Identifying Same-Cell Contours in Image Stacks: A Key Step in Making 3D Reconstructions

Abstract

Identification of contours belonging to the same cell is a crucial step in the analysis of confocal stacks and other image sets in which cell outlines are visible, and it is central to the making of 3D cell reconstructions. When the cells are close packed, the contour grouping problem is more complex than that found in medical imaging, for example, because there are multiple regions of interest, the regions are not separable from each other by an identifiable background and regions cannot be distinguished by intensity differences. Here, we present an algorithm that uses three primary metrics-overlap of contour areas in adjacent images, co-linearity of the centroids of these areas across three images in a stack, and cell taper-to assign cells to groups. Decreasing thresholds are used to successively assign contours whose membership is less obvious. In a final step, remaining contours are assigned to existing groups by setting all thresholds to zero and groups having strong hour-glass shapes are partitioned. When applied to synthetic data from isotropic model aggregates, a curved model epithelium in which the long axes of the cells lie at all possible angles to the transection plane, and a confocal image stack, algorithm assignments were between 97 and 100% accurate in sets having at least four contours per cell. The algorithm is not particularly sensitive to the thresholds used, and a single set of parameters was used for all of the tests. The algorithm, which could be extended to time-lapse data, solves a key problem in the translation of image data into cell information.

Current automated 3D cell detection methods are not a suitable replacement for manual stereologic cell counting

“…Such methods are inappropriate for dealing with the thick tissue sections and 3D images used in stereology. The second class of 2D approaches first seek to identify cell profiles in 2D image slices and then aggregate information found in multiple planes to build a full 3D segmentation (Indhumathi et al, 2011; Leung et al, 2011). The aggregation methods are largely based on heuristics (e.g., linearity of centroids and analysis of convexity) rather than complete models of cell appearance.…”

The role of adhesion energy in controlling cell–cell contacts

“…While recent advances in image analysis of 2-dimensional time-lapse movies of contacting cells in vivo allow automatic tracking of cell–cell contact dynamics in high spatiotemporal resolution [33,65,66], methods for automatic 3-dimensional tracking are less advanced [67]. Moreover, theoretical models used to determine adhesion energy on the basis of cell–cell contact dynamics either still lack experimental confirmation [58,68,69] or rather provide information about cell–cell interfacial tension which is only partially determined by adhesion energy [17,24,45,70].…”

Cell-level 3D reconstruction and quantification of the Drosophila wing imaginal disc

“…A contour grouping algorithm based on three geometric metrics is proposed to determine the correspondences between cell outlines in adjacent images in an image stack (Leung et al, 2011). This algorithm, which could be used during the 3D reconstruction of individual cells, achieved high accuracy when tested with synthetic data sets and with contours produced by manual segmentation of confocal images of gastrulating zebrafish cells.…”