An Image Processing System using Deep Learning CNN Model for Early Identification of Diabetic Retinopathy: A Review

Abstract

Diabetic retinopathy (DR) is a disease which can be observed in patients who have diabetes. Due to this, causes slowly detriment to the retina and light-sensitive nerves i.e photoreceptors, used to identify light at back to eye. DR is not identified in its early stage as no direct symptoms found in early stage, but as the patient progresses, the patient may experience blurred vision or overall loss of vision. People with diabetes can be 25 times more blind than the average person. Identification at initial stage and treatment of diabetic retinopathy (DR) are extremely important in forestall to vision loss among individuals with diabetes. However, existing screening methods, which rely on manual evaluation of retinal images captured by non-mydriatic fundus cameras, are time-intensive and error-prone.  Our proposed methodology employs convolutional neural networks (CNNs) in juxtaposition with advanced preprocessing systems to mine life-threatening features from retinal images. To address challenges such as high dimensionality and computational complexity, the model integrates Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA) for feature optimization. The system classifies retinal images into five stages of DR—Normal, Mild Non-Proliferative, Moderate Non-Proliferative, Severe Non-Proliferative, and Proliferative—achieving robust multiclass classification. To overcome limitations of traditional deep learning models, such as vanishing gradients and prolonged training times, the CNN architecture leverages dense residual U-network that enable feature reuse and enhance model efficiency. By integrating low-level features from shallow layers with high-level dense features, the model significantly improves detection accuracy. This research may mark a significant advancement in diabetic retinopathy screening, addressing key challenges in clinical practice and contributing to improved patient outcomes.