FORGE: Functionally-guided OCT Representation for Glaucoma Endophenotyping

Glaucoma endophenotyping remains challenging due to disease heterogeneity and the limitations of single-modality structural imaging. We introduce FORGE (Functionally-guided OCT Representation for Glaucoma Endo-phenotyping), a cross-modal contrastive learning framework that integrates visual field (VF) functional signals as privileged supervision during training to produce functionally informed macular RNFL (mRNFL) representations, while enabling OCT-only inference via a learned null embedding. In 5,372 paired MEEI examinations, FORGE identified 9 clinically distinct mRNFL endophenotypes with divergent progression rates (MD slopes -0.2 to -1.8 dB/year, P<0.001), improving clustering over OCT-only baselines by 22% (FCM) and 11% (GMM). External evaluation across 74,077 UK Biobank images confirmed generalizability, with improved clinical risk association (r=-0.33 vs r=0.04). Genetic analyses identified 12 additional genome-wide significant glaucoma loci compared with OCT-only phenotyping. FORGE demonstrates that functionally-guided representation learning yields clinically and genetically coherent endophenotypes, with broad applicability to multimodal diseases requiring population-scale structural inference.