Artificial intelligence-enabled mobile-based fundus imaging for diabetes screening

Executive Summary:

Retinal imaging and artificial intelligence (AI) driven oculomics are novel, promising, non-invasive tools for clinic and community based screening for conditions such as type 2 diabetes (T2D), which often remains undiagnosed until major complications occur. Even when screening programs exist, individuals with newly diagnosed T2D often present with vascular damage, including some degree of diabetic retinopathy (DR), showing the importance of screening to prevent disease progression. AI driven screening tools integrated into mobile phones could overcome screening barriers by providing tools that are easy to use and disseminate in low and middle income countries like India. A Pilot study from this group showed that fundus imaging and oculomics has high sensitivity, specificity, and accuracy for detecting T2D. Our multidisciplinary team from the US and India proposes to expand upon this early work and existing research partnership to develop, validate, and field test the addition of AI-driven screening for T2D to existing AI-integrated fundus imaging on mobile phones developed for DR screening (Remidio Fundus-on-Phone, for DR screening).

This project will be done in two phases: Phase 1 and Phase 2:

In phase 1, we aim to 1. Develop the AI software for T2D detection (using 120 existing and newly collected smartphone-based retinal images in people without and with diabetes); integrate and beta-test with retinal imaging in a sample of 60 patients in the existing retinal imaging device and 2. Evaluate, using mixed methods, the human-centric usability of the tool for T2D screening screeners (healthcare providers using the tool in a hospital setting).

We will then test the feasibility, scalability, and effectiveness in a field-based study at community outreach centres (phase 2) by: 3. comparing the diagnostic yield of T2D of AI-enabled smartphone-based fundus imaging compared to conventional, standard of care T2D screening in India (random capillary glucose testing followed by confirmatory fasting plasma glucose testing) as well as estimating the burden of DR among undiagnosed T2D cases; and 4. evaluating using mixed methods, the feasibility, scalability, and costing of AI-enabled smartphone-based fundus imaging compared with conventional T2D screening in community outreach centers. If successful, this innovative application of oculomics for T2D detection would bridge the screening gap by providing a low cost, noninvasive screening tool that is easy to use, disseminate, and sustain, thereby improving early diagnosis of T2D and DR and reducing the burden of T2D and its complications.