A Novel approach to multi-level image classification of Retinal Diseases

Abstract

Diseases of the retina, like diabetic retinopathy, age-related macular degeneration, and glaucoma, are major causes of vision loss and blindness around the world. Early identification and labelling of these illnesses are necessary to act quickly and effectively treat them. Although conventional image classification techniques are useful, they may locate minute features in retinal images poorly, which results in the incorrect diagnosis. This work proposes a novel multi-level image classification system for eye disorders' detection and categorisation. It makes the diagnosis more accurate by use of sophisticated deep learning approaches. Shallow and deep feature extraction layers of a multi-level convolutional neural network (CNN) structure define our desired approach. This combined design allows one to present complex objects at upper levels and capture minute details at lower ones. Along with comments outlining the nature and stage of every illness, the model is taught on a vast library of retinal photographs including both healthy and diseased images. The model is made more steady using several pre-processing techniques like image normalisation, augmentation, and noise reduction. We evaluate our proposed approach against present deep learning models and standard machine learning techniques. to approach performs much better, according to data, in raising classification accuracy, precision, recall, and F1-score. When it comes to distinguish between many phases of eye illnesses, our multi-level approach outperforms conventional models. This makes it a more exact and comprehensive testing instrument. We also discuss whether the model can be used in real-life healthcare environments and in other datasets. Though it has certain issues—an uneven collecting, difficult-to-understand models, and the need for a lot of computational power—the proposed technique shows potential.