Detection of Diabetic Retinopathy Using a Multi-Decision Inception-ResNet-Blended Hybrid Model

Abstract

Diabetic retinopathy (DR) is a severe complication of diabetes that affects the retinal structures and can lead to significant visual impairment or even blindness. Early diagnosis is crucial for reducing and preventing the progression of this condition. However, detecting DR’s early stages remains challenging due to subtle symptoms that are difficult to recognize independently. Our proposed model leverages 172 weighted layers to analyze both sequential and non-sequential fundus images for effective DR detection. By incorporating a multi-layered transfer learning approach, 86 layers are used for processing color fundus images, while the remaining 86 layers focus on grayscale images. The model undergoes thorough pre-processing and testing phases, utilizing eight layers of convolutions at each stage to handle various data matrices and integrate global and specialized features. The chi-square testing mechanism refines the evaluation of test cases, contributing to the model’s overall performance. Using multi-decision hybrid techniques, the model achieves a detection accuracy of 98.1%, outperforming other existing models.

Keywords

• Adam optimization
• clustered class
• convolutional neural network
• deep learning
• dual image
• diabetic retinopathy