Wednesday, April 23, 2025
4/23/2025
A generalizable deep learning platform for unifying quantification of experimental autoimmune uveitis - Posterior uveitis accounts for up to 15% of severe visual impairments in the United States. Mostly affecting individuals in childhood or young adulthood, uveitis is comparable to diabetes and macular degeneration in terms of years of visual morbidity. The disease etiology is largely unknown. Experimental autoimmune uveitis (EAU) is an animal disease that shares essential pathological features with human uveitis and is a valuable model for studying mechanisms of human uveitis and for evaluating efficacy of new therapies and diagnostic methodologies. However, a significant limitation with the experimental models including EAU is that assessment and evaluation of the disease and its severity is subject to substantial inter- and intra-observer variability and results from lab to lab are rarely reproducible. This shortcoming directly impacts quantification and accurate assessment of disease, which in turn is crucial for assessing new medical therapies. Artificial Intelligence (AI) has shown promising applications in ophthalmology and many deep learning models have been successfully applied to the screening and diagnosis of a growing list of clinical ocular conditions. Yet, AI models, such as deep learning algorithms, have not been extensively studied in experimental research such as EAU. We hypothesize that emerging foundation AI models based on self-supervised learning and vision transformer units will be able to provide a robust generalizable platform for experimental research thus improve the objectivity and reproducibility as well as accuracy of quantification and disease assessment. The overarching aim of this study is to provide AI platforms to unify experimental disease assessment with an application on EAU. We propose to 1) generate high-quality and well-annotated multi-modal datasets from three independent centers with fundus, optical coherence tomography (OCT), histological, and immunohistochemical images from EAU models and 2) develop a foundation model based on emerging AI models to quantify and assess uveitis images with potential generalization to other experimental diseases. We will validate the AI models using independent subsets of images to assure reproducibility and generalizability. To achieve these objectives, we have assembled a team of experts with integrated and complementary skills in uveitis, imaging, and AI with access to large multi-modal uveitis datasets generated previously at NIH and two other institutes in New Zealand and UK. We will publish datasets at TVST or Ophthalmology Science journals and make datasets and annotations publicly accessible to the community. We will also publish AI tools and make them publicly accessible to the vision community based on the NEI guidelines.
HHS’ Tracking Accountability in Government Grants System (TAGGS) website provides access to detailed descriptions of grants, loans, aggregated direct payments and other types of financial assistance awarded by the HHS Operating Divisions (OPDIVs) and the Office of the Secretary Staff Divisions (STAFFDIVs).