GWO-Optimized Stacking Deep Learning Framework for Diabetic Retinopathy Detection Using Enhanced Retinal Imaging.

Saved in:

Bibliographic Details
Title:	GWO-Optimized Stacking Deep Learning Framework for Diabetic Retinopathy Detection Using Enhanced Retinal Imaging.
Authors:	Nibedita, Arpita¹ anibeidta@gmail.com, Sahu, Prabhat Kumar² prabhatsahu@soa.ac.in, Patnaik, Srikanta³ srikantapatnaik@iimt.ac.in
Source:	IAENG International Journal of Computer Science. Jul2026, Vol. 53 Issue 7, p2506-2525. 20p.
Subjects:	Diabetic retinopathy, Grey Wolf Optimizer algorithm, Computer-aided diagnosis, Computer-assisted image analysis (Medicine), Retinal imaging, Image enhancement (Imaging systems), Ensemble learning, Deep learning
Abstract:	Diabetic Retinopathy (DR) is a vision-threatening complication of diabetes that damages retinal blood vessels, leading to irreversible blindness if untreated. This study presents a metaheuristic-optimized deep learning system for automated DR classification, achieving unprecedented diagnostic accuracy through Grey Wolf Optimization (GWO) of a novel stacking ensemble framework. Our methodology integrates hybrid preprocessing (CLAHE with Histogram Equalization and Gaussian-Wiener denoising) with strategic data augmentation to enhance APTOS dataset retinal images. Five architectures: Custom CNN, MobileNetV3, EfficientNetV2B1, VGG16, and ResNet50V2 were optimized using GWO to simultaneously tune preprocessing parameters, model hyperparameters, and ensemble weights. The GWO-optimized ensemble achieves state-of-the-art performance with 97.85% accuracy, 0.9982 ROC-AUC, and balanced 0.98 F1-scores for both DR-positive and negative cases, representing a 1.38% accuracy improvement over the baseline ensemble. Key advancements include 22.1% loss reduction, 98.2% early detection recall, and 98.2% referral precision while maintaining real-time deployment capability (3.5s inference per 100 images). Ablation studies confirm the critical role of our preprocessing pipeline (17.84% accuracy gain) and demonstrate that GWO optimization unlocks significant performance improvements without architectural modifications. This research establishes a new benchmark for AI-assisted DR screening by harmonizing diagnostic excellence with clinical practicality. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Diabetic Retinopathy (DR) is a vision-threatening complication of diabetes that damages retinal blood vessels, leading to irreversible blindness if untreated. This study presents a metaheuristic-optimized deep learning system for automated DR classification, achieving unprecedented diagnostic accuracy through Grey Wolf Optimization (GWO) of a novel stacking ensemble framework. Our methodology integrates hybrid preprocessing (CLAHE with Histogram Equalization and Gaussian-Wiener denoising) with strategic data augmentation to enhance APTOS dataset retinal images. Five architectures: Custom CNN, MobileNetV3, EfficientNetV2B1, VGG16, and ResNet50V2 were optimized using GWO to simultaneously tune preprocessing parameters, model hyperparameters, and ensemble weights. The GWO-optimized ensemble achieves state-of-the-art performance with 97.85% accuracy, 0.9982 ROC-AUC, and balanced 0.98 F1-scores for both DR-positive and negative cases, representing a 1.38% accuracy improvement over the baseline ensemble. Key advancements include 22.1% loss reduction, 98.2% early detection recall, and 98.2% referral precision while maintaining real-time deployment capability (3.5s inference per 100 images). Ablation studies confirm the critical role of our preprocessing pipeline (17.84% accuracy gain) and demonstrate that GWO optimization unlocks significant performance improvements without architectural modifications. This research establishes a new benchmark for AI-assisted DR screening by harmonizing diagnostic excellence with clinical practicality. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X