Research Of Preoperative Lateral Cervical Lymph Node Metastasis Prediction Models Based On Cervical Contrast Enhanced CT Radiomics

Objective:To establish a method for preoperative prediction of lateral cervical lymph node metastasis(LCLNM)of papillary thyroid carcinoma(PTC)based on cervical contrast enhanced CT.To build and improve a more intelligent auxiliary diagnosis system,so as to provide a new means for the noninvasive and accurate clinical diagnosis of PTC patients with LCLNM before surgery.Methods:This study prospectively collected neck enhanced CT images and clinical data of PTC patients who underwent radical thyroidectomy +multifunctional neck lymph node sparing dissection for the first time in Thyroid Surgery Department,General Surgery Center,First Hospital of Jilin University from December 2020 to November 2022 in strict accordance with inclusion and exclusion criteria.Target lymph nodes were accurately located by Three-Dimensional(3D)CT images before surgery,and one to one accurate sampling was performed during surgery.A total of 102 lymph node samples were obtained from 43 patients.There were 73 samples with metastatic Lymph Node(+),LN(+)] and 29 samples with non-metastatic lymph nodes [LN(-)].Lymph node samples were randomly divided into a train set(n=71)and a test set(n=31)at a ratio of 7:3.Then,volume of interest(VOI)segmentation and feature extraction were carried out.T test and least absolute shrinkage and selection operator(LASSO)algorithm have been used for dimensional reduction screening of radiomics features(RFs)in cross-validation set.Then,the clinical data were sorted out and statistically analyzed to screen out the clinical features related to LCLNM.Finally,2 logistic regression(LR)prediction models were established to predict lymph node properties respectively.Finally,receiver operating characteristic(ROC)curves were drawn,and the prediction ability of each model was evaluated according to the area under the curve(AUC).Decision curve analysis(DCA)was used to evaluate the clinical practicability of the models.Results:1427 RFs were extracted from each VOI.After T test and LASSO algorithm screening,14 RFs most associated with LCLNM were retained to construct the radiomics prediction model.The AUC(95%CI)of the train set was 0.958(0.913-1.000).The AUC(95% CI)of the test set =0.905(0.801-1.000).After statistical analysis of clinical data,one clinical feature related to LCLNM was obtained,that is,gender.It was fused with14 RFs to construct the clinical fusion model together.In the clinical fusion model,the AUC(95%CI)of the train set and the test set were0.969(0.931-1.000)and 0.914(0.815-1.000),respectively.The predictive ability of clinical fusion model is slightly improved compared with that of radiomics model,but the difference is not significant.Finally,DCA was carried out on the 2 prediction models respectively,and the results showed that both models had good clinical practicability.Conclusions:1.The radiomics model and clinical fusion model developed in this study based on neck contrast enhanced CT can predict the situation of lateral cervical lymph node metastasis in patients with PTC before surgery.2.These models are helpful to improve preoperative non-invasive prediction and risk assessment of lateral cervical lymph node metastasis in patients with PTC.