De Luca Alberto, Leemans Alexander, Bertoldo Alessandra, Arrigoni Filippo, Froeling Martijn
PROVIDI Lab, Image Sciences Institute, UMC Utrecht and Utrecht University, the Netherlands.
Department of Information Engineering, University of Padova, Italy.
NMR Biomed. 2018 Nov;31(11):e3965. doi: 10.1002/nbm.3965. Epub 2018 Jul 27.
The diffusion-weighted magnetic resonance imaging (dMRI) signal measured in vivo arises from multiple diffusion domains, including hindered and restricted water pools, free water and blood pseudo-diffusion. Not accounting for the correct number of components can bias metrics obtained from model fitting because of partial volume effects that are present in, for instance, diffusion tensor imaging (DTI) and diffusion kurtosis imaging (DKI). Approaches that aim to overcome this shortcoming generally make assumptions about the number of considered components, which are not likely to hold for all voxels. The spectral analysis of the dMRI signal has been proposed to relax assumptions on the number of components. However, it currently requires a clinically challenging signal-to-noise ratio (SNR) and accounts only for two diffusion processes defined by hard thresholds. In this work, we developed a method to automatically identify the number of components in the spectral analysis, and enforced its robustness to noise, including outlier rejection and a data-driven regularization term. Furthermore, we showed how this method can be used to take into account partial volume effects in DTI and DKI fitting. The proof of concept and performance of the method were evaluated through numerical simulations and in vivo MRI data acquired at 3 T. With simulations our method reliably decomposed three diffusion components from SNR = 30. Biases in metrics derived from DTI and DKI were considerably reduced when components beyond hindered diffusion were taken into account. With the in vivo data our method determined three macro-compartments, which were consistent with hindered diffusion, free water and pseudo-diffusion. Taking free water and pseudo-diffusion into account in DKI resulted in lower mean diffusivity and higher fractional anisotropy values in both gray and white matter. In conclusion, the proposed method allows one to determine co-existing diffusion compartments without prior assumptions on their number, and to account for undesired signal contaminations within clinically achievable SNR levels.
体内测量的扩散加权磁共振成像（dMRI）信号源自多个扩散域，包括受阻和受限水池、自由水以及血液伪扩散。由于诸如扩散张量成像（DTI）和扩散峰度成像（DKI）中存在的部分容积效应，未考虑正确的成分数量会使从模型拟合中获得的指标产生偏差。旨在克服这一缺点的方法通常对所考虑的成分数量进行假设，而这些假设不太可能适用于所有体素。有人提出对dMRI信号进行频谱分析以放宽对成分数量的假设。然而，目前它需要具有临床挑战性的信噪比（SNR），并且仅考虑由硬阈值定义的两个扩散过程。在这项工作中，我们开发了一种方法来自动识别频谱分析中的成分数量，并增强其对噪声的鲁棒性，包括异常值剔除和数据驱动的正则化项。此外，我们展示了该方法如何用于在DTI和DKI拟合中考虑部分容积效应。通过数值模拟以及在3T场强下采集的体内MRI数据对该方法的概念验证和性能进行了评估。通过模拟，我们的方法在SNR = 30时可靠地分解出三个扩散成分。当考虑受阻扩散之外的成分时，DTI和DKI衍生指标中的偏差显著降低。利用体内数据，我们的方法确定了三个宏观隔室，它们与受阻扩散、自由水和伪扩散一致。在DKI中考虑自由水和伪扩散会导致灰质和白质中的平均扩散率降低以及分数各向异性值升高。总之，所提出的方法允许在不对其数量进行先验假设的情况下确定共存的扩散隔室，并在临床可实现的SNR水平内考虑不期望的信号污染。
