Prediction Of Diabetic Retinopathy Using Weighted Fusion Deep Learning Model

Abstract

Diabetes arises from consistently elevated blood glucose levels, which can lead to vascular complications and vision loss. Timely diagnosis signifies  a crucial role in minimizing risk of advanced disorder of blood vessels of retina  and associated severe visual impairment. Hence, the classification of DR stages holds significant importance. This proposed novel study introduces a weighted fusion deep learning network designed for exigently extracting essential features and characterize  DR  stages using retinal images. The suggested system intends to iidentify retinopathy symptoms present in these images. Fundus-related features are extracted by fine-tuning the Inception V4 and VGG-19 models. The outputs of these fine-tuned models are combined utilizing a weighted fusion methodology and the ultimate recognition outcome is calculated by using softmax classifier. The suggested network exhibits an elevated degree of accuracy for recognizing DR phases, based on experimental results. The suggested approach specifically obtains an accuracy score of 99.18% and sensitivity of 97.5% when assessed on the Messidor dataset. Our suggested novel weighted fusion deep learning model  network has equivalent performance when compared to other models, thus supporting its efficiency.