The Value Of Radiomics Based On MRI FS-T2WI In The Differentiation Diagnosis Of Parotid Gland Tumors

Objective:The purpose of this study was to explore the application value of imaging omics based on MRI FS-T2WI in the identification of parotid gland tumors,(i.e.,pleomorphic adenoma,Warthin tumor,malignant parotid gland tumors),improve the correct diagnosis rate of different parotid gland tumors before operation and guide the formulation of surgical plans.Methods:A retrospective analysis of the pre-treatment MRI routine images of 148patients with parotid gland tumors confirmed by the Shandong Medical Imaging Research Institute and Shandong Provincial Hospital from October 2014 to September 2020,including 63 cases with pleomorphic adenomas and 45 cases with Warthin tumors,40 cases with malignant parotid gland tumors.The radiomic features of the tumor area were extracted from the FS-T2WI,and all patients were divided into training groups(43cases of pleomorphic adenoma,30 cases of Warthin tumor and 25 cases of parotid malignant tumor)and validation groups(20cases of pleomorphic adenoma,15 cases of Warthin tumor and 15 cases of parotid malignant tumors).Python Pyradiomics was used to perform feature extraction on all patients,the random forest to select the features of the training groups and build the models,and use the validation groups to verify the models.Statistical analysis of the three groups were performed to find out the statistical significance of each selected feature.The differential efficiency of model built was determined using a receiver operating characteristic curve(ROC)analysis.Results:Ten features with the highest importance rankings were extracted from the radiomics features in tumor area of FS-T2WI.Among the top ten features in the grouping of pleomorphic adenomas and Warthin tumors(including Square-firstorder-Uniformity、Square-firstorder Robust Mean Absolute Deviation、Square-firstorder-Median、Square-glcm-Sum Entropy、Square-gldm-Dependence Entropy、Square-glrlm-GrayLevelNonUniformityNormalized、log-sigma-5-0-mm-3D-firstorder-RootMeanSquared、Exponential-firstorder-InterquartileRange、Square-firstorder-InterquartileRange、Exponential-firstorder-Median),Square_firstorder_Uniformity was the most important single feature.This feature measures the sum of the squares of each intensity value.This is a measure of the homogeneity of the image array,where a greater uniformity implies a greater homogeneity or a smaller range of discrete intensity values.The uniformity(median(interquartile range))of multiline adenoma was 0.065(0.049,0.098),and the value(median(interquartile range))of Warthin tumor was 0.151(0.124,0.244).The AUC,optimal critical value,sensitivity,and specificity of the ROC curve of this feature were 0.853,0.122,78.1%,and 82.5%,respectively-Among the top ten features in the grouping of pleomorphic adenomas and malignant parotid tumor(including log-sigma-2-0-mm 3D-firstorder-Median.wavelet-LH-firstorder RootMeanSquared、log-sigma-3-0-mm-3D-firstorderMeanAbsoluteDeviation、log-sigma-2-0-mm-3D-firstorder_RootMeanSquared、Square-gldm-GrayLevelNonUniformity、log-sigma-4-0-mm-3D-firstorder-MeanAbsoluteDeviation、wavelet-LH-firstorder-Median、lbp-2D-furstorder-TotalEnergy、log-sigma-3-0-mm-3D-firstorder-RootMeanSquared、log-sigma-2-0-mm-3D-firstorder-Mean),the most important single feature was log-sigma-2-0-mm-3D_firstorder_Median.This feature represents the median gray level intensity within the ROI.The value(median(interquartile range))of pleomorphic adenoma was-46.318(-70.534,-35.456)and the value(median(interquartile range))of malignant parotid gland tumor was-20.706(-28.748,-13.220).The AUC,optimal critical value,sensitivity,and specificity of the ROC curve of this feature were 0.830、-28.893、77.1%、86.0%,respectively.Among the top ten features in Warthin tumor and malignant parotid tumor grouping(including log-sigma-5-0-mm-3D-glszm-GrayLevelNon Uniformity、Original-shape-LeastAxisLength、Exponential-glszm-GrayLevelNon Uniformity、Original-shape-VoxelVolume、Original-gldm-GrayLevelNon Uniformity、Original-shape-Maximum 2D Diameter Row、Logarithm-glcm-ClusterShade、wavelet-LL-gldm-GrayLevelNonUniformity、Original-shape-Mesh Volume、log-signa-2-0-mm-3D-glszm-GrayLevelNonUniformity),the most important single feature was log-sigma-5-0-mm-3D_glszm_GrayLevelNonUniformity.This feature measures the variability of gray-level intensity values in the image,with a lower value indicating more homogeneity in intensity values.The value(median(interquartile range))of Warthin tumor was 12.288(6.489,19.696)and the value(median(interquartile range))of malignant tumor was 29.531(14.568,47.726).The AUC,optimal critical value,sensitivity,and specificity of the ROC curve of this feature were 0.743、20.617、65.7%、78.1%,respectively.After the model was constructed,the AUC of the receiver operating characteristic curve in the training group in the above three groups were 0.93±0.05、0.84±0.16、0.79±0.11 respectively,and the AUC in the validation group were 0.74、0.86、0.87respectively.Conclusion:1.Based on MRI FS-T2WI for the three groups of pleomorphic adenoma,Warthin tumor and parotid malignant tumor,the most important features of individual features in radiomics are square_firstorder_Uniformity、including log-sigma-2-0-mm 3D-firstorder-Median、log-sigma-5-0-mm-3D-glszm-GrayLevelNon Uniformity.2.Pleomorphic adenoma,Warthin tumor and parotid gland malignant tumors have the best cut-off value for the most important single feature of the three groups,which have a high sensitivity and specificity for distinguishing the three.3.Radiomics based on MRI FS-T2WIis effective in the differentiation diagnosis of different parotid gland tumors.