Purchase this article with an account.
S. Bearelly, S. Raghunath, A. A. Khanifar, S. W. Cousins, S. Farsiu; Quantitative Evaluation of Fundus Autofluorescence Images to Predict Geographic Atrophy Progression. Invest. Ophthalmol. Vis. Sci. 2010;51(13):531.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We assessed the ability of fundus autofluorescence (FAF) imaging to predict rate of geographic atrophy (GA) progression using semi-automated software.
Two FAF images were acquired at least 12 months apart from 45 eyes of 45 subjects with GA secondary to age-related macular degeneration (AMD). GA areas were segmented manually by expert ophthalmologists (EO) as gold-standard and semi-automatically by our in-house developed software. We used a modified automatic GA segmentation algorithm described by N. Lee, et al (Asilomar Conf, 2008) followed by manual correction to match at least 90% of the markings by EO. Rim area was defined as the 500 µm-wide margin bordering GA. Rim area on baseline FAF images was assessed for extent of hyper-autofluorescence (HAF). In Method-1, HAF area was segmented by our software using a modified implementation of the algorithm described by J.C. Hwang, et al (IOVS, 47(6), 2006). We divided the subjects into three groups of 15, based on the total area of HAF (1: <0.16mm2, 2: between 0.16 and 1.12 mm2, 3: >1.12mm2). In Method-2, we estimated the HAF area in the central 3000 µm-diameter region. Again we divided the subjects into three groups of 15, based on the total area of HAF (1: <1.05mm2, 2: between 1.05 and 1.88 mm2, 3: >1.88mm2).
Correlation of GA growth rate and total area of HAF was 0.71(P<0.0001) and 0.57 (P< 0.0001) for Method-1 and Method-2, respectively. The median GA growth for the three categories were 0.54, 2.86, 3.3 mm2/year; and 0.87, 2.86, 2.98 mm2/year using Method-1 and Method-2 respectively. The Kruskal-Wallis test for Method-1 and Method-2 (p=2.9*10-5 among categories 1,2; p=8.2*10-6 for category 1 compared to 3) and (p=0.007 among categories 1,2; p=0.002 for category 1 compared to 3), respectively.
Extent of HAF area in the rim may be used to differentiate faster and slower progression of GA. Despite some inaccuracies in HAF segmentation, Method-1 outperformed the other technique. Development of fully-automated more accurate HAF segmentation algorithms may further increase the reliability of the proposed technique.
This PDF is available to Subscribers Only