Evaluating three different adaptive decomposition methods for EEG signal seizure detection and classification

Signal processing and machine learning methods are valuable tools in epilepsy research, potentially assisting in diagnosis, seizure detection, prediction and real-time event detection during long term monitoring. Recent approaches involve the decomposition of these signals in different modes or functions in a data-dependent and adaptive way. These approaches may provide advantages over commonly used Fourier based methods due to their ability to work with nonlinear and non-stationary data. In this work, three adaptive decomposition methods (Empirical Mode Decomposition, Empirical Wavelet Transform and Variational Mode Decomposition) are evaluated for the classification of normal, ictal and inter-ictal EEG signals using a freely available database. We provide a previously unavailable common methodology for comparing the performance of these methods for EEG seizure detection, with the use of the same classifiers, parameters and spectral and time domain features. It is shown that the outcomes using the three methods are quite similar, with maximum accuracies of 97.5% for Empirical Mode Decomposition, 96.7% for Empirical Wavelet Transform and 98.2% for Variational Mode Decomposition. Features were also extracted from the original non-decomposed signals, yielding inferior, but still fairly accurate (95.3%) results. The evaluated decomposition methods are promising approaches for seizure detection, but their use should be judiciously analysed, especially in situations that require real-time processing and computational power is an issue. An additional methodological contribution of this work is the development of two python packages, already available at the PyPI repository: One for the Empirical Wavelet Transform (ewtpy) and another for Variational Mode Decomposition (vmdpy).