FIGURE 4: (a) 3D rendering of structural MRI data used for coregistration with (Fig 3a) µCT volumetric data. Somas and dendrites of neurons in the medial vestibular nucleus (e) and somas in (g) dentate gyrus were traced from the µCT data. While these somas are much smaller than the resolution of the MRI data, they provide possible underlying sources of contrast in the T2* weighted images.

FIGURE 2: The same brain was imaged using (a, b) MRI (50 mm isotropic voxels), (c, d) µCT (1.2 mm isotropic voxels), and (e, f) EM (3 nm in-plane resolution). Panels (b) and (c) show the same FOV from the MRI and µCT imaging (the red ROI in (a)). Panels (d) and (e) show magnified µCT and EM data (the green ROI in (c)). Yellow arrows indicate corresponding blood vessels, and a single neuron is colored purple. Panel (f) shows an individual somatic synapse (white arrows) on that soma, colored orange (the blue ROI in (e)).

Figure 1: Sequence employed for the 3D SPEN DTI acquisitions, including a 180˚ chirp pulse acting in the presence of a gradient that encodes the more artifact-prone, low bandwidth dimension, a pre-encoding Ta/2 delay introduced for achieving SPEN’s full-refocusing condition where variable-orientation diffusion-weighting gradients (red) are placed, and gradients for interleaving and phase-encoding procedures. A final, 2D echo-based reacquisition of the k=0 PE line was included for all interleaves, for correcting motion phase distortions in these experiments.

Figure 3: Comparison between in vivo EPI and SPEN images extracted from 3D DTI acquisitions on a 2-days-old live mouse. The in-plane fields targeted in both experiments were 10x10 mm.

Figure 1. A) Representative ADC’(tm) map in slice thickness of 1mm, 1.5mm and 3mm with no crusher and 1.5mm with negative crusher. B) Normalized ADC’(tm)-vs-tm from 5 ms to 305 ms in ROI of brain (yellow). C) Normalized ADC’(tm)-vs-tm with mixing time from 5 ms to 105 ms. D) AXR estimated from 5-105 ms. E) ADC’(tm) in a phantom with filter gradient and readout gradient as 100 s/mm².

Figure 2. A) Representative MD’(tm) map with mixing time from 5 ms to 105 ms. B) MD’(tm)-vs-tm curve in cerebral cortex and corpus callosum with mixing time from 5 ms to 105 ms. C) shows the structural reference T2-RARE images along with fitted voxel-wise filter efficiency map and AXR with mixing time from 5 ms to 105 ms.

Fig. 4: Simulated ADC in the PyCA1. a) ADC as a function of decreasing intracellular fraction (ICF), for values between the SBEM-derived control and SE data and at a constant size distribution. b) ADC as a function of a cell volume, with swelling of a constant number of cells between the control (1x) and SE (1.95x) size distributions. Distinct curves are shown for cell vs. cytoplasmic swelling. c) ADC vs. frequency curves for the combined effects with simultaneously decreased ICF and cytoplasmic swelling between the control and SE SBEM values, reflecting the observed histopathology.

Fig. 1: PGSE and OGSE dMRI of the control and SE rat hippocampi. 0 Hz is the PGSE data point. a) Frequency dependent ADC curves in the following gray matter ROIs show characteristic differences with increasing frequencies: pyramidal cell layer of the CA1 (PyCA1); cortex; granule cell layer of the dentate gyrus (GCL); stratum lacunosum-moleculare (SLM); and stratum radiatum (SR). Shown are the average curves (solid) and the individual rat curves (dashed). b) The ROI delineations, shown overlaid on the group average ADC map at 180 Hz for control rats.

Figure 2. QTE contrasts. A,C) Examples of denoised images with B-tensor encoding (b=2800 s/mm²) in a single slice from one representative animal per experimental condition. Yellow rectangles indicate enlarged areas in adjacent panels. B,D) QTE metrics for intact and injured optic nerves.

Figure 4. DTD imaging. A) and B) show examples of DTD representations obtained in regions as in Figure 2. C) and D) show the overlap of the DTDs for all voxels contained in the slice.

Figure 4: Comparison of microstructural maps that result from different myelin sensitive contrasts when combined with outputs from the CODIVIDE model. These MVF maps report lower values than the ones derived from NODDI outputs. While these AVF maps are quite similar between myelin-metrics, they are strikingly different than the ones from the NODDI output. Here, the contrast of the AVF maps has flipped with WM presenting larger values than GM.

Figure 5: Boxplot comparisons of g-ratio values calculated from different myelin-sensitive MRI metrics, within several WM tracts. Each plot presents the g-ratio values derived from one myelin-sensitive MRI metric with either NODDI or CODIVIDE modelling of the diffusion data. The ROIs chosen are the genu (A) and splenium (B) of the corpus callosum, the corona radiata (C), the internal capsule (D), and the stratum calcarinum (E). The red line indicates a g-ratio of 0.7.

Figure 2: Spectral content for the STE from Fig. 1 color coded based on 3 frequency bands with equal encoding power determined from $$$s(\omega)$$$ and its cumulative sum (inset). Projections 1-3 along the SPAS (dotted lines), relative to the laboratory frame (XYZ), have the most (along $$$\bf{u}_\mathrm{LF}$$$) to the least power in the low frequency band. The orange lines illustrate an ODF and its main direction $$$\boldsymbol{\mu}$$$ for which the angle $$$\sigma$$$ can be calculated.

Figure 3: STP differentiates sticks from cylinders. Noiseless calculations for Watson ODF with varying OP (columns) of axisymmetric restrictions with varying $$$D_\Delta$$$: sticks ($$$d = 5\mathrm{\mu m}, D_\Delta = 1$$$), prolate ellipsoids ($$$R_3/R_1 = 5\mathrm{\mu m}/1\mathrm{\mu m}, D_\Delta = 0.99$$$), cylinders ($$$R = 5\mathrm{\mu m}, D_\Delta = 0.38$$$), oblate ellipsoids ($$$R_3/R_1 = 1\mathrm{\mu m}/5\mathrm{\mu m}, D_\Delta = -0.5$$$). In all cases, $$$D_0 = 2\times 10^{-9} m^2/s$$$. Compare the marked examples (dotted) with experimental results (Fig. 4)

Figure 2: Raw DWI b1000 (pre-scan normalize off) and smoothed b-tensor-derived diffusion metric maps are shown for two acute stroke patients. In both cases, the low region of mean diffusivity (MD) shows no change on standard tensor fractional anisotropy (FA), but greatly elevated microscopic FA (µFA), anisotropic, isotropic, and total kurtosis (MK_A, MK_I, MK_T) within the ischemic white matter.

Figure 3: The mean ± SD over 21 acute stroke patients are shown for regions over multiple slices of WM and GM within the lesion (‘L’) and contralateral (‘C’) brain. (A) Lesion MD was reduced by 39% in WM and 38% in GM. (B) FA was 14% lower in the lesion relative to contralateral WM. (C-F) In contrast, the b-tensor derived parameters were all elevated in the lesion WM by 9% for microscopic FA (µFA), 54% for anisotropic kurtosis (MK_A), 75% for isotropic kurtosis (MK_I), and 58% for total kurtosis (MK_T). Similar changes were found in ischemic GM.

Fig.1: (A) Color-coded FA, (B) principal eigenvector, and (C) RI maps overlaid on FA, showing a band of low FA and a radial diffusion orientation in most cortical regions, but a tangential orientation and lower RI in the post-central gyrus. (D) Axial and (E) 3D views of cortical columns connecting the pial surface mesh and the WM/GM surface mesh. (F) Single-column FA/RI vs. cortical depth profiles showing an FA local maximum and minimum (separated by FA_diff) and a single RI maximum (RI_max).

Fig.5: (A,B) FA/RI vs. curvature at different cortical depths. (C,D) FA_diff/RI_max vs. curvature. (E,F) FA/RI vs. cortical depth at different curvatures. The denser profiles in A correspond to the FA and FA_diff peaks in E and C (white/green/magenta arrowheads). RI decreases from crown to fundus more drastically at larger cortical depths, while RI_max occurs at smaller cortical depths towards the crown (orange arrowheads). The post-central ROI shows a lower RI/RI_max and FA_diff (red arrowheads).

I. a) Mean kurtosis b) tractography showing branching between EC and EmC (left) and thalamic connectivity (right). Ca: caudate. Cl: claustrum. EC: external capsule. Emc: extreme capsule. IC: internal capsule. GPi/GPe: internal/external globus pallidus. Th: thalamus. Pu: putamen. TR: thalamic radiation.

II a-b) CSD a-b1,2,3: in a-b. 1) CSD-based maps (0-th order SH) showing main hippocampal formation structures. 2) tractography showing internal hippocampal connectivity. 3) lower spatial resolution .

Primary eigenvectors of diffusion tensor imaging on top of mean kurtosis maps (a-c) from granular cortices where cortical fibers that are tangential to the cortical surface exist, shown in regions of interest selected from the primary somatosensory (d, red box), primary auditory (e, green box) and primary visual (c, blue box) cortex.

Co-registered axial images of mid-pons using FGATIR, standard and super-resolution fractional anisotropy with overlaid segmentations of corticospinal tract (PON; dark yellow), pontine reticular formation (PRF; light yellow) and medial longitudinal fasciculus (MLF; red) created by neuroanatomy expert using FGATIR contrast propagated to the diffusion data. The round shape of the PON segmentation is better preserved with super-resolution. There is less CSF contamination of the MLF from the 4th ventricle with super-resolution.

Overview of the pipeline. Images are labelled as solid blue blocks and processing steps are labelled as light blue blocks. The pipeline relies on denoised FGATIR data that undergoes diffeomorphic registration through a 3D T2 weighted image to a super-resolution enhanced diffusion weighted dataset.

Figure1: A) Probabilistic (n = 12) atlas label of nine brainstem nuclei overlaid on the group averaged FA or T₂w image. We observed very good (up to 100 %) spatial agreement of labels across subjects indicating the feasibility of delineating these nuclei based on 7 Tesla MRI contrast and neighborhood rules⁷. B) For each nucleus, the inter-rater agreement and internal consistency of nuclei labels were below (p < 0.05) the linear imaging resolution (1.1 mm), thus validating the nuclei atlas. Except for sMRt, volumes of nuclei labels in-vivo did not differ (p < 0.05) from literature values⁷.

Figure 4. Top) Structural connectome and Bottom) streamline density (n=19) of two pontine brainstem nuclei: A. right LC-r and B. left LDTg-CGPn (-log₁₀(p-value) displayed, Wilcoxon test). LC showed connectivity with bilateral iMRt, sMRt, HTH, striatum and widespread cortical connectivity including limbic areas in line with previous studies². LDTg-CGPn, described previously as the most extensive group of the rostral braintem² showed connectivity with HTH, VTA, PAG, basal forebrain, frontal and limbic areas as expected from previous literature^2,7.

Figure 3. Mean and standard deviation of dot product across each gradient direction. (A) Same plot as in figure 2. Depicting mean dot product by gradient direction across all participants and hippocampi (N = 192). (B) Mean dot product within each gradient direction across each subfield. Error bars represent one standard deviation. One-way ANOVA and Tukey’s post hoc test revealed significant differences of mean dot product between subfields within gradient direction. Legend depicts significant correlations. (C-E) Standard deviation of dot product for each gradient direction.

Figure 1. Hippocampal morphology and microstructure. (A-C) Laplacian potential fields in anterior-posterior, proximal distal, and inner-outer dimensions. (C) Coronal slice. (D-F) Gradient vectors from first derivation of potential field in (A-C). (G) Sagittal slice - NODDI neurite vectors on a hippocampal mask. (H) Rectangle in (G) – zoom of NODDI neurite vectors. (I) Square in (H) representing one voxel and one neurite vector. Red, purple, and blue lines depict gradient vectors at that voxel for anterior-posterior, proximal-distal, and inner-outer, respectively.

Figure 2 Mapping of the mean diffusion metrics over 17,646 subjects on the cortex: fractional anisotropy (FA), mean kurtosis (MK), and axonal water fraction (intra) from SMT MC model.

Figure 4. Brain age predictions using FA, MK, and SMT MC intra diffusion metrics estimated over Desikan-Killiany atlas of cerebral cortex. The predicted brain age values have been corrected for age bias. The predictions were estimated using XGBoost algorithm for linear model with sex and scanner site variables as covariance.

Figure 2. Cortical column analogues (a) derived from surface normal vectors spanning between white matter and the pial surface. For analysis, the right mesial temporal lobe is shown as a single region (b) and partitioned into 15 regions (c). To observe depth dependence of diffusion anisotropy, cortical FA maps from DTI data acquired at 0.8mm (d) and 0.9mm (e) are projected onto the column analogues in each region of interest.

Figure 3. Cortical profiles of fractional anisotropy (FA) projected onto all column analogues in the right mesial temporal lobe (a) and in each of 15 sub-regions (b). Profiles from 0.8mm (blue) and 0.9mm (pink) DTI data show an increase in FA with cortical depth from the pial surface (PS) to the white matter (WM) boundary, with local extrema in the middle. Similar profiles for mean diffusivity (MD) (c-d) show a reduction in diffusivity across the cortex from the PS to the WM boundary, with no local extrema.

Pearson correlation coefficient (r) obtained by pair-wise tests between parameters measured from cortical region and tract-weighted maps for 10 subjects (N = 10). Correlation coefficient was statistically significant in each test reported above (p < 0.05). The statistical tests were performed on the data projected on the mid-cortical surface (so that each data set has the same number of data points) to assess the relative correspondence of each parameter on others. Only the correlation coefficients > 0.3 are reported.

Group average (N = 10) representation of AFD_total (a, b), T1w/T2w (c, d), FA (e, f), track-weighted FOD_amplitude (g, h) and track-weighted FA (i, j) maps displayed on inflated surfaces for the left hemisphere (column 1) and right hemisphere (column 2). The volumetric data (a - f) were averaged over 70% of the cortical thickness and projected at the mid-cortical surface. The track-weighted data (g-j), computed with fwhm tract length of 40 mm, were sampled at the GM-WM boundary and were projected at the mid-cortical surface.

Figure 4. Generalized anisotropy profiles for WM bundles and subcortical GM structures in the representative slice.

Figure 1. Coronal view of the ex vivo sample (left), FA map (middle), and segmented WM and GM structures from one slice (right).

Figure 1: Co-registration of the anterior and posterior diffusion MRI data.

Figure 2: Main vector orientation modulated by the FA map for the newborn (A) and the adult (B) marmoset brains.

Figure 3: for the validated DWI-model + $$$AWF$$$-model. NODDIDTI has the largest $$$AIC$$$ out-of-the-box, i.e. with no adjustment between biomarker and axonal water fraction. While in general, all DWI models benefit significantly from additional calibration parameters, WMTI appears to benefit the most from an additional offset and both NODDI-models benefit from an additional scaling. The introduction of a fully linear calibration does not lead to significant further improvements in terms of $$$AIC$$$.

Figure 1: Schematic of the relationship between three-compartment WM tissue model and the DWI. Here, WM is assumed to be composed of three tissue compartments: axonal ($$$AVF$$$), myelin ($$$MVF$$$), and extracellular volume fraction ($$$EVF$$$). Since myelin is DWI-invisible, the axonal compartment accessible via DWI is the axonal water fraction ($$$AWF$$$). Hence DWI-based biomarkers for the axonal compartment have to be rescaled by to yield an estimate

$$f_A=(1-\mu)AWF\qquad (1)$$

of the $$$AVF$$$. This information has to be acquired from another technique.

Figure 4 Linear correlation plots between RMSE (ordinate) and MK (abscissa). Different colors represent different ROIs, all of the four subjects were plotted on the same figure. RMSEs are calculated from discrepancy between FA derived from high (b=6000 s/mm²) and low (b=4000 s/mm²) b-value datasets. PLIC of subject 1 and subject 4 was excluded in this analysis due to their insufficient SNR.

Figure 2 FA values in the five ROIs at three b-values (blue: b=2000; orange: b=4000; green : b=6000 s/mm²) of four subjects (A-D). The solid lines show the median of the distribution, the shaded regions indicate the range (25%, 75% percentile) across voxels in each ROI.

RGB maps from the multiple reconstruction methods. Note the higher contrasts present in the DIAMOND map, from which the effect of free water is removed, leading to higher intensities of the underlying FA, and thus higher contrasts.

Peaks extracted from (A) DTI, (B) fODF, and (C) DIAMOND. Whereas DTI cannot reconstruct the multi-peak populations, two-way crossings were successfully extracted from both fODF and DIAMOND. The latter presents a greater number of crossings.

Figure 1: Diffusion kurtosis maps. (a) Mean diffusivity (µm²/ms). (b) Fractional anisotropy. (c) Mean kurtosis.

Figure 3: Animated GIF showing fODFs with and without GNL correction in the frontal lobe. Subtle but distinct differences can be seen after correction.

Figure 2: We isolate the segmentation mask of the intra-axonal volume. The pixel centers are used to compute a concave hull⁶ and extract a boundary. The boundary is used to compute the heuristic diameters: perimeter, area, biggest inscribed-circle, smallest enclosing-circle and ellipse fit. The mask is used as a boundary for a MC simulation of diffusion. The MC trajectories are used to compute the directional MSD of the shape, which is converted to a ground-truth diffusion-specific diameter.

Figure 3: Axon diameter count distribution for all heuristics and MC based ground-truth. The densities were generated with a gaussian kernel density estimator. The proposed method approximates very accurately the ground-truth distribution in the case of high resolution TEM data. For the SEM data, all the distributions are close, except for the heavily underestimating method of inscribed-circle and ellipse short-axis.

Fig. 3: NOGSE contrast $$$\Delta M$$$as a function of the restriction length $$$l_c$$$ and the gradient strength $$$G$$$, considering $$$N = 8$$$ and a constant value for $$$T_EG$$$ by properly choosing $$$T_E$$$ for each gradient. The considered dynamic range for the gradient strength $$$G$$$ is achievable with current technologies.

Fig. 4: (a) Two images based on NOGSEc of the Corpus Callosum of an ex-vivo mouse brain for two gradient strengths. For both images $$$N = 2$$$ and $$$T_E = 21.5$$$ ms. (b) Average NOGSEc signal (symbols) as a function of the gradient strength for the ROIs marked in (a). The solid-lines is the fitting of our theoretical model for a log-normal distribution. The fitted parameters are the median $$$1.34 ± 0.04$$$ and $$$2.22 ± 0.02$$$ μm and geometric standard deviation $$$1.82 ± 0.02$$$ and $$$1.82 ± 0.02$$$ μm for ROI 1 and ROI 2 respectively. We considered $$$D_0 = 0.7 μ$$$m$$$^2/$$$ms.

Fig. 4: Comparing MRI-based axon radius index (ARI) estimation with the histological manually labeled electron microscopy (mlEM) and automatically labelled large-scale light microscopy (lsLM) for five region of interests. (a): Depicted is a scatter plot of the estimated ARIs from MRI against the effective axon radius from mlEM. (b): Depicted is scatter plot of the estimated ARIs from MRI against the effective axon radius from lsLM. The RMSE in mlEM is 0.49 µm whereas it is 0.1 µm in lsLM.

Fig. 5: Assessing the accuracy of MRI-based axon radius index (ARI) estimation by comparison with large-scale light microscopy (lsLM) across eighteen regions of interests in the corpus callosum. From the eighteen investigated regions, only thirteen were used in this analysis. The remaining five regions were insufficient tissue or data quality, either in lsLM or MRI.

Figure 2. Results for representative individual voxels (crosses on the S(b = 0) map) in an ex vivo rat brain at 90 µm³ isometric resolution. The D(ω)-distributions, shown as projections onto the 2D plane of isotropic diffusivity D_iso and squared normalized anisotropy D_Δ² with gray scale of contour lines given by the frequency ω, are estimated from the b(ω)-encoded signals (circles: measured, points: fit) by Monte Carlo inversion (45,48). Segmentation into tissue types is performed by defining bins in the D_iso-D_Δ² plane and calculating per-bin signal fractions f_bin1, f_bin2, and f_bin3.

Figure 4. Per-voxel statistical descriptors E[x], Var[x], and Cov[x,y] over the D_iso and D_Δ² dimensions of the D(ω)-distributions for two selected frequencies ω/2π = 80 and 180 Hz (top and middle rows) and the rate of change with frequency Δ_ω/2π of the various metrics (bottom row). The white arrow indicates the hippocampus with elevated values of Δ_ω/2πE[D_iso] in the pyramidal and granule cell layer.

Figure 4. Maps for $$$f$$$, $$$p_2$$$ and $$$D_a$$$ of the SM at the level of the optic nerves for one animal at each time point. For 7 and 30 days post-injury, these maps show clear difference between the intact/left (L) and injured/right (R) nerve.

Figure 5. Spearman correlations between histology derived metrics and SM parameters, bold values indicate p<0.05. $$$f$$$, $$$p_2$$$ and $$$D_a$$$ correlate significantly with most histology values. In particular all three show high correlations with axon density. The histology metric most directly related to the SM model, $$$f_{hist}$$$, correlates positively and significantly with its SM counterpart $$$f$$$, indicating axonal loss.

Figure 4. Preliminary data (one mouse per row). MD dependence on frequency is shown 48H-post 5-mTBI and 48H-post CHI-RF in A) the corpus callosum and B) the PFC. All error bars represent the standard deviation in the ROI (in one mouse) and the dotted lines are the least square fits to frequency^0.5. ΔMD_base (ΔMD at baseline) and ΔMD_post (ΔMD post-TBI) are reported for each case. Baseline and 48H-post TBI data are also shown for µA (C), K_LIN (D), and K_ST (E) in the corpus callosum (CC) and PFC.

Figure 3. Parameter maps from the same mouse taken at baseline and 48H-post CHI-RF. The direction-encoded color (DEC) maps were acquired using the b = 2000 s/mm² linear acquisitions from the µA protocol. Microscopic anisotropy (µA), linear kurtosis (K_LIN), and spherical tensor kurtosis (K_ST) were estimated from the µA protocol. From the OGSE protocol, mean diffusivity (MD) maps are shown at 0 Hz and 190 Hz.

Figure 1: Left: Axon from the corpus callosum of a vervet monkey brain (segmented from 3D x-ray nano-holotomography). Right: Cross-sections of the perturbations $$$\Delta B$$$ of an applied field of 7 T computed with 0.1 μm isotropic resolution with an in-house software for parallel (blue) and perpendicular (orange) orientation w.r.t. the field. The field is most affected when fibers are oriented perpendicular to the field. The gradients are stronger perpendicular to the axon than parallel to it.

Figure 4: 1st, 2nd, and 3rd row shows the signal from CC, CING, and IC, respectively. 1st, 2nd, and 3rd column show the signal from the b-vectors closest to x, y, and z, respectively. Hence, the diagonal shows the signal measured parallel to the fibers, and the off-diagonal shows signal measured perpendicular to the fibers. Each point represents the average of all voxels of interest. Number of voxels varied across scans: 681 to 816, 209 to 283, and 64 to 93 for CC, CING, and IC, respectively.

Figure 4. Scatter plot of estimated myelin content at the medial croups callosum of demyelinated mice vs. healthy controls. Values were calculated based on immunohistochemical (IHC) staining vs. MRI based myelin water fractions (MWFs) fitted using the suggested mcT₂ technique. Good agreement is shown between the two techniques, attesting to the ability to classify the two groups and detect demyelination based on MRI derived MWF values.

Figure 3. Example myelin water fraction (MWF) maps based on MRI derived MWF values. MWF values are presented along the croups callosum (CC) for cuprizone (left) and (right) control mice showing lower values for the cuprizone mouse. Maps are shown on top the T₂-wighted image (5^th echo) obtained using multi-echo-spic-echo MRI protocol and with the same colormap.

Figure 3. Significant group-by-time interaction observed in GFA and FA.

Clusters with significant interaction in the 2 × 2 mixed-design ANOVA (uncorrected P< 0.001, cluster threshold= 100 voxels) of GFA (upper low) and FA (lower row). The clusters are overlaid on the MPRAGE images. The look-up table indicates F-values for group × time interaction. Cool color indicates post-training assessment < pre-training assessment, training < control, and hot color indicates pre-training assessment < post-training assessment, control< training.

Figure 4. Alterations of resting-state functional connectivity in significant GFA Clusters.

The brain voxels which are functionally connected to the right inferior parietal lobule cluster (Blue) reveal a significant interaction in GFA using a 2 × 2 mixed-design ANOVA. The rsFC between this region and the left frontal pole, the inferior frontal gyrus and the right lateral occipital cortex, increases upon cognitive training (FWE-corrected P< 0.05). Significant clusters are shown on ch2better.nii template using MRIcron. The look-up table indicates the T-value.

Figure 1. Surface analysis output: the t-statistic is reported vertex-wise for each of the diffusion metrics. The direction of the learning related changes is color-coded, bluish for decrease, reddish for increase of the parameter value. Each row corresponds to a different metric. We report only those metrics with at least a significant cluster after cluster-wise correction (figure 2). From top to bottom: DTI’s mean diffusivity (MD); NODDI’s neurite density index (NDI) and free water fraction (FWF); CHARMED’s mean diffusivity of the hindered compartment (hMD).

Figure 2. Surface analysis - significant clusters: same as figure 1 but reporting the clusters of vertices found significant after cluster-wise correction. The cluster-forming threshold was set at p<0.001. We report clusters with p_FWE<0.05. We use different colours to help comparing groups of clusters across metrics. We identified an occipital group (yellow), a sub parietal group (cyan), a temporal group (red), a precentral sulcus group (magenta) a central sulcus group (green) and a postcentral sulcus group (blue).

Figure 2. Changes in FA from voxel-wise analyses. a) Decrease in FA in the LRN group in WM tracts underlying S1 during overall learning (d1-d5). b) Decrease in FA in LRN and increase in SMP in WM tracts underlying M1 during overall learning (d1-d5). c-d) Mean changes in FA across time in both groups in the right S1 (c) and in the left M1 (d). Expressed as relative changes from baseline (d1). LRN: learning group (in blue); SMP: control group (in orange).

Figure 3. Changes in WM microstructure in the ROI underlying the right supplementary area (SMA) where changes in functional connectivity were found (unpublished). a) The right SMA ROI from resting-state analyses (in red) and the WM ROI (in blue; overlaid on the WM mask in white) are both overlaid on the MNI152 template. b-d) Mean changes in DTI metrics from baseline (d1) in both groups: FA (b) and AD (c) decreased in the LRN group between d1 and d2 and remained lower at d5 and d17. RD increased between d1 and d2 in LRN and remained higher at d5 and d17 (d).

The cluster in the left middle frontal gyrus (MNI coordinates: X=-36, Y=54, Z=-6) showing a decrease in μFA upon the 4-week neurocognitive training. The cluster is shown as an overlay on a 3D-MPRAGE template.

Scatterplots showing significant negative correlation between the change in μFA of the left middle frontal gyrus and the changes in the response time as assessed by the orienting attention network test (r=-0.508, P=0.037). The straight and curved lines indicate the mean and 95% confidence interval.

Figure 2. Sagittal view of a 3D representation of the regions of interest described in this report : arcuate fasciculus (in blue), inferior fronto-occipital fasciculus (in green), inferior cerebellar pedunculus (in yellow), middle cerebellar pedunculus (in orange) and superior cerebellar pedunculus (in red).

Table 2. Statistically significant differences between the control (C) and dyslexic (D) populations per region and per metric. The models from which the metrics are obtained are mentioned : Diffusion Tensor Imaging (DTI), Microstructure Fingerprinting (MF) or Diamond (DMD). The correlation observed between each metric and the spelling and reading performances in both populations are also provided in the leftmost columns.

To estimate CFA in a voxel, a 3D patch of size 5 around that voxel is considered. The diffusion signal in each voxel is interpolated onto a fixed spherical grid of size 200. This results in a matrix of interpolated signals, X, where each of 125 rows is the signal for one of the voxels in the patch. The signals are first embedded into a smaller space of dimension equal to 20, where a self-attention module learns the correlation between the signals from neighboring voxels. A series of fully-connected layers are then applied to estimate the CFA for the voxel.

Comparison of our proposed method and WLLS-DTI on three fetal scans. In each row, the left image is the reference CFA image reconstructed with WLLS-DTI using the full DW-MRI measurements. The middle image is the CFA image reconstructed with WLLS-DTI using 20% of the measurements. The right image is the CFA image reconstructed with the proposed deep learning method using 20% of the measurements.

Table 1 The comparison of nonparametric test of FA values in four groups

Fig.1 The multiple comparisons’results of four groups (†, p < 0.05; *, p < 0.001)The multiple comparisons between each HIBD subgroup (mild, moderate and severe group) and the control group were analyzed using Bonferroni method with significantly difference denoted by † p < 0.05, * p < 0.001.

Figure 1. Relationships between age and NDI are shown for all tracts. Linear fit lines for the entire dataset are shown in black (where there is no significant age*sex interaction) or red and blue (separated for females and males, where there is a significant age*sex interaction). Linear fit lines for each individual subject are shown in thinner red (girls) and blue (boys) lines. Individual data points are shown in grey. Significance of linear age term *** p < 0.001, ** p < 0.01, * p < 0.05

Figure 3. Relationships between age and qihMT (1/ms) are shown for all tracts. Linear fit lines for the entire dataset are shown in black (where there is no significant age*sex interaction) or red and blue (separated for females and males, where there is a significant age*sex interaction). Linear fit lines for each individual subject are shown in thinner red (girls) and blue (boys) lines. Individual data points are shown in grey. Significance of linear age term *** p < 0.001, ** p < 0.01, * p < 0.05

Figure 2. Cortical regions in which the overall 3-tissue composition differed significantly (p<0.05, FDR-corrected) between preterm-born and term-born children, when performing compositional data analysis via multivariate statistical analysis on the isometric log-ratio transformed 3-tissue compositions.

Figure 4. For a region in the sensorimotor (right paracentral) cortex, this shows: a scatter plot of the isometric log ratios (ilr), highlighting the overall group difference in tissue compositions (left); a ternary plot of the relative fractions of WM-like (T_W), GM-like (T_G), and CSF-like (T_C) signal per participant, highlighting the relative shift from T_G towards fluid-like (T_C) composition (right). Ellipses are 95% confidence areas (the same area takes on a different shape in the ternary plot visually).

Figure 3: Display of significant adjusted GM-like signal fraction model slopes from ROIs in the JHU WM atlas colored by slope and displayed on the cohort specific template. ROIs located in the posterior parts of the brain appear to be more strongly positively associated with PDSS score than regions elsewhere.

Figure 2: Display of significant adjusted WM-like signal fraction model slopes from ROIs in the JHU WM atlas colored by slope and displayed on the cohort specific template. ROIs located in the posterior parts of the brain appear to be more strongly negatively associated with PDSS score than regions elsewhere.

Harmonization performance on two datasets: one dataset was selected as a reference, two other datasets (from the rest of the 44) were selected as target (dataset 1, dataset 2) to be harmonized to the reference. To evaluate the performance of the harmonization, average fractional anisotropy along the whole brain white matter was computed for the reference, as well as target datasets before and after harmonization. Unpaired t-tests showed large between-dataset differences (p=0.0017) prior to harmonization. Statistical differences were removed after harmonization (p=0.53).

Brain masking of the dMRI data of the ABCD study using our software (https://github.com/pnlbwh/CNN-Diffusion-MRIBrain-Segmentation). Red outlines the brain of the five subjects from the ABCD study (fsl’s slicesdir was used for visualization). Each row shows a different subject brain and each column depicts the different brain slices of each subject.

Figures 2. MRI+ patients vs controls. Showing only MK and AWF profiles for left and right ILF (A, top row) and corresponding p-values corrected for multiple comparisons using Bonferroni, age and sex (A, bottom row). Heat maps show z-scores for all DKI derived maps, for ILF (B, top row) and Unc (B, bottom row). To visualize the profile differences at corresponding anatomical locations along the bundles (0-100), the p-values are rendered onto the respective fiber bundles, showing here for MK only. Notice, only results from the side of lesions are shown for each bundle.

Figures 1. Anatomically constrained tractography, showing the two WM bundles of interest (inferior longitudinal and uncinate fasciculus) for a representative healthy subject.

Figure 2. Comparison of ALPS indexes in periventricular white matter between HIV-infected individuals and healthy controls. ALPS indexes increased significantly (p<0.05, FDR-corrected) in middle-aged HIV-infected patients stably adhering to cART.

Figure 3. Results of correlation analyses in middle-aged HIV patients. (A) The ALPS indexes of the right perivascular space were correlated with attention/working memory. The duration in which middle-aged HIV-infected patients were on antiretroviral therapy correlated positively with the performance scores of (B) abstract and executive function and (C) Learning and recall. Note, the correlation was significant at p <0.05.

Figure 2. (a) (b) (c) The ROIs were selected form DMN, antinociceptive pathway and EAN, respectively. (d) Comparison of DKI parameters showed the significant differences between HC and FM. In FM, decrease MK values shown in DLPFC (HC: 0.655 ± 0.00386, FM: 0.617 ± 0.0178 [*p = 0.032]), ACR (control: 0.967 ± 0.036, FM: 0.919 ± 0.042 [**p = 0.007]), fornix (control: 0.907 ± 0.036, FM: 0.88 ± 0.016 [*p = 0.037]) and genu of corpus callosum (HC: 0.897 ± 0.012, FM: 0.803 ± 0.017 [*p = 0.042]. (e) The MK map of whole brain shows that decrease MK values.

Figure 1. The visualization of ICA after dual regression. (a) The IC map of DMN, antinociceptive network and left and right EAN. (b) FM showed significant higher activation in DMN, antinociceptive network and EAN.

Figure 1. Left: Population-specific FA-template for the group under study. Right: white matter mask overlaid on mean FA map.

Figure 2. Scatter plots showing some significant correlations: A) FATBWM with tARCS (r=0.412), B) RDTBWM with Memory (r=-0.467), C) ADWML with EDSS (r=0.329) and D) MDWML with DD (r=0.494).

Figure 3. Comparison of VBM, DTI (MD, AD, and RD), and FW imaging (MDt, ADt, RDt, and FW) indices between patients with CBS and PSP and healthy controls.

The results showed reduced gray matter volume (blue–right blue voxels) and increased FW, MD, MDt, AD, ADt, RD, and RDt (red–yellow voxels) in CBS patients relative to controls (A). GBSS results showed increased FW, MD, MDt, AD, ADt, RD, and RDt (red–yellow voxels) in PSP patients relative to controls (B). GBSS results showed increased MD, MDt, AD, ADt, RD, and RDt (red–yellow voxels) in CBS patients relative to PSP patients (C).

Figure 2. Gray matter regions of interest.

Six gray matter regions, which are described in the previous literature as vulnerable to tau pathology, based on the Desikan–Killiany cortical and subcortical atlas are shown in one representative case.

Figure. 1 BRAVO, FA, MD, FAk, MK (A) and CBF (B) maps of BD patients and healthy subjects. 26 kinds of ROIs were drawn along with the area of brain in BD and 18 kinds of ROIs in healthy subjects.

Figure. 2 Comparison of DKI measurements among the three groups and CBF between the different brain regions corresponded in BD and control groups.

Note: L-BD, lesions of BD; N-BD, non-lesions of BD; NC, normal controls; L/R, the left/right-hemispheric side; AV, average of bilateral ROI measurements;GCC/SCC, genu and splenium of the corpus callosum; FWM/PWM/TWM, frontal, parietal, and temporal WM; BV-FWM/BV-OWM, lateral ventricle around frontal and occipital WM; Cau, caudate nucleus; Tha, thalamus; Hip, hippocampus. Significant difference: *p < 0.05; **p <0.001; ***p < 0.0001.

Figure 1. Pre- and post-surgery DTI quantitative maps of one patient with positive surgical outcome. Slice No.2 represents a vertebral level without visible compression while slice No.1 represents a compressed level before surgery. WM ROI (blue) and GM ROI (red-yellow) are automatically registered and labeled by SCT.

Figure.5 correlation between pre-surgery (A) FA and (B) MD at C2 level and JOA recovery rate (JOA Rec) in the following ROIs: dorsal columns (DC), lateral funiculi (LF), ventral funiculi (VF) and gray matter (GM). The r and P values are marked in each diagram. Correlation is considered significant if P<0.05.

Fig.1 Sequence diagram of diffusion-relaxation matrix (DRM) sequence consists of dual-echo single-shot EPI-DWI without prolongation of acquisition time. Bipolar diffusion gradient was inserted as a pre-pulse of T2-map acquisition. To obtain the dual echo of different TE, this technique uses two 180° refocusing pulses for one 90° excitation pulse.

Fig.2 The ΔT2 map(C) was also created by differencing the T2 map : b-1000(A) from the T2 map : b-0(B)

Fig. 1. Sciatic nerve fractional anisotropy. (a) T2-weighted image of the right thigh with encircled sciatic nerve. (b) Corresponding fractional anisotropy color map. (c) Reconstructed fibre track of the right sciatic nerve.