MOSAIC：用于三维钆延迟增强磁共振成像的自监督动态多编码重建框架
MOSAIC: A Self-supervised Dynamic Multi-encoding Reconstruction Framework for 3D Late Gadolinium Enhancement MRI
浏览论文内容
中文总结 AI 辅助
该研究提出MOSAIC自监督动态多编码重建框架，实现加速因子超1000的运动分辨自由呼吸双回波3D LGE MRI，在图像质量上优于现有方法。
中文摘要 AI 辅助
目的：开发并评估一种用于高度欠采样双回波三维钆延迟增强（3D LGE）磁共振成像的自监督动态重建框架。方法：MOSAIC直接从采集的欠采样数据中联合建模多回波图像内容、线圈灵敏度图和逐拍非刚性运动，无需完全采样的训练数据集或精确的预计算灵敏度图。与现有将采集数据分箱为不同运动状态（无论是否进行运动补偿）的方法不同，MOSAIC从每次心跳中重建运动分辨的3D LGE图像。该方法使用带有模拟心肌瘢痕的数字体模以及体内动物和人体研究进行评估。结果：在体模实验中，MOSAIC的峰值信噪比和结构相似性指数测量值均高于低秩深度先验重建方法及MOSAIC的消融变体。在动物和人体研究中，MOSAIC的盲法专家图像质量评分高于在线图像导航压缩感知和低秩深度先验重建方法。结论：MOSAIC证明了在加速因子超过1000时实现运动分辨自由呼吸双回波3D LGE MRI的可行性，相较于现有对比方法，其细节保留能力和伪影抑制能力均得到提升。
英文摘要
Purpose: To develop and evaluate a self-supervised dynamic reconstruction framework for highly undersampled dual-echo three-dimensional late gadolinium enhancement (3D LGE) MRI. Methods: MOSAIC jointly models multi-echo image content, coil sensitivity maps, and beat-specific nonrigid motion directly from acquired undersampled data, without requiring fully sampled training datasets or accurate precomputed sensitivity maps. Unlike existing methods that bin the acquired data into different motion states, with or without motion compensation, MOSAIC reconstructs a motion-resolved 3D LGE image from each heartbeat. The method was evaluated using digital phantoms with simulated myocardial scars and in vivo animal and human studies. Results: In phantom experiments, MOSAIC achieved higher peak signal-to-noise ratio and structural similarity index measure than low-rank deep image prior reconstruction and ablation variants of MOSAIC. In animal and human studies, MOSAIC achieved higher blinded expert image-quality scores than inline image-navigated compressed-sensing and low-rank deep image prior reconstructions. Conclusion: MOSAIC demonstrated the feasibility of motion-resolved free-breathing dual-echo 3D LGE MRI at acceleration factors exceeding 1,000, with improved detail preservation and artifact suppression relative to the state-of-the-art comparison methods.
