Scholarlink

<p>Electroencephalography (<i>EEG</i>) provides critical insight into neural activity but remains difficult to interpret due to its multidimensional and noisy nature. This study presents a novel unsupervised DL framework that integrates clustering, statistical analysis, and Convolutional Neural Networks (<i>CNNs</i>) to extract interpretable neural representations from raw EEG signals. The proposed pipeline applies K-means clustering after dimensionality reduction using Principal Component Analysis (<i>PCA</i>) to identify distinct cognitive-state groupings. Statistical validation through t-tests highlights the most discriminative EEG channels, including <i>FC</i>1, <i>C</i>3, and <i>P</i>8, confirming their role in cluster formation. A lightweight CNN architecture is then employed to learn hierarchical spatial–temporal features, achieving classification accuracy exceeding <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(99\%\)</EquationSource> </InlineEquation>. Analysis of feature maps and filter activations reveals interpretable relationships between CNN filters and EEG channels, linking computational representations to cognitive and motor processes. Overall, the proposed framework bridges unsupervised deep learning (<i>DL</i>) with neuroscientific interpretability, providing a reproducible and scalable approach for EEG-based cognitive state analysis and future real-time brain–computer interface 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

Linking neural filters to cognitive states: a framework for EEG cluster analysis using neural networks

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Syeda Noor Fathima,
  • K. Bhanu Rekha,
  • S. Safinaz,
  • Akram Pasha,
  • Syed Ziaur Rahman

摘要

Electroencephalography (EEG) provides critical insight into neural activity but remains difficult to interpret due to its multidimensional and noisy nature. This study presents a novel unsupervised DL framework that integrates clustering, statistical analysis, and Convolutional Neural Networks (CNNs) to extract interpretable neural representations from raw EEG signals. The proposed pipeline applies K-means clustering after dimensionality reduction using Principal Component Analysis (PCA) to identify distinct cognitive-state groupings. Statistical validation through t-tests highlights the most discriminative EEG channels, including FC1, C3, and P8, confirming their role in cluster formation. A lightweight CNN architecture is then employed to learn hierarchical spatial–temporal features, achieving classification accuracy exceeding \(99\%\) . Analysis of feature maps and filter activations reveals interpretable relationships between CNN filters and EEG channels, linking computational representations to cognitive and motor processes. Overall, the proposed framework bridges unsupervised deep learning (DL) with neuroscientific interpretability, providing a reproducible and scalable approach for EEG-based cognitive state analysis and future real-time brain–computer interface applications.