QIM+CIL Workshop 2025

This topic is meant for sharing information about the CIL Reconstruction Workshop – Understanding the Impact on Image Analysis

Monday 20 Oct

12:00 Arrival, lunch and coffee/tea

13:00 Introduction to filtered back-projection, preprocessing and CIL

14:45 Coffee & tea

15:15 Iterative reconstruction with CIL

17ish End of training

18:00 Dinner in Lyngby for those who registered separately

Tuesday 21 Oct

08:45 Coffee & tea

09:00 Introduction to two notebooks on reconstruction/image analysis interplay

09:15 Time to explore notebooks with CIL/qim3d

10:15 Coffee & tea

10:30 Time to explore notebooks with CIL/qim3d

11:30 Wrap-up, discussions and where to go from here

12:00 Lunch and goodbye

Jupyter Launcher

CPU only

GPUs

To get the copy of CIL demos with paths set to DTU system, please use:

cd workspace
mkdir yourname
cd yourname
git clone --depth 1 --branch qim3d https://github.com/TomographicImaging/CIL-Demos.git

For the notebooks used by qim3d use this repository:

cd workspace/<USERNAME>/
git clone https://github.com/qim-center/qim-notebooks.git

Today’s slides can be found here: cil.stfc.ac.uk/DTU_CIL_training.pdf

Documentation for qim3d

Feedback on day 1 from post-it notes:

Green (one thing you liked):

  • I liked the opportunity to try the software
  • Combination of theory and practice
  • Very hands-on.
  • Well-explained notebooks, good descriptions.
  • The notebooks were informative and intuitive.
  • Very good mixture of lectures/notebook work
  • Discussions about the details of how reconstruction works.
  • Well prepared setup - no installations etc.
  • There was enough time to work calmly with the notebooks! Good!
  • tailored levels to different backgrounds.
  • Many notebooks to try
  • Theory explanation on reconstruction methods
  • There is a good trade-off between how much presenting and time for exercises.
  • Intuitive and clear instructions about how to use CIL

Red (one things to change)

  • I would have liked to be challenged to solve a task.
  • A bit overwhelming for beginners.
  • Include more discussions between the tables
  • Follow time scheme!
  • The notebook should encourage more work.
  • Get even more time for testing of the notebooks.
  • Shift-Enter work / Run & go.
  • Missing intro to the framework.
  • The Jupyter launcher
  • The Jupyter launcher needs to work more smoothly!
  • Perhaps something more “challenge like” i.e. notebooks that does not just “run” - but requires some active interaction and also them smaller.
  • Air quality
  • Perhaps have a general summary/walk-through of each notebook after everyone is done + Discussion!
  • More “setup” for the notebooks or more time to go through.

Ideas from closing session:

If you were ten times bolder - what big idea would you recommend on interplay between reconstruction and image analysis?

What first step would you take to get started?

  • 22 - I would integrate DVC into the pipeline of QIM. The first step that I would do is to implement a tool for the registration of volumes.
  • 19 - Idea: an interactive reconstruction GUI that shows a live volumetric rendering while the reconstruction is till being processed. Users can tweak parameters and see changes live in 3D.
  • 19 - A system that allows all project members to utilize their expertise.
  • 16 - Using image analysis to automatically choose the best reconstruction out of a set instead of manual judging. Find some IA (image analysis) method that gives a metric.
  • 16 - Doing all the image analysis in the reconstruction step: Specifically: Doing segmentation as part of the reconstruction.
  • 15- Notebook / pipeline for handling & reconstructing large datasets
    • perform “coarse” reconstruction (binning) limited n-projections and visualize to user.
    • Cut data in chunks and do high quality reconstruction (automatically)
  • 15 - an easy way for visualization of time-resolved 3D-data of big sizes CT dataset - 1 timestep ca = 80 GB; also changing visualization eg segmented vs raw; vol vs clip. Step 1: easy handling of big data, both for visualizing and analysis.
  • 14 - Include qim3d in CIL Library (or vice versa) such that e.g. segmentation/central component outputs can be seen using a single (or few) functions working directly on the raw data.
  • 14 - Specialized reconstruction for specific tasks ie a reconstruction that lends itself to local thickness for example.
  • 14 - ASTRA, TIGRE, tomopy, etc. There is now many open source softwares for tomography, and the same goes for 3D analysis platforms (cf ILASTIK). What international efforts are undertaken to merge this, ensuring that resources are not wasted implementing the same things over and over again? Can QIM and CIL grow by joining forces?
  • 14 - Combined and/or sequential pipelines to perform image reconstruction and analysis live during beamtime. Leave the beamtime with publication ready figures. First step: Do pre-scanning of the given sample with lab-based micro-CT (or 1 test scan with SR [synchrotron radiation?] micro CT in advance) and set up the pipeline. Then do final beamtime.
  • 13 - Have an optical camera in the (CT) scanner. Take pictures during acquisition. Use computer vision/image analysis to extract object exterior surface geometry and appearance. Find calibration between CT source/detector & camera to enhance reconstruction.
  • 12 - A system for brain motion simulation.
  • 12 - A complete CT scan of the earth and build a virtual replica of it, so can explore every corner of it.
  • 12 - Run concurrent training as data analysis is not independent of data processing. First step: Ease of software installation.
  • 9 - Automatic or learned changes in discretization for coarse/fine features and prior expectations.
  • 8 - What would you implement? Unet segmentation (neural network segmentation) AI.