Human Embryonic Stem Cell Detection by Spatial Information and Mixture of Gaussians

Human Embryonic Stem Cells (HESCs) possess the potential to provide treatments for cancer, Parkinson´s disease, Huntington´s disease, Type 1 diabetes mellitus etc. Consequently, HESCs are often used in the biological assay to study the effects of chemical agents in the human body. However, detection of HESC is often a challenge in phase contrast images. To improve the accuracy of HESC colony detection, we combine spatial information and the outcome of a mixture of Gaussians model. While a mixture of Gaussians generates reasonable labels for various regions of HESC images, it lacks spatial details and connectivity. Sets of spatially consistent candidate labeling are generated by median filtering the image at different scales followed by thresholding. An optimal combination of filter scale and threshold which maximizes the correlation coefficient between the spatial information and the mixture of Gaussians output is obtained. The paper validates the method for various HESC videos.