Automated optimisation of cell segmentation parameters in phase contrast using discrete mereotopology

It has been shown previously that the number of epithelial cells in a monolayer can be determined in vitro using phase contrast microscopy by subtracting images mean-filtered with two different kernel radii and then thresholding to segment cells. Careful selection of filter sizes was essential to ensure the number of segmented regions corresponded accurately with the number of cells in the image, however manual parameter selection and verification is time-consuming and prone to human error. We propose an intelligent imaging approach for evaluating the success of filter size combinations for cell detection using discrete mereotopology to compare segmentations with ground truth binary images of stained cell nuclei. Applying this approach to phase contrast images of H400 epithelial monolayers with varying levels of confluency, a region in the parameter space could be identified where more than 90% of cells were correctly detected.