Drawing annotations using stackivew¶

In this notebook we use stackview.annotate to create a label image with manual annotations.

In [1]:

import stackview

import numpy as np
from skimage.data import cells3d

The cells3d dataset from scikit-image serves as example image. One of the two channels shown nuclei.

In [2]:

image = cells3d()[:,1]

stackview.insight(image)

Out[2]:

shape	(60, 256, 256)
dtype	uint16
size	7.5 MB
min	0
max	65535

Before annotating the nuclei, we create an empty label image of type unsigned 32-bit integer. This type is necessary so that stackview views the image correctly as labels. Compatible types are: int32,uint32,int64,uint64.

In [3]:

labels = np.zeros(image.shape).astype(np.uint32)

The function stackview.annotate creates a user interface where we can draw labels on top of the original image. Click and drag for drawing. Hold the ALT key for erasing.

In [4]:

stackview.annotate(image, labels)

Out[4]:

VBox(children=(HBox(children=(HBox(children=(VBox(children=(ImageWidget(height=256, width=256),)),)), VBox(chi…

You can also annotate in a different plane than the Z-plane by providing an axis (0=z,1=y,2=x).

In [ ]:

stackview.annotate(image, labels, axis=1)

Once annotations are drawn, we can view them as usual using the insight() function.

In [ ]:

stackview.insight(labels)

Just for fun we use the apoc library to create an instance segmentation of nuclei from our annotation.

In [ ]:

import apoc

segmenter = apoc.ObjectSegmenter(positive_class_identifier=1)
features = apoc.PredefinedFeatureSet.small_quick.value

segmenter.train(image=image, ground_truth=labels, features=features)
result = segmenter.predict(image)

stackview.insight(result[30])

In [ ]: