The Predictive Value Of Multimodal Ultrasound And Radiomics In Cervical Lymph Node Metastasis Of Papillary Thyroid Carcinoma

ObjectiveTo explore whether there are independent risk factors for cervical central lymph node metastasis(CLNM)in patients with papillary thyroid carcinoma(PTC)in clinical features,BRAF gene,multimodal ultrasound features,and radiomics features.Based on this,we aim to establish an optimal preoperative prediction model and nomogram to visually assess the risk of cervical CLNM in PTC patients,thereby assisting clinicians in individualized preoperative decision-making and prognosis evaluation of patients.MethodsPatients with thyroid nodules with Imaging Reports and Data Systems ≥ grade 4b and proposed for surgical treatment who attended Yantai Yuhuangding Hospital from September2020 to December 2022 were selected.129 patients with PTC who underwent preoperative multimodality ultrasonography and fine needle aspiration to obtain the BRAF were included.A random sampling method was used to divide the patients into a training group and a validation group according to 7:3,and the patients were further divided into a metastatic group and a nonmetastatic group according to the postoperative lymph node pathology in the central region of the neck.The clinical characteristics,BRAF gene and multimodal ultrasound features of PTC nodes were collected from the patients.The multimodality ultrasound images of PTC nodes were imported into 3D-Slicer software,the region of interest was outlined manually,and then the radiomics features were extracted using the Pyradiomics software package.Features with inter-observer and Intraclass correlation coefficient(ICC)>0.75 were retained,and the Least absolute shrinkage and selection operator(LASSO)was used to filter the image.The radiodomics score(Rad-score)was constructed by screening the features for non-zero features using the LASSO.The variables with differences(p<0.05)in the baseline data were included in univariate and multivariate logistic regression to determine the independent risk factors for CLNM,different prediction models were constructed based on the independent risk factors,and the prediction model with the best diagnostic efficacy was selected to construct the nomogram.Results1 Analysis of baseline data: There were 90 patients in the training group and 39 patients in the validation group out of 129 PTC patients.The differences in baseline data in both groups were not statistically significant(p>0.05).In the training group,there were statistically significant differences in gender,number of nodules,microcalcifications,perineural invasion,enhancement intensity and peak intensity between the metastatic and non-metastatic groups(p<0.05).No differences were found in age,serum thyroid stimulating hormone(TSH),BRAF gene,maximum diameter,site,morphology,border,aspect ratio,internal echogenicity between the metastatic and non-metastatic groups(p>0.05).2 Multimodal ultrasound radiomics features screening and Rad-score construction: 1409 radiomics features were extracted from images of conventional ultrasound,SWE and CEUS,and after screening out the stable features with ICC>0.75,8,11 and 17 features were selected by LASSO regression for the construction of multimodal ultrasound Rad-score.3 Comparison of the construction and efficacy of different prediction models: after univariate and multivariate logistic regression analysis,male,multifocality,perineural invasion,equal-high enhancement,and multimodal ultrasound imaging radiomics scores were independent risk factors for CLNM in the neck of PTC patients(p<0.05).In both the training and validation groups,the combined model constructed based on all independent risk factors predicted the diagnostic efficacy of cervical CLNM in PTC(AUC:0.934 in the training group and AUC: 0.774 in the validation group)better than the clinical multimodality ultrasound feature model constructed based on sex,number of nodules,tegmental invasion and intensity of enhancement(AUC:0.841 in the training group and AUC:0.633)and the multimodal ultrasound imaging radiomics model based on three ultrasound modalities Rad-score(training group AUC:0.829,validation group AUC:0.732).4 Construction and performance evaluation of nomogram: nomogram were constructed based on the joint model with the optimal diagnostic efficacy.In the training and validation groups,the calibration curve showed that the column line graph had good predictive ability for the occurrence of neck CLNM in patients with PTC;the decision curve showed that the net benefit of the nomogram was higher than that of the clinical multimodal ultrasound feature model and multimodal ultrasound imaging radiomics model within a reasonable risk threshold.Conclusions1 The clinical multimodal ultrasound feature model has good predictive efficacy for neck CLNM in PTC patients.2 The multimodal ultrasound radiomics model was not significantly better than the clinical multimodal ultrasound feature model in predicting CLNM in patients with PTC.3 The combined model constructed by adding multimodal radiomics score to the clinical multimodal ultrasound feature model has significantly improved diagnostic efficacy,and has high sensitivity and specificity,as well as certain calibration ability and high clinical application value,which is expected to provide an objective basis for accurate clinical formulation of individualized treatment plans and assessment of prognosis.4 Preoperative serum TSH and BRAF gene obtained by puncture cannot be used as a complementary diagnostic tool to predict the risk of CLNM.