Classification And Recognition Of Exfoliated Cells Based On Deep Learning

Because of its high incidence and severity in women,cervical cancer has attracted more and more attention to this malignant tumor.If early cervical cancer can be detected in time,it can be better controlled and treated.Therefore,the classification and identification of exfoliated cells plays a vital role in the diagnosis of early cervical cancer.Artificial classification and recognition is a common method of classification and recognition of exfoliated cells.The epithelial cells,lymphocytes and central granulocytes in exfoliated cells can be distinguished by artificial observation.The morphological observation of epithelial cells is an important basis for detecting cancerous cells.However,the accuracy and reliability of traditional manual observation classification will be accompanied by the influence of doctors' subjective emotions,and the classification of exfoliated cells is also a difficult problem that can not be ignored in terms of workload and cost.It is easy to cause doctors' visual fatigue and affect the effect of classification and recognition.In recent years,machine learning SVM has been widely used in the field of medical image classification and recognition.However,on the one hand,SVM-based classification and recognition of exfoliated cells has higher requirements for classification images,on the other hand,it needs complex preprocessing of classification images,and the final classification and recognition effect is deficient in efficiency and accuracy.In this paper,Faster R-CNN technology is used to realize the classification and recognition of exfoliated cells based on Tensorflow's in-depth learning framework.The system is operated in Windows environment.The main implementation is divided into three steps.Firstly,the gray images of standard cervical stained exfoliated cells were collected by photoelectric microscope and CCD camera as data sets,and Papanicolaou staining method was used here.Secondly,the standard cervical staining exfoliated cell data set was labeled to store cell types and location information.Finally,the training set is trained with VGG-16 network model by using Faster R-CNN method modified Anchor and RPN in Tensorflow deep learning framework,and the network model generated by the training is used to detect the target of cervical dyeing exfoliated cells classification and recognition.The classification and recognition of exfoliated cells' stained nuclei is the basis for further work to locate and identify aneuploidy cells more accurately from the perspective of DNA ploidy analysis and provide the basis for assistant diagnosis of tumors.The AP values of standard epithelial cells,lymphocytes,granulocytes and adherent epithelial cells on actual sample datasets using this method are 0.6055,0.5688,0.3678 and 0.2994 respectively.The experimental data show that the proposed method has better classification and recognition effect for small targets such as exfoliated cells in the image and has higher efficiency.