Scholarlink

<p>Autism Spectrum Disorder (ASD) is a developmental condition marked by challenges in attention, communication, and social engagement. Electroencephalography (EEG), owing to its fine temporal resolution, provides an effective means to study brain activity in individuals with ASD. The BCIAUT_P300 dataset offers EEG signals collected during P300-based Brain–Computer Interface (BCI) experiments designed to support joint-attention training. In this study, we introduce a Graph Sample and Aggregation (GraphSAGE) model to differentiate P300 responses from non-P300 activity. Here, each EEG channel is treated as a node within a graph, while functional interactions between channels form the edges. By aggregating information from neighboring nodes, GraphSAGE learns spatial as well as temporal dependencies, enabling the derivation of highly discriminative features. Applied across seven EEG recording sessions, the approach produced an average accuracy of 98%, surpassing several established deep learning methods. These findings demonstrate the strength of GraphSAGE in decoding event-related potentials and point toward its utility in creating adaptive BCI frameworks to aid therapeutic interventions for ASD.</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

Graphsage model for robust P300 EEG classification in Autism spectrum disorder

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Sreeraj S.,
  • Palanisamy P.,
  • Varun P. Gopi,
  • Bibin Francis,
  • Anju Thomas,
  • Saptorshi Sarkar Oyshi

摘要

Autism Spectrum Disorder (ASD) is a developmental condition marked by challenges in attention, communication, and social engagement. Electroencephalography (EEG), owing to its fine temporal resolution, provides an effective means to study brain activity in individuals with ASD. The BCIAUT_P300 dataset offers EEG signals collected during P300-based Brain–Computer Interface (BCI) experiments designed to support joint-attention training. In this study, we introduce a Graph Sample and Aggregation (GraphSAGE) model to differentiate P300 responses from non-P300 activity. Here, each EEG channel is treated as a node within a graph, while functional interactions between channels form the edges. By aggregating information from neighboring nodes, GraphSAGE learns spatial as well as temporal dependencies, enabling the derivation of highly discriminative features. Applied across seven EEG recording sessions, the approach produced an average accuracy of 98%, surpassing several established deep learning methods. These findings demonstrate the strength of GraphSAGE in decoding event-related potentials and point toward its utility in creating adaptive BCI frameworks to aid therapeutic interventions for ASD.