Combining Neural Networks and Signal Processing for Improved Electroencephalography Artifact Removal

Abstract

Electroencephalography (EEG) plays a vital role in both clinical diagnostics and neuroscience research, offering valuable insights into the dynamics of brain activity. However, the accuracy of EEG interpretation is often compromised by ocular artifacts. These artifacts—typically caused by eye blinks, eye movements, or other forms of ocular activity—can significantly distort the recorded signals and hinder reliable analysis. The difficulties of such challenges are the focus of this research; it gives way to sophisticated Blind Source Separation (BSS) methods designed for the efficient removal of ocular aberrations from electroencephalography recordings. This paper proposes three new models: an autoencoder, an autoencoder combined with Long Short-Term Memory (LSTM) networks, and a Convolutional Recurrent Neural Network (CRNN). Our models leverage temporal and geographical dependencies that enhance artifact removal, considering an 80-20 split dataset for testing and training. Added to this, the results of some evaluation measures, including Root Mean Square Error (RMSE) and Euclidean Distance (ED), hint at the CRNN model having higher efficiency in the removal of artifacts compared to the rest. On the other hand, the autoencoder and autoencoder-LSTM models present useful information related to model performance versus model complexity. These findings highlight the potential of deep learning methods to improve the interpretability of EEG signals and boost recognition performance in applications such as emotion recognition and human–computer interaction.