Neural fields for tissue attenuation curve reconstruction in sparsely sampled time-resolved CT

Lucas de Vries, Rudolf Leonardus Mirjam Van Herten, P. Matthijs van der Sluijs, Ivana Isgum, Bart J. Emmer, Charles B.L.M. Majoie, Henk Marquering, Efstratios Gavves

Keywords: conditional neural fields, CT perfusion, multiphase CT angiography, acute ischemic stroke

Abstract: Time-resolved CT imaging can aid acute ischemic stroke diagnosis by visualizing contrast agent transport through the brain (micro)vasculature. CT perfusion imaging, while widely used for stroke diagnosis, requires approximately 30 sequential scans, leading to extensive radiation exposure and motion sensitivity. As an alternative to CTP perfusion imaging, some hospitals opt for multiphase CT angiography for time-resolved analysis with reduced radiation dose. However, multiphase CT angiography lacks standardized perfusion analysis capabilities, making it more challenging to interpret than CT perfusion imaging. We present Sparse Temporal Attenuation Reconstruction (STAR), a novel approach using conditional neural fields that reconstructs tissue attenuation curves from sparse observations, allowing for reduced radiation exposure and motion sensitivity with CT perfusion, while enabling perfusion analysis from multiphase CT angiography. Our method generates full tissue attenuation curves using only 4 out of 30 observations. The results show that perfusion maps from reconstructed data match the reference perfusion maps, potentially reducing radiation and allowing recovery of motion-corrupted images. Moreover, STAR enables perfusion analysis in centers using multiphase CT angiography. Consequently, STAR has the potential to improve the stroke imaging work-up while making perfusion analysis more widely accessible.

Primary Subject Area: Unsupervised Learning and Representation Learning

Secondary Subject Area: Learning with Noisy Labels and Limited Data

Paper Type: Methodological Development

Registration Requirement: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., All authors and co-authors are correctly listed with proper spelling and avoid Unicode characters., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 50