EVIDENCE BASED PREDICTION AND PROGRESSION MONITORING ON RETINAL IMAGES FROM THREE NATIONS

The aim of this work is to demonstrate how a retinal image analysis system, DAPHNE, supports the optimisation of diabetic retinopathy (DR) screening programmes for grading colour fundus photography

DAPHNE, developed by the Department of Computer Science, University of Surrey, UK, automatically analyses characteristics of DR and DMO using both International and UK grading schemes (ICDRS and NSC) to facilitate the use of such system in various clinical settings. In DAPHNE, deep learning architectures were established to learn about different levels of features related to DR severity as well as the key markers associated to DMO. It first measures the quality of the input images then reports the findings through multiple detections. Every images will be analysed even if it is assessed as being of poor quality. This is to accommodate the fact that many poor quality images often result from significant pathology. DAPHNE outputs the analysis of the image on the quality level, grading level based on both ICDRS and NSC and measurement of suspected DMO.

Retinal image sets, graded by trained and certified human graders, were acquired from Saudi Arabia, China and Kenya. Each image was subsequently analysed by the DAPHNE automated software. The sensitivity, specificity, positive and negative predictive values for the detection of referable DR and/or diabetic macular oedema were evaluated, taking human grading or clinical assessment outcomes to be the gold standard. The automated software's ability to identify co-pathology and to correctly label DR lesions was also assessed.

In all three datasets the agreement between the automated software and human grading was between 0.84 to 0.88. Sensitivity did not vary significantly between populations (94.28% to 97.1%) with specificity ranging between 90.33% to 92.12%. There were excellent negative predictive values above 93% in all image sets. The software was able to monitor DR progression between baseline and follow-up images with the changes visualised. No cases of proliferative DR or DMO were missed in the referable recommendations.

The DAPHNE automated software demonstrated its ability not only to grade images but also to reliably monitor and visualise progression. Therefore, it has the potential to assist timely image analysis in patients with diabetes in varied populations, and also help to discover subtle signs of sight-threatening disease onset.