动态增强MRI影像组学特征预测乳腺癌腋窝淋巴结转移的价值

Dynamic contrast-enhanced MRI radiomic features predict axillary lymph node metastasis of breast cancer

目的探讨增强MRI影像组学分析预测乳腺癌腋窝淋巴结(ALN)转移的价值.方法前瞻性收集2016年5月至2017年12月浙江大学医学院附属杭州市第一人民医院疑似乳腺癌患者196例并行乳腺MRI动态增强检查,分析72个ALN的增强MRI图像,其中转移阴性35个,转移阳性37个.由1名副主任医师及1名住院医师分别进行图像评估,并进行一致性分析.根据临床手术或淋巴结穿刺术的病理结果将淋巴结分为转移阴性和阳性2组,临床及影像资料也对应分组.以3∶1的比例通过分层抽样的方法将两组淋巴结分为训练集及验证集,提取6大类共385个影像组学特征(直方图参数、形态学参数、纹理特征参数、灰度共生矩阵参数、游程矩阵参数、灰度区域大小矩阵参数),通过特征降维算法选择具有统计学意义的参数.使用二分类logistic回归建立预测模型,通过混淆矩阵对验证集进行验证,用ROC分析评价模型的诊断效能.结果降维选择后得到均匀度、全角度集群突出方差、全角度相关性、长行程优势及表容比5个影像组学特征(P<0.05),ROC分析得出其独立预测ALN转移的曲线下面积为0.747~0.931.影像组学特征的组间一致性较好,组内相关系数为0.841~0.980.基于上述特征构建乳腺癌ALN转移预测模型,该模型的ROC下面积、敏感度、特异度、准确度分别为0.953、0.893、0.926、92.6%(50/54).通过混淆矩阵对验证集进行验证,得到ROC下面积、敏感度、特异度、准确度分别为0.944、0.900、1.000、88.9%(16/18).结论基于影像组学特征构建的预测模型能无创地对乳腺癌ALN转移风险做出有效评估.

Objective To investigate the prognostic value of radiomics analysis in predicting axillary lymph nodes (ALN) metastasis of breast cancer based on dynamic contrast-enhanced MR imaging (DCE-MRI). Methods One hundred and ninety-six patients with suspected breast cancer were prospectively collected for dynamic breast DCE-MRI. Enhanced MR imaging data of 72 axillary lymph nodes were evaluated separately by a chief radiologist and a resident, and the consistency analysis was performed. Lymph nodes were dichotomized according to the pathology results derived from operation or biopsy under real-time virtual sonography based on MRI data. Clinical and imaging data were also divided into corresponding groups.(Imaging) Data from both groups were respectively classified as training set and testing set by stratified sampling in proportion with 3∶1. AK software was applied to extract 6 major categories of 385 features (including histogram, morphology, texture parameters, gray level co-occurrence matrix, run-length matrix and grey level zone size matrix from imaging), and a set of statistically significant features were subsequently obtained by dimension reduction. The prediction model was established through binary classification logistic regression and employed to externally test the validation set by the method of confusion matrix. Meanwhile, ROC analysis was applied to assess the diagnostic performance of the model. Results Of the 72 axillary lymph nodes, 35 were metastatic negative and 37 were positive. The consistency of enhanced MRI radiomics features was good, between 0.841 and 0.980. Uniformity, ClusterProminence_AllDirection_offset1_SD, Correlation_AllDirection_offset1, LongRunEmphasis_angle90_offset7 and SurfaceVolumeRatio were statistically significant differences (P<0.01), the area under the ROC between 0.747 and 0.931. In the training and testing group, the areas under the ROC, sensitivity, specificity and accuracy of the model were 0.953, 0.893, 0.926, 92.6%(50/54) and 0.944, 0.900, 1.000, 88.9%(16/18) respectively. Conclusion The prediction model based on radiomic features may provide a non-invasive and effective approach to the assessment of the risk of ALN metastasis of breast cancer.