Answers are grounded in this paper and cite their sources.
Background: Brain-computer interfaces (BCIs) enable direct interaction between human cognition and external devices by interpreting brain activity via electroencephalogram (EEG) signals. However, the performance of motor imagery (MI)-based BCIs is often limited by the efficiency and accuracy of EEG feature extraction methods. Challenges such as class separability, noise resilience, and computational complexity. Aim: This study systematically evaluates various feature extraction methods used in MI-based BCIs, with a specific focus on their performance in binary and multi-class classification tasks. Methods: A literature database from resources like Web of Science, PubMed, and Google Scholar, conducting statistical and network meta-analyses. Results: Results highlight a strong preference for hybrid methods for their robustness and versatility. The Common Spatial Pattern (CSP) method remains the most common single method, often integrated into hybrids. The network meta-analysis shows that while wavelet transforms perform well in specific contexts, they struggle with task complexity and variability. Hybrid methods, in contrast, demonstrate superior stability, achieving a surface under the cumulative ranking curve (SUCRA) value of 0.811, making them ideal for high-precision tasks. Conclusion: This study highlights the critical role of effective feature extraction, advocating for the benefits of hybrid methods while underscoring the need to further refine traditional approaches.
No takes yet. Share an insight, caveat, or question.
Cheng et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: