PROLIFERATIVE DIABETIC RETINOPATHY CLASSIFICATION FROM RETINAL FUNDUS IMAGES USING FRACTAL ANALYSIS

• Main Author: Naufal Taris, Gusna
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50851
• Institution: Institut Teknologi Bandung

Diabetic retinopathy is a form of diabetes mellitus complication that causes damage to the retinal tissue in the eye. This disease is one of the causes of blindness in the world. Proliferative diabetic retinopathy (PDR) is the most dangerous grade of diabetic retinopathy. The formation of neovascularization is a unique PDR symptom. Which if detected, the patient must meet an ophthalmologist within one week. If the symptoms not handled properly, this will cause vision loss due to blood leak. In clinical practice, the classification method still carried out by the ophthalmologist directly. The number of ophthalmologists compared to the Indoneisia’s population is 1: 170,000, this number is far from the WHO standard which is 1: 20,000. During its development, there have been many studies that capable to automatically classifying PDR. In general, the PDR detection method has two different approaches, the retinal vascular structure approach and the retinal tissue texture characterization approach. In this study, a blood vessel characterization approach was used to detect neovascularization on images. This approach is applied using fractal analysis. In this study, the wavelet transform segmentation method with 2D-Gabor wavelet was used to provide optimal fractal feature values to classify PDR. This study also used the maximum red lesions probability feature to detect PDR symptoms other than neovascularization. The most impactful feature is the shanon entropy from fractal analysis that named H(r) in combination with maximum red lesion probability that gave results AUC = 0.9335±0.03, Se. 93.38%, Sp. 81.17%. This method also provides test results that PDR classification will be stable as the image resolution decreases and PDR classification will deteriorate with poor image quality.