Mapping the impact: AI-driven quantification of geographic atrophy on OCT scans and its association with visual sensitivity loss.

Background/aims: To examine the association between artificial intelligence (AI)-driven segmentation of geographic atrophy (GA) on optical coherence tomography (OCT) and visual sensitivity loss quantified by defect-mapping microperimetry, a testing strategy optimised to quantify the spatial extent of deep visual sensitivity losses.

Methods: 50 individuals with GA secondary to age-related macular degeneration underwent defect-mapping microperimetry testing within the central 8° radius region in one eye. GA on OCT was automatically segmented with an AI-based multiclass classification and segmentation model, and GA on fundus autofluorescence (FAF) images was manually annotated. Their extent in the topographically corresponding region sampled on microperimetry was derived, and structure-function relationships were examined based on Spearman correlation coefficients (ρ). The distance of each test location from the OCT-defined and FAF-defined GA margin was also derived and used in prediction models of non-response on defect-mapping microperimetry.

Results: There was a strong correlation between the proportion of locations missed on defect-mapping microperimetry and the corresponding percentage of the central 8° radius region with GA on OCT (ρ=0.85) and FAF (ρ=0.89). Prediction models for non-response at individual test locations using GA derived from OCT and FAF imaging had a sensitivity of 59% and 62% (p=0.310), respectively, at 95% specificity.

Conclusions: AI-driven, automated quantification of GA on OCT showed a strong correlation with the global extent of visual sensitivity loss, comparable with those based on manual annotations on FAF imaging. These findings affirm the expected functional relevance of OCT-derived GA measurements and their clinical utility for monitoring disease progression in those with GA.