CornOrb: A Multimodal Dataset of Orbscan Corneal Topography and Clinical Annotations for Keratoconus Detection
Authors:
- Mohammed El Amine LazouniMLMohammed El Amine Lazouni
- Leila Ryma LazouniLLLeila Ryma Lazouni
- Zineb Aziza ElaouaberZEZineb Aziza Elaouaber
- Mohammed AmmarMAMohammed Ammar
- Soufyane ZeharSZSoufyane Zehar
- Mohammed Youcef Bouayad AghaMAMohammed Youcef Bouayad Agha
- Ahmed LazouniALAhmed Lazouni
- Amel FerouiAFAmel Feroui
- Ali H. Al-TimemyAAAli H. Al-Timemy
- Siamak YousefiSYSiamak Yousefi
- Mostafa El Habib DahoMDMostafa El Habib Daho
Open Access
Abstract
Keratoconus assessment relies on corneal topography to characterize curvature, elevation, and thickness. Public datasets remain limited, particularly for Orbscan 3 and underrepresented geographic populations. CornOrb is a multimodal dataset of Bausch & Lomb Orbscan 3 examinations retrospectively collected at an ophthalmology center in Algeria. It contains 1454 examination records from 744 patients: 889 clinically normal and 565 keratoconus records. For each eye, four anonymized PNG maps are provided: axial curvature, anterior elevation, posterior elevation, and pachymetry. Native pixel-wise quantitative curvature, elevation, and pachymetry matrices are not available in the current release. A structured CSV file provides age, sex, astigmatism, maximum keratometry, central and thinnest pachymetry, and anterior and posterior asphericity. Diagnostic labels correspond to the final adjudicated binary clinical diagnosis at the examination level and were established from clinical records together with Orbscan 3 findings. Only clinically definite normal and keratoconus cases were retained, whereas borderline, suspected, forme-fruste, subclinical, or otherwise diagnostically uncertain cases were excluded from the binary dataset. No independent quantitative diagnostic or staging index was used as a corroborating criterion, and inter-rater reliability could not be retrospectively quantified. Accordingly, performance on CornOrb should be interpreted within this adjudicated binary task and should not be extrapolated to early-stage, subclinical, or borderline keratoconus detection. CornOrb is intended to support reproducible research on keratoconus detection and analysis, including image-based classification, multimodal learning combining topographic maps and tabular variables, and reproducible comparative evaluation of machine-learning methods within the CornOrb cohort. To the best of our knowledge, CornOrb is the first publicly accessible African Orbscan 3 dataset dedicated to keratoconus analysis.Keywords
All Related Versions
Funding Information
- Agence Nationale de la Recherche
This publication has 20 references indexed in Scilit:
- Machine learning-assisted early detection of keratoconus: a comparative analysis of corneal topography and biomechanical dataScientific Reports, 2025
- Cochrane corner: artificial intelligence for keratoconusEye, 2024
- Unsupervised learning for large-scale corneal topography clusteringScientific Reports, 2020
- Corneal topography in clinical practiceJournal Français d'Ophtalmologie, 2019
- Keratoconus Natural ProgressionOphthalmology, 2019
- Age-specific Incidence and Prevalence of Keratoconus: A Nationwide Registration StudyAmerican Journal of Ophthalmology, 2017
- Keratoconus-integrated characterization considering anterior corneal aberrations, internal astigmatism, and corneal biomechanicsJournal of Cataract & Refractive Surgery, 2011
- Orbscan computerized topography: Attributes, applications, and limitationsJournal of Cataract & Refractive Surgery, 2005
- Live-epikeratophakia for keratoconusJournal of Cataract & Refractive Surgery, 1998
- KeratoconusSurvey of Ophthalmology, 1998