Automated analysis of digital fundus autofluorescence images of geographic atrophy in advanced age-related macular degeneration using confocal scanning laser ophthalmoscopy (cSLO)

Abstract Background Fundus autofluorescence (AF) imaging using confocal scanning laser ophthalmoscopy (cSLO) provides an accurate delineation of areas of geographic atrophy (GA). Automated computer-assisted methods for detecting and removing interfering vessels are needed to support the GA quantification process in longitudinal studies and in reading centres. Methods A test tool was implemented that uses region-growing techniques to segment GA areas. An algorithm for illuminating shadows can be used to process low-quality images. Agreement between observers and between three different methods was evaluated by two independent readers in a pilot study. Agreement and objectivity were assessed using the Bland-Altman approach. Results The new method (C) identifies vascular structures that interfere with the delineation of GA. Results are comparable to those of two commonly used procedures (A, B), with a mean difference between C and A of -0.67 mm2 (95% CI [-0.99, -0.36]), between B and A of -0.81 mm2, (95% CI [-1.08, -0.53]), and between C and B of 0.15 mm2 (95% CI [-0.12, 0.41]). Objectivity of a method is quantified by the mean difference between observers: A 0.30 mm2 (95% CI [0.02, 0.57]), B -0.11 mm2 (95% CI [-0.28, 0.10]), and C 0.12 mm2 (95% CI [0.02, 0.22]). Conclusion The novel procedure is comparable with regard to objectivity and inter-reader agreement to established methods of quantifying GA. It considerably speeds up the lengthy measurement process in AF with well defined GA zones.