Corneal confocal microscopy enables in vivo visualisation of the sub-basal nerve plexus; however, most available datasets provide only centreline annotations, and open pixel-level annotations are comparatively scarce, limiting broader pixel-level segmentation research. Existing resources vary in quality, annotation strategy, and clinical coverage, complicating objective algorithm comparisons and reducing their utility in developing robust models. To address these gaps, we introduce a new dataset of 410 high-quality images from 88 participants, each paired with an expert-reviewed pixel-level nerve mask. These images span two independently collected subsets with distinct acquisition conditions and participant groups. Demographic information is available for each participant, and one subset contained detailed clinical and laboratory data. The dataset offers detailed, manually curated delineations of nerve fibres, consistent formatting, and clear provenance. It is intended to support the training and validation of segmentation models, assess generalisability across imaging conditions, and explore the associations between image-derived metrics and clinical characteristics.
Qiao et al. (Fri,) studied this question.