force detect_foreground to pick up small cells

[jazberna1]

Hello,

This not really an issue. It happens to me that I have very small objects that are not picked up by the detect_foreground function.
Is there a way to force that smaller cells are detected? I have played with different values of sigma and min_foreground but that made of effect on picking up those small objects. This is a Napari screenshot of my data there I mark a detected cell (blue arrow) and non-detected cell (red arrow).

Thanks!!!
Jorge

[JoOkuma]

Hi @jazberna1, if they are smaller than config.segmentation_config.min_area they will be ignored.

Sorry for the delay, I was on vacation.

[jazberna1]

Hi @JoOkuma,

Thanks for answering back to me.

I have tried with config.segmentation_config.min_area = 4
but then I get this error when running the 'Auto detect foreground and contours from image' workflow
sys:1: RuntimeWarning: RuntimeError in aborted thread: Region too small. Size of 1 found.

Should I change something else in the configuration to avoid the error?

Thanks
Jorge

[JoOkuma]

@jazberna1 could you increase the min_area a bit.
I forgot about a recent change where min_area removes everything that's smaller min_area / 4

[jazberna1]

Hello @JoOkuma ,

I got it to work with min_area = 8, below that I get the RuntimeError in aborted thread: Region too small. Size of 1 found.
These are my combined foreground and contour layers. I still see cells not detected but I suppose if reached already the minimun area that Ultrack can detect I would have to preproces my images somehow, right? i.e. upscaling.

Thanks for your help
Jorge

[JoOkuma]

@jazberna1 Since you're detecting other small objects, I think they are not being selected because the ILP considers them false objects.

Have you checked if the movement is within linking_config.max_distance?
If that's the case, I suggest tuning the tracking_config weights.

We provide some guidelines here, https://royerlab.github.io/ultrack/optimizing.html

[jazberna1]

Hi @JoOkuma,

Yes, I agree I am detecting objects that seem to be as small as the ones that I don't detect.

I undestood however that cells not included in the detection layer are not tracked no matter how I set the tracking parameters, right?

Nevertheless I checked the movement is within my setting of linking_config.max_distance. If I start setting it higher I eventually see the tracking links different cells.

Then I made more negative the appear, disappear and division tracking weights. I see those changes altered the tracking of the already detected cells but still have most of them undetected. Is there any other place you could think I am missing those cells? Any advice will be much appreciated.

Thanks
Jorge