Processing timelapse data#

This notebook demonstrates how to process timelapse data frame-by-frame.

from import imread, imsave
import pyclesperanto_prototype as cle
import numpy as np

First, we should define the origin of the data we want to process and where the results should be saved to.

input_file = "../../data/CalibZAPWfixed_000154_max.tif"
output_file = "../../data/CalibZAPWfixed_000154_max_labels.tif"

Next, we open the dataset and see what image dimensions it has.

timelapse = imread(input_file)
(100, 235, 389)

If it is not obvious which dimension is the time dimension, it is recommended to slice the dataset in different directions.


Obviously, the time dimension is the first dimension (index 0).

Next, we define the image processing workflow we want to apply to our dataset. It is recommended to do this in a function so that we can later reuse it without copy&pasting everything.

def process_image(image, 
                  # define default parameters for the procedure
    """Segment nuclei in an image and return labels"""
    # pre-process image
    background_subtracted = cle.top_hat_box(image, 
    # segment nuclei
    labels = cle.voronoi_otsu_labeling(background_subtracted,

    return labels

# Try out the function
single_timepoint = timelapse[50]
segmented = process_image(single_timepoint)

# Visualize result
cle.imshow(segmented, labels=True)

After we made this function work on a single timepoint, we should program a for-loop that goes through the timelapse, applies the procedure to a couple of image and visualizes the results. Note: We go in steps of 10 images through the timelapse, to get an overview.

max_t = timelapse.shape[0]
for t in range(0, max_t, 10):
    label_image = process_image(timelapse[t])
    cle.imshow(label_image, labels=True)
../_images/51c6a9e598304f16847f3f1dfa098cc73ac8ee7cc392437e3ff8dca75bab9b5d.png ../_images/ab3bcb4e330a3a2eb539f33bf994b3f49b2f28fbe65f8a9536b59154eb9d7af7.png ../_images/c7b5a1ab97a72cff10faab3b10775ab52170e3cdec6c3679bf095e3199fae549.png ../_images/eaeba58f06f14bcd20ed9acc8f1d120e95a15aafe5a33046c09671057ec8b456.png ../_images/da2a4535ff6354b5b969b5aa928862976ae89f852f8210a45173e5e9a1d57bed.png ../_images/ce59c91bb9c93b4882f2dc7c145ab7e13a8b4d46c58793967833f7388be507d1.png ../_images/782db570f2761bb79105d0bd8d09f1c006db76309e0f73198f731951b7a890f7.png ../_images/a67f022642078a4ad8366a1d480c9de82cc69d9b6d359394eecfca0f5a4a310b.png ../_images/cb912bd877a26a7d486345389c8be2a466864e5802172087952024f04f799539.png ../_images/3d0b815c7f235394fb18a80db92db9d3d309ca81049f0be54e4d677eef4d11c8.png

When we are convinced that the procedure works, we can apply it to the whole timelapse, collect the results in a list and save it as stack to disc.

label_timelapse = []
for t in range(0, max_t):
    label_image = process_image(timelapse[t])
# convert list of 2D images to 3D stack
np_stack = np.asarray(label_timelapse)

# save result to disk
imsave(output_file, np_stack)
C:\Users\rober\AppData\Local\Temp\ipykernel_27924\ UserWarning: ../../data/CalibZAPWfixed_000154_max_labels.tif is a low contrast image
  imsave(output_file, np_stack)

Just to be sure that everything worked nicely, we reopen the dataset and print its dimensionality. It’s supposed to be identical to the original timelapse dataset.

result = imread(output_file)

(100, 235, 389)
(100, 235, 389)