CT纹理分析鉴别成釉细胞瘤及牙源性囊肿的价值  

Value of CT texture analysis in the differentiation of ameloblastoma and odontogenic cyst

作　　者：宋庆博 任继亮[1] 陶晓峰[1] SONG Qingbo;REN Jiliang;TAO Xiaofeng(Department of Radiology,Shanghai Ninth People's Hospital,Shanghai Jiaotong University School of Medicine,Shanghai 200011,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：《实用放射学杂志》2021年第11期1767-1771,共5页Journal of Practical Radiology

基　　金：国家自然科学基金项目(91859202,81771901)。

摘　　要：目的探讨CT纹理分析鉴别成釉细胞瘤及牙源性囊肿的价值。方法回顾性分析经病理证实的22例成釉细胞瘤和28例牙源性囊肿(17例角化囊肿及11例含牙囊肿)的术前CT平扫图像。使用ITK-SNAP软件手动勾画2组病例的全病灶感兴趣区(ROI),利用LifeX软件提取出38个纹理参数,包括6个直方图、7个灰度共生矩阵(GLCM)、11个灰度游程长度矩阵(GLRLM),3个邻域灰度差异矩阵(NGLDM)和11个灰度区域长度矩阵(GLZLM)参数。采用Mann-Whitney U检验比较2组病例CT纹理参数的差异。采用二元逻辑回归筛选出独立预测因子并建立联合模型,绘制受试者工作特征(ROC)曲线评价独立预测因子及其联合模型的效能。采用Pearson相关性系数评价有统计学意义的纹理参数间的相关性。结果10个纹理特征有显著的组间差异(P<0.05),包括2个GLRLM参数、2个NGLDM参数及6个GLZLM参数。筛选出2个独立预测因子,分别为短区域因子(SZE)、灰度不均匀度(GLNU)。GLZLM_SZE、GLZLM_GLNU及其联合模型曲线下面积(AUC)分别为0.87、0.72及0.92,敏感度分别为86.4%、90.9%及86.4%,特异度分别为85.7%、50.0%及85.7%。66.67%(30/45)的纹理参数间存在明显相关性(|r|≥0.5)。结论CT纹理分析可提供更多量化信息,成釉细胞瘤及牙源性囊肿的部分CT纹理参数存在差异,可为鉴别2种疾病提供一种新方法。Objective To explore the value of texture analysis in differentiating ameloblastoma from odontogenic cyst on CT images.Methods 22 cases of ameloblastomas and 28 cases of odontogenic cysts(including 17 odontogenic keratocysts and 11 dentigerous cysts)confirmed by pathology were analyzed retrospectively.All the patients underwent pre-treatment unenhanced CT examination.For the two groups,region of interest(ROI)was delineated manually to cover the whole lesion by using ITK-SNAP software.CT texture parameters were extracted by using LifeX software,including 6 histogram parameters,7 grey level co-occurrence matrix(GLCM)parameters,l1 grey-level run-length matrix(GLRLM)parameters,3 neighborhood grey-level different matrix(NGLDM)parameters and 11 grey-level zone length matrix(GLZLM)parameters.Mann-Whitney U test was used to compare the differences of CT texture parameters between two groups.Binary Logistic regression was performed to determine the independent predictors and to build combined model.Receiver operating characteristic(ROC)curves were generated to evaluate the diagnostic performance of the independent predictors and the combined model.Pearson coefficients were used to determine the correlations among statistically significant texture parameters.Results Ten texture parameters were significantly different between the two groups(P<0.05),including 2 GLRLM parameters,2 NGLDM parameters and 6 GLZLM parameters.Two texture parameters including short-zone emphasis(SZE)and grey-level non-uniformity(GLNU)were selected as independent predictors in Logistic regression analysis.'The GLZLM_SZE,GLZLM_GLNU and the combined model achieve area under the curve(AUC)of 0.87,0.72 and 0.92,the sensitivities of 86.4%,90.9%and 86.4%,the specificities of 85.7%,50.0%and 85.7%,respectively.Significant correlation(|r|≥0.5)was observed in 66.67%(30/45)texture parameters.Conclusion CT texture analysis can provide more quantitative information,and some CT texture parameters differ between ameloblastoma and odontogenic cyst,which may provid

关 键 词：纹理分析 成釉细胞瘤 牙源性囊肿 计算机体层成像 

分 类 号：R739.8[医药卫生—肿瘤] R780.1[医药卫生—临床医学]