Addressing Health Disparities through AI-Enhanced Prediction of Diabetic Retinopathy

PROJECT SUMMARY/ABSTRACT We propose to develop a reliable, clinically proven diagnostic imaging and monitoring technology which will mitigate the impacts of diabetic retinopathy (DR). The proposed technology reduces the cost of DR treatment, mitigates its overall social and public burden, and helps save the vision of millions of people, the vast majority of which are from underserved communities with disparities in access to high quality medical care. DR is among the leading causes of vision loss in the world, affecting more than 230 million people. In the US, ~38 million people are affected by diabetes and ~30% go on to develop DR. This devastating complication of both type I and II diabetes results in structural damage to the sensitive retinal vasculature. Minority and medically underserved populations are particularly susceptible to the complications of diabetes, including DR. DR causes 17% of vision loss in African Americans compared with 8% in non-Hispanic whites. Limited data on the prevalence of DR in Native American populations suggests that in some tribes it is as high as 45%. Some studies show as high as a 200% increased risk of DR in Latinos as compared to non-Hispanic whites. Our proposed intervention aligns with the NIH’s Health Disparities Research Framework by promoting health equity in NIH- defined health disparities populations (i.e., Black, Latino, Native American, and others) by increasing access to quality of eye care and improving vision health outcomes with early interventions that are effective in minority populations. While there is no cure for DR, there are a variety of treatments that must be applied as early as possible to prevent further damage. Recent studies have shown that small changes in the retinal vasculature’s oxygen saturation are a reliable indicator of pre-stage and early-stage DR—before structural damage occurs. Since there is no clinical non-invasive technology capable of achieving sufficiently high resolution to detect these changes, a major need exists for the development of advanced retinal oximetry technologies with proven clinical utility. Bioxytech delivers for the first time a non-invasive technology capable of providing pre-stage and early-stage DR detection before structural eye damage occurs. This Phase I project is based on a novel non-invasive imaging technology which leverages spatially modulated light to capture a retinal oxygenation image in one snapshot. Further, the proposed technology is well-suited to characterize the multi-layered structure of the retina, where absorbance is a limiting factor for other technologies. Initial human clinical studies using this technology have found that it can detect DR at least 4-13 months prior to conventional diagnosis. Building on these promising results, the proposed project: 1) further develops the technical and software components of the technology, 2) performs a clinical evaluation study at two sites which serve almost exclusively minority communities, and 3) integrates a novel machine learning platform to enhance the accuracy of early diagnosis and classification, as well as to introduce autonomous image analysis. This will enable more accurate classification of DR and facilitate patient risk stratification for more appropriate monitoring.