Title: A hybrid convolutional neural network model for detection of diabetic retinopathy

Authors: Musa Alshawabkeh; Mohammad Hashem Ryalat; Osama M. Dorgham; Khalid Alkharabsheh; Mohammad Hjouj Btoush; Mamoun Alazab

Addresses: Computer Science Department, Al-Balqa Applied University, Salt, Jordan ' Computer Science Department, Al-Balqa Applied University, Salt, Jordan ' Computer Science Department, Al-Balqa Applied University, Salt, Jordan; School of Information Technology, Skyline University College, University City of Sharjah, P.O. Box 1797, Sharjah, UAE ' Software Engineering Department, Al-Balqa Applied University, Salt, Jordan ' Computer Science Department, Al-Balqa Applied University, Salt, Jordan ' College of Engineering, IT and Environment, Charles Darwin University, Darwin, Northern Territory, Australia

Abstract: Diabetic retinopathy causes vision loss. Regular eye screening has to be done to provide the appropriate treatment and for vision loss prevention. Globally, patients with DR are increasing, which leads to work pressure on specialists and equipment. Fundus images are a key factor in effective retinal diagnosis. In this paper, a deep-learning approach is proposed to detect DR from retinal images. The proposed approach involves a combination of four effective techniques: image augmentation, contrast limited adaptive histogram equalisation, CNN and transfer learning and ensemble classification. The results show the proposed approach obtained high values of accuracy (93%), precision (95%) and recall (96%), and more stability compared with other approaches.

Keywords: deep learning; diabetic retinopathy; eye diseases; retinal diagnosis; retinal images; convolutional neural networks; medical applications; ensemble classification.

DOI: 10.1504/IJCAT.2022.130886

International Journal of Computer Applications in Technology, 2022 Vol.70 No.3/4, pp.179 - 196

Received: 09 Dec 2021
Received in revised form: 08 Apr 2022
Accepted: 15 Apr 2022
Published online: 13 May 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article