Automated cell segmentation in phase-contrast images based on classification and region growing

Cell segmentation in phase-contrast microscopy images remains a challenging problem because of the large variability in subcellular structures and imaging artifacts. In this paper, we present an approach to the automatic segmentation of tightly packed cells in phase-contrast images. We combine the classification of superpixels with the region-growing method to locate cell membrane boundaries. We demonstrate that such a combined approach is able to perform the task of cell detection and segmentation with a high level of precision. On the presented dataset, we achieved 90% precision with 78% recall. The results indicate that this method is suitable for real biological applications.