Scholarlink

The diagnosis of autism spectrum disorder (ASD) is still based on clinical observation, as there are no validated biomarkers for use in clinical practice. Although the first suspicions may appear as early as 12 months, the diagnosis may be delayed until 3 to 6 years. We propose an early screening procedure for autism starting at 9 months of age, which will allow paediatricians to objectively detect the presence of ASD risk indicators, facilitate immediate access to a specific preventive action program, and minimize the effects of the disorder. Our system utilizes a series of videos specifically designed to detect ASD risk indicators combined with machine learning classifiers to predict ASD risk. Using Random Forest, SVM, MLP, kNN, and AdaBoost, we obtained a 0.9005 ROC AUC, 75.2% sensitivity with the best classifier, which was SVM, when comparing typical development (TD) to ASD levels 1, 2, and 3. The results were up to a ROC AUC score of 0.9508 and a sensitivity of 87.64% with Random Forest, when comparing TD to ASD levels 2 and 3.

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

Early Objective ASD Screening System Based on Eye-Tracking and Machine Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sara Vecino,
  • Gloria Acevedo-Diaz,
  • Daniel Fernandez-Lanvin,
  • Javier De Andres,
  • Martin Gonzalez-Rodriguez

摘要

The diagnosis of autism spectrum disorder (ASD) is still based on clinical observation, as there are no validated biomarkers for use in clinical practice. Although the first suspicions may appear as early as 12 months, the diagnosis may be delayed until 3 to 6 years. We propose an early screening procedure for autism starting at 9 months of age, which will allow paediatricians to objectively detect the presence of ASD risk indicators, facilitate immediate access to a specific preventive action program, and minimize the effects of the disorder. Our system utilizes a series of videos specifically designed to detect ASD risk indicators combined with machine learning classifiers to predict ASD risk. Using Random Forest, SVM, MLP, kNN, and AdaBoost, we obtained a 0.9005 ROC AUC, 75.2% sensitivity with the best classifier, which was SVM, when comparing typical development (TD) to ASD levels 1, 2, and 3. The results were up to a ROC AUC score of 0.9508 and a sensitivity of 87.64% with Random Forest, when comparing TD to ASD levels 2 and 3.