Physiological artifacts removal in EEG signals: a comprehensive overview of conventional to deep learning methods to support brain health monitoring

Electroencephalography (EEG) is a crucial technique for interpreting neurocognitive functions and the underlying neural information. Despite its advantages, such as high portability, cost-effectiveness and exceptional temporal resolution, EEG is susceptible to contamination by various artifacts arising from nonneural sources, including ocular movements, cardiac activities, muscle contractions and environmental factors. These artifacts significantly hinder the accurate interpretation of neural activity and can lead to the misinterpretation of brain information. Therefore, effective artifact removal is vital for enhancing the EEG signal quality. Although numerous techniques have been developed to address this issue, no single method has been considered optimal for all applications. This comprehensive review traces the evolution of artifact removal methods from conventional to deep learning techniques, providing valuable insights for researchers. The study also highlights the challenges associated with artifact elimination. In addition, this work proposes a conceptual hybrid framework integrating convolutional neural networks (CNNs) and recurrent neural networks (RNNs) as a potential future direction for EEG artifact removal. Furthermore, this review discusses open-source datasets and performance metrics commonly employed to evaluate the efficacy of artifact removal techniques. This study can serve as a foundation for future research and the development of EEG signal processing.