Standardising ocular surface photography for artificial intelligence: current limitations and a proposed roadmap

Slit-lamp colour photography remains a cornerstone of anterior segment documentation and longitudinal monitoring. However, conventional slit-lamp photography has seen relatively limited technological advancement compared with the rapid development of quantitative anterior-segment modalities such as AS-OCT, Scheimpflug tomography and confocal microscopy. Despite the wide diffusion of digital cameras, smartphones, and high-end slit-lamp systems, ocular surface photography (OSP) remains hampered by the absence of standardised acquisition and processing protocols. The convex, aspheric geometry of the ocular surface, off-axis illumination, peripheral defocus, automatic colour correction, and pupil-driven exposure variation in retroillumination limit reproducibility, inter-device comparability, and clinical utility. In this review we critically appraise current OSP applications across major ocular surface conditions (dry eye disease, conjunctival hyperaemia, infectious keratitis, ocular surface burns, and ocular surface neoplasia), mapping the hardware, software, and reproducibility limitations that constrain clinical practice. We then examine how these limitations propagate into artificial intelligence (AI) development, where the absence of large, well-annotated, demographically diverse OSP datasets, together with the lack of reproducible quantitative phenotypes and reliable labels, is now the principal bottleneck to clinically deployable models; we discuss patient-level data partitioning and external validation, and outline how deep learning, federated learning and foundation models can both utilise and improve OSP. Finally, we propose a minimum acquisition and dataset specification, tiered by clinical setting from research to community screening, and a staged roadmap that combines hardware standardisation (focus-stacked, illumination-controlled, colour-calibrated acquisition), interoperable metadata-rich storage (DICOM where infrastructure permits), quantitative phenotype definition, and diverse international dataset collaborations, so that the ocular surface can benefit from quantitative imaging and AI as other areas of ophthalmology have begun to.