Accurate detection of seizure types is one of the important research in the field of epilepsy for effective treatment and management. The identification of potential biomarkers is difficult and challenging as the nonlinear and nonstationary characteristics of the electroencephalogram (EEG) vary within a seizure type and among the seizure types. In this work, a novel variational mode decomposition (VMD) based feature fusion approach is proposed to detect the evolution of seizures into electrographic seizures, complex partial seizures, and focal to bilateral tonic-clonic seizures. The scalp EEGs of these seizures are decomposed using VMD. The features, namely, zero crossing, Shannon entropy, approximate entropy, and root squared zeroth moment (RSZM) are extracted from bandlimited oscillatory modes. These features are employed to design a three-class classification model using random forest. In order to enhance the performance further, a fusion of RSZM measures is adapted. The proposed framework is experimented with ictal EEGs of Temple University and Hospital database. The results indicate that the six-level VMD performs better in terms of differentiating seizure types. Among the features, the RSZM is found to effectively recognize the seizure types from the first few seconds of ictal EEG. Further the fusion of RSZM results with a maximum accuracy of 96.91% for ten-fold cross-validation and 85.2% for patient-wise cross-validation. In addition, the performance of our approach is also found to be promising for the differentiation of ictal and normal EEGs when we tested on wider datasets. Therefore, the proposed framework has great potential in epilepsy diagnosis.

In this work, an attempt has been made to develop an automated system for detecting electroclinical seizures such as tonic-clonic seizures, complex partial seizures, and electrographic seizures (EGSZ) using higher-order moments of scalp electroencephalography (EEG). The scalp EEGs of the publicly available Temple University database are utilized in this study. The higher-order moments, namely skewness and kurtosis, are extracted from the temporal, spectral, and maximal overlap wavelet distributions of EEG. The features are computed from overlapping and non-overlapping moving windowing functions. The results show that the wavelet and spectral skewness of EEG is higher in EGSZ than in other types. All the extracted features are found to have significant differences (p < 0.05), except for temporal kurtosis and skewness. A support vector machine with a radial basis kernel designed using maximal overlap wavelet skewness yields a maximum accuracy of 87%. In order to improve the performance, the Bayesian optimization technique is utilized to determine the suitable kernel parameters. The optimized model achieves the highest accuracy of 96% and an MCC of 91% in three-class classification. The study is found to be promising, and it could facilitate the rapid identification process of life-threatening seizures.