Could you perhaps supply either the original file or a multi-channel image. The tiff here is unfortunately RGB and not usable...

The workflow looks solid so far, but I have some small comments:

- I would move step 2 between 5 and 6. So also threshold the debris channel and subtract the two binary images. It's very hard to ensure that the intensity values of the debris particle in both channels is exactly the same and if they are not, you might introduce a mistake while solving another

- Step 4 can be averted by checking "dark background" in the thresholding window

- Manual thresholding is not recommended, especially when measuring the fluorescence. This way, any differences you measure in fluorescence intensity and area might just be bias that was introduced during this step. There are many different thresholding methods available apart from default. The Plug-In BioVoxxel can help you with choosing one (never tried it though, this is a recommendation of my former Master student). If none of the automated thresholds work, you should reconsider the preprocessing instead.

- Are you set on measuring the fluorescence intensity? The signal-to-noise ratio in your images is rather bad, so intensity may not be the best option

Some thoughts on the high background and thresholding:

- Your tile images have uneven shading, which causes troubles for automated thresholds, because it messes with the histogramm. Does your microscope provide shading correction? There are e.g. autofluorescent Chroma slides which help you with that.

- The black values around the circle of the well might be another problem for the automated thresholding, because of shifting the histogramm. If there is only 0 as color value in these parts, you could use the color picker, choose representative background and assign the zeroes background values instead. You can also measure a defined area of your background and assign the zeroes the averaged or mean background value by the Flood Fill Tool.

- A more elaborate way to improve the images by noise correction is using deep learning. No worries, there already exist plug-ins for this, e.g. Noise2Void --> https://imagej.net/N2V

Install it by Help --> Update... --> Manage Update sites --> CSBDeep.

If you want to train on a GPU (way faster) you also need Tensorflow (and a decent graphics card), but they give a nice step-by-step instruction on their website. I played around with normal CPU and it also worked, you just need to have it run over night/ the weekend.

*Edit: you probably need to solve the uneven shading before using Noise2Void. The tool is created for *independet* noise

Good luck!

Thank you so much for your answer! This helps me a lot. Very excited about Noise2Void. The uneven shading is what bothers me the most, but hopefully these steps will get me further. I'm glad my steps were in the right direction!

Hello

It's really an interesting project. I have some experience with this condition as a Civil Engineer, In my project I faced with similar situations and problem.

I worked with both ImageJ and MATLAB for Image processing; first of all try your operations with MATLAB, It's truly powerful and results aren't the same for a certain operation on a single Image with your results from ImageJ.

Your image quality is good enough for capturing what you want. I recommend that you first do an image enhancement before thresholding. When proteins edges aren't very clear, you need to make them easy to observe. There are also various methods for edge enhancement and you need to try them and find the best one for your work.

You could remove red debris by their size or colour. Bandpass filters are good for removing red and keeping green. you need to just define two frequencies for this filter.

Additionally for the thresholding step, I think Adaptive threshold method is better for you.

For the final step you could add a counter for counting green particles in images or just use the absolute sum of green pixels in your images.

Best Wishes