基于卷积神经网络的细胞图像分割与类型判别

组织细胞图像形态各异、大小不一、纹理变化多样等特点,导致难以精准地分割细胞区域的问题,对此提出了一种基于卷积神经网络(CNN)和边缘聚类方法的新算法.对原始切片采用染色校正预处理,提高色彩对比度,利用CNN得到初步分割结果,结合边缘聚类方法提升初步分割结果的连续性和完整性.在此基础上,结合计算机视觉技术,获得分割图像中细胞颗粒的基本属性特征,并使用Softmax分类器判别细胞类型.实验结果表明:相较于经典的卷积神经网络、阈值分割、模糊聚类等细胞图像分割算法,该算法在分割结果的完整度方面提升了6.15个百分点.

Segmentation of cell images and the discrimination of cell type based on convolutional neural network

The characteristics of tissue cells in different shapes,sizes and varied textures make it difficult to segment cell areas precisely.We propose a new algorithm for the segmentation of cell images combined with convolutional neural networks(CNN) and edge clustering method.The original slices were stained with pretreatment to improve the color contrast.The convolutional neural network was used to obtain the initial segmentation results.Edge clustering method was used to improve the continuity and integrity of the initial segmentation results.On this basis,we combine computer vision techniques to obtain the basic attribute features of cell particles in the segmented images.And Softmax classifier was used to determine the cell type.The results of experiments show that compared with classic convolutional neural networks,threshold segmentation,fuzzy clustering and other cell image segmentation algorithms,the cell segmentation method proposed in this paper improves the integrity of the segmentation result by 6.15%.