A bioinformatic pipeline for image analysis of varroa related traits in honeybees comb images

In honeybees, Apis mellifera, hygienic behaviour is the uncapping and removing of dead and diseased larvae and pupae from uncapped brood cells. In times of honeybee declining worldwide, beekeepers study hygienic behaviour manually quantifying removal of freeze-killed larvae from uncapped cells. Manually counting uncapped cells in comb images is time-consuming and prone to error. Focus of this study is to design an automated pipeline for the segmentation of honeybee comb images. For this purpose, honeybee comb images were acquired, selected, and analysed through digital image processing techniques, which must handle problems due to uncontrolled illumination conditions, differing colours, rotations, scaling, and comb sizes. More precisely, for simultaneously handling poor illuminations and differing colour conditions several colour normalization algorithms have been experimented, ranging from unsupervised colour-enhancement models to colour normalization technique used in digital histology. Next rough segmentation of the area of interest (AoI), and the cells in that area, have been obtained by clustering followed by Hough transform for finding the circular AoI, and by binary operations for detaching attached cells in the AoI. Analysis of the histogram plots describing the connected components in the AoI allowed estimating the mean cell areas and therefore computing an estimate of the cell counts. Among the 127 comb images, 80 images containing limited artifacts and acquired under acceptable illumination conditions were selected as test images and allowed obtaining a correlation with manual counted cells of 0.948. The remaining 47 images, containing strong artifacts and bad illuminations conditions, were used for development and resulted in a lower correlation. The hereby generated pipeline yields an estimation of honeybee comb cells correlating with manual counted cells.