MaxIV & QIM workshop

Felipe Delestro

Senior Research Software Specialist @ QIM Center

Danmarks Tekniske Universitet (DTU)

 

fima@dtu.dk 

 

   

slides.com/delestro/qim-maxiv-2025

QIM Center

Center for Quantification of Imaging Data from MAX IV

supported by

qim.dk/about

Anders Bjorholm Dahl

Head of the QIM Center

Jon
Sporring

Deputy Head of the QIM Center

Stephen
Hall

Professor at Lund university

Rebecca
Engberg

Center Manager

What can be achieved in collaboration with the QIM center?

Data acquisition at MaxIV

Start jupyter server

Create project space

Set up permissions

Upload data to DTU servers

Convert data to OME-Zarr

Mount data locally

Data visualization

Export ROI

Load data

Run analysis

Data acquisition at MaxIV

Start jupyter server

Create project space

Set up permissions

Upload data to DTU servers

Convert data to OME-Zarr

Mount data locally

Data visualization

Export ROI

Load data

Run analysis

Projects at the QIM center

  • A project is a space in our servers with attached metadata.
     
  • Users can store their raw data and results from analysis.
     
  • Access to the directory is restricted to users who are part of the project.
     
  • Some management aspects are still done by contacting us.
  • A centralized space for image analysis at the DTU Computing Center (DCC)
     
  • Easier access to High Performance Computing (HPC) resources
     
  • Projects and resources  management
     
  • https://platform.qim.dk/

SFTP data transfer

https://filezilla-project.org/

Any SFTP client can be used, but FileZilla is a good and reliable option

transfer.gbar.dtu.dk

port 22

Globus and large data transfer

https://platform.qim.dk/tutorials/globus-maxiv

open-source Python library, that focuses on 3D volumetric data

 Getting started is easy:
pip install qim3d

Documentation at platform.qim.dk/qim3d/

Data handling

Volumetric data can be stored in a variety of file formats.

This can make the very first step of importing the data more complex than it should be

load and save

Data handling

Data handling

Volumetric data can be stored in a variety of file formats.

This can make the very first step of importing the data more complex than it should be

load and save

qim3d.io.load and qim3d.io.save can handle the most common data formats:

  • Tiff (including file stacks)
  • HDF5
  • TXRM/TXM/XRM
  • NIfTI
  • PIL (including file stacks)
  • VOL/VGI
  • DICOM
import qim3d

vol = qim3d.io.load("path/to/image.tif", virtual_stack=True)

Data handling

We make a distinction between load/save and import/export when the operation includes transformations to the data.

import and export

Currently we're implementing a 3D import/export following the OME-Zarr standard, which
uses chunked data with a
ulti-scales scheme.

Visualization

Visualization

Easily exploring volumetric data may not be so straightforward. The library provides helper functions this task simple and easy, even within Jupyter Notebooks

import qim3d

vol = qim3d.examples.bone_128x128x128
qim3d.viz.slicer(vol)

Visualization

import qim3d

vol = qim3d.examples.shell_225x128x128
qim3d.viz.slices_grid(vol, num_slices=15)

Visualization

import qim3d

# Generate synthetic collection of objects
num_objects = 15
vol, labels = qim3d.generate.volume_collection(num_objects = num_objects)
# Visualize synthetic collection
qim3d.viz.volumetric(vol)

Visualization

import qim3d

vol = qim3d.generate.volume(base_shape=(128,128,128),
                          final_shape=(128,128,128),
                          noise_scale=0.03,
                          order=1,
                          gamma=1,
                          max_value=255,
                          threshold=0.5,
                          dtype='uint8'
                          )

mesh = qim3d.mesh.from_volume(vol, step_size=3)

qim3d.viz.mesh(mesh.vertices, mesh.faces)

Visualization

import qim3d

vol = qim3d.examples.cement_128x128x128
binary = qim3d.filters.gaussian(vol, sigma = 2) < 60
labeled_volume, num_labels = qim3d.segmentation.watershed(binary)

color_map = qim3d.viz.colormaps.segmentation(num_labels, style = 'bright')
qim3d.viz.slicer(labeled_volume, slice_axis = 1, color_map=color_map)

qim3d.viz.colormaps.segmentation creates an colormap designes specifically for segmentation tasks 

The parameter min_dist allows us to control how different neighbour colors are

Image analysis

Image analysis

qim3d covers from basic functionalities to more complex methodologies.

this will allows the library to be a central point for building volumetric data pipelines. 

Image analysis

qim3d.filters contains implementations for common filter operations, including the option to run them on chunked data using Dask

 

qim3d.filters.Pipeline allows the creation of classes that act as filter pipelines

import qim3d
from qim3d.filters import Pipeline, Median, Gaussian, Maximum, Minimum

# Get data
vol = qim3d.examples.fly_150x256x256

# Show original
fig1 = qim3d.viz.slices_grid(vol, num_slices=5, display_figure=True)
# Create filter pipeline
pipeline = Pipeline(
    Median(size=5),
    Gaussian(sigma=3, dask = True)
)

# Append a third filter to the pipeline
pipeline.append(Maximum(size=3))

# Apply filter pipeline
vol_filtered = pipeline(vol)

# Show filtered
fig2 = qim3d.viz.slices_grid(vol_filtered, num_slices=5, display_figure=True)

Image analysis

The modules qim3d.detection and qim3d.segmentation contain wrappers for common methods that are useful when building larger pipelines

import qim3d

# Get data
vol = qim3d.examples.cement_128x128x128
vol_blurred = qim3d.filters.gaussian(vol, sigma=2)

# Detect blobs, and get binary_volume
blobs, binary_volume = qim3d.detection.blobs(
    vol_blurred,
    min_sigma=1,
    max_sigma=8,
    threshold=0.001,
    overlap=0.1,
    background="bright"
    )

# Visualize detected blobs
qim3d.viz.circles(blobs, vol, alpha=0.8, color='blue')

Image analysis

The modules qim3d.detection and qim3d.segmentation contain wrappers for common methods that are useful when building larger pipelines

import qim3d

vol = qim3d.examples.cement_128x128x128
bin_vol = qim3d.filters.gaussian(vol, sigma = 2)<60

fig1 = qim3d.viz.slices_grid(bin_vol, slice_axis=1, display_figure=True)
labeled_volume, num_labels = qim3d.segmentation.watershed(bin_vol)

cmap = qim3d.viz.colormaps.segmentation(num_labels)
fig2 = qim3d.viz.slices_grid(labeled_volume, slice_axis=1, color_map=cmap, display_figure=True)

Image analysis

qim3d.processing is meant for more complex methods, that perform quantification tasks that are specific but common for volumetric data

The current implementations are:

  • qim3d.processing.structure_tensor
  • qim3d.processing.local_thickness

Image analysis

Structure tensor

import qim3d

vol = qim3d.examples.NT_128x128x128
val, vec = qim3d.processing.structure_tensor(vol, visualize=True, axis=2)

Image analysis

Local thickness

import qim3d

vol = qim3d.examples.fly_150x256x256
lt_vol = qim3d.processing.local_thickness(vol, visualize=True, axis=0)

The library is an open-source project, and you're all welcome to contribute!

github.com/qim-center/qim3d​

documentation and
installation instructions:

platform.qim.dk/qim3d

☕

🧑‍💻 Practical session

Final thoughts

  • The platform and the qim3d library are constantly being developed, so your feedback is essential!
     
  • Always feel free to reach out with any questions
    • Felipe Delestro ⇨ fima@dtu.dk
    • Rebecca Engberg ⇨ reen@dtu.dk