The Research On A Non-rigid Based Mammographic Image Registration Method

Mammography has been the most reliable and effective screening tool for the early detection of breast cancer.Registration of mammographic images plays an important role in computer-aided detection(CAD)systems for detecting breast cancer in mammograms.In order to obtain sufficient deep learning training data sets,it is usually necessary to establish a large database containing thousands of mammographic images,and then analyze local image blocks of different women under the same framework.Therefore,we need to register the mammograms of different individuals,and align the points or regions with the same anatomical structure between different individuals on the image,which is convenient for the abnormal detection and classification of the later deep learning model.Image registration is one of the most important tasks when integrating and analyzing information from various sources.This paper first introduces the research background and research significance of the subject,and then introduces the current research status from two aspects:feature registration based technology and intensity registration based technology.After that,the system introduced the relevant theoretical basis of medical image registration.The mathematical expressions of medical image registration are given,and the image registration classification standard is given.The basic framework,process and basic steps of medical image registration are generally introduced.Finally,four important modules and registration quality evaluation criteria for medical image registration are introduced in detail.This paper mainly studies the registration method of mammogram images of different individuals,and proposes a non-rigid registration method to analyze the mammogram images of different women under the same framework.This makes it possible to model normality or abnormality based on different individuals.However,the traditional free-form model based registration algorithm cannot register mammographic images of different individuals accurately,and the registration efficiency is low.In this paper,a non-rigid image registration method based on an integration of global coarse registration and local fine registration is presented,the mean squared gray-level difference is used as the similarity measure,the Limited-memory Broyden Fletcher Goldfarb Shanno(L-BFGS)optimization algorithm is employed to optimize the registration parameters,and ten pairs of mammographic images of different cases selected from the Mammographic Image Analysis Society(MIAS)database are used for experiments.Experimental results show that,compared with the traditional free-form model based algorithm and the Diffeomorphic Demons registration algorithm,it is indicated that the proposed method improves the registration accuracy and efficiency of multi-individual mammographic images.