DCE-MRI图像纹理分析对乳腺良恶性病变的诊断价值

Diagnostic value of texture analysis based on DCE-MRI images on benign and malignant breast lesions

目的:利用基于灰度共生矩阵的纹理分析,回顾性分析乳腺动态对比增强(DCE)-MRI图像,探讨纹理参数与乳腺良恶性病变的相关性,并建立乳腺癌的诊断模型,提高其诊断准确性。方法:选取2016年11月—2018年11月行乳腺DCE-MRI扫描的连续性病例资料,收集纳入病例的基本临床信息及DCE-MRI图像,分为乳腺良性病变组和恶性病变组。利用MatLab软件提取基于灰度共生矩阵的17个纹理参数,比较两组病例的临床基本信息(包括年龄、绝经状态、病灶部位、体质量指数及病灶大小)和纹理参数的差异。结果:符合纳入标准的病例有136例乳腺恶性病变和67例良性病变,单因素分析中,乳腺良恶性病变组中13个纹理参数有统计学差异,且参数间存在共线性关系,通过逐步回归分析法筛选有意义的影响因素,结果提示Contrast、IMC1、Sum Average和Correlation等4个纹理参数是乳腺良恶性病变鉴别诊断中的主要影响参数。建立的诊断模型曲线下面积(AUC)为0.906 (95%置信区间:0.857~0.942,P<0.001),约登指数为0.667,最佳截断值为0.65,积分大于0.65诊断乳腺恶性病变的敏感性和特异性分别为84.56%和82.09%。验证组包括49例恶性病变和24例良性病变,对诊断模型进行验证,诊断恶性病变的敏感性和特异性分别为81.63%和87.50%。结论:基于灰度共生矩阵的纹理分析可用于乳腺良恶性病变的鉴别诊断。

Objective:To retrospectively analyze the breast dynamic contrast enhanced(DCE)-MRI images by using texture analysis based on gray level co-occurrence matrix,to explore the correlation between texture parameters and benign and malignant breast lesions,and to establish a diagnostic model of breast cancer.Methods:From December 2016 to November 2018,the continuous cases′ clinical and breast DCE-MRI data were collected and divided into benign breast lesions and malignant lesions groups.Seventeen texture parameters based on gray level co-occurrence matrix were extracted by MatLab software.The clinical information(including age,menopausal status,lesion location,BMI,and lesion size) and texture parameters were compared between the two groups.Results:There were136 breast malignant lesions and 67 benign lesions included in this study.By the univariate analysis,13 texture parameters were statistically different between the benign and malignant breast lesions,and there was collinearity between the parameters.After screening of meaningful factors,the results suggest that Contrast,IMC1,Sum Average,and Correlation were influencing factors in the differential diagnosis between benign and malignant breast lesions.The established diagnostic model area under the curve(AUC) was 0.906(95%CI:0.857-0.942,P<0.001),the Youden index was 0.667,and the optimal cutoff value was 0.65.When the score was greater than 0.65,the sensitivity and specificity of the diagnostic model were 84.56% and 82.09% respectively.The validation group consisted of49 malignant lesions and 24 benign lesions,and the sensitivity and specificity of validation group were81.63% and 87.50%,respectively.Conclusion:Texture analysis based on gray level co-occurrence matrix can be used to differentiate breast malignant lesions from benign lesions.