Research
We apply machine learning to eye disease, from corneal transplant outcomes and retinal imaging to glaucoma forecasting and eye trauma triage.
Cornea and eye banking
We study how donor endothelial cell density is measured and how well it predicts graft survival in the Cornea Preservation Time Study, forecast each eye's cell loss after keratoplasty, and test AI for keratoconus screening, keratitis on confocal microscopy, corneal nerve measurement and corneal biomechanics.
Ocular surface
We segment meibomian glands on public meibography sets, set the first pediatric reference intervals for gland structure, and test whether image patches leak between training and test sets in goblet cell analysis.
Retina
We study diabetic retinopathy treatment and outcomes in DRCR Retina Network trial data and macular degeneration progression, test general and ophthalmic foundation models on fundus, OCT, OCT angiography and laser scanning images, and show how published data splits inflate results for retinopathy of prematurity.
Glaucoma and neuro-ophthalmology
We forecast and impute visual fields with self-supervised models, test how fellow-eye pairing and data splits affect glaucoma screening and eye-tracking results, and study papilledema detection on optical coherence tomography.
Surgery, trauma and orbit
We are building a triage classifier for open globe injury from external eye photographs, for settings without an ophthalmologist on site. Other work covers capsulotomy after cataract surgery, endogenous endophthalmitis in Klebsiella liver abscess, and segmentation of orbital imaging.
Evaluating AI
Across these areas we test how much reported accuracy depends on often-overlooked implementation details, such as how the data are split and whether images leak between training and test sets.