Yilmaz, CerenYesil, SinemOzkan, Elif IlkayOkur, Erdem2026-03-272026-03-272025-10-26979833155565897983315556652687-7775https://hdl.handle.net/20.500.14365/8905https://doi.org/10.1109/TIPTEKNO68206.2025.11270114Age-related Macular Degeneration (AMD) is a leading cause of vision loss in the elderly, where early detection plays a critical role in slowing disease progression. Deep learning approaches have shown strong potential for automated AMD diagnosis from retinal images; however, their performance can be hindered by image quality variations, illumination inconsistencies, and artifacts. In this study, we propose a custom contrast enhancement mask to improve lesion visibility in fundus images prior to classification. Using the publicly available ADAM challenge dataset, three deep learning architectures-YOLOv8n-cls, InceptionV3, and a modified U-Net encoder-were trained and evaluated on both the original and enhanced datasets. Experimental results demonstrate that the enhancement method substantially improves classification performance across all models, with YOLOv8n-cls achieving the highest accuracy with 91.70%, and specificity with 96.76%. These findings highlight the importance of preprocessing in medical image analysis and suggest that lightweight models, when combined with effective enhancement techniques, can achieve high accuracy suitable for clinical deployment.eninfo:eu-repo/semantics/closedAccessImage ProcessingAMDDeep LearningAge-Related Macular DegenerationNeural NetworksConference Object10.1109/TIPTEKNO68206.2025.112701142-s2.0-105030543508