience of the doctor. Therefore, a novel multi-step spike detection algorithm based on average reference (AV) channel and bipolar (BP) channel BECT EEG is proposed, including candidate spike detection algorithm, false positive spike (FPS) elimination, spike feature extraction and random forest (RF) classification. The proposed method is evaluated using 7 routine EEG recordings. This paper shows that the sensitivity (Sen), specificity (Spe), selectivity (Sel) and accuracy (AC) obtained by the proposed method are 97.4%, 96.5%, 96.6% and 96.9%, respectively. Experimental results show that the proposed method is capable of detecting BECT spikes efficiently.

Epilepsy syndromes are typical childhood nervous system diseases, usually containing several different seizure types commonly seen. There exist around 20 known childhood epilepsy syndromes and accurate classifying these epilepsy syndromes plays a vital guidance to diagnosis and treatment. Current research mainly focused on seizure detection, while few have been presented on epilepsy syndrome classification. In this paper, a novel two-stream 3D attention module based deep network (TSA3D) is developed to exploit multi-channel time-frequency and frequency-space features of interictal EEGs for epilepsy syndrome classification. Two multi-channel montage transforms are applied for EEG feature optimization to reduce artifact affection. In TSA3D, both the channel-wise (CA) and dual spatial attention (DSA) based 3D attention modules are developed to enhance the EEG feature learning capability. The study is conducted on the long-term interictal EEGs of 115 subjects from the Children’s Hospital, Zhejiang University School of Medicine (CHZU). These data belong to 7 typical childhood epilepsy syndromes and 1 normal control group. Experimental results show that TSA3D can achieve an overall accuracy of 99.52%, the highest among compared state-of-the-art (SOTA) methods.

Simultaneous childhood epilepsy syndrome classification and seizure detection are both significant in epilepsy analysis. Current research mainly focuses on a single task, mostly on seizure detection. In this paper, a novel dual-stream multi-task network (DSMN) exploiting multi-channel scalp electroencephalograms (EEGs) is developed to simultaneously perform epilepsy syndrome classification (ESC-Task) and seizure detection (SD-Task), in short as DSMN-ESS. The close correlation between ESC-Task and SD-Task is explored to achieve better performance. To improve the performance, an information sharing gate module is designed in DSMN to enable both tasks to fully obtain the useful information. Meanwhile, a channel weight update module is developed to well extract the internal spatial relationship between multi-channel EEGs. Further, an area-under-the-curve (AUC) based loss is proposed to address the data imbalance issue in epilepsy analysis. Studies on EEG data recorded 49 patients from the Children’s Hospital, Zhejiang University School of Medicine (CHZU), are carried out to show the effectiveness of DSMN-ESS. The results show that DSMN-ESS can achieve the highest AUC, 99.95% and 99.78% in ESC-Task and SD-Task, respectively, which are superior over several state-of-the-art (SOTA) methods.

Detection of epileptic seizure from offline electroencephalogram (EEG) is of great significance in clinical diagnosis. Traditional epileptic seizure detection methods are usually based on the basic assumption that the training and testing data are sampled from data sets with the same distribution. However, in the context of epilepsy diagnosis, the EEG data vary from subject to subject, and the generalization performance of a classifier trained on data of multiple subjects typically degrades when applied to new subjects. To address this issue, we propose a cross-subject transfer learning framework for epileptic seizure detection to improve the classification performance on new subjects with unlabeled EEG samples (target domain) by transferring useful information from multiple subjects with labeled EEGs (source domain). In detail: 1) An adversarial strategy is used to identify a set of source-domain EEG samples that are most suitable for transfer learning and thus are selected as training samples for the follow-up domain adaptation. 2) A noval domain adaptation method, the Joint-Probability-Discrepancy-Based Domain Adaptation (JPDDA), is proposed to predict the labels associated with the target-domain samples. Specifically, joint probability distribution discrepancy that measures the transferability between domains and discriminability between classes is proposed to learn a domain-invariant classifier jointly with structural risk and manifold consistency. 3) The epileptic seizure detection framework based on JPDDA is validated on the Children’s Hospital, Zhejiang University School of Medicine (CHZU) dataset. Experimental results show that the proposed JPDDA can achieve high cross-subject detection accuracy, which reveals the good transferability of JPDDA.