Tables and relations

Every label image carries a table, one row per object, stored with the labels (image.zarr/labels/<name>/table): size (area_voxels, area_um3), centroid, bounding box, shape (length, elongation, main axis), and mean and standard deviation of intensity for the channels you ask for. The cluster workflow writes it during the merge, from what the segment jobs measured per tile, without reading the labels again.

from patchworks import read_table

cells = read_table("results/image.zarr/labels/cyto_labels")  # pandas DataFrame

For a store without tables, or to add intensities:

patchworks tables results/image.zarr --channels 0,1

Relating two label images

Which cell is each nucleus in? label_relations maps every object of one label image to the object of another it overlaps most, reading both chunk by chunk:

from patchworks import label_relations

table = label_relations(
    "results/image.zarr/labels/nuclei_labels/0",
    "results/image.zarr/labels/cyto_labels/0",
    a_component="",
    b_component="",
)
table[2]  # {'match': 3, 'overlap_voxels': 4821, 'overlap_fraction': 0.94}

The two must be segmentations of the same image at the same chunking. To relate several children to one parent, reading it once, patchworks.label_relations_many({"nuclei": ..., "cilia": ...}, cells).

To keep the relation in the tables instead, as columns of the child's table (cyto_labels_id, the overlap), where patchworks review corrections apply to it:

patchworks tables results/image.zarr --relate nuclei_labels:cyto_labels
patchworks tables results/image.zarr --relate cilia_labels:cyto_labels --max-distance 1.0

--max-distance (µm) gives an object touching no parent the nearest one within that distance. The cluster workflow does all of this from the relations: of a multi config, and writes a workbook per pair.

Exporting

patchworks review results/image.zarr --export tables/ --format parquet  # or csv, xlsx

writes every table with the review corrections applied. napari-chunked-regionprops (in patchworks[napari]) reads them, and measures any label layer in napari out of core.