摘要
目的:探讨CT平扫图像的一阶纹理分析特征参数对腮腺多形性腺瘤和Warthin瘤的鉴别价值.方法:收集2019年7月至2021年12月经临床病理证实的腮腺多形性腺瘤和Warthin瘤患者各35例的CT平扫图像,在纹理分析软件中勾画全瘤灶的感兴趣区,得到一阶纹理特征参数.应用统计学分析以上参数在腮腺多形性腺瘤和Warthin瘤中差异有无统计学意义,并对有统计学意义的纹理特征参数制作ROC曲线,求得其诊断效能.结果:腮腺CT平扫图像的一阶纹理特征的18项参数中,以下9项参数有统计学意义(P＜0.05):Percentile10(P=0.000)、Percentile90(P=0.000)、Energy(P=0.000)、Maximum(P=0.000)、Mean(P=0.000)、Median(P=0.000)、RootMeanSquared(P=0.000)、TotalEnergy(P=0.000)、Variance(P=0.046)其中 Percentile90、Root-MeanSquared、Mean、Median 等 4 项参数诊断效能较好,AUC 分别为 94.80％、94.20％、93.80％、93.80％.结论:腮腺CT平扫一阶纹理分析在多形性腺瘤与Warthin瘤中的有一定鉴别价值.
Abstract
Objective:To investigate the value of the first-order texture analysis feature parameters of CT plain scan images in the identification of pleomorphic adenoma and warthin tumor of the parotid gland.Methods:The CT plain scan images of 35 patients with pleomorphic adenoma of the parotid gland and 35 patients with Warthin tumor confirmed by clinicopathology were collected,and the region of interest of the whole tumor was delineated in the texture analysis software to obtain the first-order texture feature parameters.Statistical analysis was used to analyze whether the above parameters were significantly different between parotid gland pleomorphic adenomas and warthin tumors,and ROC curves were made for the texture parameters with statistical significance to obtain their diagnostic efficacy.Results:Among the 18 parameters of the first-order texture features of the CT plain scan images of the parotid gland,the following 9 parameters were statistically significant(P＜0.05):Percentile10(P=0.000),Percentile90(P=0.000),Energy(P=0.000),Maximum(P=0.000),Mean(P=0.000),Median(P=0.000),RootMeanSquared(P=0.000),TotalEnergy(P=0.000),Variance(P=0.046)among which Percentile90,RootMeanSquared,Mean,Median The diagnostic efficiency of the other four parameters was good,and the AUC were 94.80％,94.20％,93.80％,and 93.80％,respectively.Conclusion:The first-order texture analysis of parotid CT plain scan has a certain value in distinguishing pleomorphic adenoma and Warthin tumor.
维普
万方数据