Retinal Vessel Computational Fluid Dynamics

Authors: Giancarlo Paternoster, Dr. Umar K. Mian, Dr. Roy Chuck

Project Completed: Ongoing

Our research focuses on developing a new strategy to forecast areas of vascular compromise in the retina before overt damage becomes visible on standard imaging. Traditional methods rely on static snapshots that capture only morphological abnormalities, leaving the underlying fluid dynamics largely unexplored. By incorporating personalized flow simulations into the analysis, we can capture the unique hemodynamic environment of each patient’s retinal vasculature. Our premise is that subtle changes in flow velocity, shear stress, and vessel geometry can serve as early indicators of a future disease event, allowing clinicians to intervene before irreversible damage occurs.

To achieve this, we merge computational fluid dynamics (CFD) with machine learning (ML) to create a robust predictive pipeline. After segmenting and mapping out each patient’s retinal vessels, we simulate blood flow under various physiological conditions. These CFD-derived parameters—combined with morphological features and clinical annotations—form the training set for an ML classifier that flags high-risk vessels. Our pilot studies suggest that adding hemodynamic data substantially boosts predictive performance compared to image-based methods alone. Going forward, we plan to expand our data collection, refine the CFD simulations for greater accuracy, and ultimately validate this integrated model as a tool for proactive retinal disease management.