Fig. 1

Outline of the process to automatically segment crystalline cones. The method is illustrated with images of Apis mellifera as an example. a Reorientation of the eye (volume rendering) and the labelled external cornea (red) into the ‘PCA space’. b Division of the cornea surface voxels into subregions (each plotted with a different colour) and modelling of each subregion using polynomial fitting in its ‘PCA space’ (top). c Extraction and unfolding of a subvolume of the eye data. Note that the cornea surface in the unfolded subvolume is almost flat, so that the pattern of crystalline cones in the cross-section is regular. d Auto-segmentation of the raw cones using a texture-based approach. The results of this step are displayed on one slice from the unfolded images (top) and in the whole eye subvolume (bottom). The white arrow indicates the location of a circular artefact that locally prevented the segmentation of a few cones. e Back-transformation of the raw auto-segmented cones into the original eye volume (3D rendering). f Post-processing of the raw cones labels (right subfigure) to eliminate noisy detections (red on the right subfigure) and retain valid cones (green on the right subfigure - left subfigure)