AUTHOR=Walther Dominik , Viehweg Johannes , Haueisen Jens , Mäder Patrick 

TITLE=A systematic comparison of deep learning methods for EEG time series analysis

JOURNAL=Frontiers in Neuroinformatics

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf.2023.1067095

DOI=10.3389/fninf.2023.1067095

ISSN=1662-5196

ABSTRACT=Analyzing time series data like EEG or MEG is challenging due to noisy, high-dimensional, and patient-specific signals. Deep learning methods have been demonstrated to be superior in analyzing time series data compared to shallow learning methods utilizing handcrafted and often subjective features. Especially, recurrent deep neural networks (RNN) are considered suitable for analyzing such continuous data and have been previously studied, but are computationally expensive and difficult to train. In contrast, feed-forward networks (FFN) have previously mostly been considered in combination with hand-crafted and problem-specific feature extractions, such as short time Fourier and discrete wavelet transform. A sought-after are easily applicable methods that efficiently analyze raw data removing the need for problem-specific adaptations. In this work, we systematically compare recurrent as well as feed-forward topologies to provide an update and guideline for researchers dealing with the automated analysis of EEG time series data. We compare the different FFN and RNN topologies on multiple datasets and found that a recurrent LSTM architecture with attention performs best on the less complex tasks, while the temporal convolutional network (TCN) outperforms all the recurrent architectures on the most complex dataset yielding a 14% accuracy improvement. In general, we found the attention mechanism to substantially improve classification results of RNNs. Towards a light-weight and online learning-ready approach, we found extreme learning machines (ELM) to yield comparable results for the less complex tasks.