Improving sensitivity and accuracy of filamentous structure identification

Hello CellProfiler team and users,

We are interested in studying some microtubule binding proteins, which, when unperturbed, localize on microtubules and form filamentous structures in cells. Our goal is to find conditions that perturb the interaction between these proteins and microtubules.

To do so, I have built a pipeline to identify filament-like structures in cells and calculate the ratio of area occupied by filament-like structures in each cell. The current version I have kind of does what I want, but it does’t seem to be particularly sensitive, and it seems to be biased toward picking up brighter areas as “filament-like” regardless how the images actually look like. Hope I will be able to learn some tips so I can improve my pipeline. Thanks a lot in advance!

Best,
Meredith

MK014_v1_1.cpproj (87.8 KB)

Hmmm, these are tricky as your filaments are very thin and there’s a lot of what appears to me to be cytosolic background. I tinkered with a few things, but not knowing the system I’m not sure which is the best in your case- try any or some combination of these and see if it helps:

-Measuring the intensity of the cells after IDSecondary, discarding the bright ones, and then masking the image with MaskImage to analyze only the areas that contain cells in the intensity range of interest
-Smoothing the MAP2 image (a Gaussian of around 10 seemed good to me but again, YMMV) and then subtracting it from the original MAP2 image using ImageMath to bring out the filaments and help decrease the original intensity variation issue
-Instead of using ApplyThreshold, use IdentifyPrimaryObjects as your final step in pulling out the the filaments- it can be a bit less of a blunt instrument. You can then just use MeasureObjectSize on both the filament objects and the cells, followed by RelateObjects and your Calculate module.

I hope one or more of those is helpful. Good luck!