Regional properties of label

In this notebook we charcterize labels according to the mean and standard deviation of their propteries, such as size. If similarly sized objects are neighbors, the standard deviation of their size is low. If labels of different size are neighbors the standard deviation of their size is higher. This could be used to identify regions in tissues where cells of different size meet.

from skimage.io import imread
import pyclesperanto_prototype as cle
import stackview

The data we use here was derived from maize_clsm.tif was taken from here, an image shared by David Legland under CC-BY 4.0 license

image = imread("../../data/maize_clsm.tif")

stackview.insight(image)

shape (640, 639) dtype uint8 size 399.4 kB min 0 max 255

A corresponding label image looks like this:

labels = imread("../../data/maize_clsm_labels.tif")
labels = cle.exclude_small_labels(labels, maximum_size=200)
labels = cle.extend_labeling_via_voronoi(labels)
labels

cle._ image shape (640, 639) dtype uint32 size 1.6 MB min 1.0 max 254.0

Measuring size

First we need to quantify size of the objects. We can immediately visualize these measurements as parametric image.

size_map_image = cle.pixel_count_map(labels)

size_map_image

cle._ image shape (640, 639) dtype float32 size 1.6 MB min 201.0 max 14283.0

Regional properties

We can now summarize those measurments locally, e.g. by measuring the mean size of every cell an its corresponding touching neighbors.

cle.mean_of_touching_neighbors_map(size_map_image, labels)

cle._ image shape (640, 639) dtype float32 size 1.6 MB min 369.33334 max 7611.5

We can also compute the standard deviation of size, which highlights the borders between the regions with cells of different size.

cle.standard_deviation_of_touching_neighbors_map(size_map_image, labels)

cle._ image shape (640, 639) dtype float32 size 1.6 MB min 43.538486 max 3917.1946