Health professionals are invited to participate in a survey to help evaluate the use of artificial intelligence (AI) to improve diabetic retinopathy (DR) screening.
Conducted by researchers from the University of Melbourne and CERA, the survey will involve grading 45 retinal photographs and should take around 30 minutes to complete. Participation is anonymous and all who complete the study will receive a Diabetic Retinopathy Referral Summary Sheet as well as DR educational resources in appreciation.
The DR deep learning algorithm (DLA) being used for this study was developed then validated using 58,790 and 8,000 de-identified retinal photographs, respectively. These photographs were initially graded by a panel of ophthalmologists, with a criterion standard grading for each image assigned when consistent grading outcomes were achieved by at least three graders. The formal gradings were then used as the ‘ground truth’ to allow the deep learning algorithm to learn patterns in the retinal photographs, and in doing so, learn how to identify images with DR. The sensitivity and specificity of the DLA for correct referral has been assessed to be 92.3% and 93.7%, respectively.1
the survey will involve grading 45 retinal photographs and should take around 30 minutes to complete
The grading of DR is important to guide management. It is hoped that the use of an AI tool will speed screening and diagnosis for this disease, and increase accuracy at a time when the prevalence of diabetes and therefore the risk of DR is significantly increasing world-wide.
The one year risk of progression to proliferative DR (PDR) – a sight-threatening stage of the disease that requires laser treatment – varies according to the severity of retinopathy 1-5% of for those with mild non-proliferative DR (NPDR);12-26% for those with moderate NPDR and ~50% of those with severe NDPR. Accordingly, rescreening intervals and referral recommendations differ on the basis of the severity of retinopathy: mild NPDR rescreen in 12 months; moderate NPDR review by ophthalmologist within three months; severe NPDR refer to an ophthalmologist within one month.
Approval of this project has been obtained from the Research Governance Unit at St Vincent’s Hospital, Melbourne, Australia (Preocjt ID 50023, Local reference number LRR 082/19).
Find Out More
Information about the Diagnostic assistance for diabetic retinopathy using a deep learning algorithm and integrated visualisation maps: a prospective study for optometrists, orthoptists and trainees is here. (link to: https://redcap.link/ipnlckus)
Information for GPs, Aboriginal Heath Educators, nurses, and medical students is here. (link to: https://redcap.link/grp7bnjs)
Reference
- Keel S, Lee PY, Scheetz J, Li Z, Kotowicz MA, MacIsaac RJ, He M. Feasibility and patient acceiptyta obfi l a novel
