Patient-specific seizure detection method using nonlinear mode decomposition for long-term EEG signals.

The automated detection technique becomes the inexorable trend of medical development of the world. The objective of the work is to explore a feasible approach for patient-specific seizure detection in long-term electroencephalogram (EEG) recordings. For this purpose, a novel method based on nonlinear mode decomposition (NMD) has been proposed in this study. A sliding window is used on the multi-channel EEG, where four selected channels have been segmented into a series of successive short epochs with a 2-s duration. Then, the EEG is decomposed into a set of nonlinear modes (NMs) by the NMD algorithm and one type of statistical parameter named fractional central moment (FCM) is calculated over the first two NMs constituting the input feature vector to be fed to three common classifiers. The proposed features, when using K nearest neighbor (KNN), are able to detect seizures with high sensitivity values across all patients consistently. We have explored the ability of the FCM in NMD domain for classification of seizure and non-seizure EEG signals. Our approach has achieved the average sensitivity, specificity, and accuracy values as 98.40%, 99.10%, and 98.61%, respectively, over all the data groups on CHB-MIT database. The experimental results have indicated that the proposed method is not only quite reliable in diagnosing seizure with single type of feature yielding satisfied performance but also robust to variations of seizure types among patients. In this regard, it can be expected that our proposed method is endowed with promising prospects for the use of an expert software application in real-time automated seizure detection. Graphical abstract.