Neuro-Cardiac Coupling: A Classification Framework Using EEG and ECG Signals for Epilepsy Detection and Cardiac Risk Assessment

Detection of epileptic seizures is still a difficult task when monitoring patients. The proposed work is aimed at developing a multimodal system that will utilize data from electroencephalography and electrocardiography in order to perform seizure detection and assessment of the cardiac risks associated with seizures. Data collected via the use of the Siena Scalp EEG dataset have been preprocessed, segmented into frames of equal size, and 784 features, which include both statistical and wavelet-related metrics, have been selected from these data, then reduced to 150 features by applying Fisher Score.
For seizure detection, a classifier of SVM-RBF type has been used and trained via a 10-fold cross-validation. Accuracy of 90.00%, sensitivity of 85.60%, and specificity of 96.55% have been achieved in the process of validation. Furthermore, HRV analysis indicated autonomic abnormalities during seizures, such as increased sympathetic activity (SDNN: 36 ± 14 ms vs. 54 ± 17 ms, p < 0.01).