PRIME Lab @ UCSF

Precision Retinal Imaging & Machine Learning in Eye Health

Geographic atrophy growth rate varies in different topographic locations

To quantify the variation in GA lesion growth rate, we utilized the euclidean distance map and quantified GA border expansion rate at each retinal eccentricity in the macula demonstrating a progressive in crease of GA border expansion rate as the retinal eccentricity increases. Reference: https://iovs.arvojournals.org/article.aspx?articleid=2778030

Bayesian Entry Time Realignment

To study the long-term disease natural history, we developed a novel Bayesian entry time realignment (BETR) method that can reconstruct the decades-long disease natural history by synthesizing short-term data. By performing the BETR method to the AREDS data, we horizontally translated individual datasets to convert the X-axis from Time after enrollment to inferred time after disease onset. Here, we can see that the square root of GA area in individual eyes increases linearly over approximately 30 years but at different progression rates. Reference: https://www.nature.com/articles/s41598-022-08919-1

01 / 06

Geographic Atrophy Lesions Have Different Configurations

We demonstrated that the growth rate of geographic atrophy area is directly proportional to the lesion perimeter. Based on this finding, we proposed GA perimeter-adjusted growth rate (mm/yr). Reference: https://doi.org/10.1016/j.oret.2020.12.002

Go to link

Geographic atrophy growth rate varies in different topographic locations

Bayesian Entry Time Realignment

Liangbo Linus Shen, MD

Vitreoretinal Surgeon & Clinician-Scientist

Incoming Assistant Professor at UCSF Department of Ophthalmology (September 2027)

The PRIME Lab develops clinically grounded artificial intelligence and computational imaging approaches using multimodal retinal imaging, longitudinal disease progression data, and advanced statistical modeling to improve diagnosis, monitoring, and therapeutic development in retinal disease.

Contact Dr. Shen

ADVANCED IMAGING • BAYESIAN MODELING • MULTIMODAL DATA • ADVANCED IMAGING • BAYESIAN MODELING • MULTIMODAL DATA • ADVANCED IMAGING • BAYESIAN MODELING • MULTIMODAL DATA •

Large-scale Longitudinal Datasets

Expert Image Annotations

Bayesian & Spatial-Temporal Modeling

We leverage high-resolution multimodal imaging data from thousands of patients to train robust AI models for early disease detection and monitoring.

Our team of clinician-scientists provides meticulous manual annotations of retinal layers and pathologies, ensuring clinical fidelity in every model.

We utilize advanced statistical modeling to understand the complex progression of retinal disease and predict disease progression.

Translational AI Applications

Advanced Imaging Technologies

Multi-institutional Collaborations

We bridge the gap between research and the clinic by developing AI systems that are ready for integration into clinical trial protocols and monitoring.

Our lab utilizes cutting-edge AOSLO and swept-source OCT/OCTA systems to capture the highest fidelity retinal textures for computational analysis.

We partner with leading academic institutions to build large-scale datasets that drive innovation in computational medicine and AI research.