OPTIMIZATION RETINA BLOOD VESSEL SEGMENTATIONTO SUPPORT MEASUREMENT OF ARTERIOVENOSERATIOIN HYPERTENSIVE RETINOPATHY DIAGNOSIS

• Main Author: MAURISNA ASHFAHANI, FAATIHAH
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/69554
• Institution: Institut Teknologi Bandung

Hypertensive retinopathy is a disease caused by high blood pressure in the retinal blood vessels. Microvascular changes such as arteriolar narrowing are an indication of a patient suffering from hypertensive retinopathy and can be used for early detection by measuring the arteriovenous ratio (AVR), which is the ratio between the diameters of arteries and veins. The process of segmentation of blood vessels is needed to identify the structure of blood vessels in the retina so that the diameter of the vessels can be calculated. Poor image contrast, different branching patterns, and other clinical signs such as red lesions and exudates are obstacles to the retinal blood vessel segmentation process. In a previous studybyKhotimah (2021), a blood vessel segmentation method has been developed to support the diagnosis of hypertensive retinopathy, but there are still blood vessels that have not been completely detected which are characterized by a fairly low sensitivity value. Therefore, this study proposes a more optimal blood vessel segmentation method by going through several processes including the pre-processing stage, the blood vessel segmentation stage, and the post-processing stage. The process of segmenting blood vessels is carried out with a more optimal method by adding candidate features using a 2D Gabor Wavelet filter, using the NN classifier, and eliminating the rest of the false positive candidates. The purpose of adding the Gabor filter is because it has good orientation direction so that can provide a clearer direction of blood vessels. The segmentation results provide an increase in the sensitivity value from 51.50%to 55.42%ontheAVRDB dataset without significantly reducing the precision value. Performance calculations are also carried out in the area of 1.5-3 from optic disc radius and provide an increase in the sensitivity and precision values to 62.46%and 82.25%on the AVRDB dataset. This study also includes the measurement of the AVRvalue of the results of blood vessel segmentation. AVR measurements were also carried out using the Knudtson method giving an average result of 0.71 with a Pearson and Spearman correlation of 0.629 and 0.547 for ground truth segmentation. These results are better than the results obtained fromKhotimah'sresearch (2021).