Registration And Segmentation Of Fluorescent Cervical Images With Multiple Dark Areas

With the development of medical technology,a variety of methods have been used for screening cervical cancer.Colposcopy is essential for cervical inspection equipment in each hospital,automatic identification by using the computer equipment in the cervical colposcopy images collected the lesion area in order to realize the automation of cervical cancer screening,so as to reduce the burden of work and the doctor to reduce the leakage detection and error detection probability,has become the development direction of colposcopy.Single modal medical images usually cannot respond to lesion features because of the simple data they provide.Therefore,the automated technology of cervical cancer screening needs to detect the multiple dark areas in fluorescent images from the contrast of fluorescent cervix images and white light cervical images.In this paper,the automatic segmentation of the precancerous lesions of the cervix is realized by combining the registration and segmentation techniques.The image registration method in this paper is based on grayscale information.Combined with the spatial transformation method and genetic algorithm,the registration of fluorescence and white light images is realized.The similarity measure is used as the basis of genetic algorithm optimization.On the basis of the registration image,a segmentation method based on the gray ratio and the level set algorithm is proposed.Firstly,the cervix region was extracted from the white cervix image by using k-means clustering algorithm,and it enables each of the next steps to be implemented in the cervix region,rule out the interference caused by the complex background;Then the ratio of gray level to the corresponding position is obtained by using different pixels with different gray level in the same grayscale map,and the ratio image of the fluorescence and white light is obtained.Obtaining ratio image is for the difference between segmentation object and background pixel average value;finally,based on the ratio images,the extra dark area in the fluorescent cervix image was segmented by using the level set algorithm,this paper uses the C-V model level set algorithm,it is required to be obviously of the difference between segmentation object and background pixel average value,the ratio of the image just realize this principle.Some morphological processing methods have also been applied to subsequent segmentation image optimization,filled with holes,and got rid of some small areas and points.The segmentation results obtained by the presented method are compared with the FCM clustering algorithm and no registration.The results show that the average sensitivity of the presented method is improved by13.32% and the average specificity of the presented method is improved by 14.36%,comparing with no registration;the average sensitivity is improved by 15.31%,and the average specificity is improved by 14.40%,comparing with FCM clustering algorithm.When using the Jaccard Index(JI),this presented method is 10.08% and18.19% higher than the above two methods.The above experimental results show that the proposed algorithm is a better segmentation method for the detection of multiple dark areas in fluorescent cervical images.