Deep neural network for diabetic retinopathy disease severity classification

Date

2017

Authors

Philip, Mitha Ann

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Deep neural network has made way to smarter technologies for fields like health care, transportation and manufacturing. Applying a deep neural network in retinal imaging would help in the early detection of diseases that progressively leads to vision loss. Retinal surface of the eye is known to possess information to evaluate / monitor human health. Diabetic Retinopathy is a diabetic complication that affects the eyes which has the potential to cause vision impairment leading to blindness. The presence of the disease will lead to development of abnormalities like exudates, hemorrhages, micro-aneurysms in the retina. Deep neural network is an artificial neural network that has the ability to learn to recognize complex patterns/features and extract knowledge from data. It facilitates deeper understanding, classification on various extracted features. Applying this technique to retinal fundus images, facilitates automation of disease diagnosis / monitoring of disease progression. Treatment for diabetic complication at an early stage has significant chances of reversing the vision impairment. The proposed approach uses Inception, a deep and wide convolutional neural network for efficient and accurate classification of retinal information. This 5 class disease classification approach classifies the images as: 0 indicating absence of diabetic retinopathy (healthy/normal images), 1 indicating mild, 2 indicating moderate,3 indicating severe and 4 indicating proliferative presence of the diabetic retinopathy. Inception network layers uses concatenated layers of convolutions, pooling and dropout to extract the retinal feature information and a fully connected softmax layer for disease classification. The approach was successfully trained and validated on a balanced Kaggle retinal fundus image dataset and has achieved a sensitivity of 94.6% and precision of 93.7%.

Description

This item is available only to currently enrolled UTSA students, faculty or staff. To download, navigate to Log In in the top right-hand corner of this screen, then select Log in with my UTSA ID.

Keywords

Deep Learning, Machine Learning, Retinal Imaging

Citation

Department

Electrical and Computer Engineering