Scholarlink

<p>ASD as a neurodevelopmental issue disrupts social abilities and communication skills together with producing repetitive behavioral patterns. Early detection of ASD allows healthcare providers to implement proper treatment methods that deliver superior outcomes for development. Recent research shows that distinct facial differences allow professionals to distinguish ASD children from normally developing children. Researchers developed an autism spectrum disorder-detecting framework through deep learning since existing facial image analysis methods lack efficiency. The proposed methodology merges multiple deep learning operations using convolutional neural networks (CNNs), dense blocks, and bottleneck layers for obtaining effective feature extraction. These features are further refined through an attention mechanism and feature fusion strategy to enhance discriminative capability. The combined representations are then run through a deep neural network (DNN) classifier that includes batch normalization, dense layers, and dropout regularization, ending with a SoftMax layer for making predictions. The evaluation utilized the publicly accessible 2,940 facial image Mendeley dataset. The evaluation demonstrates that this newly designed ensemble model attains superior performance compared to numerous pre-trained models by reaching 98.20% accuracy. The developed approach presents an optimistic instrument to help healthcare professionals and caregivers recognize autism spectrum disorder early, which enables tailored intervention programs.</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

AFDL-Net: An Attention Mechanism-based Feature Fusion for Facial Image-based Autism Detection in Children via Hybrid Deep Learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月2日

查看全文
  • Abeer Aljohani

摘要

ASD as a neurodevelopmental issue disrupts social abilities and communication skills together with producing repetitive behavioral patterns. Early detection of ASD allows healthcare providers to implement proper treatment methods that deliver superior outcomes for development. Recent research shows that distinct facial differences allow professionals to distinguish ASD children from normally developing children. Researchers developed an autism spectrum disorder-detecting framework through deep learning since existing facial image analysis methods lack efficiency. The proposed methodology merges multiple deep learning operations using convolutional neural networks (CNNs), dense blocks, and bottleneck layers for obtaining effective feature extraction. These features are further refined through an attention mechanism and feature fusion strategy to enhance discriminative capability. The combined representations are then run through a deep neural network (DNN) classifier that includes batch normalization, dense layers, and dropout regularization, ending with a SoftMax layer for making predictions. The evaluation utilized the publicly accessible 2,940 facial image Mendeley dataset. The evaluation demonstrates that this newly designed ensemble model attains superior performance compared to numerous pre-trained models by reaching 98.20% accuracy. The developed approach presents an optimistic instrument to help healthcare professionals and caregivers recognize autism spectrum disorder early, which enables tailored intervention programs.