Age-related macular Degeneration (AMD) could become the leading cause of long-lasting vision impairment in individuals aged 50 and above if not properly treated. The paper describes an effective real-time strategy for recognizing and classifying macular problems using deep learning, with the goal of improving AMD care. The procedure entails creating a dataset in conjunction with medical retina specialists, preprocessing photos, extracting critical characteristics, choosing significant features with particle swarm optimization, and detecting bolt loosening using the densenet169 approach. The various steps involved in the process are: a) Generating a dataset requires the collaborative work of a subcommittee operating within The Royal College of Ophthalmologists’ Informatics and Audit Sub-committee. This subcommittee is composed of a varied group of medical retina experts hailing from various healthcare institutions across the United Kingdom. b) Preprocessing using image masking, color normalization, illumination normalization, contrast enhancement and identified blood vessels c) Feature extraction of quintessential features from the inputs using an autoencoder, d) selection of features from the extracted version using particle swarm optimization and finally e) detection of bolt loosening using densenet169 technique. According to the results derived from the experimental investigation, the proposed system surpasses existing cutting-edge models across various aspects, including but not restricted to achieving remarkable outcomes in terms of accuracy (0.95), sensitivity (0.97), and specificity (0.97).