Deep learning model for early detection of diabetic retinopathy from retinal images in type 2 diabetes

Abstract

Type 2 diabetes is a chronic condition characterized by elevated blood sugar levels, leading to various complications, including diabetic retinopathy—a major cause of vision loss if left untreated. Traditional methods of detecting diabetic retinopathy involve clinical examination techniques and specialized imaging modalities, such as dilated eye exams, visual acuity tests, fundus photography, fluorescein angiography, and optical coherence tomography (OCT). While effective, these methods can be invasive, time-consuming, and require specialized equipment and expertise. This study aims to explore a non-invasive and potentially more accessible approach to detecting diabetic retinopathy by leveraging machine learning techniques. The study proposes utilizing retinal images captured by smartphone-based retinal imaging devices and applying a machine learning model for detecting diabetic retinopathy. The proposed model leverages the feature extraction and classification capabilities of a pre-trained EfficientNetB0 convolutional neural network (CNN). By analyzing the features extracted from these images, the model aims to offer a more convenient and scalable solution for early detection and monitoring of diabetic retinopathy. this research has demonstrated the feasibility of using machine learning algorithms for diabetic retinopathy detection and providing actionable insights for individuals living with type 2 diabetes. These will facilitate early intervention strategies, such as lifestyle modifications or timely medical interventions, to mitigate the risk of diabetic retinopathy progression and associated vision loss. The developed deep learning model, based on the EfficientNetB0 architecture, exhibits promising results in diabetic retinopathy detection. It accurately classifies retinal images from smartphone-based devices with an impressive 88.57% accuracy and 90.31% precision. The model's reliance on convolutional neural networks and the efficient EfficientNetB0 design allow for sensitive and specific identification of diabetic retinopathy across various severity levels. This advancement has the potential to significantly enhance early detection and management of the condition, leading to improved patient outcomes and better-informed clinical decision-making in the field of diabetic eye care.

Key words: Type 2 diabetes, High sugar levels, Deep learning, CNN, EfficientNetB0, Retinal Images, Forecasting