Quantification of two dimensional progression velocity in geographic atrophy and its relationship with autofluorescence at the borders

Abstract

Purpose : Patterns in blue autofluorescence (BAF) imaging seem to be correlated with progression rate of geographic atrophies (GA). We introduce an automated technique for two-dimensional estimation of the progression velocity of atrophy borders to investigate GA progression in relation to local BAF intensity.

Methods : We retrospectively analysed 14 pairs of follow up BAF images acquired with the Heidelberg HRA (Heidelberg Engineering, Germany). BAF images were acquired at different time points from 9 eyes of patients with non-neovascular AMD and in presence of GA. Each pair of images was aligned and areas of GA were manually segmented. Velocity was estimated by image processing the border shrinkage of the GA at the second time point until the border of the first GA was met. Number of required shrinkage steps divided by the time interval between scans gave an estimated border velocity (Figure 1). The average BAF intensity of the external neighbouring areas to the initial border was also calculated. Concave borders were not used in the analysis to exclude areas that merged during the interval, preventing the velocity estimation. Linear regression was used to estimate the relationship between BAF intensity and velocity.

Results : A significantly positive correlation (p<0.0001) was found between border velocity and BAF intensity in the analysed dataset (Figure 2).

Conclusions : Progression velocity at borders of GA, as determined by a novel automated technique, is associated with BAF intensity. Our new technique offers an automated way for exploring determinants of pathophysiological changes in GA as revealed by autofluorescence at the borders, and may have promise as an objective measure for determining progression in patients.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

 

Figure 1. a) A pair of successive BAF images from the dataset showing the manual segmentation of the GA at time points t1 and t2. b) A map of the shrinkage steps required for the border at t2 to reach the border at t1. c) The border velocity calculated as the ratio of the shrinkage steps and the time interval (t2-t1 = 9 months).

 

Figure 2. Scatter plot of mean velocities and mean intensities of the neighbouring areas for each segment used in the analysis. Segments from the same GA are represented with dots of the same colour. Results of the linear regression are shown with a red line with its confidence intervals in blue.