| Objectives:To investigate the value of ultrasound(US)radiomics nomogram,combined with clinical risk factors and rad score based on two-dimensional ultrasonography(2D-US)and strain elastic ultrasonography(SE-US),for the preoperative prediction of thyroglobulin(Tg)levels in lateral cervical lymph nodes in papillary thyroid carcinoma(PTC).Methods:1.In this retrospective analysis,202 patients with suspicious PTC who received ultrasound-guided fine needle aspiration Tg(FNA-Tg)in lateral cervical lymph nodes from March 2021 to January 2022 in our Hospital.After the strict inclusion and exclusion criteria,153 patients were selected in this study.All patients underwent total thyroidectomy and bilateral cervical lymph node dissection,with complete clinical data and preoperative 2DUS and SE-US images.The optimal critical value of Tg was determined with the presence of lateral cervical lymph node metastasis in surgical pathology as the gold standard,and was divided into high level Tg group(n=99)and low level Tg group(n=54).At a ratio of 7:3,a total of patients was randomly split into a training cohort(n=107)and a validation cohort(n=46).2.Clinically relevant risk factors of the patients were collected,including sonographic features of the primary tumor(lesion,size,composition,echogenicity,shape,margin,echogenic foci,adjacent capsule,elasticity,and multiple),sonographic features of the lymph nodes(size,cystic change,calcification,hilum disappeared,and circular)and age,sex.Independent sample t test or Chi-square test were used for univariate analysis,and logistic regression analysis was used to determine clinical independent risk factors for high Tg levels in PTC lateral cervical lymph nodes,and to construct the clinical prediction model,and evaluating their diagnostic performance.3.A sonographer with more than 5 years of clinical experience manualed outline the region of interest(ROI)of the images of 2D-US(maximum horizontal and vertical planes)and SE-US(maximum vertical plane)of the nodules in each patient applied ITK-SNAP software.The original images and ROI images of each patient were imported into FAE 0.3.6software for extracting the radiomics features,with a total of 3560 radiomics features for each patient,including first-order features,shape features,texture features and wavelet features.Synthetic minority oversampling technique(SMOTE)balance and mean normalization to pretreat the features,Pearson correlation coefficients(PCC)and Kruskal Wallis(KW)test screened the most relevant features.Logistic regression(LR)classifier and10-fold cross-validation of the training cohort was used to construct the radiomics model and calculate the rad score.4.Clinical risk factors combined with rad score were incorporated into multiple logistic regression to construct the ultrasound radiomics combined model.Draw the receiver operator characteristic(ROC)curve of clinical model,radiomics model and combined model.De Long test was used to evaluate the discriminant performance of the three prediction models.5.Draw the ultrasound radiomics nomogram.Calibration curves and HosmerLemeshow tests were used to evaluate the consistency between predicted Tg levels and actual Tg levels.The clinical benefits of clinical model,radiomics model and combined model were evaluated with decision curve analysis(DCA).Integrated discrimination improvement(IDI)was calculated to evaluate the clinical incremental value of radiomics.Results:1.FNA-Tg with 24.72ng/m L as the diagnostic threshold showed the best performance in predicting PTC lymph node metastasis in the whole cohort(sensitivity: 95.5%,specificity:89.0%,AUC: 0.938).2.Lymph node size,lymph node cystic change and lymph node hilum disappeared were clinical risk factors for high Tg level of lateral cervical lymph nodes in PTC(P < 0.05).Therefore,a clinical prediction model was established,with an AUC of 0.800 in the training cohort and 0.769 in the validation cohort.3.Selecting the radiomics features,thirteen of the most relevant features to construct the radiomics prediction model.The AUC of the model in the training cohort and the validation cohort were 0.799 and 0.840,respectively.4.The diagnostic performance AUC of the ultrasound radiomics combined model constructed by lymph node size,lymph node cystic change,lymph node hilum disappeared and rad score to predict Tg level of PTC lateral cervical lymph nodes was 0.883 and 0.885 in the training cohort and validation cohort,respectively.De Long test showed statistically significant differences between clinical model and combined model in training cohort and validation cohort(training cohort: 0.800 vs.0.883,P = 0.018;validation cohort: 0.769 vs.0.885,P = 0.034).5.The ultrasound radiomics nomogram visualizes the predicted risks of different risk factors,and calibration curves and Hosmer-Lemeshow test show that the model has a well goodness of fit.DCA showed that when the threshold probability was in the range of 4%-86%,the preoperative prediction of Tg levels of lateral cervical lymph nodes in PTC by using ultrasound radiomics nomogram had a higher clinical benefit than the application of radiomics features or clinical risk factors alone.The IDI was 17.78% after the addition of the rad score,indicating that the diagnostic performance of the ultrasound radiomics nomogram was significantly improved compared with the clinical prediction model alone.Conclusions:1.The radiomics model based on 2D-US(maximum horizontal and vertical planes)and SE-US(maximum vertical plane)images extracted more tumor heterogeneity information in the depth of images,which could predict the Tg level of PTC lateral cervical lymph nodes more objectively,comprehensively and noninvasive before surgery,helping physicians to improve the diagnostic accuracy.2.Ultrasound radiomics nomogram has realized the visualization of risk prediction of different risk factors,which is conducive to clinicians to enacted individualized and precise treatment plans,and provides a non-invasive prediction tool for clinical prediction of Tg level of PTC lateral cervical lymph nodes.In clinical working,the addition of radiomics analysis can provide more medical support for clinical practice,help to enhance doctors’ confidence in diagnosis,make reasonable and effective surgical decisions,improve patients’ quality of life. |
PDF Full Text Request