Research On The Method Of Cervical Cell Image Recognition

Cervical cancer is a grave threat to the women’s health and lives.If we can receive effective treatment in the early stage of cervical cancer disease,it will have a very high survival rate.Therefore,regular cervical cancer screening among adult women is of great significance for the prevention and treatment of cervical cancer.The main method of current early screening is the pathological examination of cervical cytology.In the screening process,it mainly depends on the pathologist observing under the microscope to find out whether there are abnormal cells in the cell smear image.On the one hand,this screening method has a huge workload and is time-consuming and labor-intensive.On the other hand,it requires pathologists to have rich experience and high technical level.Therefore,it is of great practical significance to realize cervical cell image recognition based on image processing technology.The object of this study is the cervical cell image obtained based on cervical cell smear technique.The research mainly includes cervical cell segmentation and cervical cell classification.The accurate segmentation of cytoplasm and nucleus in the process of cervical cell recognition can provide basis for diagnosis and treatment,while cell classification can directly determine the existence of abnormal cells.Specific studies include:(1)A set of cell segmentation algorithm based on watershed algorithm is designed for cervical cell images with overlapping regions.In the segmentation of nucleus,a series of morphological processing including triangle threshold segmentation and ellipse fitting algorithm are used to obtain the segmentation results of nucleus.In the segmentation of overlapping cytoplasm,the marking function based on Euclidean distance transformation and geodesic distance transformation is redesigned,and the possibility of over-segmentation and under-segmentation in overlapping cervical cell segmentation is reduced by H-minima transformation,and the of segmentation overlapping cytoplasm is realized.(2)Use deep learning algorithm to segment a single cervical cell image,based on the classical medical image segmentation network U-Net network model,the convolution residual module is introduced,and the pre-trained ResNet34 network is used as the coding structure.The image feature extraction ability of U-shaped network structure is improved.At the same time,in order to further improve the segmentation effect on the edge of cell image,a post-processing method based on fully connection condition random field model is introduced to further improve the accuracy of cell segmentation.(3)In the multi-classification task of cervical cell image,the performance of classical classification network in cervical cell image classification task is compared.ResNet50 is taken as an example to verify the effectiveness of transfer learning in cervical cell classification tasks.At the same time,data enhancement,test enhancement,label smoothing and voting fusion are used to improve the accuracy of cervical cell image classification.