|Doctor of philosophy - Electrical engineering - XX ciclo - Andrea Anzalone|
TITLE OF THE THESIS:
Multiscale analysis for optimized vessel segmentation of fundus retina images
Automated segmentation of the vascolature in retinal images is important in the detection of a number of eye diseases. Some diseases, e.g retinopathy of prematurity, affect the morphology of the vessel tree itself. In other cases, e.g. pathologies like microaneurysms, the performance of automatic detection methods may be improved if regions containing vascolature can be excluded from the analysis. Another important application of automatic retinal vessel segmentation is in the registration of retinal images of the same patient taken at different times. Therefore the automatic vessel segmentation forms an essential component of any automated eye-disease screening system.
In this thesis an algorithm for the segmentation of the vessels in the images of the fundus of the human retina is developed.
In the first chapter we introduce some notations about the eye, the imaging technology and the archives of images. In the second chapter we show the state of art of the techniques proposed in the literature about vessel extraction. Since retinal vessels have a range of different sizes, it is a natural choice the use of an algorithm based on the multiscale analysis, so in the third chapter we deal in detail with the multiscale paradigm, and we discuss a mathematical framework to face this kind of problems using a differential and variational approach.
In the fourth chapter we talk about the algorithm developed to achieve the segmentation of the retinal vessels. The algorithm is modular and is made up of two fundamental blocks. The former is devoted to vessel enhancement, using a linear multiscale analysis for ridge detection, the latter provides a binary image by resorting to both a thresholding procedure and cleaning operations. The optimal values of two algorithm parameters are found out by maximizing proper measures of performances able to evaluate from a quantitative point of view the results provided by the proposed algorithm. The choice of the measure of performance allows one to tailor the solution to the specific image features to be emphasized. Some simulation results are presented and the performances of the algorithm are compared with those of other methods proposed in the literature.
In the fifth chapter we show the result improvements obtained using a nonlinear multiscale analysis (Total Variation Motion) instead of a linear technique.
Andrea Anzalone was born in Genoa, Italy, in 1979. In 2004 he received his "Laurea" (M.Sc.) five-year degree (Summa Cum Laude) in Electronic Engineering and the Ph.D. degree in Electrical Engineering in April 2008, both from the University of Genoa, Italy. From 2004 to 2008, he joined the Non-linear Circuits And Systems Group (NCAS) at the Department of Biophysical and Electronic Engineering (DIBE) of the University of Genoa. His main research interests are in the area of image processing, with emphasis on: use of parallel architectures to solve variational problems in early vision applications, use of archives of images in problems of classification or segmentation, biomedical applications.