Scholarlink

<p>Autism spectrum disorder (ASD) is a common neurological condition marked by difficulties in social communication and the presence of repetitive behaviors. Early and accurate diagnosis is vital but remains a significant clinical challenge in ASD. This study investigates an artificial intelligence approach to differentiate between autistic and typically developing children. Deep learning models were trained on publicly available eye-tracking data from 59 participants (29 ASD, 30 TD), focusing on eye movement patterns as a potential diagnostic tool. The methodology involved comprehensive image preprocessing using histogram equalization to enhance visual feature representation, while data augmentation techniques were used to address common dataset limitations in ASD research. Transfer learning with custom layers was further employed to optimize model performance. Using state-of-the-art architectures including DenseNet169, DenseNet201, VGG16, VGG19, ResNet50, ResNet50V2, ResNet152V2, MobileNet, MobileNetV2, InceptionV3, and NASNetMobile, various classification accuracies were achieved: DenseNet169 96%, DenseNet201 96%, VGG16 96%, VGG19 95%, ResNet50 93%, ResNet50V2 92%, ResNet152V2 94%, MobileNet 96%, MobileNetV2 94%, InceptionV3 85%, and NASNetMobile 91%, with corresponding sensitivities ranging from 82 to 97% and specificities from 87 to 97%. An ensemble model combining optimized VGG16, MobileNet, DenseNet169, and Vision Transformer ViT architectures achieved a classification accuracy of 98% with 98% sensitivity and 97% specificity. The results demonstrate the potential of combining advanced deep learning techniques with eye-tracking technology for developing more accurate and objective ASD diagnostic tools.</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

Effectiveness of Histogram Equalization and Ensemble Deep Learning Techniques for Detecting Autism Using Eye-Tracking

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Zeyad A. T. Ahmed,
  • Theyazn H. H. Aldhyani,
  • Mosleh Hmoud Al-Adhaileh,
  • Eidah M. Alzahrani,
  • Eid Albalawi,
  • Mohammad H. Algarni,
  • Mukti E. Jadhav,
  • Saleh N. M. Alsubari,
  • Ahmed Samir Morsy,
  • Ali Mehdi

摘要

Autism spectrum disorder (ASD) is a common neurological condition marked by difficulties in social communication and the presence of repetitive behaviors. Early and accurate diagnosis is vital but remains a significant clinical challenge in ASD. This study investigates an artificial intelligence approach to differentiate between autistic and typically developing children. Deep learning models were trained on publicly available eye-tracking data from 59 participants (29 ASD, 30 TD), focusing on eye movement patterns as a potential diagnostic tool. The methodology involved comprehensive image preprocessing using histogram equalization to enhance visual feature representation, while data augmentation techniques were used to address common dataset limitations in ASD research. Transfer learning with custom layers was further employed to optimize model performance. Using state-of-the-art architectures including DenseNet169, DenseNet201, VGG16, VGG19, ResNet50, ResNet50V2, ResNet152V2, MobileNet, MobileNetV2, InceptionV3, and NASNetMobile, various classification accuracies were achieved: DenseNet169 96%, DenseNet201 96%, VGG16 96%, VGG19 95%, ResNet50 93%, ResNet50V2 92%, ResNet152V2 94%, MobileNet 96%, MobileNetV2 94%, InceptionV3 85%, and NASNetMobile 91%, with corresponding sensitivities ranging from 82 to 97% and specificities from 87 to 97%. An ensemble model combining optimized VGG16, MobileNet, DenseNet169, and Vision Transformer ViT architectures achieved a classification accuracy of 98% with 98% sensitivity and 97% specificity. The results demonstrate the potential of combining advanced deep learning techniques with eye-tracking technology for developing more accurate and objective ASD diagnostic tools.