Scholarlink

<p>Medical imaging biomarkers have been widely used for diagnosing attention deficit hyperactivity disorder (ADHD). However, diagnosis is often limited by the physician's subjective experience and specific medical devices. In addition, most existing research primarily focuses on distinguishing between ADHD and non-ADHD individuals, ignoring its subtype discrimination. This paper proposes an objective, convenient, and efficient method for classifying ADHD subtypes in children by integrating appearance features, including facial expression and gaze distribution, from raw videos. To better extract these appearance features, an attention mechanism-based facial expression recognition network is constructed to emphasize key facial regions. Secondly, a Transformer-enhanced CNN is employed for global feature integration to improve gaze estimation performance. The proposed appearance feature networks are tested on the public datasets. The facial expression recognition rates on the Fer2013, CK +, and RAF-DB datasets are 75.82%, 98.81%, and 91.82%, respectively. The average gaze estimation errors on the Gaze360 and MPIIFaceGaze datasets are 10.13° and 3.64°, respectively. Subsequently, we extract the spatiotemporal information of each appearance feature using accumulative histograms and introduce attention to establishing dynamic feature aggregation based on the temporal convolutional network (TCN), which can further improve the accuracy of classification. Finally, classification results on our ADHD video dataset (ACVD) demonstrate the effectiveness of the proposed method, achieving an accuracy of 81.7%, specificity of 93.9%, and sensitivity of 81.7% in distinguishing ADHD subtypes.</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

Classification of Children with ADHD Subtypes Using Appearance-Based Features from Videos

  • 期刊论文

  • 发表日期：

    2025年9月12日

查看全文
  • Zihao Cheng,
  • Hongmei Huang,
  • Wei Zhu,
  • Tao Wang,
  • Changqing Wang,
  • Qian Wu

摘要

Medical imaging biomarkers have been widely used for diagnosing attention deficit hyperactivity disorder (ADHD). However, diagnosis is often limited by the physician's subjective experience and specific medical devices. In addition, most existing research primarily focuses on distinguishing between ADHD and non-ADHD individuals, ignoring its subtype discrimination. This paper proposes an objective, convenient, and efficient method for classifying ADHD subtypes in children by integrating appearance features, including facial expression and gaze distribution, from raw videos. To better extract these appearance features, an attention mechanism-based facial expression recognition network is constructed to emphasize key facial regions. Secondly, a Transformer-enhanced CNN is employed for global feature integration to improve gaze estimation performance. The proposed appearance feature networks are tested on the public datasets. The facial expression recognition rates on the Fer2013, CK +, and RAF-DB datasets are 75.82%, 98.81%, and 91.82%, respectively. The average gaze estimation errors on the Gaze360 and MPIIFaceGaze datasets are 10.13° and 3.64°, respectively. Subsequently, we extract the spatiotemporal information of each appearance feature using accumulative histograms and introduce attention to establishing dynamic feature aggregation based on the temporal convolutional network (TCN), which can further improve the accuracy of classification. Finally, classification results on our ADHD video dataset (ACVD) demonstrate the effectiveness of the proposed method, achieving an accuracy of 81.7%, specificity of 93.9%, and sensitivity of 81.7% in distinguishing ADHD subtypes.