Abstract: Because of the very special type of contrast in phase-contrast images, it is almost impossible to perform fully automated single-cell analysis and quantification successfully. Because fluorescent dyes are highly toxic, phase-contrast images are commonly used to monitor live cells. In this paper, we present a method for the fully automated segmentation, classification and quantification of individual cell morphology in phase-contrast images. We calculate the confluence of the cell population and quantify the degree of toxic damage to each individual cell following phenol incubation. The results are then compared to standard cytotoxicity assays.
Phase contrast images are commonly used in order to study cell migration, cell
tracking and cell behavior like cell division under different conditions. The
morphology of each individual cell correlates with the degree of damage caused
by, for example, toxic substances in cell cultures or endogenous toxins.
Corresponding calculations of such toxicity damages are usually based on manual or
semi-automated methods. Automated methods of differentiated live cell
monitoring, based on image analysis algorithms, are limited; they depend upon the
degree of confluence of the corresponding cells and on the morphological
complexity. Sophisticated segmentation algorithms are applied in order to achieve
a robust segmentation of in vitro cell culture images observed with a standard
phase-contrast microscope or with video microscopy. Although these algorithms
provide a good separation of cell regions from the image background, they are
not able to classify cells on a differentiated level, e.g. they do not reliably separate
attached cells automatically [
To quantify toxicity in fibroblast cell cultures (Fig. 1), a data set of phasecontrast images was acquired. Cell cultivation was based on the ISO-Norm 10993 protocol. Phenol was used for the induction of cytotoxicity. Images were derived from a Zeissmicroscope, with 10x and 20x objectives, as 8-bit TIFFS. In order to evaluate the quantification of cell damage based on morphological criteria, standard cytotoxicity assays were applied in parallel: 1. LDH assay: The degree of cell damage correlates with the concentration of lactate dehydrogenase (LDH) in the cell culture medium released by damaged cells. The LDH content is quantified by the use of a color reaction induced by LDH. 2. Live/Dead assay: In this differentiating fluorescence based staining Propidiumiodid, which permeates only the membrane of damaged cells is used to stain the nucleus.
For the image analysis we used Definiens XD, which is an application of the
Definiens Cognition Network Technology. The Cognition Network Language
(CNL) is the corresponding graphical user interface meta-language, which
allows efficient development of rule-based algorithms. CNL consists of four basic
data structures: processes, domains, image objects and image object classes and
supports the use of specific expert knowledge within rule sets. We developed
an image analysis solution (CNL rule set), which uses different “maps”, where
the same phase-contrast image is copied into different instances (maps) for the
application of independent different processing procedures. Various algorithms
for segmentation and classification were applied to different maps in order to
achieve in their combination an optimal segmentation results. Analysis results
from one map were used as context for the analysis of objects in other maps;
in this way, the same phase contrast image could be analyzed simultaneously
differently by concentrating on different aspects and by combining those aspects
stepwise into one final result. Context-neutral and context-sensitive features are
defined in order to describe the individual properties of cells, the relationships
of cells to their neighborhoods and the cell organelles. The degree of confluence
of the cell population in an image was calculated as the ratio of the total area
of all cells in an image to the sum of the overall area of cell regions plus
background. The separation of the corresponding image objects-such as cell regions
and image background-requires a precise segmentation of these objects.
Therefore, we used multi-resolution segmentation and applied pixel-based filters, such
as the edge detection filter, on such individual image objects [
DFSC f1(protrusions) + f2(vacuoles) + f3(halos) + f4(roundness) (1) The coefficients f1, f2, f3, f4 are calculated automatically using the developed rule set and according to the numerical values of the corresponding morphological parameters of individual cells. The mean value of the DFSC over all individual cells in an image is defined as the “damage factor” for the image. The ratio of this “damage factor” to the confluence of the cells in the image is defined as “damage index” . The damage index was calculated at different time points during cell culture monitoring and for different phenol concentrations (Fig. 4). 3
Fifty images were used to develop the CNL rule set; this rule set was then applied to an automatic analysis of 200 test images. In the first analysis, the cell region was separated from the background and the quotient of the cell region to the whole image area was calculated; this is a measurement of the confluence of the cell population in an image (Fig. 1, Fig. 2). In the next step, sub cellular structures such as vacuoles, halos and protrusions were segmented and classified in each of the cells (Fig. 3). The damage factor was automatically calculated for each cell and individual cells were classified as damaged (red), partially damaged (magenta) or healthy (green or yellow). An overall damage factor for each of the images was calculated (Fig. 2, Fig. 3) and the damage index for varying phenol concentrations at different time points during cell culture monitoring was also determined. Fig. 4 shows the results for the damage index plotted as a graph (in blue) and the corresponding results from the cytotoxicity assays.
Athelogou et al.
The above work shows that CNL image analysis reproduces the results of the established cytotoxicity assays. Our method can therefore be used to analyze the morphology of single cells in phase contrast images and to quantify the degree of toxic damage of individual cells and the corresponding cell cultures during cell monitoring, as an alternative to semi-automated or manual methods. Acknowledgement. The authors are grateful to State of Bavaria and Landesgewerbeanstalt Bayern (LGA) for their financial support and Paul Brookes for his help for the manuscript.