Allen Cell Discussion Forum

3D image analysis after classic segmentation

Hi everyone,
I am new to aics-segmentation. I have two questions about the classic segmentation pipepline.

  1. How to avoid the over-segementation. like the image shown below, signals in top and bottle layers seem to be processed as background or noise. And the structure seems incomplete.
  2. I am also wondering if there is an image analysis pipeline. I am trying to analyze original images by use the segmentation as mask, like what people ususally do in Fiji.

Thanks a lot.

@ashapeng , thanks for your interest!

  1. it is important to avoid over-segmentation. “How to avoid” really depends on the image and the algorithm you use. For example, if you uses spot3d filter or filament3d filter in segmenter, reducing cutoff value is a simple way to avoid over-segmentation. Based on the image above, I cannot quite tell the issue. It would be more helpful if you can label which part exactly you mean by over-segmentation.

  2. for analysis, we are taking a different approach than Fiji. We don’t use graphical interfaces. We have a lot of analysis related use cases, but all implemented in pure python. The main reason is that we want to have all the flexibility to do any analysis most suitable for the problem. You can find one example here:
    aics-segmentation/bridging_the_gap_between_binary_image_and_analysis.ipynb at main · AllenCell/aics-segmentation · GitHub


Hi Jianxu,
thanks a lot. I used the spot3d filter for my tissue(nucleolus). I tried to manipulate the cutoff(lower the value, because initially I used a higher value to reduce the noise but yielded less segmentation) and sigma(also lowering it, I had used a higher value, I guess it sacrified my resolution to distinguish smaller objects ), it seems getting better. The “over sgemented” part is like this, I circled in the new view.

the top plane seems to be cut off, and the structure looks like incomplete. Wondering could this be that I didn’t cover the the whole range of structure when taking images.
2. For the second question, thanks for the suggetion, I will play with the script and modify based on my cases.