Research Article
Automatic Detection and stage classification of Diabetic Retinopathy Using Convolutional neural network with densenet120
@INPROCEEDINGS{10.4108/eai.7-12-2021.2314696, author={E. Balaji}, title={Automatic Detection and stage classification of Diabetic Retinopathy Using Convolutional neural network with densenet120}, proceedings={Proceedings of the First International Conference on Combinatorial and Optimization, ICCAP 2021, December 7-8 2021, Chennai, India}, publisher={EAI}, proceedings_a={ICCAP}, year={2021}, month={12}, keywords={fundus diabetic retina convolutional neural network densenet 120 resnet 50 confusion matrix}, doi={10.4108/eai.7-12-2021.2314696} }- E. Balaji
Year: 2021
Automatic Detection and stage classification of Diabetic Retinopathy Using Convolutional neural network with densenet120
ICCAP
EAI
DOI: 10.4108/eai.7-12-2021.2314696
Abstract
Diabetic retinopathy (DR) is a leading eye disease which damages the blood vessels in the retina. Initially DR is asymptomatic and eventually ends up with severe or complete vision loss. The main reason of the DR is diabetic mellitus (DM) which is a type of diabetic where the pancreases fail to produce enough amount of insulin in the blood to maintain the blood sugar and glucose. Due to lack of insulin production the sugar and glucose levels are unmanageable. The person with DM is always at a higher rate of acquiring DR. the prevalence DM diabetes affects the retinal blood vessels which lead to complete blindness. According to the recent statistics of The International Diabetes Federation (IDF) around 463 million people are affected with diabetes mellitus worldwide [1]. The early detection of Diabetic Retinopathy which can postpone the progression of the blindness hence in this research work we put forwards the convolutional neural network (CNN) deep learning model to identify the different stages of diabetic retinopathy using fundus images from the kaggle database. The proposed CNN model trained with two different networks resnet 50 and densenet120 respectively. The proposed CNN model outperforms compare with the previous work reported on the deep learning methodology and yields maximum accuracy of 82.58% in multiclass classification.
