MRI纹理参数对乳腺癌分子分型鉴别的初步探讨

Preliminary study on the classification of molecular subtypes in breast cancer by the texture parameters

目的:探讨乳腺癌MRI增强扫描的纹理参数鉴别乳腺癌分子分型的可能性。方法:收集79例乳腺癌患者的MRI动态增强扫描图像,其中42例为人表皮生长因子受体-2(HER-2)阳性Luminal B型乳腺癌患者,37例为HER-2阳性非Luminal B型乳腺癌患者。共采集43个影像学纹理参数,通过Mann-Whitney检验评判MRI纹理参数在两类乳腺癌中的差异性,并通过ROC曲线评价差异有统计学意义的纹理参数对两类乳腺癌鉴别诊断的敏感度和特异度。结果:5个纹理参数在两类乳腺癌患者中差异均有统计学意义(均P<0.05),其中4个纹理参数的敏感度和特异度较高。回归模型对两类乳腺癌患者的整体判别准确率为88.6%,预测概率ROC曲线下面积为0.88。结论:乳腺癌MRI纹理参数可满足对乳腺癌HER-2阳性Luminal B型与HER-2阳性非Luminal B型的鉴别诊断,有望成为乳腺癌的临床辅助检查手段,基于乳腺癌MRI纹理参数建立的Logistic回归模型以较高的敏感度和特异度实现了对两类分子分型乳腺癌的鉴别诊断。

Objective:This study was aimed to evaluate the classification of molecular subtypes in breast cancer by the texture features derived from breast MRI images.Methods:In this research,texture features from breast MRI images,including first-order parameters,morphological parameters,gray level co-occurrence matrix parameters and Haralick parameters,were used to differentiate the HER-2 positive luminal-B breast cancer and HER-2 positive non-luminal-B breast cancer.79 cases of breast cancer patients with dynamic contrast enhanced MRI images were involved,including 42 cases with HER-2 positive luminal-B breast cancer,37 cases with HER-2 positive non-luminal-B breast cancer.Results:43 texture parameters were obtained from each patient,Mann-Whitney test were carried out on the two types breast cancer cases,ROC was also carried out to evaluate the sensitivity and specificity.5 texture parameters showed a significant difference in the two types of breast cancer patients,and 4 of them showed high sensitivity and specificity.The logistic regression model was also applied to the texture parameters,and the discriminant accuracy rate of regression model was 88.6%,the forecast area under curve of ROC was 0.88.Conclusions:The texture parameters from breast MRI images could be used to distinguish the HER-2 positive luminal-B breast cancer and HER-2 positive non-luminal-B breast cancer.The logistic regression based on image texture parameters of breast cancer is established with high sensitivity and specificity to achieve the differential diagnosis of breast cancer.MRI image texture parameters could enable the ability to distinguish the molecular subtypes of breast cancer,which might be a powerful assistant tool in future.