Smartphone-based diabetic retinopathy severity classification using convolution neural networks

Abstract

With diabetes growing at an alarming rate, changes in the retina causes a condition called diabetic retinopathy which eventually leads to blindness. Early detection of diabetic retinopathy is the best way to provide good timely treatment and thus prevent blindness. Many developed countries have put forward well-structured screening programs which screens every person diagnosed with diabetes at regular intervals. However, the cost of running these programs is increasing with ever increasing disease burden. These screening programs require well trained opticians or ophthalmologist and the global shortage of health care professionals is putting a pressing need to develop screening tools with POCT (Point-Of-Care-Technology). Using smartphone-based screening tools will help process and generate a plan for the patients thus skipping the health care provider needed to just classify the disease. In this paper, we trained and validated 4 different classifiers using VGG16, Resnet50, InceptionV3 and DenseNet121 algorithms on the Retinal fundus Kaggle dataset to segment the parts of the retina. We experimented with different image preprocessing techniques and employed various hyperparameter tuning to build a good model. We achieved the best model with DenseNet161 with a kappa score of 0.9025, sensitivity 90% and specificity 87% validated on the same dataset and compared the results. We used this model in an android application to predict the severity of retinal fundus images which can later be tested in clinical environments.