Programmatic Workflow

Using workflows in napari-lattice¶

Demo to use workflows in functions for end to end image analysis.

This notebook shows how to create & save a simple workflow

In [2]:

import os
import pyclesperanto_prototype as cle
from napari_workflows import Workflow

cle.get_device()

import os import pyclesperanto_prototype as cle from napari_workflows import Workflow cle.get_device()

Out[2]:

<Intel(R) Iris(R) Xe Graphics on Platform: Intel(R) OpenCL Graphics (1 refs)>

In [3]:

#download sample data from https://zenodo.org/records/14903188
img_path = "./sample_data/RBC_tiny.czi"

#make dir if not exists
if not os.path.exists("./sample_data"):
    os.makedirs("./sample_data")

import urllib.request

url = "https://zenodo.org/records/14903188/files/RBC_tiny.czi?download=1"
if not os.path.exists(img_path):
    print("Downloading file")
    urllib.request.urlretrieve(url, img_path)
    print(f"Downloaded at : {img_path}")
else:
    print(f"File already exists at : {img_path}")

#download sample data from https://zenodo.org/records/14903188 img_path = "./sample_data/RBC_tiny.czi" #make dir if not exists if not os.path.exists("./sample_data"): os.makedirs("./sample_data") import urllib.request url = "https://zenodo.org/records/14903188/files/RBC_tiny.czi?download=1" if not os.path.exists(img_path): print("Downloading file") urllib.request.urlretrieve(url, img_path) print(f"Downloaded at : {img_path}") else: print(f"File already exists at : {img_path}")

Downloading file
Downloaded at : ./sample_data/RBC_tiny.czi

In [4]:

#absolute path for saving
save_path = "C:/Users/rajasekhar.p/nap/napari_lattice/notebooks/sample_data"

#make dir if not exists
if not os.path.exists(save_path):
    os.makedirs(save_path)

#absolute path for saving save_path = "C:/Users/rajasekhar.p/nap/napari_lattice/notebooks/sample_data" #make dir if not exists if not os.path.exists(save_path): os.makedirs(save_path)

To design end to end image analysis workflows, we can use napari-workflows

https://github.com/haesleinhuepf/napari-workflows

Lets apply a gaussian blur filter using pyclesperanto library

In [13]:

#import gaussian filter from pyclesperanto
#We initialise a workflow
gaussian_workflow = Workflow()

#For napari-lattice, the input image should always be of the name "deskewed_image"

input_arg = "deskewed_image"
#define a simple gaussian filter
task_name = "gaussian"
gaussian_blur_sigma_x = 5
gaussian_blur_sigma_y = 5
gaussian_blur_sigma_z = 2

gaussian_workflow.set(task_name, cle.gaussian_blur, 
                      input_arg, sigma_x = gaussian_blur_sigma_x,
                      sigma_y = gaussian_blur_sigma_y,
                      sigma_z = gaussian_blur_sigma_z)

#Printing a workflow will show you the arguments
print(gaussian_workflow)

#import gaussian filter from pyclesperanto #We initialise a workflow gaussian_workflow = Workflow() #For napari-lattice, the input image should always be of the name "deskewed_image" input_arg = "deskewed_image" #define a simple gaussian filter task_name = "gaussian" gaussian_blur_sigma_x = 5 gaussian_blur_sigma_y = 5 gaussian_blur_sigma_z = 2 gaussian_workflow.set(task_name, cle.gaussian_blur, input_arg, sigma_x = gaussian_blur_sigma_x, sigma_y = gaussian_blur_sigma_y, sigma_z = gaussian_blur_sigma_z) #Printing a workflow will show you the arguments print(gaussian_workflow)

Workflow:
gaussian <- (<function gaussian_blur at 0x000001D37F6E3AC0>, 'deskewed_image', None, 5, 5, 2)

In [ ]:

from lls_core import LatticeData

params = LatticeData(
  input_image=img_path,
  save_dir=save_path,
  workflow=gaussian_workflow,
)

print(params)

from lls_core import LatticeData params = LatticeData( input_image=img_path, save_dir=save_path, workflow=gaussian_workflow, ) print(params)

[INFO:2025-08-28 11:41:57,583] Processing File ./sample_data/RBC_tiny.czi

input_image=<xarray.DataArray 'transpose-1921c499e620816830b32ee3034745d3' (T: 1, C: 1,
                                                                Z: 834, Y: 118,
                                                                X: 209)> Size: 41MB
dask.array<transpose, shape=(1, 1, 834, 118, 209), dtype=uint16, chunksize=(1, 1, 834, 118, 209), chunktype=numpy.ndarray>
Coordinates:
  * C        (C) <U17 68B 'LatticeLightsheet'
  * Z        (Z) float64 7kB 0.0 0.3 0.6 0.9 1.2 ... 249.0 249.3 249.6 249.9
  * Y        (Y) float64 944B 0.0 0.145 0.29 0.435 ... 16.53 16.67 16.82 16.96
  * X        (X) float64 2kB 0.0 0.145 0.29 0.435 ... 29.72 29.87 30.01 30.16
Dimensions without coordinates: T
Attributes:
    unprocessed:  <Element 'ImageDocument' at 0x000001D332A48F40> skew=<DeskewDirection.Y: 2> angle=30.0 physical_pixel_sizes=DefinedPixelSizes(X=0.14499219272808386, Y=0.14499219272808386, Z=0.3) derived=DerivedDeskewFields(deskew_vol_shape=(59, 1828, 209), deskew_affine_transform=<pyclesperanto_prototype._tier8._AffineTransform3D.AffineTransform3D object at 0x000001D3330298D0>, deskew_affine_transform_zyx=array([[ 1.11022302e-16, -5.00000000e-01,  0.00000000e+00,
         5.90000000e+01],
       [ 2.06907692e+00,  8.66025404e-01,  0.00000000e+00,
         0.00000000e+00],
       [ 0.00000000e+00,  0.00000000e+00,  1.00000000e+00,
         0.00000000e+00],
       [ 0.00000000e+00,  0.00000000e+00,  0.00000000e+00,
         1.00000000e+00]])) save_dir=WindowsPath('C:/Users/rajasekhar.p/nap/napari_lattice/notebooks/sample_data') save_suffix='_deskewed' save_name='RBC_tiny_deskewed' save_type=<SaveFileType.h5: 'h5'> time_range=range(0, 1) channel_range=range(0, 1) deconvolution=None crop=None workflow=<napari_workflows._workflow.Workflow object at 0x000001D32285B070> progress_bar=True

[INFO:2025-08-28 11:42:00,863] build program: kernel 'affine_transform_deskew_y_3d' was part of a lengthy source build resulting from a binary cache miss (1.36 s)
[INFO:2025-08-28 11:42:01,567] build program: kernel 'gaussian_blur_separable_3d' was part of a lengthy source build resulting from a binary cache miss (0.57 s)

In [7]:

# when running isave, if params.workflow has a workflow, then it will run the analysis workflow
params.workflow

# when running isave, if params.workflow has a workflow, then it will run the analysis workflow params.workflow

Out[7]:

<napari_workflows._workflow.Workflow at 0x1d32285b070>

In [8]:

params.save()

params.save()

Timepoints:   0%|          | 0/1 [00:00<?, ?it/s][INFO:2025-08-28 11:41:59,144] Processing File <xarray.DataArray 'transpose-1921c499e620816830b32ee3034745d3' (Z: 834, Y: 118,
                                                                X: 209)> Size: 41MB
dask.array<getitem, shape=(834, 118, 209), dtype=uint16, chunksize=(834, 118, 209), chunktype=numpy.ndarray>
Coordinates:
    C        <U17 68B 'LatticeLightsheet'
  * Z        (Z) float64 7kB 0.0 0.3 0.6 0.9 1.2 ... 249.0 249.3 249.6 249.9
  * Y        (Y) float64 944B 0.0 0.145 0.29 0.435 ... 16.53 16.67 16.82 16.96
  * X        (X) float64 2kB 0.0 0.145 0.29 0.435 ... 29.72 29.87 30.01 30.16
Attributes:
    unprocessed:  <Element 'ImageDocument' at 0x000001D332A48F40>

INFO: blockdim levels (1) < subsamp levels (3): First-level block size (4, 256, 256) will be used for all levels

Timepoints: 100%|██████████| 1/1 [00:05<00:00,  5.77s/it]

This should save a deskewed image with gaussian blur applied

You can save the workflow for use in napari or programatically later on

In [14]:

from napari_workflows import _io_yaml_v1 as io_yaml

io_yaml.save_workflow("sample_data/gaussian_workflow.yaml", gaussian_workflow)

from napari_workflows import _io_yaml_v1 as io_yaml io_yaml.save_workflow("sample_data/gaussian_workflow.yaml", gaussian_workflow)