Research On Cervical Cancer Cells Recognition Based On Deep Learning

Cervical cancer is a kind of malignant tumor which threats women’s health seriously.Early screening of cervical cancer cells is helpful for prevention and treatment to patients timely.It is a hot spot in current research that computer technology recognizes cervical cancer cell images and assists doctors to complete diagnosis.However,the recognition of cervical cancer cells faces many problems,which affect the accuracy of recognition,such as overlapping cells,background containing impurities or cells unrelated to cervical cancer cells.The research on the subject about the recognition of cervical cancer cells is based on deep learning in order to deal with the problems above,.The main work is as follows:(1)An algorithm of segmenting cervical cancer cells based on optimized circular convolution was proposed.This algorithm can solve the problems of inaccurate bounding box and large amount of calculation about pixel segmentation.On the basis of the bounding box of object detection,a hybrid attention improved circular convolution to extract the features of the points on the cell octagon contour,so as to improve the accuracy of network segmentation.A hybrid attention is introduced to make the network pay more attention to the overlapping area of cells and the points far away from the boundary of cells,so as to improve the accuracy of cervical cancer cell segmentation.Finally,the data set SIPa KMe D is used to verify how the algorithm performed.The results of experiment prove that the accuracy of this segmentation algorithm is better than other algorithms.Compared with the segmentation results of professional doctors,the segmentation effect is better.(2)An optimized CapsNet for cervical cancer cell classification was proposed.VGG-16 network optimizes the convolution layer of the CapsNet to improve the feature extraction ability of the CapsNet,extracting the multi-scale features of cervical cancer cell image in this algorithm.There are redundant image features in the main capsule and VGG-16.The self-attention updating dynamic routing algorithm is used between the main capsule and the data capsule to focus on the significant features that are useful for classification,which makes classification of cervical cancer cells more accurate.Finally,the experimental results about classification algorithm validated on the SIPa KMe D after segmentation show that the accuracy of the method is higher than other algorithms.