An Intelligent Moth Flame Optimization with Inception Network for Diabetic Retinopathy Detection and Grading

Diabetic Retinopathy (DR) is a widespread illness among diabetic patients that creates lesions on the retina affecting vision. When the DR is not identified in the earlier stage, the vision can be severely affected. Since the manual detection of DR using fundus images by ophthalmologists is time consuming, expensive, and not accurate, deep learning (DL) models have become a popular tool to detect and classify DR at the beginning stages. With this motivation, this paper designs an intelligent moth flame optimization with Inception network-based DR detection and grading (IMFO-INDR) model. The goal of the IMFO-INDR model is to detect the existence of lesions in the fundus image and assign proper class labels to it. The IMFO-INDR model involves histogram-based segmentation to determine the affected lesion areas in the fundus images. In addition, Inception v4 model is applied as feature extraction and the hyperparameters involved in it are optimally tuned by the use of MFO algorithm. At last, softmax classifier is used for the allocation of class labels to the input fundus images based on the extracted feature vectors. The experimental validation of the IMFO-INDR model takes place using the MESSIDOR database and the results are examined in terms of several aspects. The resultant values demonstrate the promising performance of the IMFO-INDR model over the existing DR diagnosis models.