THE AUTOMATIC ARTERIOVENOUS RATIO CALCULATION IN RETINAL IMAGE AS AN INDICATOR OF HYPERTENSIVE RETINOPATHY

• 主要作者: Khotimah, Yunianti
• 格式: Final Project
• 語言: Indonesia
• 在線閱讀: https://digilib.itb.ac.id/gdl/view/58296
• 機構: Institut Teknologi Bandung

Hypertensive retinopathy is a disease caused by the presence of high blood pressure in the retinal blood vessels. At a very severe level, this disease may cause blindness in the eye, therefore early detection of hypertensive retinopthy is very important to prevent the severe eye damage. Manual detection can be performed by an ophthalmologist by analyzing the patient’s retinal funduscopy image. Unfortunately, this manual detection system is ineffective due to the limited number of ophthalmologists in Indonesia and the length of time required for diagnosis. Advanced development in computer technology have made automatic detection for this disease possible to be implemented through digital image processing and machine learning. One of the characteristics that indicate hypertensive retinopathy in its early stage is narrowing in retinal. This narrowing can be evaluate quantitatively through the calculation of arteriovenous ratio (AVR) in fundus image, which is the ratio between artery and vein diameter that is usually done manually by an ophthalmologist. In this study, an automatic system to calculate arteriovenous ratio is being proposed which includes the preprocessing process, blood vessel segmentation, diameter calculation, artery and vena classification, and lastly AVR calculation. The developed segmentation algorithm is able to overcome false segmentation due to the presence of exudate and red lessions, thereby increasing the precision from 65,65% to 82,19%. This study also successfully identify 18 most importat features that become the main key to distinguish artery and vein with classification accuracy 97,3% (INSPIREAVR) and 98,5% (AVRDB) using SVM model with RBF kernel. The automatic AVR measurement was developed based on Knudtson method and its correlation with manual measurements (???? = 0,649) is comparable to the correlation between manual measurements by two observers (???? = 0,645). Thus, the automatic system of AVR calculation developed in this study is considered to be an alternative solution to replace the AVR calculation that was done manually by an opthalmologist.