Title: Tensor flow-based recurrent neural network algorithm to diagnose diabetic retinopathy

Authors: T. Jemima Jebaseeli; C. Anand Deva Durai

Addresses: Department of Computer Science and Engineering, Karunya University, Coimbatore 641114, Tamilnadu, India ' Department of Computer Science and Engineering, King Khalid University, Abha 61421, Saudi Arabia

Abstract: Type 2 diabetic patients have the chance of a sight-threatening diabetic retinopathy disease. It affects the retina of the eye and causes damages to the human vision. The pathological fundus images of the patients have lesions present in the retina in the form of exudate, microaneurysm, cotton wool spots, and haemorrhages. At the later stage, it leads to retinal detachment from the eye. Segmentation technique is used to identify the lesions present in the retina for diagnosis. It makes the job of an ophthalmologist easy and to predict the disease accurately. The early detection of disease can be treated and the patients can be saved from the vision loss. The fundus images contain illumination; hence contrast limited adaptive histogram equalisation (CLAHE) is used for image enhancement. Recurrent neural network (RNN) is applied to segment the lesions from the pathological fundus images. The performance of the proposed approach achieved an average value of 98.91% sensitivity, 99.93% specificity, and 99.89% accuracy. The proposed technique is implemented using tensor flow framework for lesions segmentation and to diagnose diabetic retinopathy and validated over 109 pathological fundus images.

Keywords: lesion; exudate; microaneurysm; haemorrhage; cotton wool spots; feature detection; segmentation.

DOI: 10.1504/IJCAET.2021.118470

International Journal of Computer Aided Engineering and Technology, 2021 Vol.15 No.4, pp.516 - 528

Received: 17 Aug 2018
Accepted: 15 Feb 2019

Published online: 09 Sep 2021 *

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