Developing a Multi-Task Enabled Open-Source Foundation Model for Personalized Analysis of Age-Related Macular Degeneration Imaging Data, Leveraging the Latest and Largest Clinical AMD Database (AONGHUS).

• Applicant: Dr. Steffen Emil Künzel
• Subject Area: Ophthalmology
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 544623978

Age-related macular degeneration (AMD) is a prevalent eye condition among older adults, affecting up to seven million people in Germany. Late-stage AMD, including neovascular AMD (nAMD) and geographic atrophy (GA), impacts around 490,000 individuals in Germany, accounting for half of all severe visual impairment cases. Current diagnostic and therapeutic approaches for AMD often employ a generalized strategy, not accounting for the disease's complex biology and significant variability between individuals. This leads to frequent yet often ineffective doctor appointments and considerable delays in the necessary treatment for late-stage AMD. As a result, elderly patients face numerous medical visits, costly and burdensome treatment cycles with a recurring risk of severe, sight-threatening complications, inefficient healthcare resource utilization, and limited success in treatments. There is a pressing need for personalized medicine in this area. My research project tackles this challenge by developing an advanced Self-Supervised Learning (SSL) model, trained on the untapped AONGHUS database containing multimodal retinal imaging files from over 10 million images of more than 280,000 eyes. The aim is to apply the model for clinically significant multi-tasks, facilitating more accurate diagnosis and tailored treatment strategies for AMD. The model is designed to detect subtle patterns and changes in retinal images critical for early detection and management of AMD. The model will be trained to classify various stages and subtypes of AMD, predict disease progression and events, monitor responses to treatments, and provide personalized therapy recommendations. This involves identifying vital morphological biomarkers and risk patterns, predicting the onset of AMD in people without previous disease history, and forecasting disease progression, treatment needs, and adverse outcomes at a level suitable for clinical application. Moreover, the pre-trained model will be released globally, enabling researchers and clinicians worldwide to adapt it to their specific datasets and research questions. This initiative promotes a customized, more accurate approach to AMD research and treatment and supports fairness in healthcare by ensuring that smaller populations and minorities also benefit from our research. This global collaboration aims to enhance clinical decision-making and improve patient care worldwide.

• DFG Programme: WBP Fellowship
• International Connection: United Kingdom
• Host: Professor Pearse A. Keane, Ph.D.