AI-enabled Diabetes Diagnostic System using Retinal Fundus Image

Usually in the ophthalmology clinic the Topcon TRC-NW6s retinal camera or similar products are used. In the QBB clinical setup Topcon TRC-NW6s retinal camera (Figure 1) was used to capture retinal images. As part of this proposal, we will integrate our DiaNet model with the existing system of retinal camera (e.g., Topcon TRC-NW6s retinal camera) (Goal 1 of this proposal). Then the captured images will be fed into the DiaNet model. DiaNet will enable the ophthalmologist, general practitioners (GP) as well as the non-physicians to have a quick scan of diabetic patients. Moreover, we will recruit a small sized cohort to validate the performance of DiaNet (Goal 2 of this proposal). Compared with traditional methods of taking retinal pictures in clinical setup, DiaNet will be a faster, portable, cheaper, and fully non-invasive method for diabetes screening. To provide hint to the physician, DiaNet will also overlay the retinal images with color-coded class activation map (CAM) or attention-based mechanism to highlight the influence of the degree to which image regions influence the final prediction.

Hamad Bin Khalifa University (HBKU)