Research On Automatic Segmentation Algorithm Of T Cell Image Based On Deep Learning

T cells play an important immunoregulatory function in the human body.At present,many treatment methods require T cell culture,research and analysis.With the development of computer science,image processing is widely used in the medical field.Cell image segmentation is an important part of cell image research,and its accurate results can lay a good foundation for subsequent research and recognition.However,due to the complicated T cell microscopic images,there are a lot of noise and aperture artifacts,and many cells are stuck together,the traditional segmentation method cannot accurately segment the target from the complex cell image,while the existing segmentation method based on neural network and deep learning is hard to effectively segment the cell from the adhesion area.In response to this problem,this article conducts related research to more accurately and effectively segment the adherent cells and realize the automatic segmentation of T cell microscopic images.After denoising and uniform illumination of the image,for the segmentation of T image,this paper proposes a method based on the combination of attention and contour area.The decoding network in U-Net is divided into two parts,the contours and regions of the cells are respectively divided,and a fusion block is added between the two sub-networks to share the characteristic information in the network.At the same time,the attention mechanism is introduced in the jump connection part of the network,which makes the network more inclined to segment the target area.In order to solve the problem of uneven positive and negative samples,focal loss is used as the loss function of the contour segmentation network.Experiments show that compared with the commonly used medical image segmentation methods,this method has better segmentation capabilities for T cell images.Since there will still be some adhesion cells in the segmented image,after analyzing its characteristics,this paper designs a segmentation method based on the main pit and skeleton.This method takes the contour and the output image of the regional network as the object,and further segmentation of the remaining adhesion cells in the image.Three different criteria are mainly used to detect and extract the concave points in the image,and then match and segment according to the distance and relative position of the concave point and the skeleton point.Experiments show that this method can further segment the adhesion area more accurately.Finally,a software for T cell image segmentation is developed,and the functions of automatic segmentation of T cell images,extraction and visualization of cell features are realized on the basis of combining the algorithms in this paper.