3D Registration Of Human Retina Image Data Based On Iterative Closest Point Approach

In recent years,with the development of science and technology,computer aided diagnosis technology has become a hot research topic in the field of clinical medicine.Computer aided diagnosis technology,which makes full use of the computer of high performance computing and 3D modeling capabilities,can help clinicians to detect and diagnose diseases and improve their efficiency by visualizing and analyzing of medical images.Meanwhile,with the development of the computer graphics and image technology,modern medical imaging technology is in continuous progress and high-end medical image equipment come after another,which also provides a good foundation to the wide application of computer aided diagnosis technology.Optical Coherence Tomography is a new kind of Optical Tomography technology,Optical Coherence Tomography fundus images that acquired by Optical Coherence Tomography equipment can show information of retina and retinal fundus image structure.Medical research shows that the changes of retinal nerve fiber layer thickness can be an important index of the neurodegenerative eye disease such as macular degeneration,glaucoma.The ability of Optical Coherence Tomography equipment that obtains retinal nerve fiber layer thickness can help clinicians make early decisions about glaucoma disease.However,there are some defects of Optical Coherence Tomography equipment.First of all,in order to avoid the influence of OCT image accuracy caused by involuntary eye movement,the most of current Optical Coherence Tomography equipments persist for a short period of time during one scan,thus,the obtained volume data is relative small,which could not provide enough information to clinical physicians.Secondly,the unprocessed Optical Coherence Tomography images always have some noises,and these noises will bring certain influence to the diagnosis and analysis of the clinical physicians.Based on the above problems,this paper introduces a registration method based on iterative closest point algorithm which could provide an Optical Coherence Tomography volume that covers large field of view.The algorithm mainly includes three steps which are the extraction of point cloud data of fundus images,the initial registration step and accurate registration step.Our algorithm first adopts the Canny edge detection method for fundus image de noising and edge feature extraction.Then gather these characteristics as the point cloud data format.And then,we assigned every point to a space grid based on the minimum bounding box algorithm,and then we adapt a boundary of strategy of point cloud data to obtained boundary feature points.Finally,we use Singular Value Decomposition algorithm to complete the initial registration.In the fine registration stage,we improve the traditional iterative closest point algorithm according to the characteristics of fundus image point cloud data.Weighting method is used in this paper,every points is weighted and those lower weighted points are excluded,thereby decrease the number of points in fine registration step,and then we introduce the M-estimated to the target equation and redefine method to calculate the distance of closest point.The experimental results show that this algorithm obviously improves the precision of the conventional data registration of Optical Coherence Tomography data and reduce the time complexity of the algorithm as well.The main part of this article first give a brief review of the medical image processing technology,the development history of classic methods of medical image registration technique.And then,the whole process of this algorithm which deals with the registration of fundus Optical Coherence Tomography images is introduced.At the end of this article,fundus images and open source point cloud data are used to conduct experiment.According to the results of the experiments,we analyzed the influence pros and cons of the traditional algorithm and our algorithm on registration of point cloud data.Finally,the future work of our algorithm is prospected.