Dear Konstantin,

I ran your sample image through the PHANTAST Toolbox mentionned by Melissa and Nicolas, please see the attached outlined image. It found a confluency of 22% for that particular example. As Nicolas said, if you want to explore that approach further, please feel free to contact us.

Best

Alexandre

Hi Antonios,

thank you for this idea,

do you have any further information for me how this method works or how to arrange this in ImageJ (as i'm really new to ImageJ and Image analysis at all).

Best, 

Konstantin

ok, lets say i found a macro to evaluate contrast. What does a value of 47.53 mean in relation to 58.32? Is it, the higher the contrast value, the higher my confluency?

Because of the phase contrast, finding edges can be more difficult. I use ImageJ so I'll give you some steps in there to try first. At the end of this answer are some links to good ImageJ and MatLab plugins that are also worth a try. 

ImageJ steps:

1. Your example image downloaded as 3 channels so: Image>Color>Stack to RGB and only use the single image (not the stack)

2. Image>Type>16-bit

3. Process>Subtract Background ... rolling ball status = 15 (or 10 if too much noise) and check Light Background

4. Image>Adjust>Threshold ... Areas that will turn black after thresholding will be highlighted either red or black, make sure Dark background is unchecked, click Apply  (Note: Other thresholding options are available if this is too noisy)

5. If you want to make sure that the cell separation stays intact then go to Process>Binary> Watershed ...you will see lines appear where the cells are separate

6. Go to Analyze>Set Measurements...check all of the boxes you would like the program to measure. You probably want to include Area Fraction as it is the percentage of pixels in the image/selection that have been thresholded.

7. Analyze>Analyze Particles (Note: If you do not set the scale {Analyze>Set Scale} then it will just measure in pixels)...options in this menu include parameters for size, circularity, outlines, etc. Chose those that are appropriate for your measurement. Click Ok to get a Results window that can be saved.

Some plugins that may do an equal or better job are:

PHANTAST Toolbox (see paper below, also)

https://github.com/nicjac/PHANTAST-FIJI/   (note: Fiji plugins can be used in ImageJ)

There is a MatLab version of this as well - https://github.com/nicjac/phantast-matlab

[or]

Phase Contrast Cell Analysis Tool

http://dev.mri.cnrs.fr/projects/imagej-macros/wiki/Phase_Contrast_Cell_Analysis_Tool_(Trainable_WEKA_Segmentation)

Good luck!

-Melissa

Article Automated Method for the Rapid and Precise Estimation of Adh...

Dear Konstantin

We are still maintaining the code mentioned in the BnB paper cited by Melissa. We could give it a go if you want to send us a sample image. Both my co-author and I may be on holiday (i am), so maybe the response takes a bit of time. bests, Nicolas

Dear Konstantin,

I ran your sample image through the PHANTAST Toolbox mentionned by Melissa and Nicolas, please see the attached outlined image. It found a confluency of 22% for that particular example. As Nicolas said, if you want to explore that approach further, please feel free to contact us.

Best

Alexandre

Wow Thank you guys for all this help! I really appreciate it.

I'll definitely go through your paper, Nicolas and Alexandre!

I will tell you guys if everything worked out!

Thank you very much!!!

Glad we could be of help. bests, Nicolas

Hallo again Guys,

first of all - You guys did a great job with that plugin! It works terrific! 

But, unfortunately i have some pictures or lets say a stimulant which creates some probs. Here is a sample picture. Could you give me some advice how to use the sigma and epsilon parameter to calculate the sample pic, and could i use the same values for sigma and epsilon parameters for pictures without the precipitate?

Hi Konstantin,

I played a bit with the parameters, by increasing the scale to about 3 you get a better result (see image attached). The rule of thumb is basically the following: if the outline is mostly inside the cell boundaries, increase the scale, if it overshoots outside, then decrease it. I usually do not bother too much with the treshold parameter. But I will ask the co-author that developped the method if he can provide you with a more complete and accurate explanation of what the parameters do exactly. Regarding your sample image, I would also say that most of the issues probably stem from the quality of the image. Cell boundaries seem a bit blurrier than on your other sample. I do not know if it is a focusing issue or if the medium is murky but if you can improve on that point, the algorithm should perform much better. 

Best

Alex

Hey Alexandre,

thank you for running my picture through your plugin. It's way better than my confluence identification of this picture. I will show you a video which will my make my problems more clear. In the following video you see some black dots appearing after a few video-seconds. This "dots" have a very big impact on the estimated confluence. It might be right, to change the scale of my pictures to get a better result. Could you tell me how to change the scale?

I am very sorry for all my questions, i am really new to image analysis and processing.

Best,

Konstantin

Hi Konstantin,

so to change the scale just click on settings and it will open a new window with all the parameters to tweak (see image attached). For your small black dots, you can use the setting "remove small objects" under "Additional processing" to discard objects under a certain area (you can adjust the value of the area too). Obviously that will only work as long as these dots do not form aggregates of the same area than a cell. 

Don't worry your questions are more than welcome. We are glad that you are interested in using the method we developed!

Best,

Alex

Hi Alexandre,

Where can I download PHANTAST independent app?

Thanks