[BioImage Analysis]: Nuclei segmentation

[haesleinhuepf]

Analysis Goal (What should be done / analysed?)

I would like to segment the nuclei in this image, e.g. using Voronoi-Otsu-Labeling.

Image Upload

📎 Drag & drop your microscopy image here (JPG, PNG, GIF, e.g. 512x512 pixels, 2D only).

Python Tools

• List of Python libraries we should use for answering this:
  • numpy
  • scikit-image
  • napari-segment-blobs-and-things-with-membranes

Detailed instructions for bio-image analysis using Python (feel free to modify)

## Detailed Python Bio-image Analysis instructions If the following tasks are requested, we can adapt the code corresponding snippets: ### Viewing images using stackview When you use stackview, you always start by importing the library: `import stackview`. * Showing an image stored in variable `image` and a segmented image stored in variable `labels` on top with animated blending. Also works with two images or two label images. stackview.animate_curtain(image, labels) * Showing an animation / timelapse image stored in variable `image`. stackview.animate(image) * Save an animation / timelapse stored in variable `image` with specified frame delay to a file. stackview.animate(image, filename="output.gif", frame_delay_ms=100) * Display an image stored in a variable `image` (this also works with label images). Prefer stackview.insight over matplotlib.pyplot.imshow! stackview.insight(image) * Display an image as a label image explicitly. stackview.imshow(image, labels=True) ### Processing images using the napari-simpleitk-image-processing (nsitk) Python library. When you use nsitk, you always start by importing the library: `import napari_simpleitk_image_processing as nsitk`. When asked for specific tasks, you can adapt one of the following code snippets: * Apply a median filter to an image to remove noise while preserving edges. nsitk.median_filter(image, radius_x=2, radius_y=2) * Applies Otsu's threshold selection method to an intensity image and returns a binary image (also works with intermodes, kittler_illingworth, li, moments, renyi_entropy, shanbhag, yen, isodata, triangle, huang and maximum_entropy instead of otsu). nsitk.threshold_otsu(image) * Computes the signed Maurer distance map of the input image. nsitk.signed_maurer_distance_map(binary_image) * Detects edges in the image using Canny edge detection. nsitk.canny_edge_detection(image, lower_threshold=0, upper_threshold=50) * Identifies the regional maxima of an image. nsitk.regional_maxima(image) * Rescales the intensity of an input image to a specified range. nsitk.rescale_intensity(image, output_minimum=0, output_maximum=255) * Applies the Sobel operator to an image to find edges. nsitk.sobel(image) * Enhances the contrast of an image using adaptive histogram equalization. nsitk.adaptive_histogram_equalization(image, alpha=0.3, beta=0.3, radius_x=5, radius_y=5) * Applies a standard deviation filter to an image. nsitk.standard_deviation_filter(image, radius_x=5, radius_y=5) * Labels the connected components in a binary image. nsitk.connected_component_labeling(binary_image) * Labels objects in a binary image and can split object that are touching.. nsitk.touching_objects_labeling(binary_image) * Applies the Laplacian of Gaussian filter to find edges in an image. nsitk.laplacian_of_gaussian_filter(image, sigma=1.0) * Identifies h-maxima of an image, suppressing maxima smaller than h. nsitk.h_maxima(image, height=10) * Removes background in an image using the Top-Hat filter. nsitk.white_top_hat(image, radius_x=5, radius_y=5) * Computes basic statistics for labeled object regions in an image. nsitk.label_statistics(image, label_image, size=True, intensity=True, shape=False) * Computes a map from a label image where the pixel intensity corresponds to the number of pixels in the given labeled object (analogously work elongation_map, feret_diameter_map, roundness_map). nsitk.pixel_count_map(label_image) ### Processing images using napari-segment-blobs-and-things-with-membranes (nsbatwm) If you use this plugin, you need to import it like this: `import napari_segment_blobs_and_things_with_membranes as nsbatwm`. You can then use it for various purposes: * Denoise an image using a Gaussian filter nsbatwm.gaussian_blur(image, sigma=1) * Denoise an image, while preserving edges: nsbatwm.median_filter(image, radius=2) * Denoise an image using a percentile (similar to median, but free in choosing the percentile) nsbatwm.percentile_filter(image, percentile=50, radius=2) * Determine the local minimum intensity for every pixel (also works with maximum) nsbatwm.minimum_filter(image, radius=2) * Enhance edges nsbatwm.gaussian_laplace(image, sigma=2) * Remove background from an image using the Top-Hat filter nsbatwm.white_tophat(image, radius=2) * Remove background from an image using the Rolling-Ball method nsbatwm.subtract_background(membranes, rolling_ball_radius=15) * Uses combination of Voronoi tesselation and Otsu's threshold method for segmenting an image nsbatwm.voronoi_otsu_labeling(blobs, spot_sigma=3.5, outline_sigma=1) * Apply a Gaussian blur, Otsu's threshold for binarization and returns a label image nsbatwm.gauss_otsu_labeling(blobs, outline_sigma=1) * Binarize an image using a threshold determined using Otsu's method (also works with li, triangle, yen, mean methods) nsbatwm.threshold_otsu(blobs) * Split touching objects in a binary image nsbatwm.split_touching_objects(binary, sigma=3.5) * Identify individual objects in a binary image using Connected Component labeling nsbatwm.connected_component_labeling(binary) * Apply a Watershed algorithm to an an image showing membrane-like structures and a label image that serves as seeds for the watershed nsbatwm.seeded_watershed(membranes_image, labeled_seeds) * Apply a Watershed algorithm to an image showing membrane-like structures. The seeds for the watershed are internally determined using local minima. nsbatwm.local_minima_seeded_watershed(membrane_image, spot_sigma=10, outline_sigma=0) * Dilate labels to increase their size nsbatwm.expand_labels(label_image, distance=1) * Smooths outlines of label images by determining the most popular label locally nsbatwm.mode_filter(label_image, radius=10) * Remove labels that touch the image border nsbatwm.remove_labels_on_edges(label_image) * Skeletonize labels nsbatwm.skeletonize(labels) ### Working with Pandas DataFrames In case a pandas DataFrame, e.g. `df` is the result of a code block, just write `df.head()` by the end so that the user can see the intermediate result. ### Processing images with scikit-image (skimage) * Load an image file from disc and store it in a variable: from skimage.io import imread image = imread(filename) * Expanding labels by a given radius in a label image works like this: from skimage.segmentation import expand_labels expanded_labels = expand_labels(label_image, distance=10) * Turn a label image into an RGB image, e.g. for saving as png: from skimage import color rgb_image = color.label2rgb(label_image, bg_label=0).astype("unit8") * Measure properties of labels with respect to an image works like this: import pandas as pd from skimage.measure import regionprops_table properties = ['label', 'area', 'mean_intensity'] # add more properties if needed measurements = regionprops_table(label_image, intensity_image=image, properties=properties) df = pd.DataFrame(measurements)

[haesleinhuepf]

git-bob comment on this

[github-actions[bot]]

^{This message was generated by git-bob (version: 0.9.0, model: gpt-4o-2024-08-06, log), an experimental AI-based assistant. It can make mistakes and has limitations. Check its messages carefully.}

To segment the nuclei in the provided image using Voronoi-Otsu-Labeling with the mentioned libraries, you can use the following code snippet:

from skimage.io import imread
import napari_segment_blobs_and_things_with_membranes as nsbatwm

# Load the image
image_url = "https://github.com/user-attachments/assets/da6456bb-7bdf-4ad8-8f0d-95e38fa110e9"
original_image = imread(image_url)

# Perform Voronoi-Otsu-Labeling for segmentation
nuclei_labels = nsbatwm.voronoi_otsu_labeling(original_image, spot_sigma=3.5, outline_sigma=1)

# Visualize the results if desired (for example using stackview)
import stackview
stackview.insight(original_image)
stackview.animate_curtain(original_image, nuclei_labels)

This code will help in segmenting the nuclei from the image using the Voronoi-Otsu-Labeling approach. Let me know if you need further assistance!

[haesleinhuepf]

Ok cool. Please download the image specified by the url and save it as "human_mitosis_small.png".

Please turn this into a jupyter notebook and adapt it like this.

• Load the above mentioned png file instead of the url.
• After loading the image, make sure to extract the first channel, e.g. using the [..., 0] syntax.
• Remove the stackview.insight call. Keep stackview.animate_cutrain for result visualization only.
• By the end, in a new notebook cell, save the resulting label image a RGB png to disc.
• Oh and, please put all import statements at the beginning of the notebook.

git-bob implement this.