AI-aided segmentation of four types of drusen in volumetric OCT

Drusen are a hallmark biomarker of age-related macular degeneration (AMD), with their size, number, and morphology (type) closely linked to disease severity and progression. Accurate segmentation and classification of drusen from optical coherence tomography (OCT) images are essential for objective AMD assessment and monitoring. In this work, we present a deep learning framework that combines a convolutional neural network for automated drusen segmentation with a dedicated classification module to distinguish four clinically relevant, distinct drusen types based on segmentation output. We evaluated our approach on a comprehensive dataset and achieved a mean Dice score of 0.74 ± 0.21 for voxel-wise segmentation accuracy and a critical success index of 0.69 ± 0.24 for drusen count accuracy. This method demonstrates substantial improvements in the quantitative drusen analysis and offers a promising tool for enhanced AMD diagnosis and tracking of disease progression.