Scholarlink

<p>Recently, numerous model networks have been developed and studied to classify the electroencephalogram (EEG) signals and achieve impressive classification accuracy. Despite the advancements in classification accuracy, there is a notable lack of research into understanding how these models interpret and represent the intricate features of EEG signals. Thus, understanding the neural dynamics underlying cognitive control is essential for advancing EEG-based classification models. This study investigates the interpretability of the Sinc-EEGNet model by integrating feature visualization of the spatial filters and Sinc-kernels, as well as partial directed coherence (PDC) analysis, with a specific focus on the destination channels activated during inhibitory processes. The Sinc-EEGNet model was applied to EEG data of twenty healthy subjects doing the Stroop task, a well-known test of cognitive control, to evaluate its ability to inhibit response. The methodology involves training the Sinc-EEGNet model on preprocessed EEG signals with Sinc-based filters and applying feature visualization techniques to understand the model’s output. PDC analysis is then employed to validate the spatial filter activation and identify the neural connections during inhibitory processes. Results demonstrate that the model achieves a classification accuracy of 97%, with clear distinctions in neural activations between subjects. Integrating feature visualization with PDC enhances model interpretability, providing insights into the neural dynamics of inhibitory control and confirming the effectiveness of Sinc-EEGNet in EEG-based classification tasks.</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

Encoding analysis in inhibition response classification using Sinc-EEGNet and partial directed coherence

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月13日

查看全文
  • Noor Syazwana Sahar,
  • Norlaili Mat Safri,
  • Tarmizi Izzuddin,
  • Nor Aini Zakaria,
  • Nurul Ashikin Abdul Kadir

摘要

Recently, numerous model networks have been developed and studied to classify the electroencephalogram (EEG) signals and achieve impressive classification accuracy. Despite the advancements in classification accuracy, there is a notable lack of research into understanding how these models interpret and represent the intricate features of EEG signals. Thus, understanding the neural dynamics underlying cognitive control is essential for advancing EEG-based classification models. This study investigates the interpretability of the Sinc-EEGNet model by integrating feature visualization of the spatial filters and Sinc-kernels, as well as partial directed coherence (PDC) analysis, with a specific focus on the destination channels activated during inhibitory processes. The Sinc-EEGNet model was applied to EEG data of twenty healthy subjects doing the Stroop task, a well-known test of cognitive control, to evaluate its ability to inhibit response. The methodology involves training the Sinc-EEGNet model on preprocessed EEG signals with Sinc-based filters and applying feature visualization techniques to understand the model’s output. PDC analysis is then employed to validate the spatial filter activation and identify the neural connections during inhibitory processes. Results demonstrate that the model achieves a classification accuracy of 97%, with clear distinctions in neural activations between subjects. Integrating feature visualization with PDC enhances model interpretability, providing insights into the neural dynamics of inhibitory control and confirming the effectiveness of Sinc-EEGNet in EEG-based classification tasks.