The SD-OCT correlates of fundus autofluorescence in geographic atrophy: a comprehensive point-to-point analysis of fundus autofluorescence with SD-OCT using phase 3 trials DERBY and OAKS.

Abstract

Purpose : We carried out detailed mapping between fundus autofluorescence (FAF) and spectral domain optical coherence tomography (SD-OCT) in geographic atrophy (GA) to establish the extent to which SD-OCT biomarkers of GA can be detected using FAF.

Methods : Same-day FAF and SD-OCT imaging of 815 patients from OAKS and DERBY were analyzed by expert human graders and a fully automated, validated deep-learning model, respectively. The model used the Classification of Atrophy Meeting (CAM)-defined SD-OCT biomarkers (photoreceptor degeneration [PRD], retinal pigment epithelium [RPE] loss, hypertransmission (HTR), and RPE and Outer Retinal Atrophy [RORA]). FAF to Near Infrared (NIR) image registration used triangulation of common anatomical landmarks to topographically overlap FAF and OCT imaging, enabling deep-learning model mapping of FAF to NIR images for a pixel-accurate (0.0057 mm 2 per pixel) FAF/SD-OCT comparison. GA features totaled individual comparisons of 276,507,257 pixels across 815 distinct pairs of FAF/SD-OCT imaging. Cohen’s
Kappa coefficient was calculated to quantify the degree of pixel-level agreement between FAF and OCT on GA lesion size and topographical correlation.

Results : Overlapping FAF and SD-OCT areas were isolated and correlated at the pixel-level. Early treatment diabetic retinopathy study (ETDRS) regions collectively showed good FAF/SD-OCT agreement for GA (mean Cohen’s Kappa 0.82 SD 0.09), RPE loss (0.81 SD 0.10), HTR (0.78 SD 0.13) and moderate for PRD (0.55 SD 0.19). The predictive value of FAF was strongest in a concentric zone overlapping the inner and outer ETDRS regions and diminished when moving both peripherally and centrally . Agreement was highest in the peri-/para-foveal ETDRS subfields; 89% parafoveally, decreasing to 63% in the central bouquet. 65% of false positives and 60% of false negatives were within 100 µm of the lesion border.

Conclusions : Our pixel-level mapping of location-specific in vivo retinal layer changes quantifies the difficulty of using FAF alone in identifying both foveal-involvement and functional retina at risk of GA. Consequently, GA research and treatment decisions will benefit from the use of additional biomarkers available from SD-OCT scans.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

(a) FAF & qOCT
(b) Registration FFA/OCT
(c) FAF/OCT Agreement metrics