Localization of Exudate for Diabetic Retinopathy using Convolutional Neural Networks

Author(s): Janhvi Chauhan, Dhaval Modi

Publication #: 2505040

Date of Publication: 15.12.2023

Country: Gujarat

Pages: 1-8

Published In: Volume 9 Issue 6 December-2023

Abstract

In order to track the progression of diabetic retinopathy (DR), exudate detection is a crucial task for computer-aided diagnosis of DR. This article uses a deep convolutional neural network (CNN) to identify exudates at the pixel level. The CNN model is saved as an offline classifier once it has been trained using expert-labelled exudate picture patches. Potential exudate candidate sites are first retrieved using the morphological ultimate opening approach in order to obtain pixel-level accuracy while cutting down on computing time. The trained CNN model is then used to classify and identify the local region (64 × 64) around the candidate points. The suggested CNN architecture achieves a pixel-wise accuracy of 89.50%, sensitivity of 87.00%, and specificity of 94.23% on the test database.

Keywords: Diabetic Retinopathy, Optic Disc, exudates, Retinal Vessels

Download/View Paper's PDF

Download/View Count: 78

Share this Article