I.e. what are the pitfalls of the MDD methods/algorithms in terms of inferring the structural information of the imaged tissue?

To see this working, head to your live site.

Search

May 03

# Are there any particular scenarios in which one should be careful about the interpretation of MDD results?

Are there any particular scenarios in which one should be careful about the interpretation of MDD results?

1 answer0 replies

0

Let us tease apart two main families of signal inversion techniques here,

cumulant expansions, such as

q-space trajectory imaging(QTI),other approaches, such as

diffusion tensor distribution imaging(DTD),and later discuss the problem of voxel sizes.

Cumulant expansions of the diffusion signal are only accurate in the low b-value regime but are often fed data including rather high b-values such as

b= 2000 s/mm2. While this implies an inherent loss of accuracy, high b-values increase precision on parameter estimation, making cumulant expansions quite precise techniques. However, these approaches are not necessarily bounded to physical solutions. For instance two-term cumulant expansions (e.g., QTI) are not physically bounded, as they implicitly assume that the intravoxel distribution of diffusion tensors is Gaussian (which spans a space comprising negative definite tensors that cannot be diffusion tensors). Consequently, these approaches may give unphysical results in low-SNR voxels. Besides, microstructural details such as intravoxel fiber dispersion are contained in higher-order cumulants and are, therefore, not properly captured by these techniques. Three-term cumulant expansions, such asdiffusion skewness tensor imaging, mitigate this problem but require comprehensive acquisition schemes.DTD is physically bounded and can provide quite accurate measurements given optimization of the acquisition sampling scheme (

Reymbaut 2021) and denoising of the acquired data (Martin). However, it remains unclear to which extent DTD is able to tease apart distinct sub-voxel components, either in terms of diffusion-profile differences or in terms of minimal signal fraction.et al.2021Finally, an important but often overlooked aspect of multidimensional diffusion analysis is that the rather low SNR levels of multidimensional diffusion data sometimes requires bigger voxel sizes to maintain reasonable SNR. For instance, voxels could be made quite anisotropic (

e.g., 2x2x4 or 2x2x6 mm3) or isotropic but voluminous (e.g., 3x3x3 mm3). Using big voxel sizes can enhance partial voluming effects, either between different tissue types or, in the brain, between different fiber populations. Note that recent work has achieved a resolution of (1.6)x(1.6)x(1.6) mm3 in spherically diffusion encoded images using super-resolution techniques (Viset al.2021). Hopefully, more works tackling this issue will appear in the years to come!