Recommendations of tools for segmentation of 10x Xenium data in human brain

I'm awaiting some 10x Xenium data from human brain (glioma patients). I'm trying to figure out the strategy for segmentation as I know the default segmentation obtained from 10x's pipeline doesn't do a fantastic job of segmenting brain due to the morphology of cells.

I've identified a number of tools: GeneSegNet, Baysor, Cellpose, StarDist and a few others. While I understand that these types of analyses require a fair amount of optimisation and trying a few different tools, I also have quite a lot of data to analyse with some time constrations so I don't want to spend too much time going down a rabit hole of benchmarking segmentations.

Can anyone recommend some tools they have worked with that gave them good results on human brain? I'm looking at human material with a mix of healthy and cancer cells and have opted for the segmentation staining package that 10x provide.

Thanks in advance

xenium spatial-transcriptomics imaging single-cell

• 417 views • link

updated 5 months ago by Bastien Hervé

• written 5 months ago by yura.grabovska •

1 answer

The segmentation with the staining package is working well on healthy cells. The segmentation problem arise when the tissue is not healthy anymore and where cells are aggregating, making it impossible to even delineate cells by eye anymore. We have tried Baysor and Cellpose on MS lesions which were unable to characterize cells in these dense regions. However, elsewhere on the tissue, both methods are working quite fine. For these problematic regions, we are trying a transcript centered approach not looking at cells but transcripts proximity.

• 0 views • link

written 5 months ago by Bastien Hervé