FIJI: Segmentation + particle analysis of noisy images with uneven staining

Does anyone have advice on segmenting images like the ones I have attached? The problems start from the segmentation, when I try to manually set a threshold, but it inevitably captures the background or excludes the lower intensity parts of the cell.

Capture Capture_uneven_bg

I know I can manually draw a ROI around each cell, but I am curious if there is a way to deal with this automatically. On the Analyse particles plug-in, I tried to play around with the circularity (all my cells should have a pretty round shape which helps a lot) and by area size of course, but it still doesn’t detect only the cell.
They also have pretty uneven background, which I can try to deal with separately, but I doubt this will solve the issue with the uneven staining in the actual cell.

Any help will be appreciated

1 Like

I would generally start with some kind of blur, usually gaussian, and then if I need a measurement, take the analyze particles ROI that is generated and move that selection back onto the original image.

A median filter might also reduce some of the background noise, but you will likely want to try a few different options to see what gives you the most accurate result. If the images are much larger than the ones you have shown, and the cells are relatively sparse, you might be able to try a “rolling ball” background subtraction, though that can run into problems if you have clumps of cells close to the size of the ball.

1 Like

Hey thanks for the quick response :slight_smile:
I managed to get it to work on one image using a suitable area size threshold for the Analyze particles, I guess I miscalculated the area range the first time I tried it.
But I fear this approach alone won’t be consistent for all my images.

Thanks a lot, I totally forgot how useful filtering can be :slight_smile: I will give all of them a go.
I will have one cell per image only, I decided to crop them one by one. Because on the large field of view the illumination is very uneven, as is the background, so it would be close to impossible to analyse an entire FOV of cells.

Hi @Isq,

The ideal tool would be a level set that you initialyze at the centre of your image and make grow until it touches the edges of the cell. But I don’t know if there is a good plugin for this in imageJ.
Does someone knows ?


1 Like

Hi Isq,
Copy the image - Invert the image (Edit>Invert) - Subtract the copy from the original.
Twice the dynamic range.

1 Like

@VirtualSlide, that’s a great idea, I guess the tricky part would be for it to know where is the edge of the cells (especially since the staining is so patchy). I will look it up, thanks!

A combination of filtering and auto local threshold will probably resolve your issues. If you want to post a complete sample image I’m sure someone can help.
cell-2.tif (692.2 KB)

1 Like