Semi‐automated assessment of microaneurysm formation rate from color fundus photographs in patients with mild NPDR

Abstract

Abstract Purpose To assess the reliability of microaneurysm (MA) formation rate, computed from color fundus photographs, in type‐2 diabetic patients with mild NPDR. Methods In a 2‐years observational and prospective study of 400 type‐2 diabetic patients with mild NPDR, MA formation rate (MAFR) was computed using a new method (MA‐Tracker). This method allows graders to earmark MAs location on color fundus photographs. By using image co‐registration, it became possible to track previously identified MAs and to identify newly earmarked ones. The reliability of the procedure, assessed using 3 independent graders and a total of 235 color fundus photographs from a previous dataset, demonstrated a very good agreement between graders for the number of MAs (ICC=0.964) and for the MAFR (ICC=0.981). Results Sixty‐three of the 400 patients/eyes that completed the baseline and the 6‐month visits had MAFR computed using the MA‐Tracker. While the number of MAs earmarked at baseline (mean±SD: 2.0±2.5 MAs) and at 6‐months (2.7±3.7 MAs) showed no statistically significant increase (P=0.105), the MAFR, on the other hand, showed a high MA turnover (mean±SD: 2.7±3.7 new MAs for this 6‐months period of follow‐up). Conclusion Microaneurysm formation rate obtained from color fundus photographs using the MA‐Tracker show a very good reliability between graders making it a new tool with potential to be used as a biomarker of DR progression.