Bioimage Analysis in Jupyter Notebooks#
In this notebook, we will load an image stored in the file data/lund.tif, visualize it, process it using pyclesperanto for segmentation, and finally visualize the segmentation results.
We will use these libraries
We typically import all libraries we need at the beginning of a notebook to check if everything is installed.
from skimage.io import imread, imsave
import stackview
import pyclesperanto as cle
import numpy as np
Loading an image#
In this step, we will load the image file named lund.tif using the imread function from the skimage.io module. This function reads the image data into a NumPy array which can then be processed or visualized in the subsequent steps. For example, we can print out the size (in pixels or voxels) of the image.
image = imread('data/lund.tif')
image.shape
(100, 256, 256)
Visualization#
There are plenty of options and libraries for visualizing images. Here we will use stackview, a library specifically developed for working interactively with image stacks in Juypter notebooks.
stackview.insight will do a projection of our 3D image and visualize it next to some descriptive statistics of the image stack.
stackview.insight(image)
|
|
We will use stackview.animate to display the loaded image slice-by-slice. This method animates through the z-stack slices, allowing us to visually inspect the image frame by frame. For performance reasons, we only display every 4th slice.
# Visualize the image as animation
stackview.animate(image[::4])
C:\Users\rober\miniforge3\envs\embo25\Lib\site-packages\stackview\_animate.py:43: UserWarning: The timelapse has an intensity range exceeding 0..255. Consider normalizing it to the range between 0 and 255.
warnings.warn("The timelapse has an intensity range exceeding 0..255. Consider normalizing it to the range between 0 and 255.")
While the view above renders in the browser in teaching materials, the next will not. The following interactive tool needs to be executed on your computer to see something. We will use stackview.slice to navigate through the image stack one slice at a time and change the slice-position manually.
# Visualize the image as a slice
stackview.slice(image)
Image processing#
For demonstration purposes, we will segment this image. For learning more about image pre-processing, segmentation, post-processing and feature extraction, please refer to the BioImageAnalysisNotebooks.
We will use the pyclesperanto library, a GPU-accelerated Python library to speed-up processing of 3D data.
First, we apply the top_hat_box filter, which helps to enhance the image by removing uneven background intensity.
# Apply top hat filter to the image
processed_image = np.asarray(cle.top_hat_box(image, radius_x=10, radius_y=10, radius_z=0))
stackview.animate_curtain(image[50], processed_image[50])
Image Segmentation#
This step involves segmenting the processed image using the voronoi_otsu_labeling function from pyclesperanto. This segmentation technique combines Voronoi tessellation and Otsu’s method to effectively separate foreground objects from the background in a semi-automated manner, resulting in a labeled image with distinct segments for each object.
# Segment the image using voronoi otsu labeling
label_image = cle.voronoi_otsu_labeling(processed_image, spot_sigma=2, outline_sigma=2)
label_image
|
cle._ image
|
We can also view this label image over the background-subtracted image using a curtain.
stackview.animate_curtain(processed_image[50], label_image[50])
We can determine the maximum number of objects by taking the maximum intensity in the label image, because the labels are numbered subsequently from 1-n.
label_image.max()
285.0
Interactive browsing the resulut#
We can use stackview.curtain the original image side-by-side with the segmented labels. This interactive visualization allows for comparing the original structures with their respective segmentation directly by dragging a curtain across the images.
# Visualize the original image and its label image using curtain
stackview.curtain(image, label_image)
Saving results#
Finally, we save the resulting label image to disk. Note: When using segmented images, use file-formats such as .tif which can easily be opened again in other software and preserved the full information in the image. Avoid .png and .jpg for saving scientific data.
imsave("data/Lund_labels.tif", label_image)
C:\Users\rober\miniforge3\envs\embo25\Lib\site-packages\skimage\_shared\utils.py:328: UserWarning: data/Lund_labels.tif is a low contrast image
return func(*args, **kwargs)