Some MeasureObjectSizeShape output parameters wrong?

Hi,

I am new to Cellprofiler and try to use it to analyze my multichannel fluorescence images of tumors. It seems that it is a very powerful software and I especially like the object-oriented approach (e.g., cell objects, blood vessel objects etc.). Thank you very much for developing this software and sharing it with everyone for free!

One aspect of my work involves the analysis of microvessel-related parameters in tumor tissue, so this was the topic for my first test pipelines.

After some trial and error I managed to get to the point were results started to look promising. E.g., I have an image from a brain tumor which was stained for CD34 to highlight the tumor microvessels. I wanted to extract some basic information from this image: microvessel density and intercapillary distances. Microvessel density was no problem. I also managed to define “microvascular domains” around each microvessel (using “IdentifySecondaryObjects” with the “Distance - N” strategy starting with the microvessel objects).

Using “MeasureObjectSizeShape” I then extracted some quantitative data from the microvascular domain objects. Unfortunately, some of the results were not what I expected (too low to be biologically plausible), especially the mean radius of the objects (from which the intercapillary distance can be derived).

So called a “SaveImage” procedure and exported the microvascular domain objects and then imported them into Image Pro Premier (Media Cybernetics) to double check the results. In IPP, I was able to find the exact same image objects with identical X and Y coordinates. The results of the area measurements were also identical in the CP and IPP output. However, in keeping with my reservations regarding the radius measurements in CP, these measurements were much higher in IPP (factor 3-4) and now much more in line with my expectations.

Does anyone have an explanation for these differences? I understand that it is most likely that there is something that I am doing wrong but I checked repeatedly and found no explanantion (especially since the area measurements turn out completely identical in CP and IPP, but the radius measurements don’t). Of course, I will gladly provide the pipeline and the image I am talking about.

Thank you very much in advance.

Kind regards,

Arnulf Mayer, University Medical Center Mainz, Germany

Hello Arnulf,

Thanks for trying CellProfiler! Would you mind exporting and posting your pipeline? Along with an example image (or set, if there are >1 channel) would be helpful (you may need to zip up the images to post, or Dropbox, Google Drive, etc are fine too). Your description is good, however it’s much easier for us to assess what is going on with the exact pipeline in front of us.

Some possibilities though:
(1) The units of scale in CellProfiler are pixels, not, say, microns.
(2) MeasureObjectSizeShape’s MeanRadius is defined as :“MeanRadius: The mean distance of any pixel in the object to the closest pixel outside of the object.”. So for microvessel objects (I assume long and thin), I would expect that this particular measure would always be quite low, no? i.e. any pixel within the microvessel is, on average, pretty close to the nearest border pixel.

In any case, please post the pipeline.

Best,
David

Hello David,

Thank you for your fast reply. I have attached the pipeline and an example image.

I probably should have waited until the next day before posting because when I looked at the Cell Profiler manual again on the next day, I realized what caused the problem. Sure enough, your suggestion No. 2 was the explanation for the differences I observed. So, the observation was real, but the cause is that Cell Profiler and Image Pro Premier define the term “mean radius” differently. In IPP, “mean radius” is defined as the average of the distances between the region centroid and its boundary. Personally, I think this definition is a little more intuitive than the CP definition. Therefore, readers of this post may still benefit from my mistake and be cautioned that they have to look at the exact definitions for the measurements.

Again, thank you for your help and for a great software.

Kind regards,

Arnulf


MicrovascularDomainsProjekt.cppipe (14.1 KB)