基于2D与3D分割的T2WI直方图分析在腮腺肿瘤鉴别中的对比研究OA北大核心CSTPCD

Objective:To compare the value of fat-saturated T2 weighted imaging(Fs-T2WI)histogram based on 2D and 3D segmentation in differentiating benign from malignant parotid tumors and,among the former,differentiating pleomorphic adenoma from adenolymphoma.Methods:A retrospective analysis of 159 patients with pathologically confirmed parotid tumors was performed,including 119 benign tumors(63 pleomorphic adenomas and 43 adenolymphomas)and 40 malignant tumors.2D and 3D tumor segmentation was performed by two doctors on axial Fs-T2WI.The maximal slice and whole-tumor region of interest were obtained.Seven histogram features were extracted using FAE software,including the 10th percentile(10th),90th percentile(90th),mean,median,entropy,skewness,and kurtosis.The intraclass correlation coefficient(ICC)was used to evaluate the inter-observer agreement of the histogram parameters.Differences in histogram characteristics were compared between be-nign and malignant parotid tumors and between pleomorphic adenoma and adenolymphoma.Independent predictors were select-ed using stepwise logistic regression.The receiver operating characteristic(ROC)curve analysis and Delong's test were used to assess the efficacy of 2D versus 3D histogram features for diagnosing parotid tumors and to compare the differences in the area under the curve(AUC)between different segmentation methods.Results:The histogram parameters for 2D segmentation(ICC:0.877～0.981)and 3D segmentation(ICC:0.877～0.986)showed excellent inter-observer agreement.In differentiating benign from malignant parotid tumors,10th was an independent predictor based on both 2D and 3D segmentation,with AUC of 0.814 and 0.789,sensitivity of 0.875 and 0.725,and specificity of 0.647 and 0.765,respectively.In differentiating pleomorphic ade-noma from adenolymphoma,the median was an independent predictor based on 2D segmentation,with an AUC of 0.890,sensi-tivity of 0.857,and specificity of 0.837.Moreover,90th,entropy,mean and the combination model were the independent factors based on 3D segmentation,with an AUC of 0.942,sensitivity of 0.857,and specificity of 0.884 fo the combination model.De-long's test showed that the AUC values of 2D and 3D segmentation models for discriminating benign from malignant parotid tumors,as well as pleomorphic adenomas from adenolymphoma had no significant differences(all P values>0.05).Conclusion:T2WI histogram can be a quantitative tool for diagnosing parotid tumors.2D segmentation can be used as a preferred method.