Question: How does Xenium perform cell segmentation?
Answer: Xenium includes a cell segmentation step as part of onboard analysis. For the most accurate cell segmentation, particularly in dense regions, Xenium takes a deep learning approach. There are a few components to this:
- Staining. Xenium employs a DAPI nuclear stain to infer cell boundaries. Additional staining capabilities will be added to the platform over time.
- Training data. Xenium uses curated sets of data to train the algorithms to ensure high quality for supported tissue types.
-
Algorithms. Xenium comes with a core set of algorithms that operate on top of staining. This approach is conceptually similar to Cellpose and Omnipose, but with unique approaches to the implementation of the neural network and post processing.
- Specifically, the reference morphology image is segmented in multiple planes across the z-stack image. The nuclear boundaries are consolidated to form non-overlapping 2D objects when projected in XY in order to produce 2D masks. Masks are then expanded by 15 µm or until it encounters another cell boundary in XY to provide approximate cell segmentation. If a transcript falls inside this boundary it is assigned to a cell. Finally, simplified polygons are made for ease of use in visualization.
Xenium strives to provide robust segmentation results that can be relied upon for supported tissues. The Xenium roadmap includes improvements to trained models as new tissue panels are supported, as well as robust staining approaches to achieve full cell boundaries. Stay tuned for updates to features!
Other Resources:
- Pre-Print: High resolution mapping of the breast cancer tumor microenvironment using integrated single cell, spatial and in situ analysis of FFPE tissue
- Analysis Guide: Performing 3D Nucleus Segmentation With Cellpose and Generating a Feature-cell Matrix
Products: Xenium In Situ Gene Expression