| Objectives:In this research,we extracted the radiomics features of different modes of ultrasound images to explore the diagnostic efficacy of the prediction model based on the combination of multi-modality ultrasound radiomics features and the features of clinical and ultrasound risk factors for cervical lymph node metastasis in patients with papillar thyroid carcinoma(PTC).A normogram based on the clinical-multimode ultrasound radiomics combined model was constructed to investigate its value in preoperative prediction of cervical lymph node metastasis in PTC.And the clinical decision curve analysis was used to quantify the net benefit at different threshold probabilities to determine the clinical usefulness of the prediction model.Methods:In this study,164 patients with PTC confirmed by pathology from March 2016 to December 2021 in our hospital were collected according to the inclusion and exclusion criteria.All patients underwent conventional ultrasound,contrast-enhanced ultrasound(CEUS)and strain elastography ultrasound(SE-US)before surgery,with complete clinical data and DICOM format images.All PTCs were randomly divided into training cohorts(n=115)and test cohorts(n=49)at a ratio of 7:3.The included cases were divided into lymph node metastasis group and non-metastasis group according to pathological results.1.Retrospective analysis of the relationship between the clinical features,multimodal ultrasonographic features and cervical lymph node metastasis in patients with PTC.The independent predictors of PTC cervical lymph node metastasis were determined by univariate and multivariate analysis,and the clinical prediction model was established to observe its prediction efficacy.2.ITK-SNAP software is used to manually outline the region of interest of the cancer focus on the original ultrasound images of each PTC patient in different modes:conventional ultrasound(cross section and longitudinal section),CEUS and SE-US.FAE Pro 0.3.6 software was used for high-throughput and multifaceted radiomics features extraction,and 930 quantitative features were extracted from each mode.Two feature datasets were established according to whether clinical and ultrasound risk factors were added,which are the simple multimodal ultrasound radiomics dataset and the dataset of the clinical and ultrasound risk factors combined with multimodal ultrasound radiomics.The dataset was randomly divided into training cohorts and test cohorts at a ratio of 7:3.The data were normalized by means and Z-score standardization,principal component analysis(PCA)and pearson correlation coefficient(PCC)were used for dimensionality reduction of feature space.Recursive feature elimination(RFE)method was used to screen key features,linear discriminant analysis(LDA)and logistic regression classifier with LASSO constraints build muli-modality ultrasound radiomics prediction model and clinical-multimode ultrasound radiomics combined model respectively.All models were using a 10-fold cross-validation,the receiver operating characteristic(ROC)curves of each model were plotted and prediction efficacy of prediction models was evaluated.3.Comparing the diagnostic efficacy of three different prediction models and determining the optimal model for predicting cervical lymph node metastasis in PTC,the combined model nomogram was established by using the radiomics scores obtained from the clinical-multimodal ultrasound radiomics model combined with clinical independent predictors.The calibration curve and consistency index(C-index)were used to evaluate the calibration and discrimination performance of nomogram,and clinical decision curve was plotted to evaluate the clinical application value of each prediction model.Results:1.Among 164 patients with PTC,70 cases had cervical lymph node metastasis and 94 cases had no metastasis.Univariate analysis showed that in the training cohort,there were statistically significant differences in nodule size(P<0.001),echo(P=0.046),microcalcification(P=0.035),multifocality(P<0.001),contact range of the adjacent capsule(P<0.001)and enhancement pattern(P=0.038)between lymph node metastases and non-metastases in PTC patients,the nodule size(P=0.017),microcalcification(P=0.006),multifocality(P=0.021)and capsule contact range(P=0.003)in test cohort were statistically different between the two groups.Multivariate analysis after variable screening showed that the nodule size ≥1cm(P=0.041),multifocality(P<0.001)and capsule contact range > 50%(P=0.021)were independent predictors of cervical lymph node metastasis in patients with PTC.Therefore,when the clinical prediction model was established,the AUC of the model was 0.841 in the training cohort and 0.777 in the test cohort.2.In the multimodal ultrasound radiomics features,15 key features were finally screened to construct the radiomics prediction model and the results showed that the AUC of the multimodal ultrasound radiomics model was 0.925 and 0.768 in the training and test cohorts,respectively.The sensitivity specificity and accuracy of the model in the validation cohort were 85.7% 67.9% 75.5%,respectively.3.Combining ultrasound radiomics features with clinical and ultrasound risk factors,20 key features are selected to form the best subset of features to construct the combined prediction model.The AUC of the combined model was 0.957 in the training cohort and0.932 in the test cohort,and its sensitivity,specificity and accuracy were as high as 95.2%,89.3% and 91.8% in the test cohort,respectively.4.Among the three prediction models,clinical-multimodal ultrasound radiomics model showed the best prediction efficacy in predicting cervical lymph node metastasis of PTC.The prediction performance of the combined model in test cohort is better than simple multimodal ultrasound radiomics model(AUC,0.932 vs 0.768,P=0.008),and better than clinical prediction model(AUC,0.932 vs 0.777,P=0.012).In addition,compared with clinical or ultrasound radiomics alone in diagnosing PTC metastatic lymph nodes,the combination of clinical and radiomics can improve the sensitivity,specificity,accuracy and other diagnostic indexes of the model to varying degrees.5.In the test cohort,the consistency index of the combined model was 0.932,which more intuitively showed that the combination of clinical and radiomics had a good predictive effect on PTC lymph node metastasis,and the nomogram correction curve fitted well with the standard curve.Clinical decision curve analysis showed that clinical-multimodal ultrasound radiomics model could make PTC patients get the greatest net benefit,which is significantly better than other models.Conclusions:1.The radiomics model based on four modal ultrasound images of conventional ultrasound(cross section and longitudinal section)and CEUS and SE-US combined the characteristics of various modes to make their information complementary,thus improving the accuracy of qualitative and location analysis of metastatic lymph nodes,and helping doctors to improve the diagnostic accuracy of preoperative PTC metastatic lymph nodes by non-invasively distinguishing PTC patients with lymph nodes metastasis before operation.2.The combined prediction model based on the features of multimodal ultrasound radiomics and clinical and ultrasound risk factors can effectively evaluate the metastatic lymph nodes in patients with PTC,and has the best diagnostic efficiency compared with the clinical prediction model and the ultrasound radiomics model,which can truly achieve the efficient and accurate preoperative prediction of PTC lymph node metastasis.The combined model nomogram visualizes the results of the model,and provides a noninvasive prediction tool that combines ultrasound image features and clinical risk factors in clinical practice.3.The clinical decision curve analysis indicated that the combined model has good clinical application value in predicting cervical lymph node metastasis of thyroid papillary carcinoma,which is helpful for clinicians to make individualized and accurate treatment plan,avoid over-diagnosis and selectively perform lymph node dissection for PTC patients with different conditions according to the prediction model to improve the postoperative quality of life of patients. |
PDF Full Text Request