基于深度卷积网络的结直肠全扫描病理图像的多种组织分割

发布时间：2018-05-01 04:15

本文选题：全扫描病理图像 + 多种类型组织　；参考：《中国生物医学工程学报》2017年05期

【摘要】：结直肠全扫描图像处理困难,原因在于图像的数据量大、结构复杂、信息含量多。目前对于结直肠癌组织病理图像的研究通常包含肿瘤和基质两种组织类型,只有一小部分研究可以解决多种组织的问题,但又不是处理全扫描的结直肠病理图像。提出一种基于深度卷积网络的结直肠全扫描病理图像进行多种类型组织分割的模型。该模型使用的网络层数有8层,利用深度卷积网络学习结直肠全扫描图像中典型的8种类型的组织,利用训练好的模型对这8种类型的结直肠组织进行分类测试,其测试集分类准确率达92.48%。利用该模型对结直肠全扫描病理图像中的8种类型组织进行分割,首先对全扫描图像进行预处理,分成5000像素×5000像素大小的图像块,然后标记出每一张中的8种类型的组织,最后将所得到的标记结果进行拼接,从而获得整张结直肠全扫描病理图像的8种类型组织的标记结果。该方法对8种类型的组织分割的准确率比较高,有一定辅助诊断的帮助。
[Abstract]:The total scanning of colorectal images is difficult due to the large amount of data, complex structure, and more information content. At present, the study of the histopathological images of colorectal cancer usually includes two types of tissues and tumors. Only a small part of the study can solve the problems of various tissues, but it is not a total scan of colon disease. A model of multiple types of tissue segmentation based on deep convolution network is proposed. The number of network layers used in this model is 8 layers, and 8 typical types of tissues in the total scanning image of colorectal cancer are studied by deep convolution network, and the 8 types of colorectal groups are used by the trained model. The classification test of the fabric is carried out. The classification accuracy of the test set is 92.48%.. The model is used to segment 8 types of tissues in the total scan pathological image of the colorectal. First, the full scan image is preprocessed into 5000 pixels * 5000 pixel size image blocks, and then the 8 types of tissues in each of the images will be marked. Finally, the results will be obtained. The labeling results were spliced to obtain the results of 8 types of tissues of the entire colorectal scan. The accuracy of the 8 types of tissue segmentation was higher and the help of a certain auxiliary diagnosis.

【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;南方医科大学病理学系;
【基金】：国家自然科学基金(61771249) 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】：R735.34;TP391.41

【相似文献】