Preliminary Study Of Benign And Malignant Thyroid Nodule Prediction Based On Radiomics

Objective:Thyroid nodules are the most common clinical symptoms in thyroid diseases and may be caused by many causes.In recent years,the detection rate of thyroid nodules has been significantly increased with the extensive application of ultrasound,CT and PET in thyroid examination.Thyroid nodules were mostly unexpected findings,there is no obvious clinical symptoms.According to the report of the American Thyroid Association,the incidence of thyroid cancer has been increasing year by year.The treatment of benign and malignant thyroid nodules is different,accurate diagnosis of benign and malignant nodules is the key to the post-treatment process.A variety of imaging techniques for the diagnosis of thyroid nodules have some significance to the most widely used ultrasonic examination,but the ultrasound examination by the doctor qualifications,misdiagnosis have occurred.Fine needle aspiration(FNA)is the most accurate method for preoperative diagnosis of benign and malignant thyroid nodules.The FNA is an invasive examination and the puncture results are affected by the puncture site and the material taken.Therefore,looking for a more accurate method to identify the benign and malignant thyroid nodules,to avoid unnecessary surgical treatment is extremely important.Radiomics is a new image post-processing technology in recent years.Through the deep excavation of the features of the existing medical image texture,image decoding can decode disease-related information hidden inside the image.This study was based on the analysis of CT image texture characteristics of thyroid CT scan,evaluate the feasibility of radiomics method to identify benign and malignant thyroid nodules,and to determine the best screening methods and determination methods.Materials and Methods:A total of 714 thyroid nodules were retrospectively collected from September 2016 to December 2016 in the First Affiliated Hospital of China Medical University.A total of 336 cases of thyroid nodules were screened for selection.There were 217 benign nodules and 119 malignant nodules.Selected patients have a complete pathological data parallel thyroid CT scan.The texture analysis software Ma Zad was used to preprocess the CT images and manually sketch the region of interest(ROI)at the maximum tumor level to extract the texture parameters.The extracted texture parameters were dimensionally reduced,and the best texture parameters of benign and malignant were selected by Fisher coefficient,POE + ACC and MI.The texture discriminant analysis process uses the B11 module in the Ma Zad software.Four different discriminant methods were applied to each dimensionality reduction method to calculate the error rate,sensitivity,specificity and accuracy respectively.Using the method of discriminating the minimum error rate as a benchmark,the chi-square test was used to analyze whether the differences of post-processing methods of various texture parameters are statistically significant.Test level α = 0.05.Results:The correlations of texture parameter features S(0,2)were screened out for all three dimensionality reduction methods.Among all texture post-processing methods,the lowest error rate was 5.65% for the MI-NDA/ANN method.In the prediction of benign and malignant thyroid nodules,the highest sensitivity was Fisher/NDA method of94.12%;the best specificity and prediction accuracy were MI/NDA method,and the specificity of MI/NDA method was 98.62%.The rate is 94.34%.Chi-square test was performed between different discriminating methods for each dimensionality reduction method.The test level was α=0.05.In Fisher’s dimension reduction algorithm,the differences between discriminant methods PCA,LDA and NDA were statistically significant,with P value less than 0.05 and χ2 values as 12.50 and 5.93,respectively.In the POE+ACC dimensionality reduction algorithm,the differences between the discriminant methods RDA,PCA,LDA and this group of NDA were statistically significant.The P value was less than 0.05,and the χ2 values were 12.89,25.27,and10.27,respectively.In the MI dimension reduction algorithm,the differences between the discriminant methods RDA,PCA,LDA and this group of NDA were statistically significant.The P value was less than 0.05,and the χ2 values were 6.31,8.19,and 15.52,respectively.Conclusion: Radiomics is a reliable and quantitative method to distinguish benign and malignant thyroid nodules.It can be used as a secondary group in the diagnosis of benign and malignant thyroid nodules.