Research On Detection Method Of Cervical Lesionsbased On Deep Learning

Recent years,in order to provide doctors with effective auxiliary diagnosis information,computer-aided diagnosis(CAD)based on classical machine learning algorithm and deep learning image processing technology is becoming more and more important.The CAD system based on medical biological image plays an important role in reducing the burden of doctor's reading film,correctly identifying related disease images,improving the accuracy of disease diagnosis,reducing the rate of missed diagnosis and misdiagnosis.Research on the feature extraction method of biological image of pathological tissue,the shape,gray level and texture of image are the most commonly extracted features,on this basis,research on the specific features and quantification methods for different pathological tissue is the research hotspot of feature extraction.Classification research mainly focuses on the benign and malignant methods of pathological tissue,among which unsupervised clustering,support vector machine,decision tree and convolutional neural network(CNN)are the most commonly used classification algorithms.The latest development of machine learning algorithm in medical image processing,especially deep learning technology,is helpful to identify,classify and quantify medical images.Medical image processing based on machine learning and deep learning mainly includes image preprocessing,region of interest(ROI)segmentation,feature extraction and tumor region recognition segmentation.Its research focuses on medical image recognition segmentation,feature extraction and classification.At present,there are still many problems in colposcopy based on cervical biological imaging,which are mainly manifested in: 1)there are great differences in the diagnosis of different doctors;2)there is no exact standard for the sampling location and number of samples under colposcopy;3)the diagnosis of cervical adenocarcinoma in situ(AIS)is still a difficult problem;4)the efficiency of traditional colposcopy image diagnosis can not meet the needs of clinical growth.In view of the colposcopy image and pathological diagnosis data generated in the process of cervical cancer screening,this paper proposes a variety of image processing algorithms based on machine learning and deep learning,aiming to improve the efficiency of cervical cancer screening and the accuracy of doctors' diagnosis.The related research results are as follows:In this paper,the cervical region of interest based on K-means is segmented to realize the segmentation and extraction of the cervical region in the medical and anatomical sense.Firstly,the histogram threshold method is used to analyze the brightness histogram(y)of the colposcope image to realize the mirror reflection preprocessing operation caused by the light in the colposcope image;then the RGB color mode of the colposcope image after preprocessing is converted into HSV color space,and the V component is extracted by K-means algorithm;finally,the area filter is used to smooth the edge,so as to segment the cervix Region.110 cases of standard colposcopy images marked by experts were tested and verified by the proposed method.Objective indexes such as DIC,Ji,ACC,SPE and Sen were used to analyze the segmentation results.In this paper,we implement a cervical lesion region aided detection algorithm based on the convolution neural network,which uses the feature extraction network to extract the global depth feature of the whole colposcope image,then uses the region candidate network(RPN)to generate the region of interest(ROI)candidate frame,and finally uses the classification function to determine whether the region in the candidate frame is a disease By changing the region and regressing the candidate frame,we can accurately locate the cervical lesion region.In view of the problem that it is difficult to define the pathological and non pathological areas in the cervico colposcopy image due to the lack of similar features,the se block is used to enhance the important features and suppress the non main features,and further improve the feature extraction effect,so as to achieve the feature classification and candidate box regression ability of the regions of interest.The research shows that the target detection algorithm based on deep learning is superior to the classical machine learning model,and realizes the auxiliary location function of cervical lesions.