Comparison of fundus photography, autofluorescence, and infrared imaging in various phenotypic features of geographic atrophy

Abstract

Purpose : We aimed to evaluate the agreement between color fundus photography (CFP), fundus autofluorescence (FAF), and infrared reflectance (IR) measurements of geographic atrophy (GA) with an emphasis on relationship with phenotypic GA features.

Methods : We analyzed CFP (n=170), FAF (n=185), and IR (n=185) images from 128 eyes with GA in 77 patients in the METformin FOR the MINimization of GA clinical trial. GA area was measured by 2 independent, masked graders, and FAF images were graded manually (FAF-M) and semi-automatically (FAF-SA) using the RegionFinder software (Heidelberg Engineering). We stratified images by GA focality, homogeneity, size, and color (Figure) to determine if significant differences in measurements exist between FAF-SA and the other methods.

Results : The GA area measured in CFP, FAF-M, FAF-SA, and IR images was 9.18 ± 8.11, 8.75 ± 7.32, 7.02 ± 5.52, and 8.76 ± 7.37 mm², respectively. All 4 methods had excellent inter-grader agreement (intraclass correlation coefficients>0.99) within the same modality. GA measured with FAF-SA was lower than CFP (P=0.001), FAF-M (P=0.004), and IR (P=0.003) (Table). The relationship persisted in unifocal, multifocal, and non-homogeneous GA lesions. In homogeneous GA lesions (n=53), FAF-SA was not significantly different from CFP (difference = -0.88 ± 1.27 mm²; P=0.06), FAF-M (-0.51 ± 0.68 mm²; P=0.27), and IR (-0.45 ± 0.71 mm²; P=0.21). Gray GA lesions (n=5) had the highest cross-modality differences when compared to FAF-SA: -6.29 ± 5.18 mm² (CFP), -8.43 ± 1.96 mm² (FAF-M), and -6.41± 2.88 mm² (IR).

Conclusions : Though each method had excellent inter-grader agreement, the measurements of GA sizes vary significantly across different modalities and GA phenotypes. GA sizes measured in FAF-SA were significantly lower than in other imaging modalities. Future clinicians and researchers should use a single, consistent modality for disease monitoring and progression.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Figure. Panel 1: CFP (1A), FAF-M (1B), FAF-SA (1C), and IR (1D). Panel 2: GA phenotypes. 2A, Homogeneous, black, unifocal GA lesion. 2B, Multifocal lesion. 2C, Non-homogeneous lesion with mixed hypo- and hyperautofluorescence within the GA. 2D, Gray GA lesion.

 

Table. Top: Number of images used in cross-modality analysis. Diagonal: Mean absolute difference (MAD) between graders within the same modality. Bottom: Cross-modality difference (D) of modality A - modality B.