Research On Segmentation Algorithm Of Pleural Effusion Cell Cluster

Lung cancer is one of the most common malignant tumors in the world.At present,the "gold standard" of a lung cancer diagnosis is that experts rely on pathological sections for tumor cell examination.This work is complicated and time-consuming and is closely related to the doctor's experience and subjective factors.With the rapid development of machine learning,profound learning,computer-aided diagnosis technology has dramatically improved pathologists' diagnosis efficiency.In diagnosing lung cancer,the tumor cell cluster's appearance in pleural effusion is an important marker of lung cancer metastasis to other organs or from other organs to lung.Identifying segment cell clusters more accurately from the pleural effusion section is an essential basis for tumor cell detection.There are many kinds of cells in pleural effusion,such as heterogeneous staining,conglutination,etc.These characteristics make the segmentation of pleural effusion cell cluster become a very challenging topic.In this paper,the detection and segmentation of pleural effusion cell clusters are realized by stages.The main research work includes:(1)set up a variety of cell sources,collect the pathological image of the pleural effusion cell cluster,and establish the data set of cell cluster segmentation and cell nuclear segmentation.(2)to extract suspicious cell regions from tumor cell clusters accurately and quickly,a three-stage fusion segmentation algorithm(CMF)based on fuzzy clustering is proposed.This method combines the fluorescence staining information of cells to get suspicious cell regions.Then,the region of interest is mapped to the original UN dyed image to generate the uncolored image's coarse segmentation results.On this basis,the improved fuzzy clustering algorithm is used to obtain the accurate edge of the cell cluster,and the texture and morphology of the cell cluster in the image are preserved.(3)to solve the problem of inhomogeneity and nuclear touch in the segmentation of pleural effusion cells,a nucleus segmentation algorithm based on the attention mechanism(current model)was proposed.The model adopts the encoder and decoder structure and uses an attention module to enhance the learning of non-significant dispersion features.Moreover,the skip link in u-net is redesigned to supplement the information lost in the feature fusion process,which realizes the pleural effusion cell cluster nucleus' s accurate segmentation.