Segment closely packed branching cells from 3D microscopic images

Background

I’m trying to segment cells from this 3D microscopic image, via skimage.filters.apply_hysteresis_threshold but I get this clump instead due to close proximity of the cells.

Analysis goals

  • My goal is not to compromise any branching information of the cells and still get single segmented cells.

Challenges

  • I’ve tried increasing the thresholds but it compromises a lot of branching of cells which I don’t favor.
  • I’ve thought of applying edge filter and then using it as bounds of the branches of single cells, but the edge filters also don’t capture the thin branches well and compromise a lot of detail.
  • Bottom part of image might be connected due to blood vessel being an artifact, but the above 2 cells are of good quality and are what I hope to extract.
  • I need suggestion for maybe some post-processing technique to dissociate these cells from the clump to achieve my goal

Hi Gupta,

Could you find a way for us to upload your original image? The resolution on this one is very poor…

Sincerely,

Matthieu

1 Like

Hi Matthieu

Thanks for replying

This is the Region of Interest cropped from my original image only. The original image is of same quality. I wanted to share this problematic region specifically. Due to close proximity of cells, the signal diffuses and becomes problematic for segmentation.

Still, if it changes anything, here’s the original full image: C1-M1_LEFT _HILUS.tif - Google Drive

ROI cuboid bounds (Z, Y, X): (25, 404, 455), (59, 541, 580)

I have tried running Cellpose 3D segmentation with different parameters on this data but it doesn’t provide results even comparable to ones I got via thresholding. Maybe this structure of cells isn’t suitable for it