BIO

Evelien Zwanenburg is a Research Fellow in X-ray Imaging at WMG, University of Warwick, with over seven years of experience in the acquisition, reconstruction, and analysis of X-ray data. Her research centres on advancing high-speed imaging for manufacturing. Her previous work includes optimising acquisition parameters and the practical implementation of iterative reconstruction using CIL from CCPi. She is currently exploring methods to apply artificial intelligence in X-ray computed tomography (XCT), particularly for denoising and image segmentation while maintaining task dependent accuracy.

PROPOSAL

X-ray Computed Tomography (XCT) is vital for non-destructive internal imaging in manufacturing, from identifying weld porosity to battery design. However, slow scan speeds from 30 minutes to hours often limits XCT to prototyping. Incorporating AI into the workflow has the potential to transition XCT into pilot-line quality assessment by accelerating acquisition without compromising quality.

Evelien Zwanenburg (EZ), the proposed RTP, has spent four years optimising iterative reconstruction to support high speed imaging, achieving a 3x speed up by using only 33% of typical projection data but importantly maintaining accuracy. This has resulted in implementation with industry partners such as biscuit baking with Mondelez [1,2] requiring scans every 20s, contributions to CCPi toolboxes, and sharing best practice with the community via CCP user meetings [6,7], conferences and papers [3,4,5]. AI represents the next step change in XCT acquisition speed but while methods exist, their practical and actionable implementation for manufacturing remains unclear.

Currently in the Learn space, EZ uses commercial ML segmentation tools and has an awareness of various XCT specific open-source codes but seeks to upskill into the explore/excel space. This funding will enable her to develop methodologies for the Core Imaging Library (CIL) supported by quantified evidence of impacts on measurement accuracy, open and accessible to the whole CCPi community. STFC Users of the software include Diamond (Burca, I12), CLF (Symes, EPAC) and ISIS (Kocklemann, IMAT) demonstrating the potential reach within the facilities. The project leverages a 50% match from a Catapult project, maximising value and industrial impact for NPRAISE. EZ has demonstrated her ability to implement CCP codes in industrial environments, and the transition to AI is a natural next step. It provides an opportunity for ownership and leadership in the field, supporting women in engineering.

ML could support the XCT pipeline at different stages: projection acquisition, image reconstruction, post processing or evaluation. At the evaluation step, commercial tools focus on ML image segmentation but across the community there is a lack of transparency regarding operator influence of initial seeds and architecture comparisons. EZ will distillthese into digestible best practices for the XCT community, referencing open sourcealternatives, ensuring more accurate segmentations in published works with greater measurement accuracy.

The second focus is the preceding step in the pipeline – pre-processing image data. Higher speed image acquisitions are achieved by reducing the number of projections, but result in additional noise and image artifacts obscuring important details, leading to lower measurement accuracy. ML can be implemented at various points of the pipeline in attempts to overcome this and restore image quality and quantification to something similar to the full high-quality projection dataset. Success here would identify greater acquisition speed ups balanced with accuracy, than that found with iterative reconstruction.

[1] Zwanenburg et al. 2024 (conference) 10.13140/RG.2.2.23590.38727

[2] BBC News Article Researchers ‘crack code’ to perfect biscuit crunch, BBC June 2024

[3] Warnett et al., 2025 (journal article)  10.1016/j.precisioneng.2025.01.019

[4] Zwanenburg et al. 2023 (conference) 10.13140/RG.2.2.12697.84324

[5] Zwanenburg et al. 2026 (preprint) 10.13140/RG.2.2.22725.31202

[6] https://ccpi.ac.uk/events/first-cil-user-meeting/

[7] https://ccpi.ac.uk/events/ccpi-show-tell-and-user-support-drop-in/