Analysis Of ACR TI-RADS Classification Characteristics Of Thyroid Nodules And Study Of Benign And Malignant Judgment Based On Thyroid Ultrasound Radiofrequency Signal

Objective To explore the value of thyroid nodule characteristics analysis and thyroid ultrasound Radio Frequency(RF)data in the diagnosis of benign and malignant thyroid.Methods A retrospective analysis was performed on 701 patients with definite pathological results who underwent surgical resection or needle aspiration biopsy in Zhejiang Provincial Cancer Hospital Hospital from January 2019 to December 2021.A total of 803 nodules were classified according to the ACR TI-RADS classification criteria for the characteristics of ultrasound images,such as composition,echo,aspect ratio,and edge morphology.Calcification and blood flow score were statistically analyzed to observe the correlation between the above indexes for the judgment of benign and malignant thyroid tumors.Then,the ultrasonic radiofrequency signals of 659 patients with thyroid nodules who met the requirements of thyroid RF data collection were obtained and stored.Then do some preprocessing,and then select the appropriate neural network model;And model training;Finally,the results of the model test set and validation set were compared with the diagnosis of senior sonographers.Results In univariate analysis,age,gender,composition,echo,aspect ratio,marginal morphol ogy,calcification and blood flow score were statistically significant for the judgment of benign and malignant(P<0.05).In multivariate analysis,gender and age were negatively correlated,while echo,aspect ratio,marginal morphology and calcification were positively correlated.For single risk factor analysis,the area underthe ROC curve(AUC)was as follows:the area under the ROC curve of echo was 0.666;The area under the receiver operating characteristic curve of aspect ratio was 0.695.The area under the receiver operating characteristic curve of the edge morphology was 0.741.The area under the receiver operating characteristic curve of calcificati on was 0.578.The area under the receiver operating characteristic curve(ROC)was 0.798 whe n echo,aspect ratio,edge shape and calcification were used to predict the risk factors of tumor.Finally,a total of 659 patients with thyroid nodules based on ultrasound radiofrequency signal model were enrolled,including 181 benign nodules and 478 malignant nodules.There were 73132 radio frequency data tensors,of which 20492 were benign and 52640 were malignant.The mean age was(47.08±12.64)years(range,9-81 years).There were 513 females and 146 males.The AUC of senior sonographers in the test set was 0.825(95%CI:0.725,0.925).The AUC of senior ultrasound physicians in the validation set was 0.839(95%CI:0.744,0.935),respectively.The AUC of RF model in test set was 0.841(95%CI:0.765,0.917).The AUC of the validation s et RF model was 0.879(95%CI:0.798,0.996).Conclusion In ROC curve analysis,the AUC of multivariate analysis was significantly higher than that of single factor analysis,indicating that the reliability of the comprehensive consideration of multiple risk factors was higher than that of single risk factors in the prediction of thyroid cancer.The deep learning model based on the RF signal of thyroid ultrasound with Inception-v3 convolutional neural network as the main trunk has a high prediction and diagnosis efficiency for thyroid malignancy,and its results can be comparable to the diagnosis results of senior sonographers using the ACR TI-RADS classification guidelines.