Scholarlink

<p>Pediatric height assessment is crucial for evaluating growth and enabling timely intervention. However, traditional methods primarily assess current bone age rather than identifying future height abnormalities. This study proposes an integrated framework with two components: one uses hand X-ray images and clinical features to predict potential extreme height cases after one year; the other employs multivariate data to predict height across different time scales. To predict potential extreme cases, we employed fine-tuned Inception-ResNet-V2 as our classification model, while addressing class data imbalance through the DSEV method. For multivariate height prediction, we used XGBoost as our prediction model, incorporating various anthropometric and medical features, such as sex, age, body height, birth weight, medical diagnosis, and treatment. Results demonstrate that our classification models achieved outstanding performance in detecting unexpectedly short and unexpectedly tall cases, showing exceptional accuracy and reliability. Our multivariate prediction models demonstrated very low mean absolute error for predictions within 6 months, 1 year, 2 years, and near-final adult height, with consistent and stable performance. This integrated framework provides clinicians with more precise and comprehensive tools for evaluating children’s growth and planning interventions.</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

Integrated framework for pediatric height assessment: X-ray-based height extreme cases classification and machine learning for multivariate height prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Ya-Wen Chang,
  • Meng-Che Tsai,
  • Sun-Yuan Hsieh

摘要

Pediatric height assessment is crucial for evaluating growth and enabling timely intervention. However, traditional methods primarily assess current bone age rather than identifying future height abnormalities. This study proposes an integrated framework with two components: one uses hand X-ray images and clinical features to predict potential extreme height cases after one year; the other employs multivariate data to predict height across different time scales. To predict potential extreme cases, we employed fine-tuned Inception-ResNet-V2 as our classification model, while addressing class data imbalance through the DSEV method. For multivariate height prediction, we used XGBoost as our prediction model, incorporating various anthropometric and medical features, such as sex, age, body height, birth weight, medical diagnosis, and treatment. Results demonstrate that our classification models achieved outstanding performance in detecting unexpectedly short and unexpectedly tall cases, showing exceptional accuracy and reliability. Our multivariate prediction models demonstrated very low mean absolute error for predictions within 6 months, 1 year, 2 years, and near-final adult height, with consistent and stable performance. This integrated framework provides clinicians with more precise and comprehensive tools for evaluating children’s growth and planning interventions.