术前表观扩散系数图纹理分析预测舌和口底鳞状细胞癌组织学分级的价值

摘要

目的 探讨术前ADC图纹理分析在预测舌和口底鳞状细胞癌(SCC)组织学分级中的价值.方法 回顾性分析2015年5月至2018年6月上海交通大学医学院附属第九人民医院经术后病理证实,且组织分级明确的49例舌和口底SCC的ADC图纹理参数.入组患者Ⅰ、Ⅱ、Ⅲ级病例分别为21、21和7例,术前均行包含DWI的MRI检查.由2名医师使用3D Slicer软件勾画全瘤ROI,并提取8个直方图参数、11个灰度共生矩阵(GLCM)参数及7个灰度游程矩阵(GLRLM)参数.使用组内相关系数(ICC)评估观察者间纹理参数测量的一致性,仅对测量重复性极好(ICC>0.8)的参数进行分析.采用Mann?Whitney U检验比较Ⅰ级与Ⅱ、Ⅲ级舌和口底SCC的ADC图纹理参数的差异.使用逐步逻辑回归筛选出独立的预测因子并建立联合模型.使用ROC分析评估纹理参数或模型预测舌和口底SCC组织分级的效能.采用Pearson相关系数评价有统计学意义的纹理参数间的相关性.结果(1)69.23%(18/26)的纹理参数在观察者间测量一致性极好(ICC:0.81～0.98),包括6个直方图参数,7个GLCM参数及5个GRLM参数.(2)直方图参数中,Ⅰ级SCC的ADC值第10百分位数(ADC10)显著高于Ⅱ、Ⅲ级SCC,而能量及熵显著低于后者(P均<0.05);GLCM参数中,Ⅰ级SCC的联合熵、差熵、和熵、差方差、差均值及对比度显著低于Ⅱ、Ⅲ级SCC(P均<0.05);GLRLM参数中,Ⅰ级SCC的灰度不均匀度及游程长不均匀度显著低于Ⅱ、Ⅲ级SCC(P均<0.05).ADC10与熵为独立的预测因子,Ⅰ级SCC的ADC10、熵分别为960(913,1 178)×10?6mm2/s、4.32(4.06,4.76),Ⅱ、Ⅲ级SCC分别为888(816,987)× 10?6mm2/s、4.88(4.57,5.29).ADC10、熵及联合模型的曲线下面积(AUC)分别为0.72、0.75、0.81.(3)具有统计学意义的纹理参数中,52.73%(29/55)的参数间有明显相关性(|r|≥0.5).结论 ADC图纹理分析可提供更多量化信息,可较为准确地区分Ⅰ级与Ⅱ、Ⅲ级舌和口底SCC.%Objective To explore the value of texture analysis on ADC maps in the preoperative prediction of histological grade of tongue and mouth floor squamous cell carcinoma (SCC). Methods Forty?nine pathologically confirmed tongue and mouth floor SCC with definite grading from May 2015 to June 2018 were retrospectively analyzed, including 21 cases of gradeⅠ, 21 cases of gradeⅡand 7 cases of gradeⅢ. All subjects underwent preoperative MRI examination with DWI included. Two doctors delineated whole tumor region of interest and extracted texture parameters by the 3D Slicer software, including 8 histogram parameters, 11 grey?level co?occurrence matrix (GLCM) parameters and 7 gray?level run?length matrix (GLRLM) parameters. Intraclass correlation coefficient (ICC) was used to evaluate the inter?observer delineation agreement, and the texture parameters with excellent reproducibility (ICC>0.8) were used for analysis only. Mann?Whitney U test was used to compare the differences of ADC texture parameters between grade Ⅰ and grade Ⅱ?Ⅲ SCCs. Stepwise logistic regression was used to determine the independent predictors and to build combined model. ROC analysis was used to explore the performance of texture parameter and model in predicting histological grade of tongue and mouth floor SCCs. Pearson correlation coefficient was used to evaluate the correlation between texture parameters with statistical significance. Results (1) Excellent inter?observer delineation agreement (ICC: 0.81-0.98) was observed in 69.23% (18/26) texture parameters, including 6 histogram parameters, 7 GLCM parameters and 5 GLRLM parameters. (2) Among histogram parameters, significantly higher 10 percentile ADC value (ADC10) and significantly lower energy and entropy were shown in gradeⅠcompared with gradeⅡandⅢSCCs (all P<0.05). Among GLCM parameters, significantly lower joint entropy, difference entropy, sum entropy, difference variance, difference average and contrast were shown in grade Ⅰ SCCs (all P<0.05). Among GLRLM parameters, significantly lower gray?level nonuniformity and run?length nonuniformity were shown in gradeⅠSCCs (all P<0.05). ADC10 and entropy were identified as independent predictors. The ADC10 and entropy were 960(913, 1 178)×10?6mm2/s and 4.32(4.06, 4.76) in gradeⅠSCCs, and 888(816, 987)×10?6mm2/s and 4.88(4.57, 5.29) in gradeⅡ?ⅢSCCs respectively. The area under ROC curve (AUC) of ADC10, entropy and combined model were 0.72, 0.75, 0.81. (3) Significant correlation (|r|≥0.5) was observed among 52.73% (29/55)texture parameters with statistical significance. Conclusion Texture analysis on ADC maps can provide more quantitative information, which can be more accurately in discriminating grade Ⅰfrom gradeⅡ?Ⅲtongue and mouth floor SCCs.