Title : Integrating Manta Ray Foraging Optimization (MRFO) algorithm and EYENET deep learning weights for intelligent ocular surface analysis
Abstract:
The accurate classification of diabetic types from ocular surface images remains a major challenge due to subtle feature variations and overlapping visual patterns among different diabetic categories. To address this issue, this study introduces a robust deep learning framework that integrates Manta-Ray Foraging Optimization (MRFO) with Cyclone Aging (CA) and Chain Foraging (CF) strategies to enhance the InceptionV3 architecture. The proposed model employs a novel dataset, Di-EYENET, which provides a comprehensive and medically validated collection of ocular surface images categorized as Type-1, Type-2, and non-diabetic. By coupling MRFO with hierarchical feature learning, the framework optimizes hyperparameter configuration and fine-tunes InceptionV3 using newly developed EYENET weights, specifically designed for diabetic eye imaging. Experimental results demonstrate that the MRFO-InceptionV3 model with EYENET weights surpasses conventional ImageNet-based models, achieving a 2% improvement in classification accuracy while maintaining computational efficiency. This research underscores the potential of nature-inspired optimization algorithms and domain-specific datasets in advancing AI-based medical imaging, offering a reliable and adaptive approach for the early and precise diagnosis of diabetic eye disease.
