The Development for Diabetic Retinopathy Lesion Detection on Fundus Photography

This research aims to advance the accuracy and precision of diabetic retinopathy lesion identification on retinas through the utilization of fundus photography, particularly from the publicly available Kaggle database, consisting of a total of 35,126 images. The image processing workflow involves four steps for enhancing the quality of low-quality retinal images. These steps include histogram matching for image sharpness adjustment, contrast enhancement for improving image clarity, noise removal using a median filtering operator, and optic disk localization for identifying and removing the optic disk boundaries. Data analysis is conducted using two techniques: K-means clustering algorithm and Mathematical morphology algorithm. The results demonstrate a sensitivity of 98.75%, specificity of 96.47%, and an overall accuracy of 92.45% for the proposed algorithm. Processing the 35,126 retinal images from the database takes an average time of 12.6 seconds per image. This study highlights the efficacy of employing K-means clustering and Mathematical morphology techniques in developing a robust tool for diabetic retinopathy lesion identification on retinas. The proposed methodology has the potential to serve as a valuable diagnostic tool, reducing errors and speeding up the diagnosis and treatment of ocular diseases.