Deep Learning Model for the Detection and Classification of Diabetic Retinopathy
Samra Nawazish, Department of Software Engineering, University of Engineering and Technology, Taxila, Pakistan.
Ali Javed, Department of Software Engineering, University of Engineering and Technology, Taxila, Pakistan.
Tahira Nazir, Department of Software Engineering, University of Engineering and Technology, Taxila, Pakistan.
Corresponding Author:
Ali Javed (ali.javed@uettaxila.edu.pk)
Abstract:
Diabetic Retinopathy (DR) is an ocular disorder found in people suffering from Diabetes Mellitus. The disorder can cause partial or sometimes complete blindness. It is essential to detect DR early and accurately to lessen the likelihood of progression to proliferative retinopathy. Due to the wide range of DR severity levels, it is difficult to effectively identify and classify all DR types. Designing an automated system that has the potential to accurately detect and classify DR severity levels is necessary to meet the challenges posed by manual systems. Existing automated DR detection systems have certain limitations i.e., evaluation on a limited dataset, and high false alarms on images having variations in color, size, and several distortions. To cope with these challenges, we have proposed an enhanced Resnet50 model with a custom classification block that can detect and classify all types of DR accurately. We performed data cleaning and augmentation in the preprocessing stage. Then we fed those images to our end-to-end enhanced ResNet50 deep learning model with customized network layers structure for the detection and classification of DR into different severity levels. We confirmed the robustness of our work by assessing it on two DR datasets i.e., the DR benchmark dataset and APTOS-2019. The presented model attained better results as compared to the state-of-the-art methods.
Keywords:
Diabetic Retinopathy; Image Classification, Resnet50; Deep Learning; Fundus Images