基于深度卷积网络和结合策略的乳腺组织病理图像细胞核异型性自动评分

发布时间：2018-06-19 00:38

  本文选题：细胞核异型性 + 深度卷积网络　； 参考：《中国生物医学工程学报》2017年03期

【摘要】：细胞核异型性是评估乳腺癌恶性程度的一个重要指标,主要体现在细胞核的形状、大小变化、纹理和质密度不均化。提出基于深度学习和结合策略模型的乳腺组织细胞核异型性自动评分模型。该模型使用3个卷积神经网络,分别处理每个病例的3种不同分辨率下的组织病理图像,每个网络结合滑动窗口和绝对多数投票法,评估每个病例同一种分辨率下的图像的分值,得到3种分辨率下的评分结果。使用相对多数投票法,综合评估每个病例的最终细胞核异型性评分结果。为评估模型对细胞核异型性评分的有效性,利用训练好的模型对124个病例的测试图像进行自动评分,并把其评分结果与病理医生的评分结果作比较,进行性能评估。该模型的评分正确率得分为67分,其结果在现有的细胞核异型性评分模型中准确率排名第二。此外,该模型的计算效率也很高,平均在每张×10、×20、×40分辨率下图像的计算时间分别约为1.2、5.5、30 s。研究表明,该细胞核异型性评分模型不仅具有较高的准确性,而且计算效率高,因此具备潜在的临床应用能力。
[Abstract]:Nuclear heterogeneity is an important index to evaluate the malignancy of breast cancer, which is mainly reflected in the shape, size, texture and density of the nucleus. An automatic grading model of breast tissue nucleus heterogeneity based on deep learning and combined strategy model was proposed. The model uses three convolution neural networks to process three different resolution histopathological images of each case. Each network combines sliding window and absolute majority voting. The image scores of each case at the same resolution were evaluated and the scoring results were obtained under three resolutions. A relative majority voting method was used to evaluate the final nuclear heterogeneity score for each case. In order to evaluate the effectiveness of the model in evaluating nuclear heterogeneity, the trained model was used to evaluate the image of 124 cases automatically, and the results were compared with those of pathologist to evaluate their performance. The scoring accuracy of this model is 67, and the result is the second in the current nuclear heterogeneity scoring model. In addition, the computational efficiency of the model is also very high, and the average computing time of each image at the resolution of 10 脳 10, 脳 20, 脳 40 is about 1.2 ~ 5.5 ~ 30 s, respectively. The results show that the model not only has high accuracy, but also has high computational efficiency, so it has potential clinical application ability.
【作者单位】： 南京信息工程大学江苏省大数据分析技术重点实验室;华中科技大学附属武汉市中心医院病理科;
【基金】：国家自然科学基金(61273259) 江苏省“六大人才高峰”高层次人才项目(2013-XXRJ-019) 江苏省自然科学基金(BK20141482)
【分类号】：R737.9;TP183
，

本文编号：2037531