{"paper_id":"a50238c2-41ae-4d03-babb-eabfb1ee2da4","body_text":"Dynamic parallel transmit diffusion MRI at 7T  Page 1 \n \nDynamic parallel transmit diffusion MRI at 7T \n \nMinghao Zhang1*, Belinda Ding1,2, Iulius Dragonu2, Patrick Liebig3, Christopher T. Rodgers1 \n \n1. Wolfson Brain Imaging Centre, Department of Clinical Neurosciences, University of \nCambridge, UK \n2. Siemens Healthcare Ltd, Frimley, UK \n3. Siemens Healthcare GmbH, Erlangen, Germany \n \n* Corresponding author \nMinghao Zhang \nWolfson Brain Imaging Centre \nUniversity of Cambridge \nCambridge Biomedical Campus (Box 65) \nCambridge, CB2 0QQ \nUnited Kingdom \nEmail: mz407@cam.ac.uk \n \nKeywords: diffusion MRI, ultra high field, 7T, parallel transmit, pTx \nWORD COUNT: 4181  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 2 \n \nAbstract \nDiffusion MRI (dMRI) is inherently limited by SNR. Scanning at 7T increases intrinsic SNR but 7T \nMRI scans suffer from regions of signal dropout, especially in the temporal lobes and \ncerebellum. We applied dynamic parallel transmit (pTx) to allow whole-brain 7T dMRI and \ncompared with circularly polarized (CP) pulses in 6 subjects.  \nSubject-specific 2-spoke dynamic pTx pulses were designed offline for 8 slabs covering the \nbrain. We used vendor-provided B0 and B1+ mapping. Spokes positions were set using the \nFourier difference approach, and RF coefficients optimized with a Jacobi-matrix high-flip-angle \noptimizer. Diffusion data were analyzed with FSL.  \nComparing whole-brain averages for pTx against CP scans: mean flip angle error improved by \n15% for excitation (2-spoke-VERSE 15.7° vs CP 18.4°, P=0.012) and improved by 14% for \nrefocusing (2-spoke-VERSE 39.7° vs CP 46.2°, P=0.008). Computed spin-echo signal standard \ndeviation improved by 14% (2-spoke-VERSE 0.185 vs 0.214 CP, P=0.025). Temporal SNR \nincreased by 5.4% (2-spoke-VERSE 8.47 vs CP 8.04, P=0.004) especially in the inferior temporal \nlobes. Diffusion fitting uncertainty decreased by 6.2% for first fibres (2-spoke VERSE 0.0655 vs \nCP 0.0703, P<0.001) and 1.3% for second fibers (2-spoke VERSE 0.139 vs CP 0.141, P=0.01).  \nIn conclusion, dynamic parallel transmit improves the uniformity of 7T diffusion-weighted \nimaging. In future, less restrictive SAR limits for parallel transmit scans are expected to allow \nfurther improvements. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 3 \n \nIntroduction \nDiffusion MRI (dMRI) provides quantitative information about the Brownian motion of water \nmolecules within tissue through application of diffusion encoding gradients.[1, 2] The mean \ndiffusivity and the fractional anisotropy[3] are important biomarkers for tissue \nmicrostructure.[4] In the context of neuroimaging, diffusion MRI has been applied to the \nevaluation of ischemic stroke[5] and tumors[6], to study human brain connectivity[7] and to \nunderstand the mechanisms of neurodegenerative disorders such as Alzheimer’s disease and \nParkinson’s disease.[8-10] \ndMRI is intrinsically limited by low signal-to-noise ratios (SNR) since the diffusion weighting \nstems from signal loss due to the diffusion gradients. As a result, in a clinical setting, diffusion \nMRI scans are usually acquired with lower resolution than structural scans.[11] Although \nscanning at higher field strength improves the intrinsic SNR per unit scan time[12], the \naccompanying decrease in tissue T2 poses challenges for the acquisition protocols.[13] \nFurthermore, scanning at ultra-high field introduces well-known challenges of increased \ntransmit field (B1+) inhomogeneity and increased RF power deposition (i.e. specific absorption \nrate, SAR). \nPrevious studies have shown that imaging at 7T makes it feasible to acquire high resolution \n(1.05 mm isotropic) dMRI images[14], and that static parallel transmission (pTx) – also known \nas “RF shimming” –  reduces transmit field inhomogeneities and improves SNR, especially in the \nlower parts of the brain.[15] \nStatic pTx allows the RF amplitude and phase of each channel to vary, whereas dynamic pTx \nallows each channel to play out a different arbitrary waveform. Dynamic pTx has been \nsuccessfully implemented in many neuroimaging sequences, including structural scans[16, 17] \nand functional MRI BOLD scans[18] yielding improved data quality. We hypothesize that the \nextra degrees of freedom from dynamic pTx will also improve the quality of 7T dMRI. \nIn this study, we aimed to improve 7T dMRI coverage using per-subject optimized 2-spoke \ndynamic pTx pulses, which we compare against conventional circularly polarized (CP or \n“TrueForm”) pulses for whole brain diffusion tensor imaging (DTI) in volunteers. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 4 \n \n \nMaterial and Methods \nEquipment \nScans were performed in a 7T MAGNETOM Terra MRI scanner (Siemens Healthcare, Erlangen, \nGermany) running VE12U SP01 software with an 8Tx/32Rx head coil (Nova Medical, MA, USA) \nat the University of Cambridge. Some scans were also performed in a 7T MAGNETOM Terra MRI \nscanner at the University of Glasgow with a custom 8Tx/32Rx head coil.[19] \nPulse sequence \nThe vendor’s single-shot echo planar imaging (EPI) sequence (‘ep2d_diff’, shown in Figure 1b) \nwas modified to allow the choice of conventional circularly polarized (CP or “TrueForm”) pulses \nor 2-spoke dynamic parallel transmit pulses for excitation and refocusing. \nWe based our reference CP protocol on the Human Connectome Project’s (HCP) published 7T \ndMRI protocol,[14] since the HCP consortium are acknowledged leaders in dMRI. The HCP \nproject used MAGNETOM 7T MRI scanners (Siemens) running a previous VB17 version of \nsoftware for their study, so we matched the parameters as best as we could onto the newer \nVE12U SP01 software of our 7T Terra scanner. \nPulse design \nSubject-specific 2-spoke pulses for pTx acquisitions were calculated offline in MATLAB \n(MathWorks, USA) with a modified version of the vendor’s pTx pulse design (PPD) framework. \nMaps of ΔB0 and per-channel B1+ were acquired using the vendor-provided dual-echo gradient \necho (GRE) and magnetization-prepared turbo-FLASH[20] (TFL) sequences. A whole brain region \nof interest (ROI) mask was computed by thresholding the magnitude image from the B0 \nmapping sequence. These field maps and details of the reference voltage, prescribed slice \npositions and SAR information were exported in three MATLAB “.mat” files and used for offline \npulse design.  \nWe ran an offline pulse design for monopolar 2-spoke pulses for excitation and for refocusing in \neach slab (a total of 8 slabs covering the brain, see below). The target excitation and refocusing \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 5 \n \nflip angles were 80° and 160° to reduce the specific absorption rate (SAR). A linear \ncorrection[21] was applied to the TFL B1+ maps. The last spokes were all fixed at the center of \ntransmit k-space (“DC spokes”). The first spokes were positioned in transmit k-space (kT-space) \naccording to the Fourier transform-based method[22] based on the field maps for the central \nslice in each slab. Briefly, the DC spoke complex coefficients were optimized; the resulting \npredicted transverse magnetization was computed using the ΔB0 and per-channel B1+ maps; this \nwas subtracted from the target magnetization to give a residual; the residual was Fourier \ntransformed to kT-space; the first spoke was placed at the kT-space point with highest residual \namplitude. The complex coefficients of both spokes were then jointly optimized, initially using \nthe magnitude least squares (MLS) algorithm[23] applying the small tip angle (STA) \napproximation, and using Tikhonov regularization on the L2 norm of the RF amplitude[24] to \ncontrol SAR. MLS pulse design used the ΔB0 and per-channel B1+ maps for all slices in the slab \nbut neglected T2* decay. Two solver and initialization combinations were used for each \ncalculation, namely a least-squares with QR factorization (LSQR)[25] solver with CP mode initial \nvalues, and a Powell[26] solver with all-zero initial values. The superior MLS solution was then \nused as the starting point for the Jacobi-matrix large-tip-angle optimization with gradient \ndescent included in the vendor’s PPD toolbox. Peak channel voltage constraints were enforced. \nSAR was controlled with a Tikhonov regularization factor λ fixed at 0.025. This value was found \nduring in silico trials as the maximum energy penalty without significantly deteriorating flip \nangle uniformity. The RF spoke subpulses were Shinnar-Le-Roux (SLR)[27] optimized 90° and \n180° pulses provided by the vendor, as used in the HCP diffusion protocol. The subpulse lengths \nwere 3.84 ms for excitation and 5.12 ms for refocusing, corresponding to half of the pulse \nlengths used with the vendor’s “low SAR” sequence option. \nTo reduce SAR, the 2-spoke pulses were converted to variable-rate selective excitation \n(VERSE)[28] pulses with the minimum-time algorithm.[29] A binary search was performed for \nthe maximum voltage, so that the total duration of the 2-spoke VERSE pulse was kept the same \nas the original SLR pulse and the phase relationship between the two spokes did not need to be \nre-optimized. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 6 \n \nBloch simulations of the predicted spin-echo signal amplitude were performed after the scans \nto evaluate the CP and pTx pulse performance. The excitation and refocusing pulse \ncombinations were simulated with strong crusher gradients on both sides of the refocusing \npulse. Simulations used 9x9x9 isochromats equally spaced spatially across the target voxel to \nsimulate dephasing from the crusher gradients. The transverse magnetization at the time of the \nspin-echo was averaged for each voxel. Relaxation was ignored. \nIn vivo Acquisition \nSix healthy volunteers (3 male, 3 female, aged 25 – 42 years, mean age 33 years) gave written, \ninformed consent, and the study was approved by the Cambridge Human Biology Research \nEthics Committee [HBREC.2020.27]. The acquisition protocol is summarized in Figure 1a. \nWhole brain B0 shimming was performed with the vendor’s GRE Brain method. Interactive \nshimming was performed where necessary to reduce the water FWHM linewidth to below \n50 Hz. B0 and B1+ field maps were obtained after B0 shimming by running a short “dummy” pTx-\nenabled scan. The dual-echo GRE B0 adjustment scan were acquired at 4.4x4.4x4.4 mm3, and \nTFL B1+ scan at 4.0x4.0x5.0 mm3. The dummy scan was set up that the PPD framework \ninterpolated these maps into 3.15 mm slices, so that each slice exactly corresponded to 3 slices \nin the diffusion acquisition (detailed below) and allowed even division of the slabs, without \nunnecessarily increasing the size of the pulse design problem. \nT1-weighted structural images were acquired with MP2RAGE[30] at 1.0 mm isotropic resolution \nwith the PASTEUR universal pulse sequence[16] in sagittal orientation. Other parameters were: \n4300ms TR, 840 / 2370ms inversion times, 1.91ms TE, 250 Hz/Px bandwidth and 5°/6° nominal \nflip angles. \nDiffusion MRI was performed with our single-shot echo planar imaging sequence using \nacquisition parameters based on the published 7T HCP protocol.[14] The whole brain was \ncovered in 120 transverse slices with no gap, with 1.05 mm isotropic voxels, 71 ms TE, 1388 \nHz/px bandwidth, 3x GRAPPA with GRE reference scans and 6/8 partial Fourier in plane \nacceleration, and 14300 ms TR. The excitation and refocusing flip angles were 90° and 180° for \nthe CP scans but reduced to 80° and 160° to reduce SAR for the pTx scans. The vendor’s default \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 7 \n \n110° CP fat saturation pulse was used for all scans. The transmit voltage was set to achieve no \ngreater than 15% overflip in the center of the brain and no greater than 235V, in line with the \nHCP recommendation. \nDue to the SAR limitations with the pTx sequence (discussed below), only single-band \nacquisitions were performed. We acquired 30 diffusion directions in two shells (b=1000 and \n2000 s/mm2) in one phase encoding direction (anterior-posterior, AP). Images with no-diffusion \nencoding (b=0 s/mm2) were acquired with the same parameters, in two phase encoding \ndirections (AP and PA) to allow distortion corrections. The PA images were acquired with 10 \nrepeats to measure temporal SNR (tSNR). All subjects were first scanned in the CP mode. \nPer-subject pTx pulse optimization was performed for 8 equal-sized slabs covering the brain in \nMatlab (Mathworks Inc, Nattick, MA, USA, version 2022b) as described above. This took 6-7min \non a desktop computer (Intel Core i9-10900X, 64 GB RAM). PTx-dMRI scans were then acquired. \nFor some subjects, the SAR limits were still exceeded, in which case TR was increased until the \nscan would run.  \nRepeat scans on a coil with full VOP supervision \nThree subjects in the aforementioned CP – pTx comparison acquisitions were scanned again at \nthe Imaging Centre of Excellence, Glasgow, UK. Ethical approval was provided by the College of \nMedical, Veterinary & Life Sciences (MVLS), University of Glasgow (Project No: 200210019). \nThese scans also used a Magnetom Terra 7T scanner with VE12U SP01 software, but this time \nwith a different 8Tx32Rx head coil for which full (non-diagonal) VOP supervision was enabled. \nAll pulse design parameters were kept the same. The scanner reported SAR levels and \nachievable TR were compared with that those from the Nova head coil with 1W per channel \nand 8W total power limits in First Level mode. Data from these scans was not otherwise \nincluded in our group analyses. \nData processing \nDICOM images were converted to NIfTI format with dcm2nii[31]. Signal and tSNR comparisons \nwere carried out using the raw b=0 no diffusion images without further processing. A Gaussian \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 8 \n \nfilter of standard deviation 1 was applied to the resultant mask for smoothing. Brain masks of \nthe raw EPI images were obtained using ‘bet’ (FSL).  \nThe voxel-wise tSNR percentage difference was calculated in the EPI space by \nδs!\"#$%& = \t\ns!\"#\t–\ts%&\ns!\"# + \t s%&\n× 200%\t(1) \nwhere spTx is the tSNR of the pTx image and sCP is that of the CP image. To account for the TR \nand hence acquisition time difference in some subjects, an adjusted tSNR was calculated by \ndividing the pTx tSNR by .TR!\"# TR%&⁄ . \nDiffusion image processing was carried out with the FSL pipeline. Susceptibility-induced off-\nresonance field estimation and the distortion correction of b=0 images was performed with \n‘topup’.[32] Brain masks for the distortion-corrected EPI images were obtained with ‘bet’, \nfollowed by eddy current, distortion and motion correction with ‘eddy’[33]. Tensor fitting was \nperformed with ‘dtifit’. The probabilistic diffusion model parameters were estimated with \n‘bedpostx’,[34] which outputs information including the estimated mean of the fractional \nanisotropy, diffusivity and the orientation and its uncertainty of up to three principal fibers. \nData were registered and projected to the Montreal National Institute 152 (MNI 152) common \nspace of 1 mm resolution via the T1w MP2RAGE structural image with a linear 12 degree-of-\nfreedom affine registration (FMRIB’s Linear Image Registration Tool, FLIRT)[35] for statistical \nanalyses.  \nStatistical significance was tested with a one-tailed paired Student’s t-test with 5 degrees of \nfreedom at a significance level of 0.05.  \n \nResults \nFigure 1c shows the RF and gradient waveforms of a pair of excitation and refocusing \nmonopolar 2-spoke pTx pulses. We report TE based on the center of the second (DC) spoke. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 9 \n \nTable 2 shows pulse optimization results for all subjects in every slice. The whole brain flip angle \nRMSE decreased significantly with pTx vs CP. For the excitation pulse with target flip angle 80°, \nmean RMSE improved (15.7° 2-spoke VERSE vs 18.4° CP; 15% reduction; P=0.012). For the \nrefocusing pulse with target flip angle 160°, mean RMSE also improved (39.7° 2-spoke VERSE vs \n46.2°; 14% reduction; P=0.008). Due to the increased sensitivity to ΔB0 from VERSE conversion, \nin some subjects, the slices immediately above the frontal sinus showed slice bending. The pTx \nscans are SAR restricted, and the SAR level and the final TR used are recorded in Table 1. \nFigure 2 (a-c) compares the simulated spin-echo signal intensity of the CP and 2-spoke pulses in \none subject, masked with a transformed brain mask from the MP2RAGE image to highlight \nsignal changes inside the brain. As expected at 7T, the CP pulses showed a strong central \nbrightening effect but gave weak signals in the inferior temporal lobes, occipital lobes, and the \ncerebellum. The pTx 2-spoke VERSE pulses produced more uniform transverse magnetization \nacross the whole brain, although with a slightly decreased signal intensity in the center of the \nbrain compared to CP scans. The standard deviation of the transverse magnetization inside the \nbrain mask decreased by 14% for the inter-subject mean (2-spoke VERSE 0.185 vs 0.214 CP, \nP=0.025). \nFigure 2 (d-e) compares the raw signal intensity between the CP and the 2-spoke acquisitions of \nthe same slices. Consistent with the Bloch simulation results, the CP images shows signal \ndropouts in the inferior parts of the temporal lobes and in the cerebellum. The pTx 2-spoke \npulses produced more signal in these areas which results in a visibly more uniform intensity \nprofile in the coronal view. Note that the pTx images show intensity variations at the transition \nbetween slabs due to their using different sets of pulses.  \nThe pTx pulses improved temporal SNR (tSNR) in the cerebellum (Figure 3). Over the whole \nbrain mask, the mean tSNR of CP and 2-spoke acquisitions are 8.04 and 8.64 respectively. After \ncorrecting for the lengthened TR in some subjects, the adjusted pTx 2-spoke tSNR is 8.47, \nrepresenting a 5.4% increase from the CP images (P=0.004). \nAs a result of the higher tSNR, the dynamic pTx pulses produced cleaner and more defined \nfractional anisotropy maps especially in the inferior parts of the brain as shown in Figure 4, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 10 \n \nwhere the CP data produced substantially more noise and the pTx images show fiber details \nmore clearly.  \nTo quantify the improvement in diffusion tensor fitting, the first fiber orientation dispersion \n(uncertainty)[34] was obtained from ‘bedpost’ and transformed to the MNI space for all \nsubjects. This is plotted in Figure 5. The pTx 2-spoke scans reduced the fitting uncertainty over \nthe entire periphery of the brain, and especially in the cerebellum and inferior temporal lobes. \nThe center of the brain showed some increase in the fitting uncertainty. This is again consistent \nwith the differences in the flip angle patterns. Averaged the whole brain and all subjects, the \nprincipal fiber orientation uncertainty decreased by 6.2% (2-spoke VERSE 0.0655 vs CP 0.0703, \nP<0.001).  \nThis result is also reflected in the crossing fiber estimation performance. Figure 6 shows the \nsecond fiber volume fraction for all subjects, where the pTx data can increase the amount of \nresolved second fibers in the cerebellum, temporal lobes and all around the edge of the brain. \nThe mean second fiber volume fraction increased by 19% from 2.31% to 2.74% (P=0.003, T(5)), \nwith the second fiber orientation uncertainty showing a 1.3% decrease (2-spoke VERSE 0.139 vs \nCP 0.141, P=0.01).  \nDiscussions \nPerformance comparison \nWe have demonstrated the use of dynamic pTx pulses for dMRI acquisition in human brain. Our \n2-spoke pTx VERSE pulses obtained more homogeneous spin-echo magnetization across the \nwhole brain, improving the signal around the periphery of the brain, especially in the \ncerebellum and inferior temporal lobes. The data show that the changes in the flip angle \npatterns, the b=0 signal intensity, tSNR and the diffusion fitting uncertainties are consistent. \nWe chose to compare the performance of CP and 2-spoke dynamic pTx pulses by using the \nNova 8Tx32Rx head coil throughout to eliminate any bias caused by the coil. We have shown \nelsewhere that the Nova 8Tx32Rx head coil achieves somewhat higher tSNR than the widely \nused Nova 1Tx32Rx head coil.[36] Therefore, we report a conservative estimate of the \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 11 \n \nimprovement for moving to pTx which in practice would also include an additional \nimprovement by switching to the Nova 8Tx32Rx coil.  \nWe used the similar acquisition parameters as the HCP 7T diffusion protocol to reflect the \nrepresentative application of diffusion MRI at 7T. The number of diffusion directions and the \ntotal volumes acquired are reduced, so that the CP vs pTx comparison can be performed within \na single session to minimize subject motion and avoid any other effects that might affect \nsession-to-session reproducibility. For the same reason, we did not use dielectric pads for the \nCP scans to avoid repositioning the subject when removing the pads for the pTx scans. \nThe diffusion scans were performed with a pTx enabled sequence with minimum alteration \nfrom the vendor’s product sequence. As such, we did not implement some specific features \nused in the HCP diffusion sequence, such as the GRE GRAPPA autocalibration scans with the \nsame ramp sampling as the EPI readout. This means that our data quality may not be as good as \nthe original HCP protocols, but this does not affect our goal to compare performance for pTx vs \nCP pulses. \nWe chose to split the whole brain into 8 slabs for pulse optimization to better account for the \nB0 and B1+ variations especially in the inferior parts of the brain. We retrospectively tested the \neffect of number of slabs on predicted pulse RMSE (Supplementary Information Figure S1), \nshowing that 8 slabs is a reasonable compromise albeit that for future studies it would be \npossible to reduce optimization time by merging slabs in the superior and middle parts of the \nbrain. \n \nSAR and TR \nPrevious work suggested that designing pulses in the sagittal or coronal orientations can be \nmore SAR efficient,[15] but we have attempted the pTx pulse design in the more usual \ntransverse direction. Our results demonstrate that this is still a viable solution, but our pulse \ndesigns were limited by SAR.  \nIn the pTx mode, with the Nova 8Tx32Rx head coil, the vendor enforces a 6-minute average \npower limit of 8 W total and 1 W per channel through a ’diagonal’ virtual observation point \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 12 \n \n(VOP) matrix in the IEC First Level mode. Besides, for extra safety, the current online VOP \nsupervision system in the Terra pTx framework adds up the maximum channel power for each \npulse, instead of adding up the channel power for all pulses before taking the maximum. This \nresults in a further 10%-15% overestimation of SAR exposure. \nIn order to achieve a similar TR as the CP scan under the SAR constraints, we had to reduce the \nflip angles to 80° and 160°, and to choose a relatively high Tikhonov regularization factor \nλ=0.025 for the high flip angle optimizer, which limits the performance of the pulse. The VERSE \nconversion reduced the total SAR by 52% (inter-subject mean), at a cost of increased sensitivity \nto off-resonance. \nWith these limitations, we scanned at approximately 100% SAR level at the same TR as the CP \nmode acquisition with the Nova head coil. Specifically, for a typical set of pulses: the 110° fat \nsaturation pulses account for approximately 30% SAR; the 80° excitation pulses, averaged over \nthe 8 sub-volumes, account for ca. 19%; the remainder of the SAR allowance is taken by the \n160° refocusing pulses, ca. 51%. The detailed pulse energies for each subject are presented in \nTable 1. Note that although the fat saturation pulses always had CP amplitudes and phases, \nthey are subjected to the stricter pTx SAR limits when other pulses in the sequence utilize pTx. \nWe were not able to use the gradient reversal method[37] or the low-SAR method[38] alone to \nsuppress the fat signal, because the VERSE pulses disrupt the relationship between ΔB0 and slice \nselection gradient, and can generate significant fat artefacts.[39] \nHowever, there is reason to believe that the current vendor-provided VOP limit may be overly \nconservative. NeuroSpin reported a SAR safety workflow which allows scanning with the Nova \nhead coil with 16 W total, 3 W per channel limits and custom VOPs with no incidents on VB17 \nsoftware version.[40, 41] \nTo investigate whether our sequence would perform better when SAR exposure is \noverestimated less severely, we ran equivalent scans on the Glasgow 7T Terra scanner which \nwas equipped with a different 8Tx32Rx head coil for which full (non-diagonal) VOP models are \navailable. This showed that all scans ran at <50% of the IEC first level SAR limit. The extra SAR \nallowance therefore opens the possibility to scan the same 2-spoke VERSE designs with at least \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 13 \n \nmultiband factor 2 at minimum TR. Alternatively, the extra SAR budget could be used to \nincrease the flip angles to 90° and 180° and increase the magnetization quality, or it may be \nused to decrease the level of VERSE modification to improve resilience to off-resonance. Some \nexamples of these potential improvements are predicted in Figure 7. \n \nBipolar spokes gradient delay \nWe chose to use monopolar spokes pulses in this study despite a longer excitation duration, \nbecause it has been shown[42] that bipolar spokes pulse performance are negatively impacted \nby imperfections in MR system gradient-RF timing. This phenomenon degrades signal in the \ninferior parts of the brain, due to an effective phase difference between the first and second \nspokes in the bipolar spokes pulse that is proportional to the slice distance from the isocenter.  \nWe measured the RF—gradient timing difference on our system with two approaches: the \ndouble-navigator method proposed by Tse et al.[42] gave a delay of 2.5 μs; and the zero-flip-\nangle method of Gras et al.[43] gave a delay of 3.1 μs. Such delays can have a significant impact \non flip angle homogeneity. \nAlthough not specified in the original paper, we noticed that the excitation flip angle has an \nimpact on the accuracy of the double-navigator method. As the flip angle increases, the \nexperimentally measured timing delay is not consistent. To demonstrate this, we performed \nBloch simulations of this double-navigator sequence with varying flip angles without any timing \ndelays. A sinc pulse with TBW=4 and Hamming window was used. The slice selection gradient \nresults in a 2 mm slice being excited, and the simulation grid had a resolution of 0.05 mm in the \nslice selection direction. The transverse magnetization was summed to simulate the signal seen \nby the receive coil. The sequence diagram and the simulated signals are shown in Figure 8a.  \nEven without an actual timing delay in the system, the measured echo peaks can be almost 40 \nµs apart (calculated timing delay 19.1 µs) just due to the asymmetric magnetization response. \nTherefore, it should be noted that the excitation flip angle needs to be small when using the Tse \ndouble-navigator method to minimize the systematic error.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 14 \n \nWe sought solution to these offsets using the method of gradient trim blips,[44] where small \nslice-selection gradient blips between the spokes are added to correct for the phase difference. \nWe noted that the VERSE pulse does not have a flat gradient plateau, so the optimal blip \nmoment should be found from the average gradient strength ((total slice-selection moment – \nramp moment) / (pulse time – ramp time)). This is illustrated in Figure 8b.  \nTo validate this formula, we conducted Bloch simulations of a phantom shown in Figure 8c, at z \n= 8 cm from isocenter with 3 µs gradient delay. The above trim blip moment is shown to restore \nthe magnetization pattern of the 2-spoke VERSE pulses. \nHowever, due to the nature of the VERSE pulses, the timing offset also causes an error in the \nslice-selection direction where the slice profile is smeared out (also shown in Figure 8c). This \ncannot be mitigated by using either monopolar pulses or the trim blip correction method. Our \nsimulation results show an approximately 8% signal coming from outside the desired slice with \nthe timing-corrected VERSE pulse compared to a system with no delay. This value should \nrepresent an upper bound for this effect for whole-brain imaging. \nThe full implementation of this gradient trim blip method for bipolar spokes VERSE pulses is \nbeyond the scope of this work and will be addressed in a future study. \n \nConclusions \nWe have successfully applied subject-specific 2-spoke dynamic pTx pulses for whole-brain \ndiffusion MRI. Compared to matched scans with circularly polarized (CP) pulses, whole-brain flip \nangle RMSE reduced and whole-brain tSNR increased. Diffusion tensor fitting of the pTx data \nhad lower uncertainty.   \n \nCRediT authorship contribution statement \nMinghao Zhang: Software, Methodology, Investigation, Formal analysis, Visualization, Writing - \nOriginal Draft. Belinda Ding: Software, Methodology, Writing - Review & Editing. Iulius Dragonu: \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 15 \n \nConceptualiztion, Software, Resources. Patrick Liebig: Software, Resources. Christopher \nRodgers: Conceptualization, Methodology, Writing - Review & Editing, Supervision, Funding \nacquisition. \nAcknowledgements \nThe authors thank Prof David Porter, Dr Sydney Williams, Dr Shajan Gunamony, Dr Graeme \nKeith, Steven Winata and Janhavi Ghosalkar for facilitating the scans in Glasgow, and Dr Robin \nHeidemann for helpful discussions. \nFunding \nMZ is supported by the Medical Research Council (grant number MR N013433-1) and the \nCambridge Trust. BD was funded by Gates Cambridge Trust. CTR was supported by a Sir Henry \nDale Fellowship from the Wellcome Trust and the Royal Society (098436/Z/12/B). CTR \nacknowledges Siemens for research support.  \nThis research was supported by the NIHR Cambridge Biomedical Research Centre (BRC-1215-\n20014). The views expressed are those of the author(s) and not necessarily those of the NIHR or \nthe Department of Health and Social Care. \n \nReferences \n[1] Stejskal EO, Tanner JE. Spin Diffusion Measurements: Spin Echoes in the Presence of a \nTime-Dependent Field Gradient. The Journal of Chemical Physics 1965;42(1):288-92. \ndoi: 10.1063/1.1695690. \n[2] Diffusion MRI: Theory, Methods, and Applications. Oxford, New York: Oxford University \nPress; 2011. \n[3] Mori S. Introduction to Diffusion Tensor Imaging. Elsevier; 2007. \n[4] Beaulieu C. The basis of anisotropic water diffusion in the nervous system – a technical \nreview. NMR in Biomedicine 2002;15(7-8):435-55. doi: 10.1002/nbm.782. \n[5] Albers GW. Diffusion-weighted MRI for evaluation of acute stroke. Neurology 1998;51(3 \nSuppl 3):S47-9. doi: 10.1212/wnl.51.3_suppl_3.s47. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 16 \n \n[6] Messina C, Bignone R, Bruno A, Bruno A, Bruno F, Calandri M, et al. Diffusion-Weighted \nImaging in Oncology: An Update. Cancers 2020;12(6):1493. doi: \n10.3390/cancers12061493. \n[7] Van Essen DC, Smith SM, Barch DM, Behrens TEJ, Yacoub E, Ugurbil K. The WU-Minn \nHuman Connectome Project: An Overview. NeuroImage 2013;80:62-79. doi: \n10.1016/j.neuroimage.2013.05.041. \n[8] Goveas J, O'Dwyer L, Mascalchi M, Cosottini M, Diciotti S, De Santis S, et al. Diffusion-\nMRI in neurodegenerative disorders. Magnetic Resonance Imaging 2015;33(7):853-76. \ndoi: 10.1016/j.mri.2015.04.006. \n[9] Correia MM, Rittman T, Barnes CL, Coyle-Gilchrist IT, Ghosh B, Hughes LE, et al. Towards \naccurate and unbiased imaging-based differentiation of Parkinson’s disease, progressive \nsupranuclear palsy and corticobasal syndrome. Brain Communications \n2020;2(1):fcaa051. doi: 10.1093/braincomms/fcaa051. \n[10] Minett T, Su L, Mak E, Williams G, Firbank M, Lawson RA, et al. Longitudinal diffusion \ntensor imaging changes in early Parkinson’s disease: ICICLE-PD study. Journal of \nNeurology 2018;265(7):1528-39. doi: 10.1007/s00415-018-8873-0. \n[11] Chilla GS, Tan CH, Xu C, Poh CL. Diffusion weighted magnetic resonance imaging and its \nrecent trend—a survey. Quantitative Imaging in Medicine and Surgery 2015;5(3):407-22. \ndoi: 10.3978/j.issn.2223-4292.2015.03.01. \n[12] Vaughan Jt, Garwood M, Collins Cm, Liu W, DelaBarre L, Adriany G, et al. 7T vs. 4T: RF \npower, homogeneity, and signal-to-noise comparison in head images. Magnetic \nResonance in Medicine 2001;46(1):24-30. doi: 10.1002/mrm.1156. \n[13] Ladd ME, Bachert P, Meyerspeer M, Moser E, Nagel AM, Norris DG, et al. Pros and cons \nof ultra-high-field MRI/MRS for human application. Progress in Nuclear Magnetic \nResonance Spectroscopy 2018;109:1-50. doi: 10.1016/j.pnmrs.2018.06.001. \n[14] Vu AT, Auerbach E, Lenglet C, Moeller S, Sotiropoulos SN, Jbabdi S, et al. High resolution \nwhole brain diffusion imaging at 7T for the Human Connectome Project. NeuroImage \n2015;122:318-31. doi: 10.1016/j.neuroimage.2015.08.004. \n[15] Wu X, Auerbach EJ, Vu AT, Moeller S, Lenglet C, Schmitter S, et al. High-resolution \nwhole-brain diffusion MRI at 7T using radiofrequency parallel transmission. Magnetic \nResonance in Medicine 2018;80(5):1857-70. doi: 10.1002/mrm.27189. \n[16] Gras V, Mauconduit F, Vignaud A, Ster CL, Leroi L, Amadon A, et al. PASTeUR: Package of \nAnatomical Sequences using parallel Transmission UniveRsal kT-point pulses. 27. \n2019:4626. \n[17] Herrler J, Liebig P, Gumbrecht R, Ritter D, Schmitter S, Maier A, et al. Fast online-\ncustomized (FOCUS) parallel transmission pulses: A combination of universal pulses and \nindividual optimization. Magnetic Resonance in Medicine 2021;85(6):3140-53. doi: \n10.1002/mrm.28643. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 17 \n \n[18] Ding B, Dragonu I, Rua C, Carlin JD, Halai AD, Liebig P, et al. Parallel transmit (pTx) with \nonline pulse design for task-based fMRI at 7 T. Magnetic Resonance Imaging 2022. doi: \n10.1016/j.mri.2022.07.003. \n[19] Williams SN, Allwood-Spiers S, McElhinney P, Paterson G, Herrler J, Liebig P, et al. A \nNested Eight-Channel Transmit Array With Open-Face Concept for Human Brain Imaging \nat 7 Tesla. Front Phys-Lausanne 2021;9. doi: ARTN 701330 \n10.3389/fphy.2021.701330. \n[20] Chung S, Kim D, Breton E, Axel L. Rapid B1+ Mapping Using a Pre-Conditioning RF Pulse \nwith TurboFLASH readout. Magnetic resonance in medicine : official journal of the \nSociety of Magnetic Resonance in Medicine / Society of Magnetic Resonance in \nMedicine 2010;64(2):439-46. doi: 10.1002/mrm.22423. \n[21] Tomi-Tricot R, Sedlacik J, Endres J, Herrler J, Liebig P, Ritter D, et al. Fully Integrated \nScanner Implementation of Direct Signal Control for 2D T2-Weighted TSE at Ultra-High \nField. 29. 2021:0621. \n[22] Yip C-Y, Fessler JA, Noll DC. Advanced three-dimensional tailored RF pulse for signal \nrecovery in T2*-weighted functional magnetic resonance imaging. Magnetic Resonance \nin Medicine 2006;56(5):1050-9. doi: 10.1002/mrm.21048. \n[23] Setsompop K, Wald LL, Alagappan V, Gagoski BA, Adalsteinsson E. Magnitude Least \nSquares Optimization for Parallel Radio Frequency Excitation Design Demonstrated at 7 \nTesla With Eight Channels. Magnetic resonance in medicine : official journal of the \nSociety of Magnetic Resonance in Medicine / Society of Magnetic Resonance in \nMedicine 2008;59(4):908. doi: 10.1002/mrm.21513. \n[24] Yip C-y, Fessler JA, Noll DC. Iterative RF pulse design for multidimensional, small-tip-\nangle selective excitation. Magnetic Resonance in Medicine 2005;54(4):908-17. doi: \n10.1002/mrm.20631. \n[25] Paige CC, Saunders MA. LSQR: An Algorithm for Sparse Linear Equations and Sparse \nLeast Squares. ACM Transactions on Mathematical Software 1982;8(1):43–71. doi: \n10.1145/355984.355989. \n[26] Powell M. A View of Algorithms for Optimization without Derivatives 1. undefined 2007. \n[27] Pauly J, Le Roux P, Nishimura D, Macovski A. Parameter relations for the Shinnar-Le \nRoux selective excitation pulse design algorithm (NMR imaging). IEEE Trans Med Imaging \n1991;10(1):53-65. doi: 10.1109/42.75611. \n[28] Conolly S, Nishimura D, Macovski A, Glover G. Variable-rate selective excitation. Journal \nof Magnetic Resonance (1969) 1988;78(3):440-58. doi: 10.1016/0022-2364(88)90131-X. \n[29] Hargreaves BA, Cunningham CH, Nishimura DG, Conolly SM. Variable-rate selective \nexcitation for rapid MRI sequences. Magnetic Resonance in Medicine 2004;52(3):590-7. \ndoi: 10.1002/mrm.20168. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 18 \n \n[30] Marques JP, Kober T, Krueger G, van der Zwaag W, Van de Moortele P-F, Gruetter R. \nMP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-\nmapping at high field. NeuroImage 2010;49(2):1271-81. doi: \n10.1016/j.neuroimage.2009.10.002. \n[31] Li X, Morgan PS, Ashburner J, Smith J, Rorden C. The first step for neuroimaging data \nanalysis: DICOM to NIfTI conversion. Journal of Neuroscience Methods 2016;264:47-56. \ndoi: 10.1016/j.jneumeth.2016.03.001. \n[32] Andersson JLR, Skare S, Ashburner J. How to correct susceptibility distortions in spin-\necho echo-planar images: application to diffusion tensor imaging. NeuroImage \n2003;20(2):870-88. doi: 10.1016/S1053-8119(03)00336-7. \n[33] Andersson JLR, Sotiropoulos SN. An integrated approach to correction for off-resonance \neffects and subject movement in diffusion MR imaging. Neuroimage 2016;125:1063-78. \ndoi: 10.1016/j.neuroimage.2015.10.019. \n[34] Jbabdi S, Sotiropoulos SN, Savio AM, Graña M, Behrens TEJ. Model-based analysis of \nmultishell diffusion MR data for tractography: how to get over fitting problems. \nMagnetic Resonance in Medicine 2012;68(6):1846-55. doi: 10.1002/mrm.24204. \n[35] Jenkinson M, Smith S. A global optimisation method for robust affine registration of \nbrain images. Medical Image Analysis 2001;5(2):143-56. doi: 10.1016/s1361-\n8415(01)00036-6. \n[36] Ding B, Rua C, Rodgers CT. Comparison of Nova 1Tx vs 8Tx head coils for routine 7T \nneuroimaging. 30. 2022:2094. \n[37] Gomori JM, Holland GA, Grossman RI, Gefter WB, Lenkinski RE. Fat suppression by \nsection-select gradient reversal on spin-echo MR imaging. Work in progress. Radiology \n1988;168(2):493-5. doi: 10.1148/radiology.168.2.3393670. \n[38] Ivanov D, Schäfer A, Streicher MN, Heidemann RM, Trampel R, Turner R. A simple low-\nSAR technique for chemical-shift selection with high-field spin-echo imaging. Magnetic \nResonance in Medicine 2010;64(2):319-26. doi: 10.1002/mrm.22518. \n[39] Williams S, Dragonu I, Ding B, Liebig P, Porter D. Simultaneous Multislice pTx for \nReadout-Segmented Diffusion Imaging at 7 T. Joint Annual Meeting ISMRM-ESMRMB \nISMRT 31st Annual Meeting.0453. \n[40] Boulant N, Gras V, Amadon A, Luong M, Ferrand G, Vignaud A. Workflow proposal for \ndefining SAR safety margins in parallel transmission. 26. 2018:0295. \n[41] Delebarre T, Gras V, Mauconduit F, Vignaud A, Boulant N, Ciobanu L. CEST MRI at 7T \nusing an optimized parallel transmission saturation scheme. 30. 2022:2718. \n[42] Tse DHY, Wiggins CJ, Poser BA. Estimating and eliminating the excitation errors in \nbipolar gradient composite excitations caused by radiofrequency-gradient delay: \nExample of bipolar spokes pulses in parallel transmission. Magnetic Resonance in \nMedicine 2017;78(5):1883-90. doi: 10.1002/mrm.26586. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 19 \n \n[43] Gras V, Vignaud A, Amadon A, Mauconduit F, Le Bihan D, Boulant N. New method to \ncharacterize and correct with sub-µs precision gradient delays in bipolar multispoke RF \npulses. Magnetic Resonance in Medicine 2017;78(6):2194-202. doi: \n10.1002/mrm.26614. \n[44] Jamil R, Mauconduit F, Gras V, Boulant N. General gradient delay correction method in \nbipolar multispoke RF pulses using trim blips. Magnetic Resonance in Medicine \n2021;85(2):1004-12. doi: 10.1002/mrm.28478. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 20 \n \nFigures \n \nFigure 1 a) The pTx diffusion acquisition protocol. The pulse design was carried out during the MP2RAGE and CP mode diffusion \nscans and did not add extra time. b) The SS-EPI diffusion sequence diagram. CP or pTx pulses can be used for both excitation and \nrefocusing pulses. The minimum TE is limited by the diffusion gradient and EPI train after the refocusing pulse. c) A typical pair of \nexcitation and refocusing pulses. The pulses are converted to VERSE form to reduce SAR. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 21 \n \n \n \n \n \n \n \n \nFigure 2 (a-c) Bloch simulaton of the spin-echo transverse magnetization comparison between CP and 2-spoke VERSE pulses in \none subject. The brain mask was obtained with FSL ‘bet’ with the MP2RAGE image, to highlight the changes inside the brain and \nto calculate statistics. (d-e) The raw b=0 image comparison in the same subject between the circularly polarized (CP) mode \nacquisition and the 2-spoke pTx acquisition. The same windowing was applied to all images. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 22 \n \n \n \nFigure 3 (a-c) Comparison of the temporal SNR between the CP and 2-spoke pTx scans. The tSNR is calculated from 10 repeated \nmeasurements of the no diffusion (b=0) images. (d) The whole brain mean tSNR across all subjects. ‘pTx adjusted’ divides the \nmeasured tSNR by sqrt(TRpTx/TRCP) to account for the lengthened TR in some subjects. pTx 2-spoke VERSE acquisitions show 5.4% \nincrease in tSNR (P=0.004, T(5)) after adjustment. \n \nd) \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 23 \n \n \nFigure 4 Fractional anisotropy color-coded by the principal diffusion direction (red - left-right; green – anterior-posterior; blue – \nsuperior-inferior) in one subject. The pTx 2-spoke data are less noisy and show better defined fiber orientations in the \ncerebellum, occipital lobe and inferior temporal lobes. \n \n \nFigure 5 The difference in the first fiber orientation fitting uncertainties between CP and pTx acquisitions averaged across all \nsubjects. Data of all subjects were transformed into MNI space for comparison. MNI coordinates are shown. Over the whole \nbrain, pTx 2-spoke decreased the fitting uncertainty by 6.2% (P<0.001, T(5)).  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 24 \n \n \n \nFigure 6 The second fiber fractional volume map plotted for all subjects. Color scale between [0.02 0.2]. pTx data allow more \ncrossing fibers to be resolved around the periphery of the brain. The whole brain mean second fiber volume fraction increased by \n19% from 2.31% to 2.74% (P=0.003, T(5)). \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 25 \n \n \nFigure 7 Predicted spin-echo intensity by Bloch simulation for several SAR limit levels for various cases a) 100% SAR level with \n1W/channel limits (as provided with the Nova 8Tx32Rx coil on 7T Terra scanners with VE12 software). b) 100% SAR level with \n1.5W/channel and 8W total (as provided with the Nova 8Tx32Rx coil on Magnetom 7T scanners with VB17 step 2.3 software). c) \n50% SAR level, i.e. Normal Mode limit, for the custom coil that we used in Glasgow. This allows MB2 scans at minimum TR. d) \n100% SAR level for the Glasgow custom coil. This shows how the extra SAR budget of the VOP model could be used to increase \npulse design quality. e) Same, but without VERSE and with fat suppression disabled. This shows how a higher SAR budget could \nimprove spin-echo uniformity by reducing the sensitivity to off-resonance. Note that these figures are plotted only to compare in \nbroad terms the effect of the SAR budget and are not intended as a comment on the two coils. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 26 \n \n \nFigure 8 a) The double-navigator sequence (Tse et al.[42]) to determine the RF-gradient timing offset, and the Bloch simulated \nsignal with varying flip angles. b) The calculation of the trim blip moment to correct for the timing delay in bipolar spokes as \nintroduced in Jamil et al.[44] for VERSE pulses. The effective gradient strength is calculated with the slice selection moment \ndivided by the corresponding ‘flat top’ time. c) Bloch simulation of the transverse magnetization of VERSE excitation pulses on a \nphantom, with 3 µs gradient delay at z = 8 cm from the isocenter. The magnetization pattern distortion caused by the timing \ndelay is corrected by the trim blips, but the slice profile is smeared out and can not be mitigated with either monopolar spokes or \nthe trim blip correction. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 27 \n \n \nTables  \nTable 1 The pTx 2-spoke pulse powers in all subjects. In pTx mode on 7T Terra systems with VE12U software version, the IEC first \nlevel SAR limit for the current ‘diagonal’ VOP model for the Nova 8Tx32Rx coils is 1 W per channel, 8 W total. This is compared \nwith the full VOP supervision with a custom 8Tx32Rx head coil at University of Glasgow. Powers are calculated for TR=14.3s for \n120 slices, but where the SAR limit is exceeded, the actual TR used for the scans was increased accordingly (last row). \n \nNova 8Tx32Rx pTx Glasgow 8Tx32Rx \npTx \nCP \nmod\ne \nSubjects 1 2 3 4 5 6 1 2 3 All \nExcitation \n/ W \n1.04 0.99 0.99 1.01 1.00 0.89 1.36 1.48 1.34 \n \nRefocusin\ng / W \n2.37 2.67 3.93 2.62 2.52 2.57 3.58 4.31 3.48 \n \nTotal \npower / \nW \n5.08 5.33 6.57 5.23 5.22 5.09 7.05 7.93 7.37 \n \n \n% of 1W in the highest power channel % of VOP SAR \n \n% SAR for \nIEC First \nLevel \n93% 100% 131% 105% 101% 101% 36% 47% 39% \n \nTR used / \ns \n14.3 14.3 18.6 15.0 14.6 14.5 14.3 14.3 14.3 14.3 \n \nTable 2 Whole brain flip angle root mean squared errors for all subjects by Bloch simulation. Errors are calculated within a brain \nmask retrospectively obtained from the structural image with ‘bet’. The decrease in flip angle RMSE with pTx 2-spoke VERSE for \nboth excitation (P=0.012) and refocusing (P<0.008) pulses are significant compared to CP in paired t-tests. \n \nExcitation (target 80°) Refocusing (target 160°) \nSubject CP 2-spoke VERSE CP 2-spoke VERSE \n1 16.9 15.6 43.3 37.9 \n2 19.8 14.0 48.2 39.2 \n3 21.6 17.0 54.0 39.5 \n4 18.8 17.2 46.9 43.6 \n5 17.2 14.7 42.4 37.6 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 28 \n \n6 16.3 15.8 42.6 40.5 \nMean 18.4 15.7 46.2 39.7 \np-value 0.012 0.008 \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint \n\nDynamic parallel transmit diffusion MRI at 7T  Page 29 \n \n \n \nSupplementary information \n \nSupplementary Information Figure S1 Flip angle root mean squared error (in °) for the 80° excitation pulse computed from Bloch \nsimulation results in one subject, (a) over the whole brain and (b) per transverse slice. The workflow time consists of the \nexcitation and refocusing pulse design, visualization and data transfer to and from the scanner. Our choice of 8 slabs is a good \ncompromise between flip angle accuracy and optimization time. \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted January 6, 2024. ; https://doi.org/10.1101/2024.01.06.574463doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}