Scholarlink

<p>In recent years, decoding of brain activity using Motor Imagery Electroencephalogram (MI-EEG) signals has gained substantial interest in the development of Brain Computer Interface (BCI). However, classification of MI-EEG signals remains challenging due to their non-linear, non-stationary characteristics and inherently low signal-to-noise ratio. <i>Approach</i>: The present work proposes a novel channel selection technique, the Stockwell Transform-based Phase Lag Index-Wilcoxon signed-rank test (S-PLI-WT), to discard MI-EEG channels containing noisy and redundant information. In this method, the PLI is computed using the S-transform, which enhances time–frequency resolution, thereby capturing inter-channel MI-EEG information across both frequency and time points. A weighting procedure is incorporated into the phase lag computation, which eliminates the effects of volume conduction and noise. The Wilcoxon signed-rank test is used to identify statistically significant connections. By leveraging graph theory, brain functional networks are constructed from the selected MI-EEG channels. The global and local topologies attributes are computed to extract MI features. An explainable feature selection framework, RFE-ELI5, is proposed to identify the most significant features. It employs the Explain Like I’m 5 (ELI-5) framework of Explainable Artificial Intelligence (XAI) to gain insights into the features that contribute significantly to the decision-making process of the proposed framework. Finally, a Quantum Support Vector Classifier (SVC) is developed to classify MI-EEG signals into four classes: left-hand, right-hand, feet, and tongue. <i>Main results:</i> The proposed methodology outperforms many existing state-of-the-art models developed for classifying MI-EEG signals and achieves an average classification accuracy of 91.71% using the BCI competition IV: dataset 2A.</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 AI framework and quantum machine learning classifier based brain connectivity-feature classification for MI-EEG BCI applications

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Manvir Kaur,
  • Vinay Kumar,
  • Rahul Upadhyay

摘要

In recent years, decoding of brain activity using Motor Imagery Electroencephalogram (MI-EEG) signals has gained substantial interest in the development of Brain Computer Interface (BCI). However, classification of MI-EEG signals remains challenging due to their non-linear, non-stationary characteristics and inherently low signal-to-noise ratio. Approach: The present work proposes a novel channel selection technique, the Stockwell Transform-based Phase Lag Index-Wilcoxon signed-rank test (S-PLI-WT), to discard MI-EEG channels containing noisy and redundant information. In this method, the PLI is computed using the S-transform, which enhances time–frequency resolution, thereby capturing inter-channel MI-EEG information across both frequency and time points. A weighting procedure is incorporated into the phase lag computation, which eliminates the effects of volume conduction and noise. The Wilcoxon signed-rank test is used to identify statistically significant connections. By leveraging graph theory, brain functional networks are constructed from the selected MI-EEG channels. The global and local topologies attributes are computed to extract MI features. An explainable feature selection framework, RFE-ELI5, is proposed to identify the most significant features. It employs the Explain Like I’m 5 (ELI-5) framework of Explainable Artificial Intelligence (XAI) to gain insights into the features that contribute significantly to the decision-making process of the proposed framework. Finally, a Quantum Support Vector Classifier (SVC) is developed to classify MI-EEG signals into four classes: left-hand, right-hand, feet, and tongue. Main results: The proposed methodology outperforms many existing state-of-the-art models developed for classifying MI-EEG signals and achieves an average classification accuracy of 91.71% using the BCI competition IV: dataset 2A.