A Comparative Study Of Enhanced CT And MRI Radiomics Model For Cervical Lymph Node Metastasis In Patients With Papillary Thyroid Carcinoma

Objective:The aim of this study was to investigate and compare the predictive efficacy of radiomics features extracted from contrast-enhanced CT and MR imaging in the primary tumor of papillary thyroid carcinoma(PTC)for the presence of cervical lymph node metastasis(LNM).Methods:A total of 148 PTC patients who underwent preoperative contrast-enhanced CT and MRI imaging were included in this retrospective study.The patients were divided into LNM and non-LNM groups based on pathological results.The radiomics features were extracted and selected by variance filter,Mann-Whitney test,Spearman or Pearson correlation analysis,and maximum relevance and minimum redundancy(m RMR)analysis.Nine machine learning algorithms,including logistic regression(LR),random forest(RF),Bayes,support vector machine(SVM),K-nearest neighbor(KNN),extreme gradient boosting(XGBoost),Adaboost,gradient boosting decision tree(GBDT),and light gradient boosting machine(light GBM)were used to build models based on CT and MRI radiomics features,respectively.The diagnostic efficacy of the radiomics models was evaluated using receiver operating characteristic(ROC)curves,and the differences in the AUC values among the models were compared using the Delong test.The diagnostic efficacy of the CT and MRI radiomics models was further validated using an independent validation dataset.Results:Of the 148 patients,78(52.3%)had LNM and 70(47.3%)did not.A total of1502 radiomics features were extracted from CT and MRI images of each PTC patient,and 1219 and 1324 features were selected for CT and MR imaging,respectively.Five CT radiomics features and five MRI radiomics features were finally included in the modeling analysis.In the validation set,the AUC of the enhanced MRI radiomics model established by LR,Bayes,Light GBM,Adaboost and KNN machine learning algorithms is slightly higher than that of the enhanced CT radiomics model.The AUC value of the enhanced MRI radiomics model based on RF,SVM,GBDT and XGBoost algorithm is slightly lower than that of the enhanced CT radiomics model.LR algorithm had the largest difference in AUC values between the enhanced CT and enhanced MRI radiomics models(0.683 and 0.763,respectively),while Bayes algorithm had the smallest difference(0.696 and 0.705,respectively).However,there was no significant difference in AUC values between the CT and MRI radiomics models(P > 0.05).Among the 9 enhanced CT models,the KNN algorithm based model had the highest AUC value(0.817),the sensitivity was 68.8%,the specificity was 71.4% and the accuracy was 70.0%.Among the 9 enhanced MRI radiomics,the KNN algorithm based model also has the highest AUC value(AUC is 0.775),with a sensitivity of 56.3%,specificity of 92.9% and accuracy of 73.3%.The calibration curve showed enhanced CT and MRI radiomics models were in good agreement with the real lymph node metastasis.The clinical decision curve shows that no matter the enhanced CT radiomics model or the enhanced MRI radiomics model,the KNN algorithm has higher clinical benefits..Conclusions:The predictive efficacy of enhanced CT radiomics models and MR radiomics models for cervical LNM in patients with PTC is roughly equivalent.Among the nine machine learning algorithms,KNN algorithm has the highest AUC.In general,the AUC value of enhanced MRI radiomics model is slightly higher,but its sensitivity is relatively low and its specificity is extremely high.