Scholarlink

<p>Autism spectrum disorder (ASD) is a complex neurodevelopmental condition that manifests uniquely in individuals, typically influencing social communication, behaviour, and sensory processing. The rise of deep learning (DL) technologies has paved the way for innovative, non-invasive early detection methods. Explainable artificial intelligence (XAI) has been introduced to enhance the understanding of the decision-making processes in DL techniques. This research presents a new technique for predicting ASD by analyzing facial images using a feature pyramid network (FPN) architecture. The FPN is employed to efficiently capture and extract multi-scale features from these images, allowing the identification of subtle facial traits that may signal ASD. This approach combines FPN with the VGG16 architecture, a well-known convolutional neural network, to analyze facial features at various scales. The proposed transfer learning strategy integrates VGG16 with FPN. This approach surpasses existing techniques, achieving a 98.28% accuracy rate. To evaluate the generalizability of the proposed approach, we further validated it on sMRI data, achieving improved results. The local interpretable model-agnostic explanations, Shapley additive explanation, and gradient-weighted class activation mapping frameworks are used to analyze these predictions further and generate visual explanations that support assumptions, along with established standards in explanations. These findings suggest that automated facial feature analysis could serve as a promising, non-invasive tool for early autism screening, complementing traditional behavioural assessment methods.</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

Explainable AI for autism spectrum disorder detection in children using facial images: a multi-scale feature extraction and transfer learning approach

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Bhagya Lakshmi Polavarapu,
  • V. Dinesh Reddy,
  • Mahesh Kumar Morampudi,
  • Ashu Abdul

摘要

Autism spectrum disorder (ASD) is a complex neurodevelopmental condition that manifests uniquely in individuals, typically influencing social communication, behaviour, and sensory processing. The rise of deep learning (DL) technologies has paved the way for innovative, non-invasive early detection methods. Explainable artificial intelligence (XAI) has been introduced to enhance the understanding of the decision-making processes in DL techniques. This research presents a new technique for predicting ASD by analyzing facial images using a feature pyramid network (FPN) architecture. The FPN is employed to efficiently capture and extract multi-scale features from these images, allowing the identification of subtle facial traits that may signal ASD. This approach combines FPN with the VGG16 architecture, a well-known convolutional neural network, to analyze facial features at various scales. The proposed transfer learning strategy integrates VGG16 with FPN. This approach surpasses existing techniques, achieving a 98.28% accuracy rate. To evaluate the generalizability of the proposed approach, we further validated it on sMRI data, achieving improved results. The local interpretable model-agnostic explanations, Shapley additive explanation, and gradient-weighted class activation mapping frameworks are used to analyze these predictions further and generate visual explanations that support assumptions, along with established standards in explanations. These findings suggest that automated facial feature analysis could serve as a promising, non-invasive tool for early autism screening, complementing traditional behavioural assessment methods.