Construction Of A Predictive Model Of Lymph Node Metastasis In The Central Neck Region Of Thyroid Papillary Carcinoma Based On Multimodal Ultrasound Radiomics

Objective To construct a predictive model of PTC(thyroid papillary carcinoma)CLNM(lymph node metastasis in the central region of the neck)using multimodal ultrasound radiomicsMethods A retrospective analysis was performed on 600 cases of thyroid papillary carcinoma treated for the first time in Zhejiang Tumor Hospital from January 2020 to December 2022,including 300 cases of lymph node metastasis group and 300 cases of non-metastasis group in the central region.The features of ultrasound images were extracted and filtered by 3Dslicer software,and then the original features were reduced by using lasso regression model.After the reduction,the original features were respectively in SVM(support vector machine)RF(Random Forest),XGB(Extreme Gradient Boosting)and LGBM(Light Gradient Boosting Machine)are tested with the best parameters selected by the eight-fold cross-validation method.The sensitivity,specificity,accuracy,AUC(area under the ROC curve)and F1-score of the imaging group model,multimodal radiomics group model and conventional ultrasound model were compared.Results The AUC of multimodal ultrasound radiomics group was higher than that of radiomics group and conventional ultrasound.The AUC of multimodal radiomics group model(SVM:0.85,RF:0.79,XGB:0.79,LGBM:0.75),radiomics group model(SVM:0.79,RF:0.74,XGB:0.68,LGBM:0.66),conventional ultrasound and clinical data model was 0.70.Conclusion The predictive model of lymph node metastasis in the central region of neck of thyroid papillary carcinoma established by multimodal ultrasound combined with radiomics and different machine learning methods has high diagnostic performance,and the SVM model is better.