Dual-encoded magnetization transfer and diffusion imaging and its application to tract-specific microstructure mapping

Abstract We present a novel dual-encoded magnetization transfer (MT) and diffusion-weighted sequence and demonstrate its potential to resolve distinct properties of white matter fiber tracts at the sub-voxel level. The sequence was designed and optimized for maximal MT ratio (MTR) efficiency. The resulting whole brain 2.6 mm isotropic protocol to measure tract-specific MTR has a scan time under 7 minutes. Ten healthy subjects were scanned twice to assess repeatability. Two different analysis methods were contrasted: a technique to extract tract-specific MTR using Convex Optimization Modeling for Microstructure Informed Tractography (COMMIT), a global optimization technique; and conventional MTR tractometry. The results demonstrate that the tract-specific method can reliably resolve the MT ratios of major white matter fiber pathways and is less affected by partial volume effects than conventional multi-modal tractometry. By reducing the contamination due to partial volume averaging of tracts, dual-encoded MT and diffusion may increase the sensitivity to microstructure alterations of specific tracts due to disease, aging, or learning, as well as lead to weighted structural connectomes with more anatomical specificity.


Introduction
Magnetic Resonance Imaging (MRI) offers valuable insight into the morphology, composition, and microstructural organization of white matter fibre pathways in the brain.Many MR contrasts are sensitive to tissue myelin content, including T1 and T2 relaxation times (review (Does, 2018)) and magnetization transfer (review (Sled, 2018)).These MR markers have been shown to correlate to varying degrees with myelin density derived from histology (see review (Lazari & Lipp, 2021;Mancini et al., 2020)).Through the combination of multiple contrast mechanisms, complementary properties of the underlying microstructure can be probed (Callaghan et al., 2014;Kolind et al., 2008) and more specific microstructural indices estimated (see review in (Cercignani & Bouyagoub, 2018)).For instance, myelin imaging can be combined with diffusion imaging to estimate the thickness of the myelin sheath relative to axon caliber, known as the g-ratio, using a biophysical model (Stikov et al., 2015).Multiple myelin-sensitive contrasts can also be combined, for example T2* and magnetization transfer (Mangeat et al., 2015), to help minimize the impact of confounds, such as iron content or field non-uniformity.
When different contrasts are not only combined but co-encoded, meaning they are encoded simultaneously within the same sequence, there is the potential to disentangle the signal contribution of different microstructural compartments within a voxel.For diffusion MRI, the sensitivity to both diffusivity and orientation can help dissociate MR properties of specific microstructural compartments and fiber orientations.When diffusion and relaxation are coencoded, the relaxation times of distinct compartments and/or fiber orientations within a voxel can be measured.This is achieved by repeating the diffusion acquisition for various echo times for T2, or for various repetition or inversion times for T1.Different approaches to both the acquisition and the analysis of diffusion-relaxometry exist.For example, non-parametric signal inversion techniques sweep through a large range of experimental parameters and rely on limited assumptions to estimate the diffusivity and relaxation times of different compartments (Benjamini & Basser, 2016, 2020;de Almeida Martins et al., 2021;Kim et al., 2017) at the cost of being rather time consuming (Lampinen et al., 2020;Veraart et al., 2018).Others use compartment models to resolve the diffusivities and T2 relaxation values of the extra-and intracellular compartments (Gong et al., 2020;Lampinen et al., 2020;Veraart et al., 2018).
Biophysical and signal models can also be used to resolve the distinct properties of multiple fiber populations present within a voxel.This approach relies on assumptions about compartment properties and takes advantage of the different orientations of fiber populations to disentangle tract-specific information, such as T1 (De Santis et al., 2016;Leppert et al., 2021).As an alternative to voxel-wise fitting, tract-specific properties can be estimated at the streamline or bundle level, using global optimization frameworks that make the assumption that a microstructural property (e.g. the intra-cellular signal fraction per unit length) is constant along a streamline's length (Daducci et al., 2015).This global approach has been used to estimate tractspecific properties such as axon caliber and tract-specific intra-axonal T2 (Barakovic, Girard, et al., 2021;Barakovic, Tax, et al., 2021), as well as myelin water fraction (Schiavi et al., 2022).
The ability to disambiguate the microstructural features of crossing white matter tracts in the brain is especially appealing in the context of tractometry studies.In conventional tractometry, quantitative or semi-quantitative MRI (qMRI) maps are projected onto reconstructed streamlines for further analysis (Bells et al., 2011;Zhang et al., 2022).In some cases, the average profile of streamlines forming a tract is computed.This has been applied to various studies of white matter development (Yeatman et al., 2012) and pathologies such as multiple sclerosis (Dayan et al., 2016;Reich et al., 2008) and stroke (Li et al., 2022).In other applications, the scalar values from qMRI maps are averaged over the whole bundle (e.g., (Correia et al., 2008;Slater et al., 2019)).
Such summary qMRI measures are also used to weigh the edges of structural connectomes (Boshkovski et al., 2022;Bosticardo et al., 2022;Kamagata et al., 2019;Wei et al., 2018).The tract qMRI estimates in all these studies are likely confounded by partial volume effects from crossing and kissing fibers which are present in 60-90% of white matter voxels (Jeurissen et al., 2013), which can potentially bias the resulting measures and could reduce the sensitivity to subtle differences.Incorporating tract-specific information from dual-encoded sequences would lead to more anatomically specific tract properties and more informative connectomes.However, the diffusion-relaxometry implementations described above often require time consuming acquisitions, advanced gradient performance, and/or complex processing routines, making them less amenable for use in patient populations.
Here we introduce an efficient dual-encoded magnetization transfer (MT) and DWI sequence to estimate the MT ratio (MTR) of individual white matter tracts using a global, whole brain optimization framework.MT is a contrast mechanism that is sensitive to the properties of bound macromolecular protons, such as bound pool fraction and exchange rate.In the brain, these bound macromolecules are largely found in cellular membranes including the lipid-rich myelin sheath (see review (Sled, 2018)).Through simulations, the acquisition parameters of the dualencoded sequence were optimized to co-encode this information in a clinically acceptable scan time of 7 mins at 2.6 mm isotropic voxel size at 3 Tesla.The optimal protocol was executed and repeated on 10 healthy subjects and analyzed using Convex Optimization Modeling for Microstructure Informed Tractography (COMMIT, (Daducci et al., 2015)) to map the distinct MTR values of fiber bundles in the tractogram.The tract-specific MTR values and their scanrescan repeatability were compared to conventional MTR tractometry.

Sequence design:
The dual-encoded sequence is comprised of a spatially non-selective MT preparation module, inserted prior to the excitation of each slice in a 2D diffusion-weighted spin echo planar imaging (EPI) acquisition (Figure 1).A dual-polarity pulsed MT preparation module was used to maximize contrast while maintaining an acceptable radio-frequency (RF) power deposition (Varma et al., 2018).The following parameters can be controlled by the user at the console: the offset frequency and polarity (positive or alternating), pulse duration (t), inter-pulse time gap (Dt), number of pulses, and the flip angle of the MT pulse (FAMT), as well as the duration of the spin echo excitation (Texc) and refocusing pulses (Tref).Previously implemented MT-weighted spin echo (SE)-EPI sequences used either a single Gaussian off-resonance pulse combined with diffusion weighting (Gupta et al., 2005) or used multiple pulses without diffusion weighting (Battiston et al., 2019).In contrast to the typical 3D spoiled gradient echo sequences used for MT-weighted experiments, diffusion acquisitions typically use SE-EPI with fat saturation and multi-band pulses (MB), which all contribute to SAR and thus limit the energy deposition that can be used for MT contrast.The MTR is given by the ratio of the images with and without MT saturation (MTR=1-MTon /MToff).This means that unwanted off-resonance contributions from the fat saturation and MB pulses will limit the available contrast since they are present in both MTon and MToff images.
To maximize MT contrast, we chose to avoid MB acceleration at the cost of scan time.
Furthermore, we investigated doing fat suppression by adjusting the ratio of the timing and amplitude of the excitation and refocusing pulses (Ivanov et al., 2010) rather than using a fat saturation pulse.These changes come at a cost of a slight increase in echo time (TE).
The number of slices and diffusion directions were chosen to provide full brain coverage and sufficient angular resolution for tractography, such that we can use a global optimization framework to estimate tract-specific MTR values (Daducci et al., 2015).We chose a relatively low b-value of 1500 s/mm 2 to maintain sufficient signal from the extra-axonal compartment, where MT contrast will arise from the interactions at the surface of the myelin sheath.The MTweighted signal from myelin water has mostly decayed at the relatively long TE values (~60-70 ms) required for this diffusion weighting (van Gelderen & Duyn, 2019), whereas the MT contrast from the intra-and extra-axonal spaces will remain.

Simulations for sequence optimization:
To determine the acquisition parameters that maximize MT contrast efficiency, simulations of a 2-pool model, including a dipolar component, were carried out in MATLAB using recently presented optimization software (Rowley C, 2022) following a minimal approximation approach (Portnoy & Stanisz, 2007) (model assumptions: super Lorentzian lineshape for the bound pool, longitudinal relaxation time of the bound pool T1b = 1 s, transverse relaxation of the bound pool T2b = 1 µs, dipolar order relaxation time T1d = 3 ms, transverse relaxation of the free pool T2a = 60 ms, exchange rate between the pools R = 26 s -1 , bound pool fraction M0b = 0.1, observed relaxation time Raobs = 850 ms, Gaussian MT pulse shape).
The following parameter search space was simulated: MT offset frequency = 1-10 kHz; TRMT = 90-150 ms; number of pulses = 1-15; pulse duration = 1-12 ms.The following protocol parameters were kept fixed: Dt = 0.3 ms, resolution = 2.6 mm 3 , 62 slices, TE = 58 ms, b-value = 1500 s/mm 2 , directions = 30.The FAMT was set to the maximum within the SAR constraints for the whole sequence (3.2 W/kg for the head).In addition, the total scan time was constrained to a maximum of 10 minutes, which includes the 2 acquisitions, with and without the MT preparation module, needed to compute the MTR.MTR represents the relative decrease in signal due to the saturation pulses, which can in part be due to the direct saturation of water and may lead to an erroneous attribution of changes to the macromolecular pool.To maximize our sensitivity to the macromolecular pool, MTR efficiency (MTR/scan time) was computed using simulations with the 2-pool model, where MTR is the difference between two cases where M0b = 0.1, and M0b = 0, with the latter representing the free-water pool only.

In-vivo acquisitions
All MR images were acquired on a Siemens Prisma-Fit 3 Tesla scanner using a 32-channel head The simulation results were verified with two in vivo datasets: first using the optimal protocol that was found to maximize MT contrast efficiency (offset frequency = 3 kHz; dual irradiation; TRMT = 90 ms; number of pulses = 7; pulse duration = 1 ms), and second with a higher TRMT = 110 ms while keeping the total SAR constant at 97% of the allowable limit.For all subsequent in-vivo acquisitions, the optimal MT preparation was used.
To verify the reduction of unwanted sources of SAR and off-resonance effects that arise from the MB excitation and refocusing pulses and fat saturation that are typically used in diffusion acquisitions, three different datasets were acquired.In the first, standard parameters were used (MB = 3, TE/TR = 55/3000 ms and fat saturation), in the second MB was removed resulting in a longer TR (TE/TR = 55/6400 ms, fat saturation), and in the third the fat saturation was replaced by an adjustment of the excitation and refocusing pulse duration ratios that minimizes refocusing of the fat signal.The pulse lengths were computed according to Eqn 8 in Ivanov et al. (Ivanov et al., 2010) for a field strength of 3 T and a slice thickness of 2.6 mm (TE/TR = 58/5900 ms, Texc = 3.328 ms, Tref = 9.472 ms).In all cases, the MT preparation was kept the same (optimal based on simulations), the TE and TR were set to the minimum, and the FAMT was increased until the total SAR reached 97% of the allowable limit.All other parameters were kept constant (63 slices, GRAPPA = 2, resolution = 2.6mm isotropic, PF = 6/8, BW = 1500 Hz/px, b-value = 1500 s/mm 2 ).

Image pre-processing
Diffusion weighted images (DWIs) (both with and without MT saturation) were denoised (Veraart et al., 2016) and pre-processed to account for subject movement, susceptibility and eddy current induced distortions with a combination of MRtrix3 (Tournier et al., 2019) and FSL (Andersson et al., 2003).Image non-uniformity correction was performed with ANTs (Tustison et al., 2010), using the bias field estimated from the MToff data applied to both the MToff and MTon data; the same correction was used for both acquisitions.Subsequently, all DWIs were upsampled to 1 mm isotropic voxels, registered to the T1-weighted MPRAGE image and aligned to the Desikan-Killiany segmentation atlas using FreeSurfer v7 (Desikan et al., 2006).The brainstem was further segmented into substructures (Iglesias et al., 2015).The computed transformations were inverted and applied to the T1-weighted image and the atlas such that all further analysis was carried out in the subject's native diffusion space.Anatomically constrained probabilistic tractography was performed on the MToff data only using the iFOD2 algorithm (J. Tournier, 2010) with 3 million streamlines, as implemented in MRtrix3.Streamlines not connecting nodes of the atlas were discarded.

Tract-specific MTR analysis pipeline
COMMIT was used to estimate tract-specific MTR values as follows.COMMIT estimates a chosen parameter describing a microstructural property of a streamline from tractography with the assumption that this parameter is constant along the streamline's trajectory.This parameter is called the streamline weight, x.In the current implementation, the streamline weight is the signal per unit length of the part of the fiber bundle represented by the streamline.Ignoring sources of signal variation such as relaxation and macromolecular content, this can be interpreted as a volume per unit length, i.e., the cross-sectional area (Smith et al., 2022).Here, the compartments of the fiber bundle attributed to the streamline consists of the combined intra-and extra-axonal spaces of the fiber.The diffusion response function of this combined space can be represented by an anisotropic tensor or "zeppelin".The other space in the voxel is represented by a "ball" modeling free water (Panagiotaki et al., 2012) and can vary from voxel to voxel.It is therefore assumed that the signal contribution from the intra-axonal space and its immediate surrounding extra-axonal space is constant along each streamline.Furthermore, all streamlines throughout the brain are assumed to have the same diffusivity parameters (diffusivity parallel to the streamline direction D||= 1.7E-3 mm 2 /s; diffusivity perpendicular to the streamline direction D^ = 0.6E-3 mm 2 /s) and the ball compartment (isotropic diffusivity= 3E-3 mm 2 /s to capture contributions from free water).COMMIT was applied separately to the MTon and MToff datasets, using the tractogram computed on the MToff dataset.In both cases, the signal was normalized to the b=0 s/mm 2 of the MToff dataset.The drop in signal due to MT-weighting will lead to a proportional change in streamline weights between the two fits.The following voxel-wise equations describe the fitting process of the MTon and MToff datasets: where S(q) is the signal at each q-space location, R zeppelin represents the response function of the zeppelin compartment, rotated to align to the fiber orientation, and is scaled by the length of the streamline intersecting the voxel.Finally, the x zeppelin represent the contribution of each streamline (i) and x ball the contribution of free water in the voxel.Streamlines whose weight was greater than zero in both MTon and MToff were grouped into bundles according to the pair of grey matter regions they connect.The tract-specific MTon and MToff were then calculated as the sum of all streamline volumes in the bundle.The volumetric contribution of streamline j to the bundle is given by the product of its weight xj times its length Lj .Finally, the tract-specific MTR is computed by combining the individual connectomes with the standard equation (S.Schiavi et al., 2020): A selection of large white matter tracts that connect nodes of the Desikan-Killiany atlas were extracted from the final connectome.The selected tracts are listed in Table 1 along    The pipeline is illustrated in Figure 2.
As previously done in Schiavi et al 2022 (Schiavi et al., 2022), the proposed method was compared to what we will refer to henceforth as conventional tractometry, whereby a separate MTR map was sampled along each streamline, taking the median along its length, and averaged across streamlines within a bundle.The MTR maps used here were calculated as the average across all directions of the diffusion-weighted datasets, omitting the b=0 image (MTRdw).For both methods, subject-wise bundle MTR values were compared with a t-test, to quantify whether there were consistent differences between bundles.The full pipeline is available on github (https://github.com/TardifLab/mt-diff)and data is available via request through a formal data sharing agreement and approval from the local ethics committees.
These results point to TRMT being the parameter with the most significant impact on MT contrast efficiency.This is seen in Figure 3A and B, where for a fixed TRMT, several different combinations of offset frequencies and number of pulses lead to a similar efficiency.The simulation results were validated by acquiring 2 datasets, first using the optimal protocol with TRMT = 90 ms (red circle in Figure 3), and second at a higher TRMT = 110 ms (green circle) while keeping the total SAR constant at 97% of the allowable limit.This resulted in a FAMT of 596° and 656° respectively.
Figure 4 illustrates the contrast efficiency (MTR/time) calculated using the b=0 images with and without MT-weighting, as well as for the average of 6 of the 30 diffusion orientations (DiffAVG).
This agrees with the simulation results, where minimizing the TRMT increases the contrast efficiency.saturation pulse prior to each slice is removed (6400 ms vs 5900 ms). Figure 5B illustrates the gain in MTR contrast (68%) that can be achieved by removing MB and the standard fat saturation, whereby the FAMT can be increased while remaining within 97% of the allowable SAR limit.The results for tract-specific and tractometry MTR values across all 10 subjects are shown in Figure 6 for the selected white matter tracts listed in Table 1.As shown in (A), there are significant within subject differences between the methods for the fWM, Pons and Splen (shown with an *, p<0.01).In (B) and (C), the group-wise MTR values are shown for both methods, as well as the corresponding scan-rescan repeatability in (D) and (E).The mean scan-rescan percent difference of tract-specific MTR in the selected tracts is slightly higher than that for tractometry (~3% vs ~1%) and as expected, the bilateral differences (left (L) and right (R)) are not significant (n.s.) for both methods.However, the tract-specific MTR exhibits a higher dynamic range and a significant difference between the pons and CST which is not present for the standard tractometry method.For both methods, there is a significant difference between the CST and PrCG-Thal connections.

Discussion:
The goal of this work was to design and optimize an efficient dual-encoded sequence, which can be combined with a global optimization microstructure informed tractography framework to estimate tract-specific MTR values.One of the primary motivations is to be able to provide tractspecific myelin indices, which should in turn lead to more anatomically and microstructurally specific myelin-weighted structural connectomes.
In terms of the sequence design, our simulation results agree with previous work from Varma et al (Varma et al., 2018) where the use of dual irradiation and rapidly switching between polarities, in this case short 1ms pulses with alternating polarity, help maximize MTR contrast (Lee et al., 2011;Varma et al., 2017).In addition, we showed that by not using MB pulses and fat saturation, more power can be used for MT contrast, at the cost of an increase in scanning time.
Nevertheless, the optimized protocol with both MTon and MToff, can be acquired in under 7 minutes.
In line with previous literature (Garcia et al., 2011;Mehta et al., 1995), the range of MTR values in major white matter fiber tracts in healthy young adults is relatively small.This and the low number of subjects (10) might explain the limited number of significant differences between tracts at the subject level for both the tract-specific and tractometry methods.However, in the case of the tract-specific values, the dynamic range is larger, and the group trends are corroborated by the MTRdw map, particularly when referring to regions where a particular tract is the dominating population in the voxel.Conversely, the tractometry maps are relatively flat, likely due to the extensive partial volume effect occurring over the length of the tracts.The increase in dynamic range has also been seen in previous work using both voxel-wise and global approaches compared to more standard tractometry (De Santis et al., 2016;Leppert et al., 2021;Schiavi et al., 2022).The scan-rescan repeatability is lower in the tract-specific approach compared to the tractometry, which is expected given the increased sensitivity of per-streamline fitting compared to the blurring that sampling and averaging along an underlying scalar map entail.Sources of error that particularly affect the global optimization technique are false positives and false negatives in tractography and the sensitivity to B1 + non-uniformity.Future work will focus on exploring model-based corrections for B1 + non-uniformity (Rowley et al., 2021) and the reduction of T1 bias in the MTR maps (Helms et al., 2008).
The current implementation aims to draw advantages from these previous methods while attempting to simplify both the acquisition and analysis by using fewer measurements and a simple ratio of contrasts with global tractograms, while still providing co-encoded information.
One advantage of the global optimization technique over voxel-wise approaches is its ability to dissociate bundles that are parallel at the voxel level for some extent of their trajectory and then diverge, becoming geometrically distinct globally.The co-encoding approach presented here is expected to provide additional potential for bundle dissociation, particularly when there are large expanses of voxels with multiple bundles, which is the case for most of the voxels in the brain (Jeurissen et al., 2013).If partial volume effects were not an issue, tractometry would theoretically have given the same results.COMMIT can dissociate parallel tracts because of its global cost function, which attempts to fit the signal in all voxels simultaneously, combined with the assumption that the microstructure of a streamline is constant along its length.
The choice of the zeppelin & ball model was made primarily because the MT effect is expected to be significant in the extra-axonal space, and we want to capture all of this.Unlike other implementations of COMMIT (Daducci et al., 2015) that keep the intra-axonal compartment signal constant along all streamlines and the extra-axonal compartment varies per fixel (fiber element in a voxel), we chose to keep the intra-plus proximal extra-axonal compartments constant.This implies that changes in fiber density within a single streamline are minimal or have a minimal effect on the diffusion signal.
Although the MTR contrast reported here will correlate with myelin density to some extent, it will be modulated by the surface to volume ratio, sheath thickness, and exchange between compartments, and thus be sensitive to fiber size, the presence of other bound proton populations, orientation and packing geometry as well.In fact, the source of the MTR contrast is different than traditional MT-GRE experiments, whose short TE is geared towards weighting interactions at both the inner and outer surface of the myelin as well as within the sheath in order to estimate myelin content.With a longer TE and diffusion weighting, the MT-weighted signal is Ultimately, many of the methods discussed above provide complementary tract-specific measures of microstructure that could be combined to understand the role of tract microstructure on brain connectivity and function.Future work will include comparing their performance across different modalities and applications.

Conclusion
This work presents a novel dual-encoded MT and diffusion sequence, for which parameters have been optimized for MT contrast efficiency.The resulting 2.6 mm whole brain protocol can be acquired in under 7 minutes and is an important step towards providing tract-specific myelin indices that minimize biases due to partial volume effects with neighboring tracts.In turn, this will provide more statistical power and insight when the microstructure of specific tracts is altered, for example through disease, ageing, function, or treatment.Finally, the potential to provide more anatomically-specific connectomes could have a significant impact on brain network analysis.

Figure 1 :
Figure 1: Sequence diagram of co-encoded MT-diffusion: A pulsed MT module inserted prior to the diffusion preparation of the 2D-EPI acquisition of each slice.The duration (t), inter-pulse time gap (Dt), number of pulses, and the flip angle of the MT pulse(FAMT) can be controlled as well as the duration of the excitation (Texc) and refocusing pulses(Tref) coil at the McConnell Brain Imaging Centre of the Montreal Neurological Institute.The project was reviewed and approved by the Research Ethics Board of McGill University.

Figure 2 :
Figure 2: Processing pipeline: The MToff data is used to generate the tractogram, which is then used with COMMIT to produce the MTon and MToff connectomes.These connectomes are then combined to get tract-specific MTR values

Figure 3 :
Figure 3: Simulated results of contrast efficiency: (A) Number of saturation pulses vs TRMT (B) Offset frequency vs TRMT and (C) Saturation pulse durations vs number of saturation pulses.Red circle highlights the optimal protocol and the green, sub-optimal

Figure 5 :
Figure 5: Reducing sources of unwanted SAR and MT by avoiding MB and replacing standard fat saturation by using a ratio of pulse lengths for fat suppression.(A) diffusion-weighted image (B) the MTR of the b=0 images

Figure 7 :
Figure 7: Example of tract-specific and tractometry bundle MTR results for one subject.Open arrows highlight the regions where a single fiber population is dominating the voxel (for example in (A): the pons (yellow) and CST (red) and (B): the PrCG-Thal connection (orange)), which gives an indication of the non-partial-volumed MTRdw value along each tract (blue color scale).The overlayed MTR results (hot color scale) for the tract-specific method show betteragreement with the underlying scalar map and higher contrast between tracts compared to conventional tractometry.
reduced overall, but to a different extent for each compartment.The echo time will modulate the contribution of each compartment based on their respective T2, such that the myelin water (T2 = 10-40ms,(MacKay et al., 1994)) signals contribute the least and the extra-axonal (T2 = 30-50ms) and intra-axonal (T2 = 80-120ms)(Veraart et al., 2018) signals contribute the most.Diffusion weighting will have a greater attenuation on the extra-axonal water than the intra-axonal compartment.In addition, the MT-diffusion experiment is inherently less affected by unwanted direct saturation effects, due to the attenuation of free water by diffusion and the fact that it is modeled as a separate compartment (ball).Therefore, through the modulation of TE and b-value, the contributions of the different compartments to the tract-specific MTR measurements can be explored.Centric k-space encoding techniques and powerful gradients can be used to shorten the TE and increase the achievable b-value to gain a more detailed understanding of the origin of the MT-and diffusion-weighted signal.

Table 1 :
Selected bundle names, description, and nodes of the Desikan-Killiany parcellation that they connect