Orbscan IIz multimodal corneal topography dataset of anterior axial power maps with structured clinical parameters for unsupervised and self-supervised representation learning

Abstract

We present a multimodal corneal topography dataset comprising 3000 anterior axial power maps acquired from 3000 unique patients using the Orbscan IIz slit-scanning device (Bausch & Lomb) at a single private ophthalmology clinic in Iraq. The examinations were performed during routine clinical practice between 2007 and 2025 using the same diagnostic device throughout the acquisition period. Each image is paired with 19 structured clinical parameter fields extracted from the original device output using an automated OCR-based pipeline. These parameters include simulated keratometry astigmatism (SimK_Astig), maximum and minimum corneal power (MaxK and MinK), zone-specific irregularity indices, mean power, astigmatic power, steep and flat axis orientations across 3 mm and 5 mm optical apertures, corneal pachymetry at the thinnest point with spatial coordinates, anterior chamber depth, white-to-white diameter, pupil diameter, and angle kappa with intercept coordinates. All images were fully anonymized, cropped to the central map region, resized to 1024 × 1024 pixels, and saved in lossless PNG format. Following a quality control pipeline assessing image integrity, metadata completeness, and anatomical plausibility, 2633 image–metadata pairs satisfied the predefined inclusion criteria and constitute the quality-controlled subset characterized in this article. The dataset is released without diagnostic labels by design, positioning it as a foundation for unsupervised, self-supervised, and semi-supervised representation learning in anterior segment analysis. The complete dataset is available on Mendeley Data (DOI: 10.17,632/78wt2nc387.3) under a CC BY-NC 4.0 license.