Image Segmentation¶

This is a minimal tutorial on how to use ngio for image segmentation.

Step 1: Setup¶

We will first implement a very simple function to segment an image. We will use skimage to do this.

In [1]:

# Setup a simple segmentation function
import numpy as np
import skimage


def otsu_threshold_segmentation(image: np.ndarray, max_label: int) -> np.ndarray:
    """Simple segmentation using Otsu thresholding."""
    threshold = skimage.filters.threshold_otsu(image)
    binary = image > threshold
    label_image = skimage.measure.label(binary)
    label_image += max_label
    label_image = np.where(binary, label_image, 0)
    return label_image.astype(np.uint32)

# Setup a simple segmentation function import numpy as np import skimage def otsu_threshold_segmentation(image: np.ndarray, max_label: int) -> np.ndarray: """Simple segmentation using Otsu thresholding.""" threshold = skimage.filters.threshold_otsu(image) binary = image > threshold label_image = skimage.measure.label(binary) label_image += max_label label_image = np.where(binary, label_image, 0) return label_image.astype(np.uint32)

Step 2: Open the OmeZarr container¶

In [2]:

from pathlib import Path

from ngio import open_ome_zarr_container
from ngio.utils import download_ome_zarr_dataset

# Download the dataset
download_dir = Path(".").absolute().parent.parent / "data"
hcs_path = download_ome_zarr_dataset("CardiomyocyteTiny", download_dir=download_dir)
image_path = hcs_path / "B" / "03" / "0"

# Open the ome-zarr container
ome_zarr = open_ome_zarr_container(image_path)

from pathlib import Path from ngio import open_ome_zarr_container from ngio.utils import download_ome_zarr_dataset # Download the dataset download_dir = Path(".").absolute().parent.parent / "data" hcs_path = download_ome_zarr_dataset("CardiomyocyteTiny", download_dir=download_dir) image_path = hcs_path / "B" / "03" / "0" # Open the ome-zarr container ome_zarr = open_ome_zarr_container(image_path)

Unzipping contents of '/home/runner/work/ngio/ngio/data/20200812-CardiomyocyteDifferentiation14-Cycle1-tiny.zarr.zip' to '/home/runner/work/ngio/ngio/data/tmp'

Step 3: Segment the image¶

For this example, we will not segment the image all at once. Instead we will iterate over the image FOVs and segment them one by one.

In [3]:

from ngio.experimental.iterators import SegmentationIterator

# First we will need the image object and the FOVs table
image = ome_zarr.get_image()
roi_table = ome_zarr.get_roi_table("FOV_ROI_table")

# Second we need to derive a new label image to use as target for the segmentation

label = ome_zarr.derive_label("new_label", overwrite=True)

# Setup the segmentation iterator
seg_iterator = SegmentationIterator(
    input_image=image,
    output_label=label,
    channel_selection="DAPI",
    axes_order=["z", "y", "x"],
)
seg_iterator = seg_iterator.product(roi_table)

# Make sure that if other axes are present they are iterated over
seg_iterator = seg_iterator.by_zyx()

max_label = 0  # We will use this to avoid label collisions
for image_data, label_writer in seg_iterator.iter_as_numpy():
    roi_segmentation = otsu_threshold_segmentation(
        image_data, max_label
    )  # Segment the image

    max_label = roi_segmentation.max()  # Get the max label for the next iteration

    label_writer(patch=roi_segmentation)  # Write the segmentation back to the label

# No need to consolidate, the iterator does it automatically after the last write

from ngio.experimental.iterators import SegmentationIterator # First we will need the image object and the FOVs table image = ome_zarr.get_image() roi_table = ome_zarr.get_roi_table("FOV_ROI_table") # Second we need to derive a new label image to use as target for the segmentation label = ome_zarr.derive_label("new_label", overwrite=True) # Setup the segmentation iterator seg_iterator = SegmentationIterator( input_image=image, output_label=label, channel_selection="DAPI", axes_order=["z", "y", "x"], ) seg_iterator = seg_iterator.product(roi_table) # Make sure that if other axes are present they are iterated over seg_iterator = seg_iterator.by_zyx() max_label = 0 # We will use this to avoid label collisions for image_data, label_writer in seg_iterator.iter_as_numpy(): roi_segmentation = otsu_threshold_segmentation( image_data, max_label ) # Segment the image max_label = roi_segmentation.max() # Get the max label for the next iteration label_writer(patch=roi_segmentation) # Write the segmentation back to the label # No need to consolidate, the iterator does it automatically after the last write

Plot the segmentation¶

In [4]:

import matplotlib.pyplot as plt
import numpy as np
from matplotlib.colors import ListedColormap

rand_cmap = np.random.rand(1000, 3)
rand_cmap[0] = 0
rand_cmap = ListedColormap(rand_cmap)

fig, axs = plt.subplots(2, 1, figsize=(8, 4))
axs[0].set_title("Original image")
axs[0].imshow(image.get_as_numpy(c=0, z=1, axes_order=["y", "x"]), cmap="gray")
axs[1].set_title("Final segmentation")
axs[1].imshow(label.get_as_numpy(z=1, axes_order=["y", "x"]), cmap=rand_cmap)
for ax in axs:
    ax.axis("off")
plt.tight_layout()
plt.show()

import matplotlib.pyplot as plt import numpy as np from matplotlib.colors import ListedColormap rand_cmap = np.random.rand(1000, 3) rand_cmap[0] = 0 rand_cmap = ListedColormap(rand_cmap) fig, axs = plt.subplots(2, 1, figsize=(8, 4)) axs[0].set_title("Original image") axs[0].imshow(image.get_as_numpy(c=0, z=1, axes_order=["y", "x"]), cmap="gray") axs[1].set_title("Final segmentation") axs[1].imshow(label.get_as_numpy(z=1, axes_order=["y", "x"]), cmap=rand_cmap) for ax in axs: ax.axis("off") plt.tight_layout() plt.show()

Step 4: Masked image segmentation¶

In this example we will use a mask to restrict the segmentation to certain areas of the image. In this case we will create a simple mask for illustration purposes, but in a real case scenario the mask could come from another segmentation mask.

In [5]:

# Create a basic mask for illustration purposes
mask = ome_zarr.derive_label("mask", overwrite=True)
mask_data = mask.get_as_numpy(axes_order=["z", "y", "x"])
mask_data[:, 200:-200, 500:2000] = 1
mask_data[:, 200:-200, 3000:-500] = 2
mask_data[:, 600:-600, 1200:-1000] = 0
mask_data[:, 700:-700, 1600:-1500] = 3
mask.set_array(mask_data, axes_order=["z", "y", "x"])
mask.consolidate()

fig, axs = plt.subplots(2, 1, figsize=(8, 4))
axs[0].set_title("Original image")
axs[0].imshow(image.get_as_numpy(c=0, z=1, axes_order=["y", "x"]), cmap="gray")
axs[1].set_title("Mask")
axs[1].imshow(mask.get_as_numpy(z=1, axes_order=["y", "x"]), cmap=rand_cmap)
for ax in axs:
    ax.axis("off")
plt.tight_layout()
plt.show()

# Create a basic mask for illustration purposes mask = ome_zarr.derive_label("mask", overwrite=True) mask_data = mask.get_as_numpy(axes_order=["z", "y", "x"]) mask_data[:, 200:-200, 500:2000] = 1 mask_data[:, 200:-200, 3000:-500] = 2 mask_data[:, 600:-600, 1200:-1000] = 0 mask_data[:, 700:-700, 1600:-1500] = 3 mask.set_array(mask_data, axes_order=["z", "y", "x"]) mask.consolidate() fig, axs = plt.subplots(2, 1, figsize=(8, 4)) axs[0].set_title("Original image") axs[0].imshow(image.get_as_numpy(c=0, z=1, axes_order=["y", "x"]), cmap="gray") axs[1].set_title("Mask") axs[1].imshow(mask.get_as_numpy(z=1, axes_order=["y", "x"]), cmap=rand_cmap) for ax in axs: ax.axis("off") plt.tight_layout() plt.show()

In [6]:

from ngio.experimental.iterators import MaskedSegmentationIterator

# First we will need the masked image object
# (that contains the masking table information inside)
image = ome_zarr.get_masked_image(masking_label_name="mask")

# Second we need to derive a new label image to use as target for the segmentation
label = ome_zarr.derive_label("masked_new_label", overwrite=True)

# Setup the masked segmentation iterator
seg_iterator = MaskedSegmentationIterator(
    input_image=image,
    output_label=label,
    channel_selection="DAPI",
    axes_order=["z", "y", "x"],
)

# Make sure that if other axes are present they are iterated over
seg_iterator = seg_iterator.by_zyx()

max_label = 0  # We will use this to avoid label collisions
for image_data, label_writer in seg_iterator.iter_as_numpy():
    roi_segmentation = otsu_threshold_segmentation(
        image_data, max_label
    )  # Segment the image

    max_label = roi_segmentation.max()  # Get the max label for the next iteration

    label_writer(patch=roi_segmentation)  # Write the segmentation back to the label

# No need to consolidate, the iterator does it automatically after the last write

from ngio.experimental.iterators import MaskedSegmentationIterator # First we will need the masked image object # (that contains the masking table information inside) image = ome_zarr.get_masked_image(masking_label_name="mask") # Second we need to derive a new label image to use as target for the segmentation label = ome_zarr.derive_label("masked_new_label", overwrite=True) # Setup the masked segmentation iterator seg_iterator = MaskedSegmentationIterator( input_image=image, output_label=label, channel_selection="DAPI", axes_order=["z", "y", "x"], ) # Make sure that if other axes are present they are iterated over seg_iterator = seg_iterator.by_zyx() max_label = 0 # We will use this to avoid label collisions for image_data, label_writer in seg_iterator.iter_as_numpy(): roi_segmentation = otsu_threshold_segmentation( image_data, max_label ) # Segment the image max_label = roi_segmentation.max() # Get the max label for the next iteration label_writer(patch=roi_segmentation) # Write the segmentation back to the label # No need to consolidate, the iterator does it automatically after the last write

In [7]:

import matplotlib.pyplot as plt
import numpy as np
from matplotlib.colors import ListedColormap

fig, axs = plt.subplots(2, 1, figsize=(8, 4))
axs[0].set_title("Original image")
axs[0].imshow(image.get_as_numpy(c=0, z=1, axes_order=["y", "x"]), cmap="gray")
axs[1].set_title("Final segmentation")
axs[1].imshow(label.get_as_numpy(z=1, axes_order=["y", "x"]), cmap=rand_cmap)
for ax in axs:
    ax.axis("off")
plt.tight_layout()
plt.show()

import matplotlib.pyplot as plt import numpy as np from matplotlib.colors import ListedColormap fig, axs = plt.subplots(2, 1, figsize=(8, 4)) axs[0].set_title("Original image") axs[0].imshow(image.get_as_numpy(c=0, z=1, axes_order=["y", "x"]), cmap="gray") axs[1].set_title("Final segmentation") axs[1].imshow(label.get_as_numpy(z=1, axes_order=["y", "x"]), cmap=rand_cmap) for ax in axs: ax.axis("off") plt.tight_layout() plt.show()

In [ ]: