Scholarlink

<p>Motor Imagery (MI)-based Brain-Computer Interfaces (BCIs) depend on interpretable and robust feature extraction methods. Although Common Spatial Pattern (CSP) algorithms effectively discriminate neural patterns, their practical and clinical utility remains limited by the lack of explainability and sensitivity to noise and inter-subject variability. This paper introduces a framework known as balanced and explainable common spatial pattern (BECSP), which integrates adaptive feature selection and explainability within a unified CSP-based decoding pipeline. The proposed framework combines a RuleFit-based feature-ranking mechanism and complementary explainability techniques to identify the most meaningful spatial filters, rather than relying on a fixed number of features. Comprehensive evaluations on the BCI Competition datasets (IIIa, IVa, and IIa) demonstrate that the proposed BECSP framework achieves competitive classification performance while improving robustness and interpretability compared to conventional CSP approaches, with a mean accuracy of 79.89 ± 0.69% versus 77.93%. The proposed framework showed greater benefits for low-performing subjects, where adaptive feature selection yielded more stable and physiologically meaningful CSP representations. Quantitative analysis further demonstrates a statistically significant improvement in robustness to inter-subject variability (<InlineEquation ID="IEq1"><EquationSource Format="TEX">\(p&lt;0.05\)</EquationSource><EquationSource Format="MATHML"><math><mrow><mi>p</mi><mo>&lt;</mo><mn>0.05</mn></mrow></math></EquationSource></InlineEquation>, Wilcoxon signed-rank test). Furthermore, electrode-importance heatmaps and topographical spatial-filter visualizations consistently emphasize sensorimotor cortical regions associated with MI activity, supporting the physiological relevance of the proposed framework. For practical deployment, the proposed explainability interface assists BCI developers in identifying and analyzing potential misclassifications. Overall, the proposed framework contributes toward more reliable and interpretable ElectroEncephaloGraphy (EEG)-based BCIs for neurorehabilitation and assistive technology applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable common spatial pattern to improve the design of brain-computer interfaces

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • Kais Belwafi,
  • Sofien Gannouni,
  • Bouziane Brik,
  • Hatim Aboalsamh,
  • Hassan Mathkour

摘要

Motor Imagery (MI)-based Brain-Computer Interfaces (BCIs) depend on interpretable and robust feature extraction methods. Although Common Spatial Pattern (CSP) algorithms effectively discriminate neural patterns, their practical and clinical utility remains limited by the lack of explainability and sensitivity to noise and inter-subject variability. This paper introduces a framework known as balanced and explainable common spatial pattern (BECSP), which integrates adaptive feature selection and explainability within a unified CSP-based decoding pipeline. The proposed framework combines a RuleFit-based feature-ranking mechanism and complementary explainability techniques to identify the most meaningful spatial filters, rather than relying on a fixed number of features. Comprehensive evaluations on the BCI Competition datasets (IIIa, IVa, and IIa) demonstrate that the proposed BECSP framework achieves competitive classification performance while improving robustness and interpretability compared to conventional CSP approaches, with a mean accuracy of 79.89 ± 0.69% versus 77.93%. The proposed framework showed greater benefits for low-performing subjects, where adaptive feature selection yielded more stable and physiologically meaningful CSP representations. Quantitative analysis further demonstrates a statistically significant improvement in robustness to inter-subject variability (\(p<0.05\)p<0.05, Wilcoxon signed-rank test). Furthermore, electrode-importance heatmaps and topographical spatial-filter visualizations consistently emphasize sensorimotor cortical regions associated with MI activity, supporting the physiological relevance of the proposed framework. For practical deployment, the proposed explainability interface assists BCI developers in identifying and analyzing potential misclassifications. Overall, the proposed framework contributes toward more reliable and interpretable ElectroEncephaloGraphy (EEG)-based BCIs for neurorehabilitation and assistive technology applications.