Abstract
The recently developed deep learning models can be employed to design computer aided diagnosis (CAD) models for diabetic retinopathy (DR). Though several DR classification approaches are available in the survey, but still there is need to improve the overall DR detection performance. With this motivation, this paper design a novel metaheuristic with deep learning enabled computer aided diagnosis model for DR (MDL-CADDR) detection and grading. The proposed MDL-CADDR technique involves pre-processing stage to boost the quality of fundus images. In addition, Archimedes Optimization Algorithm (AOA) with Kapur's Entropy (AOA-KE) based image segmentation technique is applied. Moreover, Chimp Optimization Algorithm with DenseNet (COA-DN) based Feature Extraction and Spiking Neural Network (SNN) based classification processes are performed to classify distinct stages of DR. The performance validation of the MDL-CADDR technique on benchmark MESSIDOR data set pointed out the supremacy of the MDL-CADDR technique with maximum accuracy of 99.73%.