基于虚拟光学密度图像的乳腺癌近期发病预测

发布时间：2018-05-25 18:17

  本文选题：虚拟光学密度图像 + 乳房X线摄影术　； 参考：《中国医学影像技术》2017年08期

【摘要】：目的探讨对原始乳腺钼靶图像进行变换和采用机器学习算法融合不同类型的图像特征,以提高乳腺癌近期发病风险预测精度的价值。方法自匹兹堡大学医学中心的临床数据库下载185例女性受检者头足(CC)位全数字化乳腺X线摄影(FFDM)图像。首先对原始灰度图像进行乳腺区域分割并将其变换为虚拟光学密度图像,而后从原始灰度图像和虚拟光学密度图像中分别提取不对称特征。基于此不对称特征分别训练第1阶段的2个决策树分类器,再以这2个分类器输出的得分值作为输入,训练第2阶段的1个决策树分类器。对乳腺癌近期发病风险预测性能采用留一法进行验证。结果采用两阶段决策树融合方法预测乳腺癌的ROC曲线下面积(AUC)为0.9612±0.0132,敏感度为96.63%(86/89),特异度为91.67%(88/96),准确率为94.05%(174/185)。结论从虚拟光学密度图像中可提取出对乳腺癌具有较高预测力的特征,采用两阶段决策树方法对两类图像特征进行二次融合有助于提高乳腺癌近期发病风险预测精度。
[Abstract]:Objective to improve the accuracy of breast cancer risk prediction by transforming original mammary mammography and using machine learning algorithm to fuse different types of image features. Methods A total digital mammography (FFDM) image of 185 female subjects was downloaded from the clinical database of University of Pittsburgh Medical Center. Firstly, the original gray image is segmented and transformed into a virtual optical density image, and then the asymmetric features are extracted from the original gray image and the virtual optical density image. Based on this asymmetric feature, two decision tree classifiers in the first stage are trained, and the score values of the two classifiers are used as input to train a decision tree classifier in the second stage. The predictive performance of breast cancer risk in the near future was verified by one-left-one method. Results the ROC curve area under the ROC curve was 0.9612 卤0.0132, the sensitivity was 96.63 / 86 / 89, the specificity was 91.67 / 96 / 96, and the accuracy was 94.05 / 1754 / 185.Results by using the two-stage decision tree fusion method, the area under the ROC curve was 0.9612 卤0.0132, the sensitivity was 96.63 / 86 / 89, the specificity was 91.67 / 96. Conclusion the features with high predictive power for breast cancer can be extracted from the virtual optical density image. Using the two-stage decision tree method to perform secondary fusion of the two kinds of image features is helpful to improve the prediction accuracy of the risk of breast cancer in the near future.
【作者单位】： 上海理工大学医疗器械与食品学院;
【分类号】：R730.44;R737.9
，

本文编号：1934192