Scholarlink

<p>Accurate decoding of neural activity is critical for brain-computer interfaces, yet unimodal electroencephalography (EEG) or functional near-infrared spectroscopy (fNIRS) suffer from limited temporal or spatial resolution. We propose a Multi-Stream Feature Learning and Region-Aware Graph Fusion (MSFL-RGF) framework that integrates EEG and fNIRS with enhanced interpretability. Six complementary feature streams are embedded into brain graphs and refined through an Adaptive Graph Connectivity Module and an Adaptive Residual Graph Convolution Block. A region-aware pooling mechanism aggregates node features into anatomically defined brain regions, followed by intra- and inter-modality probabilistic fusion. Experimental results on a Word Generation task dataset demonstrate that MSFL-RGF achieves a classification accuracy of 95.3% while improving interpretability.</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

MSFL-RGF: multi-stream feature learning and region-aware graph fusion for EEG-fNIRS brain decoding

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Tianhao Xia,
  • Binqiang Xue,
  • Yinhua Liu

摘要

Accurate decoding of neural activity is critical for brain-computer interfaces, yet unimodal electroencephalography (EEG) or functional near-infrared spectroscopy (fNIRS) suffer from limited temporal or spatial resolution. We propose a Multi-Stream Feature Learning and Region-Aware Graph Fusion (MSFL-RGF) framework that integrates EEG and fNIRS with enhanced interpretability. Six complementary feature streams are embedded into brain graphs and refined through an Adaptive Graph Connectivity Module and an Adaptive Residual Graph Convolution Block. A region-aware pooling mechanism aggregates node features into anatomically defined brain regions, followed by intra- and inter-modality probabilistic fusion. Experimental results on a Word Generation task dataset demonstrate that MSFL-RGF achieves a classification accuracy of 95.3% while improving interpretability.