Tagged-Informed Prior for Motion Quantification in Cine CMR Using Implicit Neural Representations

Laura Alvarez Florez, Soufiane Ben Haddou, Fleur V. Y. Tjong, Ivana Isgum

Keywords: Cardiac MRI, Tagged CMR, Implicit Neural Representations, Image Registration, Cardiac Motion Quantification, Prior Initialization

TL;DR: We use tagged CMR to build a population motion prior for INR-based cine CMR registration.

Abstract: Accurate quantification of myocardial motion from cine cardiac magnetic resonance (CMR) is essential for assessing cardiac function. Although tagged CMR provides high-fidelity measurements of myocardial deformation, its longer acquisition time limits routine clinical use, making cine CMR motion estimation the more widely applicable approach. Implicit neural representations (INRs) offer a promising framework for cine-based motion estimation by modelling cardiac motion as a continuous spatio-temporal function. However, they require subject-specific optimisation and are sensitive to initialization, leading to slow convergence. Furthermore, optimisation from random initialization can lead to large number of solutions that may not guarantee biomechanically plausible motion. To address these limitations, we propose a strategy to improve and accelerate INR-based registration of cine CMR by leveraging a population-level prior derived from tagged CMR data. First, we train subject-specific INRs on the tagged cine dataset to encode characteristic myocardial deformation patterns. Second, we aggregate their parameters across subjects to form a tagged-informed population prior. Third, we use this prior initialization to warm-start the optimization of cine INRs. The resulting prior provides a physiologically meaningful starting point for cine-only INR optimisation, reducing the search space and promoting more realistic cardiac motion. We develop and test the method on the UK Biobank. Compared with standard initialization, the proposed prior enables the INR to reach near-optimal performance using only half as many optimisation steps, achieving a 4\% improvement in Dice and a 15\% reduction in Hausdorff distance. These gains also translate to a test set of 855 subjects from a different institution, encompassing different pathologies, where the prior yields smoother and more physiologically plausible strain curves. The code for this research is publicly available.

Primary Subject Area: Application: Cardiology

Secondary Subject Area: Image Registration

Registration Requirement: Yes

Reproducibility: https://github.com/qurAI-amsterdam/tagged-prior-inr

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \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)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 161