Pathological Image Cell Nucleus Segmentation Based On Deep Learning

The cell nuclei segmentation in pathological images is of great significance for cancer research,In clinical diagnosis,the stage and type of cancer are usually judged based on the spatial geometric distribution and morphological characteristics of cells.These usually rely on doctors to manually extract cell nuclei,which often requires doctors to have a wealth of Experience,and time-consuming and labor-intensive.With the continuous development of computer-aided diagnosis,deep learning technology can allow the computer to automatically extract the area that doctors need to observe from medical images and generate auxiliary diagnosis suggestions,which effectively improves the efficiency and objectivity of clinical diagnosis.This paper is based on the deep learning convolutional neural network,based on three models respectively,to construct a deep convolutional neural network according to the characteristics of the nuclear medical image and the defects of each method.Aiming at the characteristics of the nucleus data set,the nucleus has a wide range of distribution,small size,and high density.This paper uses Segnet as the basic model.By introducing an improved module of non-local mean,the model can better learn the context information.Coupled into subject features and edge features,and supplemented with corresponding supervision respectively to improve the model’s ability to distinguish between cell nuclei and subject boundaries.Aiming at the Segnet basic model does not make full use of low-level feature,this article builds a neural network on the basis of the Unet++model.First,this article replaces the convolution module of the encoder part with the residual module to improve the ability of model feature extraction,in order to avoid the irreversible loss of information due to the downsampling operation,this paper connects a highresolution branch in parallel with the original model,and combines multiple features at the same depth through the feature fusion module proposed in this paper.Scale information,and finally combine the two parallel branches through the channel and spatial attention mechanism module to improve the model’s ability to segment small objects such as nuclei.Aiming at the first two methods that require special setting of hyperparameters for different data sets to achieve the optimal effect of the model,this paper uses the nnUnet method to alleviate the differences between the two data sets of the model.This method can be automatically performed according to the characteristics of the data set.Network configuration,thus eliminating the need for artificial network configuration links,this paper uses more effective data enhancement based on it,and post-processing after the inference result is obtained to make the segmentation result graph smoother.