Computer-assisted colour fundus image analysis

Abstract

Research in automatic retina image analysis has become increasingly important in ophthalmology. The retina is the only location where blood vessels can be directly visualized non-invasively in vivo. Hence, it serves as a ‘window’ to some of the pathologies like glaucoma, cerebrovascular and cardiovascular complications. Any change in the optic disc structure, blood vessel width, blood vessel tortuosity, and presence of lesion serves as indications of the pathologies. Realizing the importance of retina image analysis in pathology detection, there has been increasing research in ophthalmology to find statistical proof of correlation between certain retinal features with certain types of pathologies. Automatic retina image analysis can play an important part in the processing of large number of data required in correlational studies. It also provides a more qualitative and standardized assessment of retinal images which is harder to achieve in direct ophthalmoscopy. In this thesis we present two automatic retinal image analysis algorithms; fovea detection and artery-vein classification. The fovea, which lies within the macula region, is responsible for high resolution vision. Hence, presence of lesions or changes in the morphology of this region may be signs of pathologies such as age-related macular disease, which can cause blindness. Automatic detection of fovea allows automatic analysis of this region for the identification of macular related diseases. Our detection framework models the fovea region as an avascular region coupled with prior anatomical information on the fovea position in the retina. The artery-vein classification algorithm aims to classify vessels for the purpose of automating the estimation of artery-vein ratio (AVR). Numerous researches have shown that AVR is a well-established indicator of cardiovascular diseases. Automated estimation of AVR is an important step to process large number of images, for diagnosis as well as for correlational studies.

Date of Award	2011
Original language	English
Supervisor	Manuel Trucco (Supervisor)