AI 中文总结
本研究提出MWF-MIMOSA方法，结合MZS-SSL与GACELLE框架训练MLP，实现高效同时弛豫测量与髓鞘水分数成像，速度提升超100倍，扫描时间短、分辨率高且鲁棒性好。
AI 中文摘要
定量磁共振成像（qMRI）相比传统对比度加权成像，对组织成分和病理改变具有更高的灵敏度与特异性。在各类qMRI生物标志物中，髓鞘水成像备受关注，因为髓鞘在脑功能中发挥核心作用，其改变与多种神经系统疾病密切相关。然而，传统髓鞘水分数（MWF）成像技术常受限于扫描时间长、空间分辨率低、信噪比（SNR）低以及比吸收率（SAR）高的问题。本研究提出MWF-MIMOSA，用于高效同时进行T1、T2、T2*弛豫测量、磁敏感源分离及MWF估计。为实现该目标，采用多对比度多切片零样本自监督学习（MZS-SSL）联合重建全脑复数值图像。为提升参数估计步骤的计算效率，在GACELLE GPU加速参数估计框架内训练多层感知器（MLP），以规避计算密集的布洛赫模拟过程，使MWF估计的计算速度提升超100倍。通过数值模拟评估MWF-MIMOSA的准确性与精度，活体实验结果进一步验证其鲁棒性。与现有髓鞘水成像方法的对比显示，MWF-MIMOSA与成熟方法高度相关，同时能提供更高空间分辨率的互补定量参数图，且扫描时间更短。值得注意的是，该方法可在5分钟内实现1毫米各向同性分辨率的同时多参数成像，在10分钟内实现0.7毫米各向同性分辨率的同时多参数成像。这些结果表明，MWF-MIMOSA在快速、高分辨率同时弛豫测量与髓鞘水成像方面具有应用潜力。
英文摘要
Quantitative magnetic resonance imaging (qMRI) provides improved sensitivity and specificity to tissue composition and pathological alterations compared with conventional contrast-weighted imaging. Among various qMRI biomarkers, myelin water imaging is of particular interest because myelin plays a central role in brain function and its alteration is closely associated with many neurological diseases. However, conventional myelin water fraction (MWF) mapping techniques are often limited by long scan times, low spatial resolution, reduced signal-to-noise ratio (SNR), and high specific absorption rate (SAR). Here, we propose MWF-MIMOSA for efficient simultaneous T1, T2, T2* mapping, magnetic susceptibility source separation, and MWF estimation. To achieve this, multi-contrast and multi-slice zero-shot self-supervised learning (MZS-SSL) was used to jointly reconstruct whole-brain complex-valued images. To improve computational efficiency of the parameter estimation step, a multilayer perceptron (MLP) was trained within the GACELLE GPU-accelerated parameter estimation framework to circumvent the computationally intensive Bloch simulation process, resulting in a >100-fold computational speed-up in MWF estimation. Numerical simulations were performed to evaluate the accuracy and precision of MWF-MIMOSA, and in-vivo results further demonstrated its robustness. Comparison with existing myelin water imaging methods showed that MWF-MIMOSA is highly correlated with established approaches, while providing complementary quantitative parameter maps at higher spatial resolution and with shorter scan times. Notably, simultaneous multi-parametric mapping was achieved in 5 min at 1 mm isotropic resolution, and in 10 min at 0.7 mm isotropic resolution. These results demonstrate the potential of MWF-MIMOSA for fast, high-resolution simultaneous relaxometry and myelin water imaging.
Comments43 pages, 16 figures
