Ex vivo multispecies dataset for three-dimensional reconstruction of the corneal stromal nerve plexus using hexapod-based confocal microscopy and deep-learning segmentation

This dataset provides three-dimensional confocal microscopy images and deep-learning-based nerve segmentations of the stromal corneal nerve plexus from ex vivo samples of multiple species (porcine, murine, ovine, and human). Corneal volumes were acquired using a confocal microscope synchronized with a programmable six-axis hexapod platform that precisely positioned the specimens, enabling large-scale volume scans of the corneal stroma. A U-Net-based 2.5D segmentation model was trained exclusively on one porcine cornea, achieving an Intersection-over-Union (IoU) of 0.92. The same model was subsequently applied to additional porcine, murine, ovine, and human corneas without retraining. The dataset includes large-area en face mosaics reconstructed from volume stitching for each depth level, along with corresponding automated nerve segmentations. Additionally, the manually annotated nerves used as ground truth for model training are provided.