EEG Microstate Features as an Automatic Recognition Model of High-Density Epileptic EEG Using Support Vector Machine

Figures

• 

  Figure 1. The technology roadmap of this paper.

• 

  Figure 2. Flow chart of EEG microstate analysis. Each EEG datum point was used to calculate the GFP curve of each data point. We plotted the potential of all electrodes at the time of local maximum of the GFP curve to generate the original topographic map. The original maps were subjected to the clustering algorithm, which grouped submitted maps into a group of clusters (4 maps) according to the terrain similarity. Finally, the clustering map was inversely fitted to the GFP curve, and each data point was marked with the most relevant clustering map. Therefore, EEG signals were converted into EEG microstate sequences.

• 

  Figure 3. Microstate category recognition process based on K-means clustering.

• 

  Figure 4. Topographies of the selected global microstate classes retrieved from the clustering algorithm for sub−bands.

• 

  Figure 5. Line charts of classification results of EEG microstate parameters in different sub-bands, in which the number of sub-bands on the x-axis is the same as in Table 4. Four criteria were used to evaluate the classifier model. Obviously, the classification effect of the classifier model in the γ-band was better than in other sub-bands.

• 

  Figure 6. The ROC curve of epilepsy classification using the SVM classifier using microstate parameters as feature inputs, plotted with recall as same as the true positive rate (TPR) on the y-axis, and specificity as the same as the false positive rate (FPR) on the x-axis. (a) The ROC curve of the EEG microstate parameters in the delta (0.5~4 Hz) frequency band using the SVM classifier; (b) The EEG microstate parameters in the θ (4~8 Hz) frequency band using SVM. The ROC curve of the classifier classification; (c) The ROC curve of the α (8~13 Hz) frequency band EEG microstate parameters using the SVM classifier; (d) The β (13~30 Hz) curve. The ROC curve of the EEG microstate parameters of the frequency band classified by the SVM classifier; (e) The ROC curve of the EEG microstate parameters of the γ (30~45 Hz) frequency band classified by the SVM classifier; (f) The ROC curve of the EEG microstate parameters in the 45~80 Hz frequency band using the SVM classifier.