A Comparative Study of Enhanced Drusen Detection and Classification in AMD Using Deep Learning Approaches

Accurate identification and classification of drusen in age-related macular degeneration (AMD) are essential for timely diagnosis and management. This study presents a relative assessment of deep learning (DL) algorithms like DenseNet (DN) 121, 169 and 201 for the automated identification and categorisation of drusen in retinal pictures. Multiple neural network models, including advanced convolutional neural networks (CNN) and transfer learning frameworks, were trained and tested on curated datasets containing diverse AMD cases. Performance was rigorously assessed using various metrics, with results benchmarked against expert clinical interpretations. The best-performing models demonstrate superior diagnostic precision, particularly in distinguishing early-stage and advanced drusen, compared to traditional methods. The comprehensive evaluation shows the transformative possibility of tailored DL pipelines for clinical AMD assessment, offering significant improvements in reliability, speed, and objectivity for large-scale screening and monitoring of retinal health.