Diffusion Imaging in the Post HCP Era

Abstract

Diffusion imaging is a critical component in the pursuit of developing a better understanding of the human brain. Recent technical advances promise enabling the advancement in the quality of data that can be obtained. In this review the context for different approaches relative to the Human Connectome Project are compared. Significant new gains are anticipated from the use of high-performance head gradients. These gains can be particularly large when the high-performance gradients are employed together with ultrahigh magnetic fields. Transmit array designs are critical in realizing high accelerations in diffusion-weighted (d)MRI acquisitions, while maintaining large field of view (FOV) coverage, and several techniques for optimal signal-encoding are now available. Reconstruction and processing pipelines that precisely disentangle the acquired neuroanatomical information are established and provide the foundation for the application of deep learning in the advancement of dMRI for complex tissues. Level of Evidence: 3 Technical Efficacy Stage: Stage 3.

Keywords: diffusion imaging; human connectome project; ultra high field.

Figure 1:

Summary of performance metrics for the HCP scan evaluation for a T₁=1200ms. (A) The SNR efficiency, defined as SNR/√TR, as a function of TR. (B) The relative loss in SNR per shell by using the TE for the highest b value shell for all shells instead of the minimum TE for each b value. (C) The reduction in signal for different shells for isotropic diffusion with a diffusivity of 0.0017 mm²/s for white matter representative of the corpus callosum. (D) SNR gain for different maximal gradient strength, and infinitely fast slew-rate.

Figure 2:

7T Tractography (Euler integration using Gaussian radial basis functions for orientation interpolation with 10,000 streamlines) from the primary fiber orientation in the primary motor cortical area. Center image shows 1000 streamlines and right image shows 5000 streamlines to elucidate different levels of details in the display of the fiber distribution. Background image is unbiased T1-weighted data.

Figure 3:

1/g maps for SMS/MB acceleration for different MB factors for simultaneously excited slices with R = 3 in the phase encode dimension. The g-factor calculations are performed for sagittal, coronal, and axial slices and slice accelerations (labeled on the left side of the figure) with k-space undersampling performed on the AP direction for sagittal and axial slices and in the LR direction for the axial slices. The data are presented as a MIP over an 80-mm sagittal slab superimposed on a silhouette of a sagittal slice. The mean and the maximal (at 98% level) g-factor numbers are given as 1/g values in the lower left corner for each figure. Images are from a representative subject, and numbers are averages over all participants. From (49)

Figure 4:

a) SNR efficiency for different TRs and for different excitation flip angles for spins that have a T₁ of 1200ms relative to the SNR for TR=8800ms with a 180° refocusing pulse. (b), and (c) relative spatial variation of the flip angle at 3T (b) and 7T (c). The 7T used 5 mm thick, high permittivity dielectric pads under the neck and on both sides of the head to improve the normally poor B₁⁺ in the cerebellum and temporal lobes and the pads are visible in the B₁⁺ map, on each side of the axial slice shown.

Figure 5:

Comparing HCP 7T dMRI protocol (HCP 1Tx) with the HCP protocol run with pTx pulses (HCP pTx protocol). Data are shown for one subject. Fractional anisotropy (FA) maps (left panel) and volume fraction maps for second fiber orientations (right panel). The color FA is FA (in the range of [0 1]) with the color representing the orientation of the principal fiber (red: left-right; green: A-P; blue: inferior-superior). The volume fraction map is shown in a colorscale of [0.05 0.2] (with yellow being high and red being low in volume fraction), overlaid on the respective FA map (in a grayscale of [0 1]). Both dMRI datasets were acquired with 1.05 mm isotropic resolutions, MB2, in-plane acceleration factor = 3, and TE = 71 ms. The TR for pTx acquisition was slightly longer (7400 versus 7000 ms). Both acquisitions used the same q-space sampling scheme (double shells, b-value = 1000/2000 s/mm²), corresponding to 143 unique image volumes, each acquired twice with anterior-posterior (AP) and PA phase encode directions. Total scan time was kept constant for both datasets (i.e., 40 min divided into 4 segments of 10 min each). Note that the use of pTx improved fiber orientation estimation performances not only in lower temporal lobe (as indicated by yellow arrows) but also in cerebellum (as highlighted by red arrows). Adapted from (38).

Figure 6:

Simulated maximal g-factor for SMS/MB 2D SE-EPI for Sp-Sg and slice-GRAPPA. Top; maximal g-factors using 99 percentile over a whole-brain acquisition as a function of the MB factor. For MB=[2,3,4] a kernel of 3×3 is used; for these MB factors the results are plotted using red and black colors, for Sp-Sg and slice GRAPPA, respectively, and they superimpose. For MB ≥4, a kernel of 7×7 is used; for these MB factors, the results are plotted in green and blue for Sp-Sg and slice GRAPPA, respectively. Bottom row, left, g-factor map using kernels calculated with Slice-GRAPPA, and bottom row right, g-factor map using kernels calculated with Sp-Sg.

Figure 7

Comparison of anatomical consistency with different model options in EDDY. Left column, the effect of spin-history in the presence of rapid movement, with a rigid body correction (top) and with an estimation for inter volume signal drop (bottom). Under large movement, and after correction, the corrected brain from different q-vectors should have the same brain outline. The static susceptibility map shows difference in the frontal areas (center column) as indicated with yellow arrows, and these are corrected using the dynamic susceptibility map (right most column). Adapted from https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/eddy

Figure 8

Sensitivity of different q-space sampling schemes in resolving (two and three-way) crossing fibers within the centrum semiovale. The amount of crossings resolved by each scheme is normalized by the maximum. (6)

Figure 9:

Comparison between BusineX, BedpostX, and Constrained Spherical Deconvolution (CSD), showing improved fiber parameter estimations by BusineX. Upper panels and lower-left panel show color coded orientation estimates (orientation distribution functions in the case of CSD) at the pons region highlighted in the inset view. The background is the sum of anisotropic volume fractions for BusineX and BedpostX, and fractional anisotropy (FA) for the CSD. The areas highlighted with arrows depict the improvements; the better detection of fiber crossings (violet and green arrows) and the lower estimation uncertainty (red arrows). Lower right panel shows the detected number of second (blue) and third (red) fiber crossings at two representative ROIs (left superior longitudinal fasciculus and posterior corona radiata), and its variation with acceleration in diffusion gradient directions (under-sampling factor). The improved estimations in BusineX is due to the data-dependent local learning of hyperparameters, at each voxel and for each possible fiber orientation, that moderate the strength of priors governing the parameter variances.

Figure 10:

Comparison of experimental SNR between 2 SMS/MB and 3D (Multislab-Multiband) high resolution acquisitions for whole brain, 0.9 mm isotropic resolution (reproduced from (110)), showing the SNR from the shortest VAT with what is feasible with 2D SMS/MB and 3D acquisitions with 8 slices/slab, all obtained with iPAT=2 to maintain a TE<100ms. The experimental thermal noise (g∙σ) is used as the hardware thermal noise, with g the encoding noise amplification and σ the system thermal noise. The 2D SMS with MB 2, 12s TR and 6 average (VAT=72s) is similar to the achievable SNR to a 3D acquisition using a TR=1.6s, 1 Average (VAT=19s). For approximately matched scan times of 72 s for SMS/MB and 76 s for 3D, the latter has 2x higher SNR. For matched scan-time the 3D with TR=1.6 s and 2 averages versus TR=3.2s and single average have similar SNR reflecting how the fact that SNR efficiency varies slowly near the optimal TR (Figure 3).

Figure 11:

The effect of MPPCA pre-processing on Life-span data. The MPPCA is applied to un-processed dMRI data. A slice from a volume with b=3000 s/mm² is show in A/ and the same slice after MPPCA noise variance reduction is shown in B. After post-processing as described in the “Preprocessing pipelines” the FA is calculated with FSL, and reproduced in C/ and D/ for the native DICOM and the MPPCA processed data respectively.