Figure 1.
Overall image analysis workflow for a typical experiment.
First, variations in illumination and staining are corrected. Nuclei are identified by thresholding, then used as seeds to identify cell edges. Finally, DNA-damage foci are identified. Schematic data shown, based on image courtesy of Scott Floyd, Michael Pacold, and Michael Yaffe. Colors of nuclei, cells, and foci are arbitrary.
Figure 2.
Thresholding by mixture models.
Mixture models derive a threshold from two density functions (one for the background, one for the foreground) fitted to the distribution of intensities in the image. Units are arbitrary. Original image from project described in Moffat et al. [80].
Figure 3.
Local thresholding methods compute the threshold τa for a pixel a from statistics of intensities of pixels {i} in a neighborhood Na of a rather than from the entire image I. Original image from project described in Moffat et al. [80].
Figure 4.
(A) Brightfield image of C. elegans worms not amenable to thresholding because of intensity variations. The color bar on the right of the image shows that brighter pixels are displayed as red and dimmer pixels as blue. Original image from the project described by Moy et al. [32]. (B) Contour plot of smooth illumination function fitted to one or a set of images such as (A). (C) Corrected image obtained by pixel-wise division of (A) by (B). (D) Worms in the corrected image are consistently darker than the background and can therefore be identified by thresholding.
Figure 5.
Splitting clusters of objects.
(A) Thresholding this image of two nuclei results in one continuous outline rather than two objects. Original image from project described in Moffat et al. [80]. (B) The local maxima in the smoothed image correspond poorly with the centers of the nuclei. (C) The local maxima (red squares) in the distance transform of the image (shown as contours) correspond well with the centers of the nuclei. (D) Seeded watershed from the local maxima in (C) divides the cluster correctly.