Research On Leukocyte Image Segmentation Method Based On Graph Neural Network

White blood cells are a key element for maintaining human immune function.In clinical diagnosis,observing the number and morphology of different types of white blood cells is an important means of diagnosing hematopoietic diseases.Therefore,the detection of white blood cells is very important in the field of clinical medicine,and white blood cell image detection includes Based on the steps of image acquisition,cell segmentation,feature extraction,and classification and recognition,white blood cell segmentation is a challenging task in the field of medical image processing.It faces the following two problems.One is that there are many types of white blood cells and their morphology is complex.Variety of changes,the traditional algorithm segmentation effect is not good,the second is that the segmentation algorithm of supervised learning often requires a large number of training samples manually labeled by professionals,which costs a lot of labor.In order to solve the above problems,this article focuses on the research and discussion of white blood cell segmentation.It mainly includes the research of white blood cell segmentation algorithm based on superpixel and graph convolutional network,the research of white blood cell segmentation algorithm based on convolutional neural network and dynamic hypergraph neural network,and the research of white blood cell segmentation algorithm based on dynamic attention map convolutional network.The main contents are as follows:(1)In order to further improve the accuracy of white blood cell image segmentation under the background of training a small amount of label information,a white blood cell segmentation algorithm based on graph convolutional network is proposed.Using super-pixel algorithm as the preprocessing method of image segmentation,the structure of the graph is constructed to solve the problem that the graph convolutional network cannot directly process the image data.When constructing the undirected feature map,we convert the color feature of the superpixel sub-graph into the feature matrix of the graph node.In order to use the spatial neighborhood relationship,the threshold judgment method is used to construct the adjacency matrix,and the obtained graph data is trained The graph convolutional network is trained for node classification,and then the node category label is linked to the superpixel subgraph to achieve the purpose of image segmentation.Through experiments on multiple types of white blood cells,the segmentation accuracy is improved to a certain extent compared with the comparison algorithm.(2)In view of the fact that the graph convolutional network does not fully mine the depth information in the construction of the feature matrix and its own poor performance due to the simple static graph structure,a white blood cell segmentation method based on the convolutional network and the dynamic hypergraph neural network is proposed,and the hypergraph is used.Replace simple graphs to better represent the one-to-many relationship between nodes.The proposed dynamic hypergraph reconstruction model reconstructs the graph structure in hierarchical convolution.In the construction of the hypergraph,the trained convolutional network is used to extract the high-order information of the node to construct the feature matrix,and the clustering algorithm is used to fully combine the adjacency relationship of the node in the local and global to construct the adjacency matrix.Use the convolution module to get the node classification prediction of the constructed hypergraph,and then turn it into the image segmentation result.(3)The graph attention network uses an attention mechanism to adaptively assign weights to different neighboring nodes to solve the problem of fixed weights assigned to neighboring nodes in the hierarchical convolution of the graph convolution network to improve performance,but in deep propagation due to noise Information and reduce accuracy.Combined with the graph attention system,we propose a white blood cell processing algorithm based on the dynamic attention graph convolutional network.The attention layer is designed before the graph convolution layer,the attention weight is imported to adjust the adjacency matrix,and the graph convolution is guided to focus on key nodes and convolve in the hierarchy.JJZ dynamically adjusts the chart structure.Experimental results show that this network has better segmentation performance than the first two networks,and has room for further improvement.