Deep-learning-based groupwise registration for motion correction of cardiac $T_1$ mapping

TL;DR

A novel deep-learning-based groupwise registration framework, which omits the need for a template, and registers all baseline images simultaneously, and shows further improved performance of registration and mapping over well-established baselines.

Abstract

Quantitative $T_1$ mapping by MRI is an increasingly important tool for clinical assessment of cardiovascular diseases. The cardiac $T_1$ map is derived by fitting a known signal model to a series of baseline images, while the quality of this map can be deteriorated by involuntary respiratory and cardiac motion. To correct motion, a template image is often needed to register all baseline images, but the choice of template is nontrivial, leading to inconsistent performance sensitive to image contrast. In this work, we propose a novel deep-learning-based groupwise registration framework, which omits the need for a template, and registers all baseline images simultaneously. We design two groupwise losses for this registration framework: the first is a linear principal component analysis (PCA) loss that enforces alignment of baseline images irrespective of the intensity variation, and the second is an auxiliary relaxometry loss that enforces adherence of intensity profile to the signal model. We extensively evaluated our method, termed ``PCA-Relax'', and other baseline methods on an in-house cardiac MRI dataset including both pre- and post-contrast $T_1$ sequences. All methods were evaluated under three distinct training-and-evaluation strategies, namely, standard, one-shot, and test-time-adaptation. The proposed PCA-Relax showed further improved performance of registration and mapping over well-established baselines. The proposed groupwise framework is generic and can be adapted to applications involving multiple images.

Figures (4)

• Figure 1: An overview of the proposed template-free groupwise registration framework. It takes the baseline image series $I^N$ through the registration module to predict $\phi^N$. The warped series $I^N \circ \phi^N$ undergoes PCA decomposition to calculate the groupwise PCA loss without a template. If the auxiliary mapping module is enabled, the relaxometry loss is also activated to refine the registration module.
• Figure 2: Illustration of cardiac motion correction in qMRI. Sampled (a) misaligned and (b) registered series. Voxel-wise intensity curve for (c) misaligned and (d) registered series. And (e) the comparison of eigenvalues of the correlation matrix.
• Figure 3: Boxplots of $T_1$ SD values in the myocardium, with lower values indicating better motion correction: (a) Pre-Gd (in-domain) and (b) Post-Gd (out-of-domain). All five scenarios were evaluated in three training-and-evaluation settings. One-sided Wilcoxon signed-rank tests were conducted to compare the performance of PCA-Relax against that of PCA and VM-G. Statistically significance ($p < 0.05$) is labeled with *.
• Figure 4: Estimated $T_1$ and SD maps of a (a) pre-contrast and (b) post-contrast sequence with the TTA strategy. The mean values of the SD maps in the myocardium are reported. We highlight the difference in the SD maps by the yellow boxes and arrows.