Scholarlink

<p>Autism spectrum disorder (ASD) often entails alterations in attention and cognitive control, which require methods capable of reliably measuring associated underlying neural. Electroencephalography (EEG) provides a direct estimate of brain dynamics, and the P300 component is well established as an index of attentional processing. Common classification methods, such as Support Vector Machines (SVMs) and convolutional models, tend to focus on isolated features and overlook the allocation of information across electrodes. To overcome this limitation, we introduce a graph-based learning strategy that represents each EEG trial as a network, where the electrodes form the nodes and their functional relationships define the connections. Using a Graph Isomorphism Network (GIN) architecture, the method is applied to the BCIAUT_P300 dataset and shows a reliable separation between P300 and non-P300 activity. Using an inter-channel structure, this framework improves the interpretation of attention-related signals and supports the development of more robust Brain–Computer Interface (BCI) solutions 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

GIN-EEG: a graph isomorphism network approach for P300-based cognitive assessment in autism

  • 期刊论文

  • 发表日期：

    2025年11月21日

查看全文
  • Sreeraj Sahadevan,
  • Palanisamy Ponnusamy,
  • Varun P. Gopi,
  • Bibin Francis,
  • Anju Thomas,
  • Snehapriya Thiagarajan

摘要

Autism spectrum disorder (ASD) often entails alterations in attention and cognitive control, which require methods capable of reliably measuring associated underlying neural. Electroencephalography (EEG) provides a direct estimate of brain dynamics, and the P300 component is well established as an index of attentional processing. Common classification methods, such as Support Vector Machines (SVMs) and convolutional models, tend to focus on isolated features and overlook the allocation of information across electrodes. To overcome this limitation, we introduce a graph-based learning strategy that represents each EEG trial as a network, where the electrodes form the nodes and their functional relationships define the connections. Using a Graph Isomorphism Network (GIN) architecture, the method is applied to the BCIAUT_P300 dataset and shows a reliable separation between P300 and non-P300 activity. Using an inter-channel structure, this framework improves the interpretation of attention-related signals and supports the development of more robust Brain–Computer Interface (BCI) solutions for ASD.