Skip to content

Segmentation and single-cell quantification

Once the ST and imaging modalities have been aligned, you can segment the images into single cells/nuclei, and then aggregate the spot locations into individual cells for subsequent analysis.

Cell segmentation from tissue image

Let's create a new Open-ST h5 object containing a cell-by-gene expression matrix. First, you will need a cell (or nuclear) segmentation mask.

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     segment \
     --model HE_cellpose_rajewsky # default model for segmentation of H&E images

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     segment \
     --model HE_cellpose_rajewsky \
     --image-in uns/spatial_pairwise_aligned/staining_image_transformed

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     segment \
     --model HE_cellpose_rajewsky

We segment cells (or nuclei) from staining images using cellpose. We provide a model that we fine-tuned for segmentation of fresh-frozen, H&E-stained tissue, here, but you can use any other model (e.g., pretrained from cellpose, like cyto2 or nuclei, or your own). Also, by default, segmentation is extended radially 10 pixels (see --dilate-px), to account for cytoplasm surrounding the nucleus as a first approximation of cell shape (might hold or not depending on the tissue).

I want to segment very small and very large cells...

You can perform an additional round of segmentation by, e.g., adjusting the diameter parameter.

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     segment \
     --mask-out uns/spatial/staining_image_mask_large \
     --dilate-px 50 \
     --diameter 50 # diameter for the larger cell type

In this case, we changed --mask-out to a different key, so we can keep both masks inside the Open-ST h5 object.

Then, you can combine the segmentation masks of both diameter configurations. This command will apply an "AND" between all images, to only preserve mask of non-overlapping, with the hierarchy provided in the --image-in argument (first has higher priority).

openst segment_merge \
     --h5-in spatial_stitched_spots.h5ad \
     --mask-in uns/spatial/staining_image_mask uns/spatial/staining_image_mask_large
     --mask-out uns/spatial/staining_image_mask_combined

Quality control of segmentation

You can assess the quality of segmentation with openst preview:

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     preview \
     --image-key uns/spatial_pairwise_aligned/staining_image_transformed uns/spatial/staining_image_mask

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     preview \
     --image-key uns/spatial/staining_image uns/spatial/staining_image_mask

This will create a napari window with two image layers. Change the mask image layer into a label layer, which is designed for displaying each integer (ID from the segmentation mask) as a different random color, with background rendered as transparent.

If you are satistied with the quality of the segmentation, you are all set to continue with single-cell quantification.

Single-cell quantification

Then, you can create a single file containing the transcriptomic information aggregated into (segmented) single-cells.

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     transcript_assign \
     --spatial-key obsm/spatial_pairwise_aligned_fine \
     --mask-in uns/spatial_pairwise_aligned/staining_image_transformed

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     transcript_assign \
     --spatial-key obsm/spatial_manual_fine \
     --mask-in uns/spatial_pairwise_aligned/staining_image_transformed

openst from_spacemake \
     --project-id openst_demo_project \
     --sample-id openst_demo_sample \
     transcript_assign \
     --spatial-key obsm/spatial_manual_fine

Expected output

After the steps above, you will have a single h5ad file with transcriptomic information per segmented cell, with spatial coordinates aligned to the staining image. The staining image and the segmented image are provided in this object, so it is possible to visualize it with squidpy or spatialdata, among other tools.

Warning

In the Open-ST h5 object, the cell with ID 0 will correspond to the background. Please remove it before proceeding with analysis.

This concludes the preprocessing of 2D spatial transcriptomics and imaging data of the Open-ST protocol. Next steps include 3D reconstruction, and downstream analysis of nD data.