Scholarlink

Background <p>Accurate preoperative discrimination between uncomplicated and complicated acute appendicitis in children is critical for optimal surgical decision-making but remains a clinical challenge. Current reliance on conventional imaging and laboratory findings lacks sufficient precision, necessitating the development of more reliable predictive tools.</p> Methods <p>We conducted a multicenter retrospective study to develop and validate a machine-learning model integrating CT radiomics and clinical variables. Data from 287 pediatric patients (aged 2–14 years) at Xiaogan Central Hospital (Jan 2017–May 2025) were randomly divided into a training (<i>n</i> = 201) and an internal-validation cohort (<i>n</i> = 86). An independent external-validation cohort comprised 164 patients from Yichang Central People’s Hospital (Nov 2023–May 2025). Preoperative CT images and clinical data were processed; features with &lt; 30% missing values were imputed. An automated modeling pipeline constructed 100 candidate models by combining 10 feature-selection algorithms with 10 machine-learning frameworks. Model performance was assessed using receiver operating characteristic curve analysis, with interpretability provided by SHapley Additive exPlanations (SHAP).</p> Results <p>The optimal model, chi square feature selection combined with AdaBoost, demonstrated robust discriminative ability. It achieved AUCs of 0.900 (95% CI: 0.832–0.958), 0.889 (95% CI: 0.841–0.937), and 0.893 (95% CI: 0.855–0.931) in the training, internal-validation, and external-validation cohorts, respectively. The model achieved a sensitivity of 87% and specificity of 85% in the external-validation cohort. SHAP analysis identified radiomic texture features (RootMeanSquared, Strength, Skewness), appendicolith presence, and maximum body temperature as the most important predictors.</p> Conclusion <p>A CT radiomics-based machine learning model integrating clinical features can accurately differentiate complicated from uncomplicated appendicitis in children, providing an interpretable framework to support preoperative surgical decision-making.</p> Trial registration <p>Not applicable (retrospective diagnostic study).</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

An interpretable computed tomography radiomics-based machine learning model for preoperative prediction of complicated appendicitis in children

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月1日

查看全文
  • Zhaopu Li,
  • Yang Li,
  • Zhongjing Zhang,
  • Jiaqing Lin,
  • Tao Xu,
  • Wei Zhu,
  • Zhaokun Guo,
  • Kun Yang

摘要

Background

Accurate preoperative discrimination between uncomplicated and complicated acute appendicitis in children is critical for optimal surgical decision-making but remains a clinical challenge. Current reliance on conventional imaging and laboratory findings lacks sufficient precision, necessitating the development of more reliable predictive tools.

Methods

We conducted a multicenter retrospective study to develop and validate a machine-learning model integrating CT radiomics and clinical variables. Data from 287 pediatric patients (aged 2–14 years) at Xiaogan Central Hospital (Jan 2017–May 2025) were randomly divided into a training (n = 201) and an internal-validation cohort (n = 86). An independent external-validation cohort comprised 164 patients from Yichang Central People’s Hospital (Nov 2023–May 2025). Preoperative CT images and clinical data were processed; features with < 30% missing values were imputed. An automated modeling pipeline constructed 100 candidate models by combining 10 feature-selection algorithms with 10 machine-learning frameworks. Model performance was assessed using receiver operating characteristic curve analysis, with interpretability provided by SHapley Additive exPlanations (SHAP).

Results

The optimal model, chi square feature selection combined with AdaBoost, demonstrated robust discriminative ability. It achieved AUCs of 0.900 (95% CI: 0.832–0.958), 0.889 (95% CI: 0.841–0.937), and 0.893 (95% CI: 0.855–0.931) in the training, internal-validation, and external-validation cohorts, respectively. The model achieved a sensitivity of 87% and specificity of 85% in the external-validation cohort. SHAP analysis identified radiomic texture features (RootMeanSquared, Strength, Skewness), appendicolith presence, and maximum body temperature as the most important predictors.

Conclusion

A CT radiomics-based machine learning model integrating clinical features can accurately differentiate complicated from uncomplicated appendicitis in children, providing an interpretable framework to support preoperative surgical decision-making.

Trial registration

Not applicable (retrospective diagnostic study).