The Research Of Cervical Cell Segmentation Method Based On Deep Convolution Neural Network

Cervical cytology smear is a widely used method for early screening of cervical cancer.Accurate segmentation of cervical cells is an important step for extracting the supporting conditions of characteristic parameters of cells and qualitative analysis of cells.In this paper,the application background of computer-assisted cervical cytology screening diagnosis and the accurate segmentation of cervical nucleus and cytoplasm in cervical smear images were studied.In the process of cervical cytological smear image examination,there are many influencing factors such as uneven dye distribution,bacterial infection,impurities or light changes and overlapping cells,which pose challenges to the automatic segmentation of cervical cells.This method mainly includes two aspects of cell individual level localization and pixel level segmentation in cervical smear image.Among them,CRPN(Cellular Region Proposal network)model was used for cell localization at the individual level to detect,locate and border regression cervical cells in smear images under the microscope,and free single cells and partially overlapped cells were effectively separated by border frame.In this paper,through the improvement and optimization of CRPN model,the recall rate and accuracy of CRPN model for cervical cells were effectively improved,and the interference of background noise and impurities in smear images were effectively overcome and excluded,which provided favorable conditions for the pixel segmentation of cervical cells.The pixel-level segmentation model for cervical cells consists of two units: extraction of down-sampling features and mapping of up-sampling features.This article will GAN(Generative adversarial network)and cell fusion segmentation model,through the cell division generator and the segmentation results of discrimination against training and learning,and effectively improve the accuracy of the segmentation of target areas.In this paper,a large number of experimental studies were conducted on the Region proposal-segment model and the current typical cervical cell segmentation methods.The experimental data showed that the algorithm model proposed in this paper could effectively improve the accuracy of segmentation of cytoplasm and nucleus in cervical cytology smear image,and provide a strong support for promoting the automatic interpretation of cervical cytology smear image and improving the reliability of diagnosis.