Automatic Seizure Detection and Prediction Based on Brain Connectivity Features and a CNNs Meet Transformers Classifier

Abstract

(1) Background: Epilepsy is a neurological disorder that causes repeated seizures. Since electroencephalogram (EEG) patterns differ in different states (inter-ictal, pre-ictal, and ictal), a seizure can be detected and predicted by extracting various features. However, the brain connectivity network, a two-dimensional feature, is rarely studied. We aim to investigate its effectiveness for seizure detection and prediction. (2) Methods: Two time-window lengths, five frequency bands, and five connectivity measures were used to extract image-like features, which were fed into a support vector machine for the subject-specific model (SSM) and a convolutional neural networks meet transformers (CMT) classifier for the subject-independent model (SIM) and cross-subject model (CSM). Finally, feature selection and efficiency analyses were conducted. (3) Results: The classification results on the CHB-MIT dataset showed that a long window indicated better performance. The best detection accuracies of SSM, SIM, and CSM were 100.00, 99.98, and 99.27%, respectively. The highest prediction accuracies were 99.72, 99.38, and 86.17%, respectively. In addition, Pearson Correlation Coefficient and Phase Lock Value connectivity in the β and γ bands showed good performance and high efficiency. (4) Conclusions: The proposed brain connectivity features showed good reliability and practical value for automatic seizure detection and prediction, which expects to develop portable real-time monitoring equipment.

Keywords: CNNs meet transformers; EEG; brain connectivity; epileptic state classification; support vector machine.

Figure 1

The block diagram of the proposed framework. FC: functional connectivity; EC: effective connectivity; PCC: Pearson correlation coefficient; PLV: phase locking value; MI: mutual information; GC: Granger causality; TE: transfer entropy; PCA: principal component analysis; SSM: subject-specific model; SIM: subject-independent model; CSM: cross-subject model; SVM: Support Vector Machine; CMT: CNNs Meet Transformers; Conv: convolution; BN: batch normalization; DW Conv: depth-wise convolution; MHSA: multi-head self-attention; FFN: feed-forward network; Avg Pool: average pooling.

Figure 2

Montage A and periods of CHB-MIT EEG data. (a) The electrode placement in Montage A. Most EEG signals were recorded using the international 10–20 electrode system, and two electrodes (FT9 and FT10) were based on the 10–10 electrode system. The EEG channels adopting bipolar montage are on the right, where each electrode’s voltage is linked and compared to an adjacent one to form a chain of electrodes. The bipolar montage can offer better artifact rejection than referential montages, and it is free of volume conduction problems [24,25,26]. (b) The definition of different EEG states. The ictal phase was extracted according to experts’ manual marks. 15 to 30 min before the onset of each seizure was defined as the pre-ictal period, so the SPH here was 15–30 min [27]. The inter-ictal state was within an interval between half an hour after the end of a seizure and before the onset of the next pre-ictal state [28].

Figure 3

Connectivity feature visualization before and after normalization. The brain connectivity feature image of 110 × 110 comprised 25 connectivity adjacency matrixes (size: 22 × 22). Every 22 columns from left to right represent the features obtained in δ, θ, α, β, and γ, respectively. Every 22 rows from top to bottom represent the features calculated by PCC, PLV, MI, GC, and TE, respectively. (a) Brain connectivity image before normalization. The ranges of connectivity values measured by different methods differed, making the difference of most feature elements unobvious. (b) Brain connectivity image after normalization. The matrix in the red rectangular box indicated the adjacency matrix computed in γ band using PCC method. All the connectivity values between each pair of channels were arranged in the form of adjacency matrix according to the channel order. The horizontal and vertical axes of this matrix represented the order and names of channels. All the connectivity values corresponding to each measure were normalized to [−1, 1]. The differences among most feature elements were more obvious for classifier learning.

Figure 4

Training and validation strategies for (a) SSM, (b) SIM, and (c) CSM.

Figure 5

Feature selection (a) strategy and (b) results. The results include the comparison of SC among connectivity methods for seizure detection (top-left) and prediction (top-right) and the comparison of SC among frequency bands for seizure detection (bottom-left) and prediction (bottom-right). The SC corresponding to each method was calculated based on the features in all five frequency bands. The SC corresponding to each band was calculated based on the features extracted by PCC and PLV methods.