Scholarlink

<p>The neurological disorder known as autism spectrum disorder (ASD) impacts the behavior of kids, socialisation, and interpersonal relationships. For prompt action and personal assistance, initial and precise detection is essential. This study presents a multimodal framework combining Handwritten Text (HT) images and Eye Tracking Ratio (ETR) data for ASD classification. The framework employs Taneja Generalised Gibbs Bell Fuzzy (T2GB-Fuzzy) for feature labelling and a novel SwishSin Average Spectop-K Convolutional Neural Network (2SASK-CNN) for classification. Experiments were conducted on two public datasets: the autism Spectrum Disorder in Children (ASDC) dataset with 569 HT images (381 ASD, 188 control) and the ETR dataset with 59 children contributing 396,298 gaze records. Subject-wise splits ensured no data leakage. The suggested approach outperformed the current CNN, ANN, and SVM techniques with an accuracy of 98.94% and an F1-score of 98.72%. The small dataset size, class imbalance, and lack of external validation are some of the limitations, even though the results show the promise of multimodal fusion for ASD detection.</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

Eye tracking and handwritten text-based autism spectrum disorder detection in children using 2SASK-CNN

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月8日

查看全文
  • S. A. Karthik,
  • H. L. Gururaj,
  • J. Shreyas

摘要

The neurological disorder known as autism spectrum disorder (ASD) impacts the behavior of kids, socialisation, and interpersonal relationships. For prompt action and personal assistance, initial and precise detection is essential. This study presents a multimodal framework combining Handwritten Text (HT) images and Eye Tracking Ratio (ETR) data for ASD classification. The framework employs Taneja Generalised Gibbs Bell Fuzzy (T2GB-Fuzzy) for feature labelling and a novel SwishSin Average Spectop-K Convolutional Neural Network (2SASK-CNN) for classification. Experiments were conducted on two public datasets: the autism Spectrum Disorder in Children (ASDC) dataset with 569 HT images (381 ASD, 188 control) and the ETR dataset with 59 children contributing 396,298 gaze records. Subject-wise splits ensured no data leakage. The suggested approach outperformed the current CNN, ANN, and SVM techniques with an accuracy of 98.94% and an F1-score of 98.72%. The small dataset size, class imbalance, and lack of external validation are some of the limitations, even though the results show the promise of multimodal fusion for ASD detection.