LAUSR

IN the past decades, image cytometry has made great progress due to the advances in optical imaging and image processing. The central problem of image cytometry in many studies is cell segmentation which has received more and more attention in the recent years. The advances of cell segmentation lie in the abundant emergence of new methods, the increased automation level and the increased accuracy. However, most state-of-the-art methods or software tools mainly rely on existing basic image processing algorithms developed many decades ago. For example, the watershed algorithm is adopted by Cellsegm (1), SMASH (2), ImageJ (3), CellProfiler (4), Alanazi’s method (5), and the recently reported method by Tsujikawa et al. (CYTOA 95:4; pp 389– 398). Classical thresholding is adopted by Cellsegm, SMASH, ImageJ, and Alanazi’s method while K-means clustering is adopted by Tsujikawa’s method. These basic image processing algorithms have been developed to solve the general computer vision problems and all of them have inherent insufficiencies. Consequently, they have limited capability to solve the challenging cell segmentation problems at large even if these challenges have been recognized world widely for a long time. The facing challenges include the ever increasing complexity, the great variety of cell types, the low image contrast, the poor image quality, the connection or overlapping of neighboring cells, the nonuniform pixel intensity and the influence of noise or clutter. It is obviously out of the capability of the existing image processing algorithms to solve them at large. Thus, we argue that more effective image processing algorithms should be developed to meet the genericity of the great variety of cells while consistently achieving specificity in solving these challenging problems in different cases.