A multi-frame network model for predicting seizure based on sEEG and iEEG data

Abstract

Introduction: Analysis and prediction of seizures by processing the EEG signals could assist doctors in accurate diagnosis and improve the quality of the patient's life with epilepsy. Nowadays, seizure prediction models based on deep learning have become one of the most popular topics in seizure studies, and many models have been presented. However, the prediction results are strongly related to the various complicated pre-processing strategies of models, and cannot be directly applied to raw data in real-time applications. Moreover, due to the inherent deficiencies in single-frame models and the non-stationary nature of EEG signals, the generalization ability of the existing model frameworks is generally poor.

Methods: Therefore, we proposed an end-to-end seizure prediction model in this paper, where we designed a multi-frame network for automatic feature extraction and classification. Instance and sequence-based frames are proposed in our approach, which can help us simultaneously extract features of different modes for further classification. Moreover, complicated pre-processing steps are not included in our model, and the novel frames can be directly applied to the raw data. It should be noted that the approaches proposed in the paper can be easily used as the general model which has been validated and compared with existing model frames.

Results: The experimental results showed that the multi-frame network proposed in this paper was superior to the existing model frame in accuracy, sensitivity, specificity, F1-score, and AUC in the classification performance of EEG signals.

Discussion: Our results provided a new research idea for this field. Researchers can further integrate the idea of the multi-frame network into the state-of-the-art single-frame seizure prediction models and then achieve better results.

Keywords: EEG; deep learning; feature extraction; multi-frame network; pre-ictal; seizure prediction.

Figure 1

Part of the data of channel 20 of chb01 in the CHB-MIT dataset, in which 0–1.5 h is the inter-ictal, 1.5–2.5 h is the pre-ictal, there is a seizure onset near 2.5 h which is marked with a red arrow, which will last for tens of seconds, and the subsequent period is the post-ictal.

Figure 2

Structure of the multi-frame seizure prediction model.

Figure 3

The heat map of the correlation coefficient matrix extracts the position information through the correlation between the electrodes. Source data is from the CHB-MIT dataset.

Figure 4

The convolutional neural network (CNN) for instanced-based feature extraction. Note that the last two dimensions of input are the number of channels. For example, the Dog_1 in the Kaggle dataset has 16 channels.

Figure 5

LSTM memory cell.

Figure 6

F1-score analysis of the three models for subjects in the Kaggle dataset.

Figure 7

The area under the curve (AUC) analysis of the three models for subjects in Kaggle dataset.

Figure 8

F1-score analysis of the three models for subjects in the CHBMIT dataset.

Figure 9

AUC analysis of the three models for subjects in the CHBMIT dataset.

Figure 10

The performance analysis of CNN-LSTM and the multi-frame network in the Kaggle dataset.

Figure 11

The performance analysis of CNN-LSTM and the multi-frame network in the CHB-MIT dataset.

Figure 12

Comprehensive performance analysis of the four models in two datasets.