Study On Ultrasound Imagomics For Predicting Lymph Node Metastasis In Thyroid Cancer

Objective: To study the two-dimensional gray-scale ultrasound image based on the lesions of patients with thyroid carcinoma(TC),build a prediction model with the method of ultrasound imageology,and predict whether the cervical lymph nodes of patients with thyroid papillary carcinoma have metastasis as the ultimate purpose,so as to provide scientific reference for clinical surgery to select appropriate lymph node dissection and scope,and avoid excessive medical treatment.Methods: From July 2020 to June 2021,528 patients with TC who were first diagnosed at the First Affiliated Hospital of Wannan Medical College(Yijishan Hospital)and underwent thyroid cancer focus resection and neck lymph node dissection were retrospectively analyzed.Based on the gold standard of surgical and pathological diagnosis,two-dimensional ultrasound gray-scale images of the patient’s satisfaction were retained.In the selection process,we should pay attention to the judgment based on the ultrasonic characteristics of the lesions,pay attention to the examination of the main location of the lesions,the maximum length diameter,and whether there is calcification,and collect the maximum long-axis section photos of each patient for archiving,and collect the patient’s age,sex,and general clinical information and store them on the personal workstation.According to the surgical and pathological characteristics,two groups were divided: clear cervical lymph node metastasis(LNM)group and clear non-lymph node metastasis(NLNM)group.For the best conventional two-dimensional ultrasound original image of each TC patient’s focus,the 3D slicer,a commonly used imageomics tracing software,is used to draw the complete region of interest(ROI)with the best focus data image displayed by manual tracing method.The extended software package "Radiomics" imageomics software is selected to extract the ultrasound imageomics features of the delineated region of interest.All the included cases were screened into training set and test set by7:3 random stratified sampling.The extracted imageomics features and other data are first standardized.In order to prevent data redundancy and over-fitting,correlation analysis and the least absolute shrinkage and selection operator(LASSO)algorithm are selected for filtering and dimensionality reduction of feature data.For general ultrasound image features,select single factor analysis,select statistically significant features,and form a new feature set with ultrasound image histology features.After removing invalid variables through logistic multiple regression analysis,build a joint prediction model.The diagnostic efficacy was objectively evaluated by drawing the receiver operating characteristic curve(ROC)of the subject.The consistency between the predicted value of the model and the actual clinical pathology selects the calibration curve for the efficiency test.Results: 1.A total of 528 cases of thyroid papillary carcinoma were collected in this study,including 338 cases in the NLNM group and 190 cases in the LNM group.Among the general clinical features,the distribution of age and sex was not statistically significant,while the diameter and boundary of the lesions were statistically significant(P<0.05).2.In the training set,in order to prevent the redundant features of big data analysis from forming unnecessary interference and other effects in data mining,it is necessary to remove the irrelevant features first.After the obtained features are filtered and dimensioned through correlation analysis and LASSO regression method,the characteristics obtained are the four image group characteristics with the highest correlation.They belong to shape features and texture features(GLCM,GLDM)respectively.3.The ultrasound prediction model built by combining the imageomics features retained after screening and dimensionality reduction,the AUC in the training set was0.888,and the AUC in the validation set was 0.691,both of which performed well.4.The calibration curve was used for validation.The results showed that the prediction results of the prediction model were in good agreement with the postoperative pathology,and the prediction model was highly reliable.Conclusion: This study proposes a convenient and novel research tool based on the original ultrasound data,and a clinical prediction model of cervical lymph node metastasis in patients with thyroid carcinoma based on the extracted ultrasound image group features and the general two-dimensional ultrasound features,It can accurately and effectively evaluate the condition of patients with thyroid carcinoma before surgery,evaluate and predict whether to expand lymph nodes and the choice of surgical methods,and provide important reference for clinicians to formulate individualized and accurate surgical plans for different patients before surgery,showing a good level of diagnostic performance.This application method based on ultrasound imageomics is expected to greatly optimize clinical decision-making,reduce excessive medical treatment as much as possible,and improve the survival and prognosis of patients.