Newby need help



This software is great - just what I needed for my project. I monitor liposome-loaded rhodamine accumulation in cells, and I’d need to measure the change in intensity over time. So far others in my lab only measured the average pixel intensity of an area which was selected so that the same amount of cells were present in each sample.
I think it’s not a good solution.
I was playing around with imageJ, but it turned out to be a nightmare to quantify each and every one of my images, not to mention that most of the technical terms are not very clear to me -so I’m not even sure if this was done right. (I got some books ordered from the library on image analysis, but it will take time to catch up, and I have to work till then.)

I need general advice on how to proceed (practically right now I’m the only one who does this kind of work, and it’s not in the profile of the lab), and some specifics about the use of cellprofiler. I really apologize, but right now I have no one else to ask. (I’ve spent the last 3 days trying to figure things out with some success.)

  1. which magnification should I use on the microscope for this kind of work: lower, which shows many small cells or higher which only shows 20-40 cells?

  2. I usually start from black to gradual increase in fluorescence; so in order to detect all the cells even when stained weakly I have to use the DAPI images, right? But how? When I only use the rhodamine signal, only 4-9 cells are recognized from about 30. How can I tell the software that the cells detected by the blue signal are there at the red as well?

  3. Should I use the average fluorecence on the image or localize the cells, and get their average? Can I correct it later with the number of cells? (Obviously 45 cells will give off more signal, than 20.)

  4. By setting the treshold to my cells, I eliminate the area of background -so from that on I only measure my cells, right? I tried to do that in cellprofiler, but I can’t see beforhand if the values are good. (In imageJ you can use a slider and see your image change)

  5. I need some help making sense of the results as well: where do I find the meaning of the terms used there?

  6. how do you report the results? :smile: What format is acceptable?

  7. it sound stupid, but how do I know how many pixels are the diameter of an object? (The microscope can put a scale on the image; how can I measure it in pixels?)

How do you think my pipeline should look like? (General description; I really don’t expect anyone to do my job for me.) What kind of modules do I need?

I really appreciate anyone’s imput. This is a very exciting field, but an unknown one for me. (I was lookin for a book like “image analysis for dummies” on amazon, but no luck so far…) Thank you very much.



I think you are on the right track. Hopefully I can help.

  1. Higher resolution is always better for this type of work. You can take more images of the same well if you need a higher sample.

  2. Use DAPI with IdentifyPrimAutomatic and get your “nuclei”. If you have a cell stain (e.g. actin), use IdentifySecondary with the “propagation” method to define a cell boundary. If not, you can use IdentifySecondary with the Distance option, and create a ring around your nuclei, any size you want. The IdentifySecondary module produces objects which you call “Cells”. Then you can check the fluorescence in your Red channel using the Cell objects.

  3. This depends on what you are looking for. You will have per-cell data so you can do whatever you want, including correcting by number of cells. If you give a better idea of what you are looking for I might be able to give some suggestions.

  4. Use the automatic thresholding options. Sometimes you need to adjust the “Threshold Correction Factor”, but in general, it works very well. (Read the help to see how to change the correction factor)

  5. Each module has help which explain the measurements. For your case, I would read the MeasureObjectIntensity module’s help.

  6. After an analysis, you create a MAT file, which is named on the bottom right corner of the CellProfiler window (e.g. DefaultOUT.mat). You can then go to Data Tools -> ExportData and select this output file. Everything can be exported to excel.

  7. Use the MeasureObjectAreaShape module. It will give Major/Minor axis of each cell. It fits the best oval around your object, and gives the widest and thinnest region of this oval.

Example Pipeline:
IdentifyPrimAutomatic (DAPI)
IdentifySecondary (Phalloidin, or use Distance)
IdentifyTertiary (if stain is not nuclear)
MeasureObjectIntensity (Red channel, Nuclei/Cells/Cytoplasm)
MeasureObjectAreaShape (Cells)

If you measure the DAPI channel in your nuclei, you can also get DNA content, and you may be able to correlate your liposome stain with the cell cycle. We usually meausure everything, then dig in to the data to find what we want, but have other measurements available if we should need them.

Hope this helps!


Thank you very much for the answers. I’ll spend a few days with the program, and there’s more questions to come. I really appreciate that you took the time to answer. (And thank you all for the program. It is incredible.)

Can an universal treshold be used with all the images within an experiment? (I have images with no staining, with faint staining, and with strong staining of cells -after all, the signal first is very weak, and then it gets really strong, so the difference between the background and the signal changes considerably. I basically take photos of the cells every 2 minutes for 35 min and watch for the accumulation of the rhodamine. The exposue time is set so that I don’t have any autofluorescence in the beginning. DAPI and VIS photos are taken at the 0 time point and at the very end to minimalize bleaching. These photos can then be overlayed to show the DAPI and the rhodamine signal on the same image. I, then, need to compare the change in rhodamine accumulation using rhodamine-loaded liposomes, so I need to compare data from several experiments.)

In fact, there’s a question I’m not sure how to address. The complete area of the cells does not show for a while in the experiment, as it takes some time for the rhodamine to accumulate to the extent that it becomes clearly visible -therefore the CellProfiler would not be able to find the outline of the cells -or it would set the borders smaller than they really are. Is it possible to first ID the cells on the later-stage photos, and then apply these borders to the early ones where only the DAPI stain is visible? This way the area where the intensity change is monitored would remain constant. (Is that important at all?)

I apologize for asking these basic questions, but our lab is a peptide chemistry-oriented one. Cells and liposomes are just extra, so there’s no one who could answer these questions. I have the maual printed out, and I’ll spend some time with the software. If I hit a wall, I’ll ask again. Thank you very much for your time and effort.


Mike may have a better reply for you on your last question but I thought I’d get you started. Yes - I am pretty sure you should be able to process the last image in your set and then measure the intensity within those objects going all the way back through the time lapse series of images (assuming the cells do not move around very much).

Here is how: Put the last image in its own folder and set up a pipeline to identify the cells as usual. Then, use ConvertToImage to convert the objects (cells) into a binary (black and white) image. Finally, use a SaveImages module to save the resulting image with a new name.

Now, set up a pipeline to use those cell outlines for your entire movie: Set up the pipeline using LoadImages to load each image in your movie, but also add a LoadSingleImage module to load the binary image you previously saved. Now, use IdentifyPrimAutomatic to identify cells in the binary image (every time around the cycle it will identify the same objects - you can have the settings be very simple here, e.g., use an absolute threshold of 0.5 without any fancy de-clumping, because you’ve already got a nice binary image). Lastly, use MeasureObjectIntensity to measure the intensity of your movie image within the cells. Note, this also answers your question about identifying objects using one channel (color) and then measuring the intensity of another channel within those borders - YES, this is what we do all the time. Look at MeasureObjectIntensity’s help for more details.

Does that work? Mike, feel free to offer other suggestions. It would be nice if you could save and load a label matrix image (that is, save and load ‘objects’ rather than an ‘image’) to avoid having to re-identify objects during each cycle but at the moment I don’t think that is a possibility. If you save an image as grayscale it will cause problems if there are more than 255 objects due to the limits of the image file format, and the LoadSingleImage module is not quite equipped to deal with objects rather than images.

To answer your other question, Yes - you can use an absolute threshold throughout the movie by typing in a number between 0 and 1 in the IdentifyPrimAutomatic module. But I agree that your idea of using the last (brightest) image in the movie is the better approach.

A couple of other things: higher resolution is always better in terms of getting precise measurements, but lower resolution can be good because you have more cells sampled; that is, better statistics. So you need to balance those two considerations in making your decision. Also, for interactive thresholding, try the method called “Set interactively” in IdentifyPrimAutomatic. This gives you a slider, as you had hoped.



Thank you all for the suggestions.
I ran into a few problems :smile:
When I work with the data it’s generally enough to use the “image” data, the averaged intensity changes, right? When it identifies 83 cells, it’s hard to work with all the data cell by cell. But the image values show a general increase in fluorescence for all the identified cells -that should be enough, right?

I’ve used the last, stronges signal to identify the cells (somehow the binary image did not work). If you take a look at the two images here, you’ll see that the software detected given cells as “clumps” of many cells, even if you say “yes” for merging small objects into bigger ones. I checked the average intensities, and the values are slightly different.
I know I should either be using the DAPI for cells -but then with no staining I have no cytoplasm visible, or the binary file -but it doesn’t really show smaller cells.

Merged version

  1. using binary image (made by ImageJ. The size of objects is the same as the previous two: 10-200 pixels.)

Average values (Integrated intensities -does that mean that the number of cells present in the image is already factored in? Are these values comaparable between different assays?)


"Merged version"

What do you think? Are these differences big enough to cause concern? What else should I change? (I can upload the pipeline somewhere if needed.)

and a last question (for now): let’s say, everything’s perfect, and I can start processing my images. How do I report my results in a paper? Integrated intensities plotted versus time? What else should I report with tthem? (Magnification, exposure time, rhodamine-loaded liposome-concentration, temperature…?)
Thank you again.


It’s too bad we don’t have time to devote to really getting you up and running on this. It looks to me from the images you posted that you need to improve the cell identification before you trust any measurements. I recommend that you read the Help for IdentifyPrimAutomatic and adjust the Smoothing fliter and the neighborhood size. You have too many ‘objects’ identified within each cell. I would turn off the Merge objects option and play with the settings on those other two options (see the help) until the cell identification looks much better (that is, each cell is one colored blob), then as a last step you can turn Merge objects on again. You never expect it to be perfect. Especially when you have cells with two ‘lobes’ of fluorescence, it’s ok that the computer recognizes it as two cells, but looking at the image I think you can do much better by smoothing the image more and increasing the neighborhood size. Please post the new results if you improve them so we can see them!

A side note, you may be wondering why your measurements show a nice increase even though the identification is not working well - it’s probably because there is a very dramatic increase overall and so it doesn’t matter that you have chopped up each cell into little pieces, because each ‘cell piece’ shows an increase. I imagine your measurements will improve and be less noisy if you improve the cell identification, though.

The meaning of the various measurements is defined within the help for MeasureObjectIntensity, so I won’t repeat the explanation here. When you export Per-image data, it tells you the MEAN of the values for the individual cells in the image, so you end up with the MEAN mean intensity or the MEAN integrated intensity.

Intensities are usually labeled ‘arbitrary intensity units’ because for your microscope you do not have a good way to convert the intensity values into a real, physical measurement (that is, you do not know a conversion constant like 50 photons per intensity unit). So, yes, plotting Integrated Intensity (arbitrary units) vs. time would be good. The experimental details should probably be reported in Materials and Methods (Magnification, exposure time, rhodamine-loaded liposome-concentration, temperature…?) In particular, you should note whether the exposure time was constant throughout the experiment (I hope it was!)