Rectal Fundus Image Recognition Using Deep Learning Ensemble CNN Models

Abstract

As a vital tool for ophthalmologists to detect potential blinding problems, retinal fundus image analysis is an essential part of medical image analysis. This work looks at retinal fundus images for the purpose of detecting diabetic retinopathy. After every model has been trained independently, the probabilities for every class are totalled to get the ultimate value. The class with the greatest value is called the decision class. Because of this, the proposed methodology consists of the following steps: gathering retinal fundus images; pre-processing images (resizing, contrast enhancement, shade correction, normalization, and image augmentation); feature extraction and classification using a group of specially trained convolutional neural networks; and assessment and evolution of the ideal weights for a more precise diagnosis of diabetic retinopathy in fundus images.