Seizure onset zone classification of intracranial EEG signals from epilepsy patients

The intracranial electroencephalographic (iEEG) signals contain information about seizures and their onset location. Several seizure onset patterns are used in the literature to detect onset zones, which have clinical significance in neurosurgery. This study introduces a supervised machine learning method to detect seizure onset patterns from iEEG signals. The iEEG signals were marked three seconds long at the time of each seizure onset pattern. Thirty-five features representing the time, frequency, and time-frequency domain characteristics were extracted from 100 seizures originating from 29 patients, collectively containing 297 seizure channels. A Linear Discriminant analysis (LDA) is applied to the features to classify seizure onset patterns. The classification output is the class of each seizure signal, which is either non-SOZ or a class of six onset patterns. The classifier performance is assessed through accuracy, precision, recall, and F-score metrics. In summary, employing machine learning to classify seizure onset patterns can offer an objective approach, and the method could help neurosurgeons objectively choose seizure onset locations based on various ambiguous patterns.