乳腺X线摄影全乳图像纹理特征与乳腺良恶性病变相关性分析

Analysis of the Correlation Betweenfull-Field Mammographic Image Texture Features and Benign or Malignant Breast Lesions

目的探讨乳腺X线摄影全乳图像纹理特征与乳腺良恶性病变的相关性。方法搜集经病理证实的乳腺良、恶性病例各50例,所有患者治疗前均进行乳腺X线摄影检查。使用MaZda软件在单侧内外侧斜位(MLO)图像上手动勾画全乳感兴趣区(ROI),进行纹理特征提取,然后对提取的纹理参数进行统计学分析。结果对30个纹理参数进行t检验,结果显示有28个纹理参数具有统计学意义(P<0.05)。FPM+NDA分类结果显示,4个恶性样本分类到良性组,5个良性样本分类到恶性组,以恶性组为阳性组,良性组为阴性对照组,计算得准确率91%、敏感度92%、特异度90%、漏诊率8%、误诊率10%。受试者工作特征曲线(ROC)结果显示直方图峰度(Kurtosis)、第99百分位数(Perc.99%)、方差(Variance)和灰度共生矩阵中和平均(SumAverg)对鉴别乳腺良恶性病变具有较高效能,以Kurtosis、Perc.99%最优。结论全乳纹理特征直方图中Kurtosis、Perc.99%、Variance及灰度共生矩阵中SumAverg与乳腺良恶性病变有较大相关性,全乳高密度和较低复杂度与较高的乳腺癌发病概率相关,纹理定量分析可用于乳腺良恶性病变鉴别诊断。

Objective To explore the correlation between full-field mammographic image texture features and benign or malignant breast lesions.Methods Fifty benign and malignant breast cases confirmed by pathology in our hospital were collected,and all patients underwent mammography before treatment.The breast region of interest(ROI)was manually delineated on the unilateral mediolateral oblique(MLO)image using MaZda software.Performing texture feature extraction,and then performing statistical analysis on the extracted texture parameters.Results T-test of 30 texture parameters showed that 28 texture parameters were statistically significant(P<0.05).FPM+NDA classification results showed that 4 malignant samples were classified as benign,5 benign samples were classified as malignant,with malignant as the positive group and benign as the negative control group.The calculated accuracy was 91%,sensitivity was 92%,and specificity was 90%,Missed diagnosis rate was 8%,misdiagnosis rate 10%.The receiver operating characteristic(ROC)curve results show that the histogram Kurtosis,Perc.99%,Variance,and the gray-scale co-occurrence matrix SumAverg have higher efficiency in identifying benign and malignant breast lesions,with Kurtosis and Perc.99%being the best.Conclusion Histogram Kurtosis,Perc.99%,Variance and GLCM SumAverg of full-field mammographic image texture features have a greater correlation with benign and malignant breast lesions.High density and low heterogeneity of full-field mammographic image are associated with a higher probability of breast cancer incidence.Quantitative analysis of mammographic image texture features could be used for the differential diagnosis of benign and malignant breast lesions.