Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Intensity-based 3D motion correction for cardiac MR images

Nil Stolt-Ansó, Vasiliki Sideri-Lampretsa, Maik Dannecker, Daniel Rueckert

TL;DR

The paper tackles inter-slice misalignment in cardiac MRI caused by breath-holds by proposing a subject-specific, anatomy-free method that jointly optimizes rigid 3D motion for all SA and LA slices to maximize intensity agreement along their intersections. It formulates the problem as a global optimization over per-slice transforms $R(\boldsymbol{\theta})$ and $T(\boldsymbol{t})$, encapsulated in $\phi \in \mathbb{R}^{N\times6}$, with a pairwise $L2$ intensity-difference loss $\mathcal{L}(\phi)$ computed on intersection lines $d^{AB}$ and optimized via Adam on GPUs. The approach demonstrates robust convergence across a wide range of synthetic misalignments on 10 UK Biobank CMR subjects, achieving better rotational correction than translational correction and operating in about 30 seconds per run. By avoiding anatomical priors and relying solely on intensity information along slice intersections, it enables consistent multi-slice alignment that can improve downstream LV measurements in CMR analyses.

Abstract

Cardiac magnetic resonance (CMR) image acquisition requires subjects to hold their breath while 2D cine images are acquired. This process assumes that the heart remains in the same position across all slices. However, differences in breathhold positions or patient motion introduce 3D slice misalignments. In this work, we propose an algorithm that simultaneously aligns all SA and LA slices by maximizing the pair-wise intensity agreement between their intersections. Unlike previous works, our approach is formulated as a subject-specific optimization problem and requires no prior knowledge of the underlying anatomy. We quantitatively demonstrate that the proposed method is robust against a large range of rotations and translations by synthetically misaligning 10 motion-free datasets and aligning them back using the proposed method.

Intensity-based 3D motion correction for cardiac MR images

TL;DR

The paper tackles inter-slice misalignment in cardiac MRI caused by breath-holds by proposing a subject-specific, anatomy-free method that jointly optimizes rigid 3D motion for all SA and LA slices to maximize intensity agreement along their intersections. It formulates the problem as a global optimization over per-slice transforms and , encapsulated in , with a pairwise intensity-difference loss computed on intersection lines and optimized via Adam on GPUs. The approach demonstrates robust convergence across a wide range of synthetic misalignments on 10 UK Biobank CMR subjects, achieving better rotational correction than translational correction and operating in about 30 seconds per run. By avoiding anatomical priors and relying solely on intensity information along slice intersections, it enables consistent multi-slice alignment that can improve downstream LV measurements in CMR analyses.

Abstract

Cardiac magnetic resonance (CMR) image acquisition requires subjects to hold their breath while 2D cine images are acquired. This process assumes that the heart remains in the same position across all slices. However, differences in breathhold positions or patient motion introduce 3D slice misalignments. In this work, we propose an algorithm that simultaneously aligns all SA and LA slices by maximizing the pair-wise intensity agreement between their intersections. Unlike previous works, our approach is formulated as a subject-specific optimization problem and requires no prior knowledge of the underlying anatomy. We quantitatively demonstrate that the proposed method is robust against a large range of rotations and translations by synthetically misaligning 10 motion-free datasets and aligning them back using the proposed method.
Paper Structure (5 sections, 1 equation, 4 figures)

This paper contains 5 sections, 1 equation, 4 figures.

Table of Contents

  1. Introduction
  2. Method
  3. Experiments and results
  4. Conclusion
  5. Acknowledgements

Figures (4)

  • Figure 1: Diagram of resulting optimization on slice alignment. Ventricle contours are hand-drawn for illustrative purposes. Top row: LA 4-chamber image with intersections lines of SA slices with original orientations (left/red) and optimized orientations (right/green). Bottom row: Nearest neighbour interpolation of LA 4-chamber view from SA slices with original orientations (left) and optimized orientations (right).
  • Figure 2: Result of intensity interpolation at slice intersections before (red) and after (green) optimization. Two slice pairs are displayed: short axis (SA) to long axis (LA) 3-chamber (left), and short axis to long axis 2-chamber (right). Red and green lines depict the intersection lines of each image pair. Each pair's interpolated intersection intensities are displayed side-by-side on the vertical block. Blue segments depict padding points outside of image bounds. Areas of significant change are shown in yellow (ventricles) and cyan (stomach).
  • Figure 3: Distributions of absolute error in parameters for all runs (with respect to motion-free parameters). Absolute errors are presented for parameters before and after optimization for both rotation and translation parameters. Results are shown under various ranges of uniformly sampled initial rigid transformations. Dataset consisted of 10 aligned subjects, each of which was randomly misaligned 10 times per condition.
  • Figure 4: Distributions of mean and maximum absolute error in parameters for a given optimization run (with respect to motion-free parameters). Absolute errors are presented both rotation and translation parameters. Results are shown under various ranges of uniformly sampled initial rigid transformations. Dataset consisted of 10 aligned subjects, each of which was randomly misaligned 10 times per condition.