Launch binder

Extract custom features

This example shows how to extract features from the tissue image using a custom function.

The custom feature calculation function can be any python function that takes an image as input, and returns a list of features. Here, we show a simple example by defining a function to calculate the mean of the images.

Custom features are calculated by using features = 'custom', which calls squidpy.im.ImageContainer.features_custom(). In addition to feature_name and channels we can specify the following features_kwargs:

  • func - custom feature extraction function.

  • additional_layers - names of image layers that should be passed to func together with layer.

  • additional keyword arguments for func.

import scanpy as sc
import squidpy as sq

Let’s load the H&E Visium dataset.

# get spatial dataset including high-resolution tissue image
img = sq.datasets.visium_hne_image_crop()
adata = sq.datasets.visium_hne_adata_crop()

Define a custom feature extraction function.

def mean_fn(arr):
    """Compute mean of arr."""
    import numpy as np

    return np.mean(arr)

Now we can extract features using mean_fn by providing it within features_kwargs.

sq.im.calculate_image_features(
    adata,
    img,
    features="custom",
    features_kwargs={"custom": {"func": mean_fn}},
    key_added="custom_features",
    show_progress_bar=False,
)

The result is stored in adata.obsm['custom_features'].

adata.obsm["custom_features"].head()
mean_fn_0
AAAGACCCAAGTCGCG-1 101.868998
AAAGGGATGTAGCAAG-1 102.959180
AAAGTCACTGATGTAA-1 107.924715
AAATGGCATGTCTTGT-1 97.444851
AAATGGTCAATGTGCC-1 102.854690


Use squidpy.pl.extract() to plot the histogram features on the tissue image or have a look at our interactive visualization tutorial to learn how to use our interactive napari plugin.

sc.pl.spatial(
    sq.pl.extract(adata, "custom_features"),
    color=[None, "mean_fn_0"],
    bw=True,
)
mean_fn_0

You can also pass more than one image layer to the custom feature extraction function. For this, specify the necessary additional layer names using additional_layers in features_kwargs. The specified image layers will be passed to the custom feature extraction function.

Here, we show this behavior by defining a feature extraction function that sums two image layers:

def sum_fn(arr, extra_layer):
    """Compute sum of two image layers."""
    import numpy as np

    return np.sum(arr + extra_layer)


img.add_img(img["image"].values, layer="extra_layer")

sq.im.calculate_image_features(
    adata,
    img,
    layer="image",
    features="custom",
    features_kwargs={"custom": {"func": sum_fn, "additional_layers": ["extra_layer"]}},
    key_added="custom_features",
    show_progress_bar=False,
)

Total running time of the script: ( 0 minutes 28.075 seconds)

Estimated memory usage: 299 MB