基于2D与3D分割的T_(2)WI直方图分析在腮腺肿瘤鉴别中的对比研究  

comparative study of T_(2)WI histogram analysis based on 2D and 3D segmentation in the differential diagnosis of parotid tumors

作　　者：史宏伟 任继亮[1] 袁瑛[1] 陶晓峰[1] SHI Hong-wei;REN Ji-liang;YUAN Ying;TAO Xiao-feng(Department of Radiology,Shanghai Ninth People's Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai 20001l,China)

机构地区：[1]上海交通大学医学院附属第九人民医院放射科,上海200011

出　　处：《中国临床医学影像杂志》2024年第6期391-395,共5页Journal of China Clinic Medical Imaging

基　　金：国家自然科学基金资助项目(编号82172049)。

摘　　要：目的:对比基于二维(Two-dimensional,2D)和三维(Three-dimensional,3D)分割的脂肪抑制T_(2)加权成像(Fat-satu-rated T_(2)weighted imaging,Fs-T_(2)WI)直方图分析鉴别腮腺良、恶性肿瘤及多形性腺瘤和腺淋巴瘤的价值。方法:回顾性分析经病理证实的腮腺肿瘤患者159例,其中良性肿瘤119例,包括多形性腺瘤63例,腺淋巴瘤43例;恶性肿瘤40例。由2名医师于轴位Fs-T_(2)WI上分别进行2D和3D肿瘤分割,获取最大层面及全瘤感兴趣区域(Regionof interest,ROI)。采用FAE软件提取7个直方图特征,包括第10百分位数(10th)、第90百分位数(90th)、均值、中位数、熵、偏度、峰度。采用组内相关系数(Intraclass correlation coefficient,ICC)评价直方图参数观察者间的一致性。比较腮腺良、恶性肿瘤之间及多形性腺瘤和腺淋巴瘤之间直方图特征的差异。使用逐步逻辑回归筛选出独立预测因子。采用受试者工作特征(Receiver operating characteristic,ROC)曲线及Delong检验评估2D与3D直方图特征诊断腮腺肿瘤的效能,并比较不同分割方式曲线下面积(Area under the curve,AUC)的差异。结果:2D分割(ICC:0.877~0.981)和3D分割(ICC:0.877~0.986)的直方图参数均具有很好的观察者间一致性。区分腮腺良、恶性肿瘤,基于2D和3D分割时,10th均是独立预测因子,AUC分别为0.814和0.789,灵敏度分别为0.875和0.725,特异度分别为0.647和0.765。区分多形性腺瘤与腺淋巴瘤,基于2D分割的独立预测因子是中位数,AUC为0.890,灵敏度为0.857,特异度为0.837;基于3D分割的独立预测因子是90th、熵与均值,模型的AUC为0.942,灵敏度为0.857,特异度为0.884。Delong检验显示2D分割和3D分割模型鉴别腮腺良、恶性肿瘤及多形性腺瘤和腺淋巴瘤的AUC值间均无显著性差异(P值均>0.05)。结论:T_(2)WI直方图分析能够为腮腺肿瘤的诊断提供一种定量工具,2D分割可作为优先选择的检查手段。Objective:To compare the value of fat-saturated T_(2)weighted imaging(Fs-T_(2)WI)histogram based on 2D and 3D segmentation in dfferentiating 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-TWI.The maximal slice and whole-tumor region of interest were obtained.Seven histogram features were extracted using FAE software,including the 1Oth percentile(1Oth),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 adenoma from adenolymphoma,the median was an independent predictor based on 2D segmentation,with an AUC of 0.890,sensitivity of 0.857,and specificity of 0.837.Moreover,90th,entropy,mean and the combination model were the independent factors based on 3D segm

关 键 词：腮腺肿瘤 磁共振成像 

分 类 号：R739.87[医药卫生—肿瘤] R445.2[医药卫生—临床医学]