From the *High Field MR Center, Department of Biomedical Imaging and Image-Guided Therapy, Medical University of Vienna; †Orthopedic Department, Evangelisches Krankenhaus Wien, Vienna, Austria; ‡Siemens Healthcare GmbH, Erlangen, Germany; §Center of Anatomy and Cell Biology, Vienna General Hospital, and ∥Christian Doppler Laboratory for Clinical Molecular MR Imaging, Medical University of Vienna, Vienna, Austria.
Invest Radiol. 2017 Aug;52(8):488-497. doi: 10.1097/RLI.0000000000000364.
The aim of this study was to investigate the origin of random image artifacts in stimulated echo acquisition mode diffusion tensor imaging (STEAM-DTI), assess the role of averaging, develop an automated artifact postprocessing correction method using weighted mean of signal intensities (WMSIs), and compare it with other correction techniques.
Institutional review board approval and written informed consent were obtained. The right calf and thigh of 10 volunteers were scanned on a 3 T magnetic resonance imaging scanner using a STEAM-DTI sequence.Artifacts (ie, signal loss) in STEAM-based DTI, presumably caused by involuntary muscle contractions, were investigated in volunteers and ex vivo (ie, human cadaver calf and turkey leg using the same DTI parameters as for the volunteers). An automated postprocessing artifact correction method based on the WMSI was developed and compared with previous approaches (ie, iteratively reweighted linear least squares and informed robust estimation of tensors by outlier rejection [iRESTORE]). Diffusion tensor imaging and fiber tracking metrics, using different averages and artifact corrections, were compared for region of interest- and mask-based analyses. One-way repeated measures analysis of variance with Greenhouse-Geisser correction and Bonferroni post hoc tests were used to evaluate differences among all tested conditions. Qualitative assessment (ie, images quality) for native and corrected images was performed using the paired t test.
Randomly localized and shaped artifacts affected all volunteer data sets. Artifact burden during voluntary muscle contractions increased on average from 23.1% to 77.5% but were absent ex vivo. Diffusion tensor imaging metrics (mean diffusivity, fractional anisotropy, radial diffusivity, and axial diffusivity) had a heterogeneous behavior, but in the range reported by literature. Fiber track metrics (number, length, and volume) significantly improved in both calves and thighs after artifact correction in region of interest- and mask-based analyses (P < 0.05 each). Iteratively reweighted linear least squares and iRESTORE showed equivalent results, but WMSI was faster than iRESTORE. Muscle delineation and artifact load significantly improved after correction (P < 0.05 each).
Weighted mean of signal intensity correction significantly improved STEAM-based quantitative DTI analyses and fiber tracking of lower-limb muscles, providing a robust tool for musculoskeletal applications.
本研究旨在探讨在激发回波获取模式扩散张量成像（STEAM-DTI）中随机图像伪影的来源，评估平均化的作用，开发一种使用信号强度加权平均值（WMSI）的自动伪影后处理校正方法，并将其与其他校正技术进行比较。
本研究获得了机构审查委员会的批准和书面知情同意。使用 STEAM-DTI 序列在 3T 磁共振成像扫描仪上对 10 名志愿者的右小腿和大腿进行扫描。在志愿者和离体（即使用与志愿者相同的 DTI 参数的人体小腿和火鸡腿）中研究了基于 STEAM 的 DTI 中的伪影（即信号丢失），这些伪影可能是由不自主的肌肉收缩引起的。开发了一种基于 WMSI 的自动后处理伪影校正方法，并与先前的方法（即迭代重新加权线性最小二乘法和基于异常值拒绝的信息稳健张量估计[iRESTORE]）进行了比较。使用不同的平均值和伪影校正，对基于感兴趣区域和掩模的分析进行了扩散张量成像和纤维跟踪指标的比较。使用 Greenhouse-Geisser 校正和 Bonferroni 事后检验的单向重复测量方差分析评估了所有测试条件之间的差异。使用配对 t 检验对原始和校正图像进行了定性评估（即图像质量）。
随机定位和形状的伪影影响了所有志愿者的数据。自愿性肌肉收缩期间的伪影负担平均从 23.1%增加到 77.5%，但在离体时不存在。扩散张量成像指标（平均弥散度、各向异性分数、径向弥散度和轴向弥散度）表现出不均匀的行为，但在文献报道的范围内。基于感兴趣区域和掩模的分析，在纤维跟踪指标（数量、长度和体积）校正后，在小腿和大腿中均有显著改善（P<0.05 每个）。迭代重新加权线性最小二乘法和 iRESTORE 显示出等效的结果，但 WMSI 比 iRESTORE 更快。校正后肌肉描绘和伪影负荷显著改善（P<0.05 每个）。
加权信号强度校正显著改善了基于 STEAM 的定量 DTI 分析和下肢肌肉的纤维跟踪，为肌肉骨骼应用提供了一种强大的工具。
