A graph-mining algorithm for automatic detection and counting of embryonic stem cells in fluorescence microscopy images

Many cell-based research studies require the counting of cells in order to understand and validate experiments through statistical analyses. Although progress in imaging technology has enabled the automation of cell counting for many different cell types, this process still has to be done manually in the case of images of embryonic stem cells. In this paper, we present a new automatic algorithm to detect and count embryonic stem cells in fluorescence microscopy images that identifies pluripotent stem cells cultured in vitro. Our approach uses luminance information to generate a graph-based image representation. The cell pattern is defined as a subgraph, and a graph-mining process is applied to detect the cells. The method is tolerant to variations in cell size and shape. Moreover, it can easily be parameterized to handle different image groups resulting from distinct differentiation protocols. The paper presents numerical results from tests made on a database with more than two hundred images, including EB cryosection, embryoid body cell migration, murine embryonic stem cell colonies under murine embryonic fibroblast, and neurosphere images. The results from our algorithm were validated by expert biologists, and provide good precision, recall and F-measure. Finally, a comparative study with the widely used watershed algorithm is presented.