Study On Registration Between Fundus Photographic Image And Contrast Image And Measurement For Vascular Wall Thickness

The fundus is the only organ in human body where blood vessels can be seen.Through the analysis of fundus image,the likelihood of developing diseases such as arteriosclerosis,hypertension and diabetes will be detected and for example,the abnormal thickness of the fundus wall can be a sign of diabetic retinopathy.An important way to analyze fundus image is fundus image registration,which is one of medical image registrations.Medical image registration is of great value in clinical medicine,which can integrate different pieces of information reflected by different images and improve the diagnosis accuracy.The different options of registration methods for medical images of different organs result in different modalities images.The thesis is to analyze the current medical image registration methods at home and abroad and find and further improve a technique suitable for multimodal fundus image registration.Since the color image of fundus shows the outer diameter of the blood vessel while the contrast image only shows the inner diameter,the thickness of the blood vessel wall can be measured after registering both the color image and the contrast image for fundus.The thesis introduces some medical image registration methods,fundus image registration methods,fundus vascular segmentation methods and fundus vascular wall thickness measurement methods.For multimodal ocular fundus image registration and fundus vascular wall thickness measurement,the topics of this thesis are as followed:1)Preprocess the multimodal fundus image,including background mask extraction,grayscale processing(color channel extraction),blood vessel segmentation,image denoising,image smoothing,skeleton extraction,etc.The purpose of preprocessing is to accurately extract useful Information,accelerate subsequent experiments and improve the accuracy of the experiment.2)Register and integrate the pre-processed images,the focus of which is to extract the characteristic structure unique to the fundus image and to dopt the classical methods such as Hessian matrix and random sampling consistency algorithm to improve the matching efficiency of the pair.Two kinds of matching methods with advantages and disadvantages are proposed,and block registration is performed on the results of preliminary registration,which further improves the registration accuracy.3)Position the image after registration,align the image and determine the thickness of the blood vessel wall based on the optic disc.In this step,through blood vessel segmentation,the blood vessel wall of fundus is subtracted from the blood vessel of fundus image as the blood vessel wall.The algorithm proposed mainly serves clinical medical image experiments.The thesis analyzes multiple sets of data and compares the methods of others.By comparing experiments,it proves that the proposed algorithm can improve the accuracy of multimodal fundus images and the ratio of fundus wall thickness to optic disc diameter is also obtained.