Specify prior segmentation

[ConstensouxAlexis]

Hello, I have 2 questions regarding the use of nucleus segmentation as prior segmentation:

1. In Merscope data, I do not see any information on nucleus in the detected_transcripts.csv; in this case, do you use the prior cell segmentation as baseline for ProSeg ?
2. In Xenium data, the old segmentation with nucleus was indeed very bad, but the new multimodal segmentation is much better. Do you plan to add the option to chose which prior segmentation can be used as baseline for ProSeg ?

Thank you very much ! Alexis Constensoux

[dcjones]

Hi Alexis,

1. That's correct, there's no compartment information in MERSCOPE, so the whole cell is used. In my paper I resegmented nuclei using cellpose to for some analyses.
2. Proseg can be informed by both. You can try setting --prior-seg-reassignments-prob: this is 0.5 by default, but if you set it to less than 0.5 in will penalize deviation from the whole cell segmentation. I haven't done enough testing to know if this improves things in general, but it's reasonable thing to try on platforms with good membrane staining.

My plan is to implement more precise pixel-level priors in a future version, but that's a little ways away.

[ConstensouxAlexis]

Thank you for your answer ! I will try playing with --prior-seg-reassignments-prob parameter on my Xenium data.