Detection Of Diabetic Retinopathy Using A Multi-decision Inception-resnet Blended Hybrid Model

Abstract

Diabetic Retinopathy (DR) Represents A Critical Complication Of Diabetes Mellitus, Leading To Progressive Vision Impairment And Potential Blindness If Left Undetected. This Research Presents A Novel Multi-decision Inception-ResNet Blended Hybrid Model For Automated DR Detection And Classification. The Proposed Architecture Integrates 172 Weighted Layers, Strategically Divided Into Dual-image Processing Pathways: 86 Layers Dedicated To Color Fundus Image Analysis And 86 Layers For Grayscale Image Processing. By Employing A Multi-layered Transfer Learning Approach With Adaptive Moment Estimation (Adam) And Stochastic Gradient Descent (SGD) Optimization Techniques, The Model Achieves Comprehensive Feature Extraction Across Both Sequential And Non-sequential Image Data. The Architecture Incorporates Eight Convolutional Layers At Each Processing Stage, Enabling The Extraction Of Both Global And Specialized Features Through Chi-square Testing Mechanisms. Evaluated On The EyePACS And APTOS Datasets, The Model Demonstrates Superior Performance With A Detection Accuracy Of 98.1%, Outperforming Existing State-of-the-art Approaches. The Multi-decision Framework Effectively Classifies DR Into Five Severity Stages: No DR, Mild DR, Moderate DR, Severe DR, And Proliferative DR, Providing A Robust Solution For Early-stage Diabetic Retinopathy Detection In Clinical Settings.

Keywords