Research On Cell Segmentation Method Of Lung Pathological Image

Cell segmentation is the basis of the entire pathological image-assisted diagnosis system,and it is also an indispensable part.Although traditional algorithms can achieve certain segmentation results,most of them rely on specific images,imaging methods,and artificially designed initialization processes.They cannot solve large-scale image segmentation problems and lack generalization.With the continuous development of deep learning technology,many network models have been applied in the field of medical image segmentation.However,the existing network model cannot solve the problem of overlapping cell segmentation.Therefore,in view of the complex background of lung cell pathological images and the data characteristics of overlapping and intertwined cells,combined with existing methods,a new structure is proposed for cell segmentation.First of all,for the problems of noise and unclear cell boundaries in the image,a method combining of bilateral filtering and Laplacian sharpening is used for preprocessing.While removing noise,it retains cell edge information as much as possible,and enlarges the target area and the contrast of the background.For the segmentation of cell regions,U-Net is used as the basic model to optimize the model based on the characteristics of the lung cell pathological image data.Since the background area of the lung pathological image is much larger than the target area,the attention mechanism is introduced to suppress the characteristic response of the irrelevant background area,and the better activation function and loss function are selected to improve the segmentation performance of the model.On the basis of model segmentation results,using area and roundness as screening conditions,a discriminant model is established to distinguish between single and overlapping cells to avoid over-segmentation of single cells in subsequent steps.There are different forms of overlapping cells.This article mainly deals with overlapping cells in tandem.Overlapping cells in series have the characteristics of curvature changes at intersections.Therefore,a comprehensive analysis of the current multiple key point detection algorithms,the use of the bottleneck detection algorithm closest to the cell overlap state to detect the separation point pairs,and the ellipse fitting algorithm for boundary correction to separate overlapping cells.In order to verify the above method,the threshold segmentation method,K-Means,and watershed algorithm are selected for experimental comparison.The results show that the algorithm in this paper can accurately segment the cell area,and has a good segmentation effect on overlapping cells,and realizes the repair of broken edges,verified the feasibility and effectiveness of the algorithm.