Diabetic Retinopathy Severity Detection Using InceptionV3 and ResNet50 Architectures

Diabetic Retinopathy (DR), a condition of eyerelated disorder. It leads to blindness and severe vessels, which consist of the macula as a centralpart and are responsible for the sharpness. In this study, we proposed the model of Inception V3 which can capture the multiscale features extraction from the retinal fundus images, and ResNet50, which provides the residual connection to enhance the capability of learning the complex features. The InceptionV3 is applicable to detect small lesions like microaneurysms and large abnormalities. The ResNet50 works based on residual learning, which helps train deeper networks by allowing gradients to flow across skip connections, thereby avoiding vanishing gradient problems. Both architectures help to improve robustness by enhancing the classification of DR levels like Proliferative Diabetic Retinopathy, Severe Diabetic Retinopathy, Moderate Diabetic Retinopathy, Mild Diabetic Retinopathy, and No Diabetic Retinopathy. Our model's performance is calculated to predict the key metrics like accuracy, precision, recall, and F1 score. InceptionV3 and ResNet50 are two models that achieve 95% and 94% respectively <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">[1] [2].</sup>Act as further resources to aid ophthalmologists in diagnosing and improving DR early.