基于卷积神经网络的病理细胞核分割
|
摘要
针对病理图像中细胞核的精准分割问题,结合全卷积网络框架和高分辨率网络框架的特点,提出一种卷积网络对细胞核进行自动精准地分割;基于稀疏非负矩阵分解的方法将具有严重颜色分布差异的病理图像进行颜色分布归一化,以归一化后的图像为输入,利用所提出的卷积网络对细胞核进行分割;该网络通过减少下采样算子的使用,使图像信息在前向计算过程中不会过分丢失,并使用扩张卷积扩大深层神经元的局部感受野尺度大小;所采用的分割方案在2017年MICCAI病理数字图像分割数据集中达到0. 848的平均dice分数;实验表明,融合全卷积网络框架和高分辨率网络框架的卷积网络在病理图像中实现了细胞核自动精准的分割,可以有效减轻影像医师的工作负担。
Aiming at the accurate segmentation of nuclei in pathological images,a neural network structure which combines the characteristics of fully convolutional network framework and high-resolution network framework is proposed to automatically and accurately segment nuclei. Due to the difference of color distribution in pathological images caused in the staining process,a method based on sparse non-negative matrix decomposition is used to normalize the color distribution of all pathological images. Then the proposed convolutional neural network is used to segment the nuclei accurately with the normalized images as input. By reducing the use of down-sampling operators,the network can make the image information not lose excessively in the forward process,and can expand the size of the local receptive field of neurons in deep layers by using the dilated convolution operators. We test our method on the 2017 MICCAI pathological digital image segmentation data set,and get the average dice score of 0.848. Experiments show that the convolutional neural network fusing fully convolutional network framework and high-resolution network framework can automatically and accurately segment nuclei in pathological images,which can effectively reduce the workload of imaging physicians.
Aiming at the accurate segmentation of nuclei in pathological images,a neural network structure which combines the characteristics of fully convolutional network framework and high-resolution network framework is proposed to automatically and accurately segment nuclei. Due to the difference of color distribution in pathological images caused in the staining process,a method based on sparse non-negative matrix decomposition is used to normalize the color distribution of all pathological images. Then the proposed convolutional neural network is used to segment the nuclei accurately with the normalized images as input. By reducing the use of down-sampling operators,the network can make the image information not lose excessively in the forward process,and can expand the size of the local receptive field of neurons in deep layers by using the dilated convolution operators. We test our method on the 2017 MICCAI pathological digital image segmentation data set,and get the average dice score of 0.848. Experiments show that the convolutional neural network fusing fully convolutional network framework and high-resolution network framework can automatically and accurately segment nuclei in pathological images,which can effectively reduce the workload of imaging physicians.
引文
[1]FORAN D J,YANG L,CHEN W,et al.Imageminer:ASoftware System for Comparative Analysis of Tissue Microarrays Using Content-based Image Retrieval,Highperformance Computing,and Grid Technology[J].Journal of the American Medical Informatics Association,2011,18(4):403-415
[2]QI X,XING F,FORAN D J,et al.Robust Segmentation of Overlapping Cells in Histopathology Specimens Using Parallel Seed Detection and Repu lsive Level Set[J].IEEE Transactions on Biomedical Engineering,2012,59(3):754-765
[3]LU Z,CARNEIRO G,BRADLEY A P.An Improved Joint Optimization of Multiple Level Set Functions for the Segmentation of Overlapping Cervical Cells[J].IEEETransactions on Image Processing,2015,24(4):1261-1272
[4]KONG J,ZHANG P,LIANG Y,et al.Robust Cell Segmentationfor Histological Images of Glioblastoma[C]//Chongqing:Biomedical Imaging(ISBI),2016IEEE 13th International Symposium on.IEEE,2016:1041-1045
[5]SIRINUKUNWATTANA K,RAZA S E A,TSANG Y W,et al.Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images[J].IEEE Transactions on Medical Imaging,2016,35(5):1196-1206
[6]XING F,XIE Y,YANG L.An Automatic LearningBased Framework for Robust Nucleus Segmentation[J].IEEE Transactions on Medical Imaging,2016,35(2):550-566
[7]LONG J,SHELHAMER E,DARREL T.Fully Convolutional Networks for Semantic Segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2015:3431-3440
[8]LI W,WANG G,FIDON L,et al.On the Compactness,Efficiency,and Representation of 3D Convolutional Networks:Brain Parcellation as a Pretext Task[C]//New York:International Conference on Information Processing in Medical Imaging.Springer,Cham,2017:348-360
[9]VAHADANE A,PENG T,SETHI A,et al.Structurepreserving Color Normalization and Sparse Stain Separation for Histological Images[J].IEEE Transactions on Medical Imaging,2016,35(8):1962-1971
[10]LECUN Y,BENGIO Y,HINTON G.Deep Learning.[J].Nature,2015,521(7553):436-444
[2]QI X,XING F,FORAN D J,et al.Robust Segmentation of Overlapping Cells in Histopathology Specimens Using Parallel Seed Detection and Repu lsive Level Set[J].IEEE Transactions on Biomedical Engineering,2012,59(3):754-765
[3]LU Z,CARNEIRO G,BRADLEY A P.An Improved Joint Optimization of Multiple Level Set Functions for the Segmentation of Overlapping Cervical Cells[J].IEEETransactions on Image Processing,2015,24(4):1261-1272
[4]KONG J,ZHANG P,LIANG Y,et al.Robust Cell Segmentationfor Histological Images of Glioblastoma[C]//Chongqing:Biomedical Imaging(ISBI),2016IEEE 13th International Symposium on.IEEE,2016:1041-1045
[5]SIRINUKUNWATTANA K,RAZA S E A,TSANG Y W,et al.Locality Sensitive Deep Learning for Detection and Classification of Nuclei in Routine Colon Cancer Histology Images[J].IEEE Transactions on Medical Imaging,2016,35(5):1196-1206
[6]XING F,XIE Y,YANG L.An Automatic LearningBased Framework for Robust Nucleus Segmentation[J].IEEE Transactions on Medical Imaging,2016,35(2):550-566
[7]LONG J,SHELHAMER E,DARREL T.Fully Convolutional Networks for Semantic Segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2015:3431-3440
[8]LI W,WANG G,FIDON L,et al.On the Compactness,Efficiency,and Representation of 3D Convolutional Networks:Brain Parcellation as a Pretext Task[C]//New York:International Conference on Information Processing in Medical Imaging.Springer,Cham,2017:348-360
[9]VAHADANE A,PENG T,SETHI A,et al.Structurepreserving Color Normalization and Sparse Stain Separation for Histological Images[J].IEEE Transactions on Medical Imaging,2016,35(8):1962-1971
[10]LECUN Y,BENGIO Y,HINTON G.Deep Learning.[J].Nature,2015,521(7553):436-444