目的:开发基于MMP7和其他血清学检测指标的机器学习诊断模型,用于早期有效地诊断胆道闭锁(BA)。
方法:对北京儿童医院2019年1月1日至2023年12月31日因病理性黄疸住院的患者资料进行回顾性分析。患者血清MMP7,肝脏硬度测量,和其他常规血清学检测也纳入研究.构建了六个机器学习模型,包括逻辑回归(LR),随机森林(RF),决策树(DET),支持向量机分类器(SVC),神经网络(MLP)和极端梯度提升(XGBoost),诊断BA。使用接收器工作特征曲线下的面积来评估各种模型的诊断功效。
结果:共98名患者被纳入研究,包括64例BA患者和34例其他胆汁淤积性肝病患者。在六种机器学习模型中,XGBoost算法模型和RF算法模型实现了最佳预测性能,训练集和验证集的AUROC接近100%。在训练集中,这两个算法模型达到了准确性,精度,召回,F1得分,AUROC为1。通过模型解释分析,血清MMP7水平,血清GGT水平,和结肠粪便被确定为诊断BA的最重要指标。基于XGBoost算法模型构建的列线图也展示了方便高效的诊断功效。
结论:机器学习模型,特别是XGBoost算法和射频算法模型,基于术前血清MMP7和血清学检测的构建可以更有效、准确地诊断BA。诊断最重要的影响因素是血清MMP7、血清GGT、和大便。
OBJECTIVE: To develop a machine learning diagnostic model based on MMP7 and other serological testing indicators for early and efficient diagnosis of biliary atresia (BA).
METHODS: A retrospective analysis was conducted on patient information from those hospitalized for pathological jaundice at Beijing Children\'s Hospital between January 1, 2019, and December 31, 2023. Patients with serum MMP7, liver stiffness measurements, and other routine serological tests were included in the study. Six machine learning models were constructed, including logistic regression (LR), random forest (RF), decision tree (DET), support vector machine classifier (SVC), neural network (MLP), and extreme gradient boosting (XGBoost), to diagnose BA. The area under the receiver operating characteristic curve was used to evaluate the diagnostic efficacy of the various models.
RESULTS: A total of 98 patients were included in the study, comprising 64 BA patients and 34 patients with other cholestatic liver diseases. Among the six machine learning models, the XGBoost algorithm model and RF algorithm model achieved the best predictive performance, with an AUROC of nearly 100% in both the training and validation sets. In the training set, these two algorithm models achieved an accuracy, precision, recall, F1 score, and AUROC of 1. Through model interpretation analysis, serum MMP7 levels, serum GGT levels, and acholic stools were identified as the most important indicators for diagnosing BA. The nomogram constructed based on the XGBoost algorithm model also demonstrated convenient and efficient diagnostic efficacy.
CONCLUSIONS: Machine learning models, especially the XGBoost algorithm and RF algorithm models, constructed based on preoperative serum MMP7 and serological tests can diagnose BA more efficiently and accurately. The most important influencing factors for diagnosis are serum MMP7, serum GGT, and acholic stools.