Wound Healing with low contrast images


I tried to set up a pipeline of wound healing, however, my images’ contrast is not very good for identify the primary objects.

I tried the pipelines in previous questions, but they did not help with my images.

Is there anyone facing the same problem?

My images are attached.

1 Like

Dear @Peter_Lu,
a lot of times the best way to analyze these kinds of situations is redo the experiment with a better acquisition setup rather than try to invent or fine-tune algorithms and parameters for weeks.

you can try to enhance a little bit the cell layers respect to the wound with Variance Filters or similar…but I will try to enhance the acquisition part, asking maybe to some microscope experts

Is it possible for you to redo the experiment?

Emanuele Martini

1 Like

What do the pipelines do exactly ? To know what did not work

Otherwise from what I have seen before something like that in Fiji :

And then selecting the full image in a rectangular ROI and Analyze>Plot Profile

From there you can find the maxima to define the size of the wound
Good luck !


Dear Emanuele,

Many thanks for your suggestion! I’m also thinking about repeating the experiment!

Peter Lu

Dear Thomas,

Thanks for your help! I’m able to get a similar binary image as you did with Fiji. However, I though the pipeline recognised the mid black dots as cells, which are false positive.

It should be the reasons of the quality of the image. I should be able to repeat the experiment to get better quality of images.

Peter Lu

1 Like

Dear Peter,
if it is possible to redo the experiment and achieve a higher image quality, that’s for sure the best option to get reliable results. An alternative, potentially even on these data, but of course also for new data, could be our ScratchAssayAnalyzer, http://mitobo.informatik.uni-halle.de/index.php/Applications/ScratchAssayAnalyzer, developd by @markusglass. It is based on entropy analysis and uses a level set approach to localize the wound region in such kind of images. The advanced segmentation method helps to get the tissue regions as a whole and avoids oversegmentation.

On one of your images I got the following prototypical result - without doing any parameter tuning that might help to better define the boundaries and improve the result.



1 Like

I think PixelClassification could also give good results in this case. The “blue” area looks what you want, right?

1 Like

On the FIJI track, I just wanted to throw out there that a 8 pixel median filter prior to using the variance filter cleans up a lot of the dirt.

1 Like

Cool, I’m also using Zen.


Hi Guys,

I think I could use IdentifyObjectsManually to select the area of the cells if the quality of images were still not good enough.

Peter Lu

1 Like

If possible I would really try to avoid any “manual” methods, wherever possible… :wink:


Dear Peter_Lu,

You could try to segment using Ilastik. Make select “Cells” as label 1 and “Wound” as label 2.
Once segmented , import the simple segmentation maps in Fiji. Use Process>Math>Multiply by 100 and you get an image that can be used in Analyze Particles. Define a minimum size as 20000 and you’ll get your wound selected. All the small dots in the middle will be discarded.
I get this:

As a result, I get the wound area.

It may be a good idea to crop your image to get rid of the top and bottom 10%, or retake your images…

Good luck!

1 Like