The Pathological Analysis Of Cervical Cells Based On Deep Learning

Various types of computer-aided diagnosis and analysis techniques have gradually entered the clinical application of medical science,on account of the insufficiency of traditional manual interpretation methods for pathological images,especially computer-aided cytological pathological analysis.The classifier in the computer-assisted analysis system of cervical cells is usually composed of three parts: target extraction,feature extraction and classification.The performance of the classifier is still insufficient,which is mainly affected and limited by the effectiveness of the artificial selected features.The convolutional neural networks that develop well recently,can automatically extract effective features,which can make up for the shortcomings of the manual selection methods mentioned above.In addition,the current computer-assisted analysis system cannot provide quantitative analysis and qualitative analysis results for a sample at the same time.In view of this situation,aiming at dual-images of pathological slides of cervical cells with Feulgen-Eosin complex staining,this paper applies methods based on deep learning to cervical cells pathology analysis.In order to advance the performance of cervical cell image classification,this paper adopts a ResNet classification network with image pairs as input.The classification network achieved a precision of more than 92% on the test set,and its recall rate was 30% higher than the SVM classifier,reaching 93%.In addition,the ResNet classification network performed equally well in comparison tests of clinical data.In order to avoid the disadvantage of the traditional way,which obtains too much irrelevant input images,this paper uses a candidate region selection method based on the SegNet segmentation network.This method can effectively detect almost all cervical nucleus in the pathological images.And,the total candidate images are greatly reduced compared to the traditional method,at least up to 50%.This paper combines SegNet segmentation network and ResNet classification network,then proposes a cervical cell detection method with deep segmentation network locating the cervical nucleus.In the test set,the detection method proposed in this paper has achieved high accuracy,and the recall rate is 32% higher than that based on SVM.The detection method in this paper can provide a reliable basis for subsequent quantitative analysis and has practical application value.