Quantifying Microtubule Distribution

Image Analysis
, , , ,
#1

Hello everyone!

I am looking for a way to measure microtubule distribution in cells for comparison with cells in other treatment groups.
I have tried using FIJI to pre-process my images (Subtracting background, thresholding, Difference of Gaussian) and running plugins such as ridge detection to find the filaments, but I can’t seem to find a method that will allow for good quantification or replication.
My thought process was to segment the cells, capture the filaments, and find a way to calculate the area of the filaments that lie within the segmented cells, but I have not had much luck with this.

I’d greatly appreciate any suggestions for methods of quantification, or information about other programs (CellProfiler maybe?) or workflows/pipelines that could accomplish this.

Thank you!

Attached below are two microscopy pictures that I want to compare.

Cells_A.tiff (3.7 MB) Cells_B.tiff (3.7 MB)

#2

Hi,

I would suggest a tandem approach using Ilasik and Cellprofiler. Ilastik is very good for segmentation and finding the filaments will be much, much better. And then that segmentation output can be imported into Cellprofiler as a mask to perform measurements (or ImageJ if you want). Cellprofiler can segment into cells and make per cell measurements also, although the accurate placement of boundaries of cells is always a challenge and won’t be perfect for all cells.

Good luck!

#3

We have a video tutorial outlining using ilastik and CellProfiler here: https://www.youtube.com/watch?v=89XPqczqhvU

and there is a written tutorial on GitHub: https://github.com/CellProfiler/tutorials/tree/master/PixelBasedClassification

Good luck!

#4

Hi everyone,

Any update if you were able to get these filaments to segment properly using Ilastik? I’m running into the same issues. I’ve attached a couple photos to show the microtubules that I’m trying to segment. Unfortunately, no matter how I train the set it can’t segment it. Which is kind of strange because when I go to “features” and use “Difference of Gaussians, Hessian of Gaussian Eigenvalues, and Lapplacian of Gaussians” it clearly outlines the tubular structures quite nicely but my data set won’t train to these values. I looked through all the tutorials online unfortunately I couldn’t find one that allowed for segmentation or skeletonization of these types of structures. Ultimately, I would want to do something similar in the green channel and quantify the amount of overlap between the two. Any help would be appreciated. Red microtubules.tif (2.6 MB)

Difference in Gaussians
Difference in Gaussians819×560 31.1 KB
Hessian of Gaussian Eigenvalues
Hessian of Gaussian Eigenvalues802×557 46.1 KB
Lapplacian of Gaussians
Lapplacian of Gaussians792×593 30.3 KB
Green 4.tif (2.6 MB)

#5

Hi @jliebert,

If you’d like to import data into CellProfiler, we recommend training a pixel-based classification project and then exporting probability maps. Could you share an example of your probability maps for the microtubule class and background class? That would help to get a better sense of what’s going on.

Once the probability maps are exported, that’s when we typically import them into CellProfiler for further analysis.

Cheers,
Pearl