Workshop

This year the workshop day will be held before the conference, 16th of June 2026.

XR Simulation and U-Net Segmentation for Real XCT Workflows

Description

This hands-on workshop provides an end-to-end workflow for X-ray Computed Tomography (XCT), combining physics-based simulation with machine-learning segmentation.

Session 1 (Prof Franck Vidal):

Participants will learn how to generate realistic, XCT projection data using the gVXR package. The focus will be on building simulation setups that mirror practical XCT scenarios and producing synthetic datasets suitable for downstream analysis and model training.

Session 2 (Dr Matthew Peter Jones):

Participants will train machine-learning segmentation models, focusing on the U-Net architecture. The session will demonstrate a practical training workflow and show how synthetic data from Session 1 can be incorporated to improve robustness, reduce labelling burden, and support generalisation to real CT volumes.

Both sessions are grounded in battery CT examples, with exercises designed to be practical, transferable, and directly relevant to other imaging workflows.

Who this workshop is for

• CT users interested in simulation for method development, protocol testing, or data augmentation
• Researchers working with XCT data who want a clear route from dataset creation to segmentation
• Practitioners new to ML segmentation with python who want a guided, practical introduction (not just theory)

Learning objectives

By the end of the workshop, you will be able to:

• Set up and run gVXR simulations to generate synthetic XCT datasets
• Understand how simulation choices (geometry, materials, noise, artefacts) affect the data you generate
• Build a practical U-Net segmentation training workflow for CT-derived data
• Use synthetic data augmentation strategically to improve segmentation performance and reliability

Pre-requisites

• Some familiarity with CT imaging and image analysis is recommended
• Some familiarity with Python is useful but not required
• No ML background required (we’ll focus on practical workflow and good habits)

Practical requirements

• Exercises will use pre-prepared cloud notebooks/scripts, with guided steps throughout
• All participants must bring a laptop (Windows/macOS/Linux) suitable for running Python notebooks

Training material

The slides, Jupyter notebook, some data and Python scripts can be found at: https://github.com/TomographicImaging/gVXR-training-dXCT2026. The training data (as of “training a ML model”) is on Zenodo.