Construction and validation of a prediction model for central lymph node metastasis of papillary thyroid carcinoma based on contrast-enhanced venous phase CT radiomics
10.16016/j.2097-0927.202503007
- VernacularTitle:基于增强静脉期CT放射组学的甲状腺乳头状癌中央区淋巴结转移预测模型构建与验证
- Author:
Xingyun HE
1
;
Chen LIU
;
Junze DU
;
Qian LI
;
Kang CHEN
;
Rui FAN
;
Jian WANG
Author Information
1. 陆军军医大学(第三军医大学)第一附属医院放射科
- Keywords:
papillary thyroid carcinoma;
machine learning;
radiomics;
prediction model
- From:
Journal of Army Medical University
2025;47(12):1367-1375
- CountryChina
- Language:Chinese
-
Abstract:
Objective To construct and validate an interpretable machine learning model integrating contrast-enhanced venous phase CT radiomics and clinical features for preoperative prediction of central lymph node metastasis(CLNM)of papillary thyroid carcinoma(PTC).Methods A case-control study was conducted on 243 histologically confirmed PTC patients.Among them,196 patients from the First Affiliated Hospital of Army Medical University were randomly allocated into a training set(n=137)and an internal validation set(n=59)at a 7:3 ratio,while the left 47 patients from No.958 Hospital of PLA Army were assigned into an external validation set.All participants completed standardized preoperative contrast-enhanced neck CT scanning.Quantitative radiomic features were systematically extracted from venous phase CT images through an open-source big data AI platform.Six machine learning classifiers,eXtreme Gradient Boosting(XGBoost),Support Vector Machine(SVM),Random Forest(RF),Logistic Regression(LR),k-Nearest Neighbors(KNN),and Decision Tree(DT)were implemented to construct clinical-radiomics integration models.The predictive performance was quantitatively assessed through receiver operating characteristic(ROC)curve analysis,with area under the curve(AUC)values calculated for training,internal validation,and external validation sets.Model interpretability was achieved using Shapley additive explanations(SHAP)framework,with particular focus on elucidating feature contributions in the best-performing model.Results The XGBoost model had an AUC value of 0.936(95%CI:0.895～0.976),0.832(95%CI:0.724～0.941),and 0.772(95%CI:0.632～0.912)in training,internal and external validation sets,respectively.SHAP analysis revealed age as the most influential clinical predictor,with younger patients showing higher CLNM risk(OR=0.968,P=0.042).Conclusion Our machine learning-based clinic-radiomic prediction model demonstrates satisfactory performance in predicting CLNM of PTC,providing valuable references for clinical decision-making.
