Scholarlink

<p>Juvenile idiopathic arthritis (JIA) commonly affects the temporomandibular joint (TMJ), leading to dentofacial deformities and orofacial symptoms. Timely diagnosis and treatment initiation are essential for optimizing patient outcomes. However, clinical examination—the primary screening method— has limited accuracy, increasing the risk of delayed TMJ involvement detection. This study develops, trains and tests an artificial intelligence (AI) model for predicting TMJ involvement in newly diagnosed JIA patients and compares its model performance with expert clinician assessments for validation. A longitudinal dataset of 6,153 standardized orofacial examinations from 1,054 patients with JIA was used to train an Extreme Gradient Boosting (XGBoost) model to predict TMJ involvement. An independent cohort of 55 newly diagnosed patients was used to evaluate the model. Twenty-six clinically relevant features were selected and preprocessed for model input. Model performance was evaluated based on classification accuracy and concordance with expert clinician assessments. Model interpretability was analysed using Shapley additive explanations (SHAP) to identify key predictive features. The XGBoost model achieved an overall accuracy of 85.5% in predicting TMJ involvement. Model predictions showed significant concordance with expert clinician assessments (<i>p</i> &lt; 0.001), although the model identified a higher prevalence of TMJ involvement than experts. The most influential predictive features were reduced condylar translation, facial asymmetry, protrusion, patient-reported orofacial pain and reduced mouth-opening capacity. The developed AI model demonstrates strong predictive performance for TMJ involvement based on clinical examination. By facilitating earlier detection, the model has the potential to support clinical decision-making, enable timely intervention, and improve patient outcomes.</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

Improving early detection of temporomandibular joint involvement in juvenile idiopathic arthritis with a clinically interpretable machine learning model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月7日

查看全文
  • Stratos Vassis,
  • Lena Todnem,
  • Dikte Straadt,
  • Christian Fischer Pedersen,
  • Beatrice Noeldeke,
  • Cory M. Resnick,
  • Mia Glerup,
  • Thomas Klit Pedersen,
  • Ruben Pauwels,
  • Peter Stoustrup

摘要

Juvenile idiopathic arthritis (JIA) commonly affects the temporomandibular joint (TMJ), leading to dentofacial deformities and orofacial symptoms. Timely diagnosis and treatment initiation are essential for optimizing patient outcomes. However, clinical examination—the primary screening method— has limited accuracy, increasing the risk of delayed TMJ involvement detection. This study develops, trains and tests an artificial intelligence (AI) model for predicting TMJ involvement in newly diagnosed JIA patients and compares its model performance with expert clinician assessments for validation. A longitudinal dataset of 6,153 standardized orofacial examinations from 1,054 patients with JIA was used to train an Extreme Gradient Boosting (XGBoost) model to predict TMJ involvement. An independent cohort of 55 newly diagnosed patients was used to evaluate the model. Twenty-six clinically relevant features were selected and preprocessed for model input. Model performance was evaluated based on classification accuracy and concordance with expert clinician assessments. Model interpretability was analysed using Shapley additive explanations (SHAP) to identify key predictive features. The XGBoost model achieved an overall accuracy of 85.5% in predicting TMJ involvement. Model predictions showed significant concordance with expert clinician assessments (p < 0.001), although the model identified a higher prevalence of TMJ involvement than experts. The most influential predictive features were reduced condylar translation, facial asymmetry, protrusion, patient-reported orofacial pain and reduced mouth-opening capacity. The developed AI model demonstrates strong predictive performance for TMJ involvement based on clinical examination. By facilitating earlier detection, the model has the potential to support clinical decision-making, enable timely intervention, and improve patient outcomes.