Application Of Deep Learning In The Description Of Liquid-based Cervical Cell Specimen

Cervical cancer occupies the first place in the incidence of female reproductive system tumors.With the promotion of liquid-based cervical cytology screening,the incidence and mortality of cervical cancer have been declining year by year.However,the domestic machine-aided diagnosis system of machine-based cervical cytology a bit less.In recent years,deep learning has produced new breakthroughs in various fields and has generated positive responses in various industries.The ability of a convolutional neural network to classify an image even exceeds the human classification ability in certain situations.Based on the two reasons,this paper hopes to use the powerful image classification ability of deep learning to assist in the intelligent screening of liquid-based cervical cytology aided diagnosis system.The classification target is the description of the specimen.The importance of the description of the liquid-based cervical cell specimen is not inferior.For the importance of the diagnosis of the specimen,the description of the specimen is directly related to the reliability of the diagnosis.(1).This paper uses CIFAR-10 dataset to evaluate the classification ability of convolutional neural networks on images,understand the basic structure of convolutional neural networks and the transfer learning method of convolutional neural networks.(2).Before using the convolutional neural network to classify the liquid-based cervical cell image for classification,a large number of data sets need to be collected.Therefore,this paper defines six important types of specimens in the first to be described in liquid-based cervical cytology: normal specimens,small number of cells,bacterial vaginosis,inflammatory specimens(divided into moderate inflammation and severe inflammation),and atrophic changes.The labeling of liquid-based cervical cells previously collected by an automated microscope platform was also performed.(3).Based on the previous convolutional neural network architecture,this paper designs a convolutional neural network that describes the classification of specimens.It trains the specimen description data sets and then tests and evaluates the trained networks.The Alex network structure was used to describe the classification of the specimens by means of transfer learning.The network for the completion of the transfer learning was tested and evaluated.Finally,we compared the differences between the two training methods studied.The experimental results show that the convolutional neural network can classify the sample description well and achieve the expected classification effect.