Segmentation And Classification Of Cancer Cells Under The Action Of Drugs Based On Deep Learning

Most cancers are fatal diseases,and it’s the leading cause of death.The rapid growth and invasion of cancer cells is the main reason for the high mortality rate of cancer.Therefore,analyzing the invasiveness of cancer cells is very practical,which can be used as a tool to guide the research of anti-cancer drugs.At present,the in vitro cell culture technology which called organ chip has made great progress.Among them,the in vitro tumor model can simulate the growth state of the tumor in the complex environment of the human body,and the analysis of the relevant parameters of the in vitro tumor model can be used as an important indicator to judge the state of cancer cells under the action of drugs.With the development of image processing technology,image-based analysis of cancer cell status is a simple and fast way.In this paper,we combined deep learning methods and image analysis of cancer cell growth status.We focused on the theme of image-based analysis of cancer cell growth status,and and study algorithms for cancer cell growth analysis.The main research contents are as follows:(1)Before obtaining the outline of the cell aggregation area,we found that the cells may not form spheres during the culture process,or the cells are scattered due to death during the culture process.Only spherical cells can be used as a biological model for drug analysis.Therefore,first of all,we carried out the classification task of whether the cells can become aggregated spheres,which can distinguish the images that do not need to be analyzed before the contour acquisition,and reduce unnecessary workload.In this paper,a dual-flow classification network is used,and a classification accuracy rate of 95.15% is obtained on the test set.It can distinguish most images that cannot be aggregated spheres.(2)In this paper,we focus on the core problem of cancer cell contour acquisition in image-based cancer cell state analysis methods.In the traditional method,the contour extraction of the cell aggregation area usually requires the combination of image processing methods and manual correction.In this paper,for the task of clustering region segmentation in cell contour extraction,based on the deep learning method,we used an U-shaped network structure as the basic architecture,and proposed an end-to-end cell clustering region segmentation algorithm,which achieves an accuracy of more than 97% in objective evaluation indicators,Under the verification of a large number of test sets,the proposed method can accurately segment the cell aggregation area,and finely divide the invasion antenna of strong invasive cells to obtain contour information that fits the periphery of the aggregation area.(3)Finally,we proposed EPI and MSEI as indicators to judge the invasion characteristics and degree of invasion of different cells.These indicators can be used to judge the invasion characteristics and degree of invasion of different cells,and can replace the analysis method through experience.After a large number of experimental calculations,we verified the correctness of the indicators we proposed.