Validating CellProfiler Object Orientation


#1

Hi,

I have cells growing on an oriented substrate and I’m looking to validate the orientation module for my pipeline. I use the MeasureObjectSizeShape module in CellProfiler 2.1.1. To that end, I started by drawing identical circles in PowerPoint, then altering their dimensions to go systematically from a 1:1 ellipse to a 2:1 ellipse, all in the same orientation. All the shapes were saved as .tif files.
My problem comes with the circles (1:1 ellipses if you will). I was expecting a random distribution of orientations, but have instead obtained a distinct mode around -36 degrees. I’m taking this to be due to the pixellation that is part and parcel of any raster file.

My question is thus in two parts. First, is it possible to have vector graphics open in CellProfiler to verify that what I’m seeing is an artefact? Second, and more fundamentally, has anyone advice on what images are a good way to validate the software modules?

Regards,

A loyal CellProfiler User


#2

Hi,

My colleagues may address this question better. In the meantime, I guess you may find certain explanation in this post:

Hope that helps a little.


#3

Thank you: that was indeed very helpful. It helps to explain why my circles are measured as having a major axis and minor axis of (63.0001524653 and 62.7680803137) pixels respectively. So, if I understand it, CellProfiler fits an ellipse that contains the object and then measures the direction of its major axis using the numpy.arctan2 function?


#4

This is the list of currently supported formats in CP- I don’t know of any off the top of my head that support vector graphics, but there may be some.

I suspect it is due to the pixelation from your raster- it’s possible you can clean up your circle images in something like Photoshop. Especially if you’re only seeing this in perfect circles and your ellipses display the expected behavior, hopefully this won’t affect your downstream application (measuring cells on an oriented substrate) too much.

ETA- yes, CP does use the numpy.arctan2 function on the bounding ellipse to calculate orientation.


#5

Using Inkscape to draw circles solved the problem beautifully and the orientation is now random. Looks like PowerPoint doesn’t actually do circular circles.