Artifacts using primary object segmentation with shape declumping option

I am segmenting images from live cell imaging experiments which have low image contrast in the nuclear channel. I typically use intensity for both of the following steps “method to distinguish clumped objects” and “method to draw dividing lines between clumped objects”. These options work fairly well for images with strong intensity gradients from the center to the perimeter of the nuclei. So, through my testing I found that shape works better than the intensity option with these images. However, I find that I get pie shaped notch artifacts in the nuclei with this declumping option. Is this a bug in the segmentation algorithm or is it to be expected from shape declumping? Here is an example segmentation pipeline using shape declumping for you to review. Do you think there are any better segmentation settings to use for this type of image?

7-8-16_LOGWithShapeDeclumping.cpproj (616.4 KB)

images.zip (2.7 MB)

Uploading…

Yeah, shape is definitely prone to that. If they’re fatal for your downstream application I got decent declumping with setting distinguish to Shape and lines to Intensity; you could also maybe try something like erode the nuclei by ~10 pixels then re-expand them. Good luck!