Identifying speckles/clusters, bright compared to the relati


So I am trying to use cellprofiler to identify clusters of E-selectin on endothelial cells, these generally show up as bright speckle-like clusters on the surface of the cells.

The pipeline I have generally works quite well on the control, WT endogenous E-selectin, except when it comes to images of cells which have been transfected with the mutant versions. The problem is that the transfection isn’t 100%, so on some cells you get fairly bright staining of the cell, along with even brighter speckles which are easily picked out. Whereas on other cells, the cell is expressing less E-selectin, so stains less bright, but still has clusters on the surface.

the problem is that the clusters on the dimmer cells are less than, or equivalent to in brightness, to the background staining on the brighter cells. They are still identifiable clusters by eye, but the program gets confused when trying to identify just by thresholding, it either identifies loads of false positives on the brighter cells, or fails to pick up the clusters at all on the dimmer cells.

I have tried to collect object information on Intensity, neighbours, radial distribution and texture, and then try and filter out all the false positives on cellprofiler analyst, but it still has problems striking the right balance.

I have considered whether it would be possible to add another stain to outline each cell, and then threshold and count clusters separately per cell, but I wanted to come here first and ask if there was anything I could really do without resorting to repeating the experiments!

cluster pipeline.cp (13.6 KB)


Unfortunately, without a cell stain (and preferably a nuclei stain to do along with it), there is not much you can do with these images in order to obtain a per-cell cluster count.

If you are just interested in an image-wide cluster identification and count, then the speckle filtering step in EnhanceOrSupress should be enough to reduce the background staining while maintaining the cluster features. You should set the speckle filter size to roughly the size of the features you want to keep, i.e, approximately the largest cluster size.