scallops.segmentation.watershed.segment_cells_watershed

scallops.segmentation.watershed.segment_cells_watershed(image, nuclei, threshold='Li', cyto_channel=None, threshold_correction_factor=1, quantile_threshold=0.2, nuclei_channel=0, rm_small_std=None, rolling_ball=False, sigma=None, closing_radius=None, t=0, chunks=None, depth=30, watershed_method='distance', at_least_nuclei=True)

Segment cells using watershed algorithm, matching cell labels to nuclei labels.

Parameters:

  • image (DataArray) – Image with the dimensions (t, c, z, y, x)

  • nuclei (ndarray) – Labeled segmentation mask of nuclei. Uses nuclei as seeds and matches cell labels to nuclei labels.

  • cyto_channel (int | Sequence[int] | None) – Index or list of indices for cyto channel(s). If not provided, all channels except for nuclei_channel are used.

  • threshold (Literal['Li', 'quantile', 'Otsu', 'Local'] | float) – One of Li, Otsu, quantile, or float value If Li, use Li’s iterative Minimum Cross Entropy method threshold_li(). If Otsu, use Otsu’s method threshold_otsu(). If quantile, use threshold_quantile()

  • threshold_correction_factor (float) – Factor to adjust the computed threshold by.

  • quantile_threshold (float) – Quantile when threshold is ‘quantile`

  • threshold – Threshold to apply to cell mask.

  • nuclei_channel (int) – DAPI channel index

  • rm_small_std (float | None) – If specified, remove objects smaller than specified number of standard deviations of identified nuclei

  • rolling_ball (bool) – If true, apply skimage.restoration.rolling_ball subtraction to mask prior to calculating threshold

  • sigma (float | None) – sigma (optional) gaussian filter sigma to smooth the image prior to computing threshold

  • closing_radius (int | None) – (optional) disk radius for closing cell labels

  • t (list[int] | int | None) – Optional list of time indices (0-based) to use for computing cell mask

  • chunks (tuple[int, int] | None) – Chunk size to perform segmentation using non-overlapping chunks.

  • depth (int) – The number of elements that each block should share with its neighbors when using dask

  • at_least_nuclei (bool) – Ensure cells occupy at least the same pixels as nuclei

  • watershed_method (Literal['binary', 'distance', 'intensity']) – Watershed method: If distance, use distance transformation of the binarized mask. If intensity, use inverted cytoplasm intensity image (same as CellProfiler WATERSHED_I) If binary, use distance from nuclei as in publication The phenotypic landscape of essential human genes <https://pubmed.ncbi.nlm.nih.gov/36347254/>.

Returns:

Cell labels matching nuclei labels and thresholds if an auto threshold method is used.

Return type:

tuple[ndarray, float | None]