| Objective:To explore the incremental value of CT(computed tomography),radiologist’s subjective diagnosis and clinical risk factors in predicting cervical lymph node metastasis of papillary thyroid carcinoma(PTC);The purpose of this study was to establish a preoperative prediction model for lymph node metastasis of papillary thyroid carcinoma.Methods:(1).319 patients with PTC confirmed by surgery and pathology in the first hospital of Shanxi Medical University from November 2017 to April 2021 were retrospectively collected.After strict screening,178 patients were included in this study.They were randomly divided into training group(n = 125)and test group(n = 53)according to the ratio of 7:3.The age,sex,body mass index(BMI),thyroid hormone,location,anteroposterior diameter,transverse diameter,anteroposterior to transverse diameter ratio,calcification,capsule,Hashimoto’s thyroiditis(HT),nodular goiter(NG)and lymph node metastasis reported by radiologists were collected.According to the surgical and pathological results,the patients were divided into lymph node metastasis group and non-lymph node metastasis group.(2).Three radiologists with clinical experience of 5,10 and 13 years independently evaluated the status of PTC cervical lymph node metastasis(including lymph node metastasis group,non-lymph node metastasis group and suspected lymph node metastasis group).Chi square test was used to compare the clinical parameters between lymph node metastasis group and non-lymph node metastasis group.The statistically different clinical risk factors were included in multiple logistic regression analysis and support vector machine(SVM)analysis to select the best model construction method to construct the clinician’s subjective diagnosis model.(3).An imaging diagnostic physician manually delineates the regions of interest(ROI)of tumor foci in conventional CT images,CT arterial phase and venous phase layer by layer using ITK-SNAP software,and extracts high-throughput radiomics features(including firstorder features,shape features,texture features and wavelet features,with a total of 2553 radiomics features for each patient).The features data were normalized by Z-score standardization and mean standardization;Principal component analysis(PCA)and Pearson correlation coefficient(PCC)dimensionality reduction methods were used to process characteristic data;Analysis of variance and Kruskal Wallis test were used to select significant features.Finally,multiple logistic regression analysis was used to establish the noncontrast model,the arterial contrast model,the venous contrast model and the threephase radiomics model.(4).Combining the characteristics of clinician subjective diagnosis model and threephase radiomics model,and incorporating multiple logistic regression analysis,the combined model integrating clinical risk factors,physician diagnosis and radiomics features was constructed.(5).Draw the receiver operating characteristic curve(ROC)of clinician subjective diagnosis model,three-phase radiomics model and combined model,and compare the diagnostic efficiency between the three models.(6).Binary logistic regression analysis was used to screen clinical risk factors.Combined with clinical risk factors and radiomics features,the nomogram of the combined model was constructed,and the correction curve was drawn to evaluate the error of the combined model in predicting the status of lymph node metastasis in patients with PTC.Decision curve analysis was used to compare the expected benefits of clinician subjective diagnosis model,three-phase radiomics model and combined model in clinical diagnosis,and Delong test was used to compare the area under curve(AUC)of clinician subjective diagnosis model and combined model.Results:(1).Age,location,capsule,anteroposterior diameter,transverse diameter,anteroposterior to transverse diameter ratio,CT reported lymph node metastasis 1 and CT reported lymph node metastasis 3 were significantly different between the cervical lymph node metastasis group and the non-lymph node metastasis group of PTC(P < 0.05),and there was no significant difference in the ratio of other clinical factors(P > 0.05).(2).No matter in the training group or the test group,the AUC value of noncontrast model,arterial contrast model,venous contrast model and three-phase radiomics model in predicting cervical lymph node metastasis was almost the same as that of clinician’s subjective diagnosis model(AUC in the training group: 0.786,0.808,0.827,0.790 vs.0.781,0.796,0.800;AUC in the test group: 0.781,0.791,0.790,0.813 vs.0.758,0.729,0.743).(3).Binary logistic regression analysis showed that age,anteroposterior diameter,anteroposterior to transverse diameter ratio and CT reported lymph node metastasis status 1were independent predictors of PTC lymph node metastasis,which could be included in the nomogram of the combined model.(4).The radiomics features were added to the clinician prediction model to construct the nomogram of the combined model.In the training group,the AUC value increased from0.781 to 0.868;In the test group,the AUC value also increased significantly,from 0.758 to0.878.Decision curve analysis showed that the combined model showed higher clinical benefits in predicting central neck lymph node metastasis in patients with PTC than the other two models.There was significant difference in the diagnostic performance between the clinician subjective diagnosis model and the combined model(P = 0.003 in the training group;P = 0.017 in the test group),and the AUC value of the combined model was higher than that of the clinician subjective diagnosis mode.Conclusion:CT radiomics can predict the cervical lymph node metastasis in patients with PTC before operation.The combination of CT radiomics into the clinical diagnosis process and clinical risk factors and imaging physician diagnosis can significantly improve the prediction accuracy.The constructed nomogram provides a potential non-invasive tool for patients with PTC to evaluate the status of lymph node metastasis. |
PDF Full Text Request