基于常规MRI纹理分析技术鉴别良恶性软组织肿瘤

The value of texture analysis derived from conventional MRI in differentiating benign and malignant softTissue tumors

目的:探讨基于常规MRI的纹理分析技术在鉴别良恶性软组织肿瘤(STT)中的价值。方法:回顾性分析2015年1月-2019年9月在我院经手术病理证实的91例STT患者的病例资料。所有患者术前行MRI检查,包括横轴面T1WI及横轴面、冠状面和矢状面脂肪抑制质子密度加权像(FS-PDWI)。采用MaZda软件进行纹理分析,在各序列图像中选取病灶最大层面及相邻的2个层面,测量肿块大小,沿病灶边界勾画ROI,提取病灶的纹理特征,包括直方图参数(均值、变异度、偏度、峰度及第1、10、50、90、99百分位数)和灰度共生矩阵(GLCM)参数(能量、对比度、相关、平方和、逆差矩、均和、和方差、和熵、熵、差方差和差熵)。采用t检验、LSD(方差齐)或Mann-Whitney U检验(方差不齐)对良恶性组的患者年龄、病灶大小及纹理参数进行比较,对差异有统计学意义的指标进一步行ROC曲线分析评估其诊断效能。采用多因素Logistic回归分析获得判断良恶性STT的独立预测因素并建模,绘制ROC曲线评估模型的鉴别诊断效能。结果:91例中良性43例、恶性48例。良恶性STT的大小及在T1WI和FS-PDWI图像上的对比度、相关、逆差矩、差方差和差熵的组间差异均有统计学意义(P<0.05);其它GLCM参数及直方图参数在两组间的差异无统计学意义(P>0.05)。FS-PDWI图像上,除差熵的诊断效能低于T1WI(AUC:0.710 vs.0.714),其它参数(对比度、相关、逆差矩、差方差)的诊断效能均优于T1WI(AUC分别为0.853 vs.0.761,0.849 vs.0.742,0.750 vs.0.714和0.807 vs.0.723)。以病理结果良、恶性为因变量,采用逐步回归法来筛选自变量,选入回归模型的变量为T1WI上的差熵及FS-PDWI上的差方差和差熵(P值分别为0.033、0.030和0.031),提示上述参数是判断STT良恶性的独立预测因素,相应模型的AUC为0.811,敏感度为82.4%,特异度为71.4%。结论:基于常规MRI的纹理分析技术有助于良恶性软组织肿瘤的鉴别。

Objectvive:To investigate the diagnostic value of the texture analysis derived from conventional MR imaging in differentiating between benign and malignant soft tissue tumors(STTs).Methods:91 patients with pathologically proved STTs from January 2015 to September 2019 were enrolled in this retrospective study.All patients underwent conventional MR imaging before surgery including axial T1WI,fat-suppression proton density weighted imaging(FS-PDWI),FS-PDWI of coronal and sagittal plane.Texture features were calculated from manually drawn ROIs by using MaZda software.The size of STTs were manually measured and calculated.The texture features,including histogram parameters(mean,variance,skewness,kurtosis,P1,P10,P50 P90 and P99)and GLCM parameters[energy,contrast,correlat,sum of squares(SumOfSqs),inverse difference moment(IDM),sum of average(SumAverg),sum variance(SumVarnc),sum entropy(SumEntrp),entropy,difference variance(DifVarnc),difference entropy(DifEntrp)]were measured and analyzed for each patient.Least significant difference ort test were used to compare the differences between the parameters of the two groups which were normal distribution and equal variance,while Mann-Whitney U test was used to compare the parameters that did not conform to normal distribution or variance.ROC curve analysis was performed regarding the statistically significant parameters and the areas under curve(AUC)were calculated.Multivariate logistic regression analysis was accomplished to obtain the independent predictive factors of benign and malignant STTs,and ROC curve was drawn to evaluate the differential diagnosis efficacy.Results:There were 43 benign and 48 malignant included in a total of 91 STTs.The size of tumors,parameters of contrast,correlat,IDM,DifVarnc,and DifEntrp of T1WI and FS-PDWI showed significantly differences between the two groups(all P<0.05).Whereas,all the histogram parameters and the rest of the GLCM parameters(energy,SumOfSqs,SumAverg,SumVarnc,SumEntrp,entropy)did not differ significantly between the two groups(all P>005).The AUC of the contrast,correlat,IDM,DifVarnc and DifEntrp of T1WI for distinguishing benign from malignant STTs were 0.761,0.742,0.714,0.723 and 0.714 respectively;the contrast,correlat,IDM,DifVarnc and DifEntrp of FS-PDWI were 0.853,0.849,0.750,0.807 and 0.710,respectively.Multivariate logistic regression analysis showed that the DifEntrp of T1WI,DifVarnc and DifEntrp of FS-PDWI were screened out as the independent variables,which suggested that they were the predictors for STTs characteristics.The AUC of the logistic regression differential diagnosis model was 0.811,sensitivity was 82.4%,specificity was 71.4%.Conclusion:Texture analysis of conventional MR imaging can provide reliable and objective basis for differentiating benign STTs from malignant ones.