Glaucoma is a progressive optic neuropathy and a leading cause of irreversible blindness worldwide. Due to its asymptomatic nature until advanced stages, a large proportion of cases remain undetected, limiting opportunities for timely intervention. We will evaluate whether pretrained deep learning (DL) models that quantify glaucomatous damage and its progression from fundus photos can accurately identify individuals with glaucoma in the UK Biobank cohort and reveal novel phenotypic and genetic associations.
Specifically, we will validate continuous, DL-derived damage and progression scores against available ground truths (including self-reported glaucoma and related clinical metrics) and examine how these scores relate to demographic, lifestyle, and clinical factors to uncover previously unrecognized risk profiles. In parallel, we will test for associations between model outputs and genetic data to illuminate potential genetic contributions to glaucomatous damage. This work aligns with the UK Biobank’s mission to advance disease understanding and improve health outcomes through innovative analytic tools. In keeping with the UK Biobank AI policy, the pretrained models will not be fine-tuned on UK Biobank data.