Advanced Signal Processing and Machine/Deep Learning Approaches on a Preprocessing Block for EEG Artifact Removal: A Comprehensive Review

With recent technological and medical advancements, there is an emergence of new EEG (electroencephalogram) signal applications such as brain computer interface (BCI), rehabilitation/restoration of patients, preventive healthcare, etc. With these new applications, the artifact handling becomes a serious problem, since the acquisition of an EEG signal is much prone to biological and non-biological artifacts. These artifacts overlap with EEG components in time and in frequency domains and even in spatial domain. The use of raw EEG can be confusing and leading to misinterpretation. For example, in the medical field, this can cause several problems in dealing with patients. In engineering, the use of raw EEG signal leads to unintentional control that reduces the accuracy and the performance of the EEG-based system. One of the main challenges of artifact handling is to remove the artifacts without bringing much distortion to the neural activity. There is then a strong urge to use advanced signal processing methods, machine learning, and deep learning to handle the EEG artifacts. In this paper, we propose to review the most used methods to detect and remove the artifacts from the EEG signals. We present different proposed methods along with their theoretical background. A comparative study between different methods based on their pros and cons is conducted. We provide some guidelines to choose an appropriate method for EEG artifact handling. We also discuss some challenges that face the use of the EEG signal in the new trends of BCI applications. The possible solutions to those challenges are also discussed.