In cardiac CINE, motion-compensated MR reconstruction (MCMR) is an effective approach to address highly undersampled acquisitions by incorporating motion information between frames. In this work, we propose a deep learning-based framework to address the MCMR problem efficiently. Contrary to state-of-the-art (SOTA) MCMR methods which break the original problem into two sub-optimization problems, i.e. motion estimation and reconstruction, we formulate this problem as a single entity with one single optimization. We discard the canonical motion-warping loss (similarity measurement between motion-warped images and target images) to estimate the motion, but drive the motion estimation process directly by the final reconstruction performance. The higher reconstruction quality is achieved without using any smoothness loss terms and without iterative processing between motion estimation and reconstruction. Therefore, we avoid non-trivial loss weighting factors tuning and time-consuming iterative processing. Experiments on 43 in-house acquired 2D CINE datasets indicate that the proposed MCMR framework can deliver artifact-free motion estimation and high-quality MR images even for imaging accelerations up to 20x. The proposed framework is compared to SOTA non-MCMR and MCMR methods and outperforms these methods qualitatively and quantitatively in all applied metrics across all experiments with different acceleration rates.

翻译：在心脏电影成像中，运动补偿磁共振重建（MCMR）是一种通过整合帧间运动信息来解决高度欠采样采集问题的有效方法。本文提出了一种基于深度学习的框架，以高效处理MCMR问题。与当前最先进的（SOTA）MCMR方法将原始问题分解为两个子优化问题（即运动估计与重建）不同，我们将此问题表述为具有单一优化的整体实体。我们摒弃了传统的运动扭曲损失（运动扭曲图像与目标图像之间的相似性度量）来估计运动，而是直接由最终重建性能驱动运动估计过程。在不使用任何平滑损失项、且无需在运动估计与重建之间进行迭代处理的情况下，实现了更高的重建质量。因此，我们避免了繁琐的损失加权因子调整和耗时的迭代处理。在43个内部采集的2D电影数据集上的实验表明，所提出的MCMR框架即使在高达20倍加速成像条件下，也能提供无伪影的运动估计和高质量的MR图像。所提出的框架与SOTA非MCMR和MCMR方法进行了比较，在不同加速率的所有实验中，该框架在所有应用指标上均优于这些方法，表现在定性和定量两方面。