基于乳腺X线图像不同区域的纹理分析鉴别乳腺肿块良恶性

Texture analysis based on mammogram region segmentation to identify benign and malignant breast masses

目的:探讨乳腺X线图像上基于肿块不同区域的纹理分析对于乳腺肿块良恶性的鉴别价值。方法:回顾性分析经乳腺X线检查发现并经手术病理证实的108例(共计118个)乳腺肿块患者的病例资料。其中良性肿瘤60个,恶性肿瘤58个。分别在每个病灶的整体、核心和边缘三个区域勾画感兴趣区(ROI),采用MaZda软件分别对3个ROI进行纹理特征的提取。比较良恶性组之间各项纹理特征的差异,对差异具有统计学意义的纹理特征绘制ROC曲线,计算曲线下面积(AUC)。采用二元logistic回归分析建立联合诊断模型并与常规乳腺X线诊断结果进行比较。结果:两组间肿块核心区域的各项纹理特征的差异均无统计学意义(P>0.05);两组间肿块整体区域和边缘区域的相关度、对比度、差方差、总和熵、熵和差异熵的差异均具有统计学意义(P<0.05),其中整体区域中的总和熵、熵及边缘区域中的对比度、总和熵、熵、差异熵的AUC均≥0.7。联合诊断模型的AUC值为0.918,显著高于常规乳腺X线诊断(P<0.05),敏感度和特异度分别为84.5%和91.7%,与常规乳腺X线诊断间的差异均无统计学意义(P>0.05)。结论:基于乳腺X线图像边缘区域的纹理分析方法对乳腺肿块的良恶性鉴别具有较高价值,能为临床诊断提供客观、可靠的依据。

Objective:To investigate the value of texture analysis using different ROI selected methods based on mammary gland molybdenum target X-ray images for distinguishing benign and malignant breast masses.Methods:108 patients with 118 breast masses found by mammary gland molybdenum target X-ray examination were retrospectively analyzed.According to their pathological results,they were divided into benign(n=60)and malignant groups(n=58).The regions of interest(ROIs)in each mass were drawn using three methods:the whole,core and edge area of the mass,then the texture features of the three ROIs were extracted by MaZda software respectively.The differences of texture features between the two groups were compared statistically,ROC curves were drawn for texture features with significant differences,and AUCs under the curves were calculated.Binary logistic regression analysis was used to establish a combined diagnosis model,and then its diagnosis results for breast masses were compared with that of conventional mammography.Results:There were no significant differences in the texture features of the core area in the breast mass between the two groups(all P>0.05).The values of correlation,contrast,difference variance,sum entropy,entropy and difference entropy based on the whole ROIs and the edge ROIs between the two groups showed significantly statistical differences(all P<0.05).The AUCs of sum entropy and entropy based on whole ROIs,and contrast,sum entropy,entropy and difference entropy based on edge ROIs were all≥0.7.The AUC value of the combined diagnosis model was 0.918,which was significantly higher than that of mammography(P<0.05).The sensitivity and specificity were 84.5%and 91.7%,respectively,and the differences were not statistically significant(P>0.05)when compared with those of mammography.Conclusion:The texture analysis method based on edge regions of mammary gland on molybdenum target images has high value in the differentiation of benign and malignant breast masses,and can also provide objective and reliable basis for clinical diagnosis.