Master's Thesis Project

In the human eyes, retinal diseases are correlated with the deformity present in the retina. These retinal diseases if not treated on time will lead to vision loss and even worse to permanent blindness. Premature infants may lose their eyesight due to abnormal development of retinal blood vessels. In clinical terms this disease is known as retinopathy of prematurity (ROP) that is one of the leading causes of irreversible blindness despite improvements in prevention over years. Although in the past there were some solutions presented for early diagnosis of retinal diseases, most of the methods have been tested on publicly available datasets which have not found prolific in the diagnosis of ROP disease. Therefore, this study focuses on major problems in ROP diagnosis that is the segmentation of retinal blood vessels in ROP fundus images followed by the length, width, and tortuosity measurement of the segmented blood vessels. The analytical results can assist ophthalmologists to early diagnose ROP along with the approximate measurements of the vessels and reduce the chance of blindness in infants. Firstly, for blood vessel segmentation in ROP images an image processing model was developed that consisted of noise reduction and contrast enhancement steps to overcome the low contrast, noise, and illumination problem. Secondly, we performed a delineation of morphological attributes of retinal blood vessels such as length, width, and tortuosity. For this centerline of vessels, branching point identification, endpoint identification, and automatic labelling of vessel segments were done using morphological operations and lookup tables. For vessel length measurement region property perimeter and geodesic distance transformation methods were compared. Finally, distance transformation was used for width measurement. Experiment results show that the proposed method outperformed the existing methods for blood vessel segmentation in ROP images. Also, the method was effective in the measurement of morphological attributes of retinal vessels.