The Diagnostic Value Of Radiomics Model Based On ZOOMit DWI Images In Papillary Thyroid Carcinoma

Objective:To investigate the value of imaging omics model based on ZOOMit DWI images for diagnosis of papillary thyroid carcinoma and prediction of cervical lymph node metastasis.Methods:The data of 129 patients with thyroid nodules from March 2021 to May2022,all patients were confirmed by histopathology,including 51 benign nodules and107 malignant nodules;91 patients with malignant nodules underwent thyroid resection and lymph node dissection,including 51 without lymph node metastasis and 40 with lymph node metastasis.The data set of benign and malignant thyroid nodules was divided into training set（126 cases）and validation set 8:2（32 cases）;the data set with lymph node metastasis was randomly divided by 7:3 into training set（63 cases）and validation set（28 cases）.All patients underwent ZOOMit DWI scans on a Siemens MAGNETOM Prisma 3.0 T superconducting MR imager.The best display level of lesions was selected on the ADC（b=800 s/mm²）map,the region of interest（Region of interest,ROI）was manually delineated,and the ADC_meanvalue was averaged in three times,and statistical analysis was conducted.The difference was statistically significant at P<0.05.ROI was manually delineated in DWI（b=800 s/mm²）images,and features were extracted by imaging radiomics software to extract features,including 18first-order features,14 shape features,and 4 texture features（including gray scale symbiosis matrix,gray scale region matrix,gray scale path matrix,and gray scale dependent matrix）.A total of 874 radiomics features in the data set of benign and malignant thyroid nodules were included after three filtering processing methods,Removing redundant features using variance threshold selection,Using an analysis of variance（ANOVA）to screen for the most valuable features,By logic,st（LR）to the prediction model,And verify in the validation set;A total of 100 radiomics features were included in the lymph node metastasis data set,Removing redundant features using variance threshold selection,Using the minimum absolute contraction and selection operator（LASSO）screened the features,Building a predictive model by LR,And were validated in the validation set.The final receiver operating characteristic（ROC）curve was drawn to evaluate the diagnostic efficacy of the LR model with the area under the curve（AUC）.Results:1.ADC（b=800 s/mm²）varied between benign and malignant thyroid nodules（P<0.05）,ROC curve analysis AUC was 0.954,sensitivity 0.953 and specificity0.863.2.The ADC（b=800 s/mm²）value was statistically significant between the groups with and without lymph node metastasis（t=-4.609,P<0.05）.ROC curve analysis yielded an AUC of 0.725,sensitivity 0.980 and specificity 0.600.3.ANOVA-LR model of benign and malignant nodules:36 features were selected to enter the model after ANOVA dimension reduction screening.The model established by LR has AUC 0.831,accuracy 0.873,sensitivity 0.729 and specificity 0.780 in the training set and AUC0.827,accuracy 0.750,sensitivity 0.818 and specificity 0.600 in the validation set.4.LASSO-LR model of PTC with lymph node metastasis dataset:19 features were selected to enter the model for AUC 0.931,accuracy 0.889,sensitivity 0.857,specificity0.914 in the training set;AUC 0.927,accuracy 0.786,sensitivity 0.833,and specificity0.750 in the validation set.Conclusion:1.Imagomics model based on MRI ZOOMit DWI（b=800 s/mm²）image has high diagnostic value for PTC diagnosis and good clinical application prospect for predicting lymph node metastasis.2.The ZOOMit DWI technology ADC value（b=800 s/mm²）has a good diagnostic efficacy for PTC diagnosis and prediction of lymph node metastasis.