An Optimal Diabetic Retinopathy Detection and Classification Approach based on integrated Hybrid Convolutional Neural Networks (CNNs)

Abstract

Diabetic Retinopathy is a retina disease caused by diabetes mellitus and it is the leading cause of blindness globally. Early detection and treatment are necessary to delay or avoid vision deterioration and vision loss. This research aims to examine the significant importance of Deep Learning (DL) that shows substantial gain in artificial intelligence applications, particularly in the medical of things (MoTs) and guarantees impactful results in the diagnosis, classification, prediction and categorization of numerous stages of diabetes using applications based on machine Learning (ML). In this study, we have proposed an ML-based diabetic retinopathy mechanism to classify diabetic issues using a Multi-Layer Neural Network (MLNN). We have used the PIMA Diabetes Dataset (PDD) to test and train the proposed ML model. To increase the accuracy of the (PDD) dataset we looked into different activation capacities, learning algorithms, precision and strategies for dealing with lost information and compared the proposed DL model with conventional machine learning approaches, specifically Random Forest (RF) and Naive Bayes (NB), to evaluate the pattern for the results. Our MLNN-based ML model outperformed the other classifiers with a significant increase of 2.27% in classification precision accuracy.

Keywords: Artificial intelligence (AI), neural networks, Multilayer feed-forward neural Network(MLNN), Naive Bayes (NB), Random Forest (RF), pima diabetes dataset (PDD)