About DCX analysis



I am just a junior for this software,it is a very well software to do some analysis. And now I am working with analysis DCX-labeled cells in the brain. Please see the attachment about this staining.
I have two factors need to analysis: 1.is the number of DCX-labeled cells ; 2. is cell layer thickness of DCX-labeled cells. And I establish a quite simple pipeline for calucate the cell numbers.
It is the pipeline:
Load Image
Color to Grey
Color to Grey
Export to Excel

The threshlod that I set up for Hoechst staining is [0.05,1]; for DCX staining is [0.2,1];
The diameter for Hoechst staining is [4,10]; for DCX staining is [5,20].
This pipeline look works for calculat the cell numbers, but the problem is: the DCX-staining isnot specific for nuclear or cytoplasm, the labeled cell has a long dentrites projection from cell body.So when i do analysis, the projection was also recognized as cells.So that could be increase aritfical error of analysis.I tried crop module,it looks works not very well. So could you give me some suggestions about how to anyalsis the cell body’s number and delete the signal of projections? Thank,

At the mean time, about how to analysis the layer thickness, i absolutely have no idea of which module I need to created. Just give me a clue. I will appreciate about that. Thanks very much!!




Hello Peifei,

You have a good start on your project - here are some suggestions:

  • First, be sure to export your images into a lossless image format, like bmp or png. Jpgs lose information
  • Typically we first use IdentifyPrimAutomatic on the nuclear stain image (Hoechst here), then use IdentifySecondary on the cell stain image (e.g. DCX) to “propagate” the cytoplasm. You could do this, but we have worked on a few neuronal projects here, and they are notoriously tough. BUT, as you only want to count the DCX-labelled cells, your problem is easier, and I will outline them below:
    *SmoothOrEnhance (optional, but may help hoechst segmentation by smoothing out the bright spots). Set “Images to be smoothed”=Hoecsht, “Method”=RemoveBrightRoundSpeckles, “Size” = 10 pixels or so for your images
    *IDPrimAuto. Images=Hoechst, Diameter=[20,65] (your nuclei seem to be ~30 pixels wide to me – not the bright spots, right?), Method=Otsu Adaptive (adaptive since brightness varies across image/nuclei), Size of Smoothing filter=~7, Suppress Local Maxima=~20
    *MeasureObjectIntensity. Images=grayDCX, Objects=whatever you called the hoechst nuclear objects

You could also add IdentifySecondary, as I mentioned above, if you want to propagate the DCX labelled cells from the nuclei you defined.

As for the (cortical?) layer, I have little to go on. There is no scale bar, so I can’t go on the absolute distance. And the full extent of the layers don’t seem to be visible, so I can’t get a relative value either. The simplest way (and I don’t know if this is possible for you) would be to orient the bottom layer of cells to be horizontal in the image, and then use the X/Y locations of the identified nuclei from ExportToExcel to calculate the distances. Note that you can set the “Pixel size” in the Main gui window of cell profiler if you need distances in real space, or just calculate the real distance afterwards in Excel.

Note also to see pixel values in CellProfiler, go to any Figure window, click on the “CellProfiler Image Tools” menu, select ShowOrHidePixelData and the intensity and position values can be read out in the lower left.

Hope this helps!