Cardiac cine magnetic resonance imaging not only requires higher imaging speed but also needs to address motion artifacts. Especially in the case of free-breathing, more motion artifacts are inevitably introduced. This poses higher demands on the reconstruction performance of the model and its ability to capture temporal information. Previous methods have not effectively utilized the temporal dimension information to compensate for motion artifacts. In order to fully leverage the spatiotemporal information and reduce the impact of motion artifacts, this paper proposes a motion-guided deformable alignment method with second-order bidirectional propagation. Furthermore, aligning adjacent frames may lead to low accuracy or misalignment issues, which are detrimental to subsequent fusion reconstruction. Previous methods have not sufficiently integrated and corrected the aligned feature information. This paper proposes a multi-resolution fusion method to further correct alignment errors or artifacts. Compared to other advanced methods, the proposed approach achieves better image reconstruction quality in terms of peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and visual effects. The source code will be made available on https://github.com/GtLinyer/MDAMF.

翻译：心脏电影磁共振成像不仅需要更高的成像速度，还需应对运动伪影问题。特别是在自由呼吸状态下，不可避免地会引入更多运动伪影，这对模型的重建性能及其捕获时间信息的能力提出了更高要求。以往方法未能有效利用时间维度信息补偿运动伪影。为充分挖掘时空信息并减少运动伪影的影响，本文提出了一种基于运动引导的可变形对齐方法，并引入二阶双向传播机制。此外，相邻帧的对齐可能导致精度降低或错位问题，这对后续融合重建不利，而以往方法未能充分整合与校正对齐后的特征信息。为此，本文提出了一种多分辨率融合方法，以进一步校正对齐误差或伪影。与现有先进方法相比，本文方法在峰值信噪比（PSNR）、结构相似性指数（SSIM）及视觉效果方面均实现了更优的图像重建质量。源代码将于https://github.com/GtLinyer/MDAMF 公开。