An integrated RF-receive/B0-shim array coil boosts performance of whole-brain MR spectroscopic imaging at 7 T

Metabolic imaging of the human brain by in-vivo magnetic resonance spectroscopic imaging (MRSI) can non-invasively probe neurochemistry in healthy and disease conditions. MRSI at ultra-high field (≥ 7 T) provides increased sensitivity for fast high-resolution metabolic imaging, but comes with technical challenges due to non-uniform B0 field. Here, we show that an integrated RF-receive/B0-shim (AC/DC) array coil can be used to mitigate 7 T B0 inhomogeneity, which improves spectral quality and metabolite quantification over a whole-brain slab. Our results from simulations, phantoms, healthy and brain tumor human subjects indicate improvements of global B0 homogeneity by 55%, narrower spectral linewidth by 29%, higher signal-to-noise ratio by 31%, more precise metabolite quantification by 22%, and an increase by 21% of the brain volume that can be reliably analyzed. AC/DC shimming provide the highest correlation (R2 = 0.98, P = 0.001) with ground-truth values for metabolite concentration. Clinical translation of AC/DC and MRSI is demonstrated in a patient with mutant-IDH1 glioma where it enables imaging of D-2-hydroxyglutarate oncometabolite with a 2.8-fold increase in contrast-to-noise ratio at higher resolution and more brain coverage compared to previous 7 T studies. Hence, AC/DC technology may help ultra-high field MRSI become more feasible to take advantage of higher signal/contrast-to-noise in clinical applications.


Results
Simulations. The effects of increasing linewidths on the spectral fitting and 2HG quantification are shown in Fig. 1 for simulated spectra. For narrow linewidths ≤ 0.1 ppm, a rich spectral pattern can be noticed in the 2.1-2.6 ppm range (Fig. 1A,C). For linewidths above 0.1 ppm this spectral pattern is gradually lost by partial signal cancelation due to increasing overlap of negative and positive peaks (Fig. 1B). The characteristic negative peak of 2HG obtained for double spin-echo TE1/TE2 = 58/20 ms is clearly visible at 2.25 ppm for linewidths up to 0.1 ppm (Fig. 1A,C,D). The ratio between estimated and true 2HG concentration (Fig. 1E) is close to 1 (± 0.1 range) for linewidths ≤ 0.1 ppm, and rapidly decreases towards 0 for increasing linewidths. The relative CRLB (Fig. 1F) remains under 20% for linewidths ≤ 0.1 ppm, but sharply increases for larger linewidths.
Phantoms. The results of the five different shimming conditions in the phantom are shown in Fig. 2. The B 0 fieldmaps ( Fig. 2A) at the top show gradual homogeneity improvement going from 2SH box to (2SH brain + AC/ DC brain ) joint shimming, with 29% narrower width of B 0 histograms (Fig. 2B) that decreased from 27.57 Hz (0.093 ppm) for 2SH box to 19.61 Hz (0.066 ppm) for (2SH brain + AC/DC brain ) joint shimming. The 2HG maps www.nature.com/scientificreports/ ( Fig. 2A) correspond better to the T1 weighted image and the known 2HG concentration in the tubes, when AC/DC shimming is superimposed on the 2SH shimming. Examples of spectra (Fig. 2C) from each tube show narrower lines and less artifacts for measurements with superimposed AC/DC shimming. In particular, artifacts coming from insufficient water suppresion and frequency shifts are visible in spectra acquired with 2SH box shimming in the 1 mM and 2 mM tubes that have the worst local B 0 distribution. In Fig. 2D is shown zoom on the spectral interval [2.2, 2.4] ppm containing the 2HG negative peak at 2.25 ppm and the Glu positive peak at 2.35 ppm. The 2SH + AC/DC shims show a clear titration with concentration for the negative 2HG peak at 2.25 ppm and a stable positive Glu peak at 2.35 ppm for constant Glu concentration, while there is more variability in the case of the 2SH shims. Correlation between the measured and ground-truth 2HG concentrations is shown in the right most panel of Fig. 2D. The highest correlation coefficient and statistical significance Figure 2. Phantom measurements performed under five shimming conditions 2SH box , 2SH brain , 2SH box + AC/ DC brain , 2SH brain + AC/DC brain , and (2SH brain + AC/DC brain ) joint . (A) B 0 fieldmaps maps of 2-HG and T1-weighted MRI with concentration of 2-HG marked on every tube 0, 1, 2, 3, 4, 5 mM (magenta Results from MRSI measurements for the same two volunteers and the same two shimming conditions are shown in Fig. 4. The metabolic maps of total NAA, choline and creatine show better agreement with the anatomical imaging and less artifacts for 2SH box + AC/DC brain shimming, which are particularly visible in 2SH box as Figure 3. Comparison of B 0 homogeneity obtained with 2SH box and 2SH box + AC/DC brain shimmings in the first two healthy volunteers. Five representative slices are shown throughout the shimmed brain slab, with the standard deviation of B 0 distribution indicated below. Histograms of B 0 distribution over the entire brain slab are shown overlaid at the bottom for both shimming methods. MEMPRAGE anatomical images are shown in the left most columns for each volunteer. missing areas due to signal dropout in the anterior frontal regions. The quality parametric maps (SNR, CRLB, FWHM) show an increase in mean SNR by 9-35%, a decrease of mean CRLB by 18-21%, and a decrease of mean linewidth by 17-37% for 2SH box + AC/DC brain compared to 2SH box . Local improvements in quality parametric maps are even higher for the anterior frontal brain areas. Overlay of spectra from voxels in anterior, middle and posterior brain areas show more narrower metabolic peaks with reduced lineshape distortion for 2SH box + AC/ DC brain shimming. In volunteer one the 2SH box spectra in the anterior frontal regions have large frequency shifts that cannot be corrected by the LCModel frequency correction, while this is not a problem for 2SH box + AC/ DC brain spectra. Additional axial slices from the 3D MRSI brain slab are shown in Supplementary Figs. S1 and S2. In Fig. 5 the real-time B 0 fieldmaps obtained with the navigator are shown for the third healthy volunteer and all five shimming conditions. The B 0 fieldmaps show a gradual improvement in homogeneity, up to 55% from a standard deviation of 24.45 Hz (0.082 ppm) for 2SH box shimming to a standard deviation of 11.17 Hz (0.037 ppm) for (2SH brain + AC/DC brain ) joint shimming. The plots of frequency correction show similar stability for the scanner shimming hardware (2SH) alone and the combined scanner plus AC/DC shimming hardware (2SH + AC/DC). Isolated spikes in the frequency plots correspond to brief head motion as shown in the translation and rotation plots. The B 0 fieldmaps obtained with the GRE sequence are presented in Supplementary Fig. 3, which also shows that B 0 histograms have narrower width for shim conditions where AC/DC is superimposed on 2SH.
The metabolic maps and the quality parametric maps obtained in volunteer 3 for all five shimming conditions are shown in Fig. 6. All the metabolic maps obtained with the three combinations of 2SH + AC/DC shims show better agreement with anatomical imaging and less artifacts compared to the two 2SH only shims. The quality parametric maps show mean improvements of 25% increased SNR, 19% decreased CRLB, and 18% narrower FWHM, between 2SH only and 2SH + AC/DC shimmings. Figure 7 shows the B 0 fieldmaps obtained with the GRE sequence in the mutant IDH1 glioma patient under four shimming conditions. Compared to the 2SH box the global B 0 homogeneity improves by 5.5% for 2SH brain , by 42% for 2SH box + AC/DC brain , and by 48% for 2SH brain + AC/DC brain shims, respectively. Marked improvements in local B 0 homogeneity are also obtained for the tumor region of interest (ROI) by 4.5%, 28% and 36%, respectively, when comparing the same shim conditions.
The tumor specific maps of 2HG to total creatine (2HG/tCr) 41 obtained in the mutant IDH1 glioma patient are shown in Fig. 8. The tumor is not diagnostically apparent in the 2HG/tCr maps obtained with the two 2SH shimmings (2SH box and 2SH brain ), which show no clear increase in this metabolic marker above background. The 2HG/tCr maps obtained with two combinations of 2SH + AC/DC shims (2SH box + AC/DC brain and 2SH brain + AC/ DC brain ) show an increase in this metabolic marker within the tumor ROI. The CRLB maps for 2HG fitting show the largest clusters of tumor voxels having CRLB < 20% and fewer false positive voxels outside the tumor for measurements obtained with 2SH box + AC/DC brain and 2SH brain + AC/DC brain shims. The improved 2HG quantification provides fewer false negative voxels in the tumor and fewer false positive voxels outside the tumor, resulting in www.nature.com/scientificreports/ better image quality with higher tumor contrast-to-noise ratio for 2SH box + AC/DC brain and 2SH brain + AC/DC brain shimmings. By comparison, the maps obtained with 2SH shims are non-diagnostic with a contrast-to-noise ratio less than one, indicating tumor 2HG signal below the background variability. Tumor heterogeneity is visible in the metabolic maps of 2HG/tCr and tCho/tCr (total-choline/creatine) obtained by 2SH brain + AC/DC brain . Examples of spectra from tumor and healthy brain show improvement in spectral resolution with better peak separation, higher SNR, and less baseline artifacts for 2SH box + AC/DC brain and 2SH brain + AC/DC brain shims. Metabolic maps of tNAA, tCho, tCr and the quality parametric maps SNR, FWHM and CRLB from mutant IDH1 glioma patient are presented in Supplementary Fig. 4. The metabolic maps obtained with the two 2SH + AC/ DC shimmings show the tumor more clearly, with less image artifacts compared to the two 2SH shims. In particular, total choline as a tumor marker shows a marked increase in the tumor only for 2SH + AC/DC maps, while areas of high choline far from tumor are visible in 2SH maps. When compared to 2SH box the: (1) SNR increases by 3% for 2SH brain , by 14% for 2SH box + AC/DC brain , and by 17% for 2SH brain + AC/DC brain ; (2) CRLB decreases by 4% for 2SH brain , by 17% for 2SH box + AC/DC brain , and by 23% for 2SH brain + AC/DC brain ; (3) FWHM decreases by 10% for 2SH brain , by 13% for 2SH box + AC/DC brain , and by 18% for 2SH brain + AC/DC brain . Quantitative results from in vivo measurements in healthy subjects and patients are summarized in Table 1.

Discussion
In this work we demonstrated that an integrated AC/DC shim array improves 3D MRSI at 7 T over a wholebrain slab when multi-coil array shimming is combined with the standard scanner second order spherical harmonics shimming. The improvements are particularly large in the anterior frontal areas and inferior slices, which are hard to shim with the standard 2SH method, but improvements are obtained thoroughout the brain. We investigated also the improvements due to optimization of 2SH shimming over the head vs brain, with the former being the manufacturer's method which includes the brain, skull and scalp in the shimmed volume. The 2SH shimming over the brain slab (2SH brain ) improves B 0 and MRSI compared to the 2SH shimming over the head slab (2SH box ), however these improvements are smaller than what is obtained by adding the AC/DC shimming to the 2SH shimming for either shim optimization (box or brain). Hence, the improvements by AC/ Figure 5. Real-time motion correction and frequency correction for 3D MRSI in the third healthy volunteer under five shimming conditions 2SH box , 2SH brain , 2SH box + AC/DC brain , 2SH brain + AC/DC brain , and (2SH brain + AC/ DC brain ) joint  www.nature.com/scientificreports/ DC shimming are largely due to the ability of the AC/DC coil to generate highly arbitrary B 0 field patterns, and secondarily due to optimizing shimming over the brain only compartment. All three 2SH + AC/DC shims that were investigated provided comparable improvements, with a slightly superior performance for the jointly optimized (2SH brain + AC/DC brain ) joint . The results from humans are summarized in Table 1.
The AC/DC shimming improved both global and local B 0 homogeneity as measured by narrower histogram distributions (55%) and spectral linewidths (29%), respectively. Global B 0 homogeneity determines how much the mean frequency of a certain MRSI voxel is shifted from the central frequency in the shim volume, and is important for frequency selective schemes such as water suppression 5,6 and spectral editing 7 . Adjustment of water suppression and spectral editing is performed globally on the central frequency in the brain, and for MRSI voxels that are shifted far from the central frequency the suppression and editing efficiency is compromised. This can lead to large residual water signal that overwhelms the metabolite signal in those voxels, but can also contaminate other voxels due to reduced point spread function of MRSI. This is true not only for water suppression schemes such as WET 5 , but also for metabolite cycling 6 that is dependent on frequency selective metabolite inversion for water subtraction. Insufficient water suppression manifests as large baseline artifacts that distort spectra and interfere with metabolite fitting. Such artifacts were visible in phantom and in-vivo MRSI measurements with 2SH shimmings, and were largely reduced by the addition of AC/DC shimming. In phantom the susceptibility difference between magnevist-doped (paramagnetic) solutions inside the tubes and the magnevist-free (diamagnetic) surrounding solution cannot be corrected well by 2SH shimming, but this is reduced by AC/DC shimming. Further improvement in coil combination 55,56 and B 1 transmit efficiency 57 can improve spatial correlation of metabolic maps at ultra-high field. Local B 0 homogeneity determines the intravoxel spectral linewidth in MRSI, which is important for spectral peak separation and metabolite fitting. AC/DC shimming provided a significant reduction (29%) in spectral linewidth and an increase (31%) in SNR, which resulted in improved accuracy for metabolite quantification by smaller (22%) Cramer-Rao lower bounds. While local homogeneity might be improved by smaller voxels with high resolution MRSI 58 , such acquisitions may not always be possible due to SNR constraints for low concentration metabolites (GABA, GSH), and better shimming may also improve high resolution data. On the other hand, challenges for water suppression and spectral editing due to global inhomogeneity do not change with image resolution and cannot be corrected with post-processing B 0 correction methods 24 , hence requiring better shimming. Furthermore, better global and local B 0 homogeneity can help MRSI in the case of spectral-spatial encoding 59 and advanced low-rank reconstruction methods 60 . Figure 6. Comparison of MRSI in the third healthy volunteer under five shimming conditions 2SH box , 2SH brain , 2SH box + AC/DC brain , 2SH brain + AC/DC brain , and (2SH brain + AC/DC brain ) joint . An inferior axial slice is shown from the stack of 3D MRSI data as indicated by the blue dashed line in the coronal and sagital views at the top. The metabolic maps of total NAA (tNAA), total Choline (tCho), total Creatine (tCr), linewidth (FWHM), signalto-noise ratio (SNR), and Cramer-Rao lower bounds (CRLB of tCr) are shown for all shimming methods. Examples of spectra from frontal (cyan box), left lateral (green box), right lateral (pink box), and occipital (black box) voxels are shown overlaid for all shimming conditions. The values under the maps indicate the mean and standard deviation calculated over the whole-brain slab. MEMPRAGE anatomical image is shown at the top, with the limits of the MRSI slab shown on the sagital and coronal slices by the red dashed lines. www.nature.com/scientificreports/ Multi-coil shimming has been explored for single voxel spectroscopy at ultra-high field at 9.4 T 16 , however to date the use of shim arrays for MRSI has not been explored beyond 3 T 33,34 . The MRSI improvements obtained by us at 7 T with AC/DC shimming are in line with previously reported improvements using 4 th -order spherical harmonics shimming 12 . Hence, AC/DC shimming may provide an alternative to high-order spherical harmonics shimming, with lighter in-bore hardware. The stability of our AC/DC hardware was monitored by real-time B 0 fieldmapping with a volumetric navigator which showed similar frequency drift for 2SH and 2SH + AC/DC Figure 7. Comparison of B 0 fieldmaps in mutant IDH1 glioma patient under four shimming conditions 2SH box , 2SH brain , 2SH box + AC/DC brain , and 2SH brain + AC/DC brain . Five representative slices are shown through the shimmed whole-brain slab with the tumor ROI contours. Standard deviation of B 0 distributions is given below for the entire slab and the tumor. Histograms of B 0 distribution over the entire brain slab are shown overlaid at the bottom for all the shimming conditions. FLAIR anatomical image is shown on the left most column. The AC/DC methodology has potential for clinical translation as demonstrated in the mutant IDH1 glioma patient. The AC/DC shimming enabled us to obtain the largest brain coverage shown to date for 2HG imaging at 7 T. The quality of 2HG imaging was improved by AC/DC shimming with fewer voxels with false results, which yielded higher tumor contrast-to-noise ratio. This patient was particularly challenging because of prior tumor surgery and chemotherapy that may have decreased the 2HG levels, but also due to metal implants used for skull repair. The metal implants can interfere with B 0 shimming and B 1 transmit efficiency which can decrease the SNR of MRSI. It was hence critical to improve the SNR for 2HG detection by using the AC/DC shimming. Figure 8. Maps of the D-2-hydroxyglutarate to total creatine (2-HG/tCr) shown in axial, sagital, coronal views in the mutant IDH1 glioma patient obtained with the four shimming conditions 2SH box , 2SH brain , 2SH box + AC/ DC brain , and 2SH brain + AC/DC brain . FLAIR image is shown at the top, and the map of total choline to total creatine (tCho/tCr) for 2SH brain + AC/DC brain shim is shown at the bottom. Cramer-Rao lower bounds (CRLB) maps are shown at the right of the metabolic maps. Below each metabolic map the mean and standard deviation of metabolic ratios in the tumor are calculated, and the contrast-to-noise ratio of the tumor. The tumor contours are indicated on the FLAIR and metabolic maps. The tumor 2HG volume detected with CRLB < 20% and the ratio of 2HG to FLAIR volume is indicated for each shimming under the CRLB maps. Examples of spectra from the tumor (B) and healthy brain (C) are shown for all shimming conditions in the right two columns. The maps of tNAA, tCho, tCr, SNR, FWHM and CRLB (of tCr) are shown in Supplementary Fig. 4. Table 1. Mean and standard deviation of B 0 -maps, MRSI spectral linewidth (FWHM), signal-to-noise ratio (SNR), and goodness of fit measure by Cramer-Rao lower bounds (CRLB of tCr) metric, contrast-to-noise ratio (CNR of 2HG), the percentage of voxels with acceptable quality (i.e. FWHM < 0.1 ppm, CRLB < 20%, SNR > 10). The CNR was calculated for tumor 2HG in the mutant IDH1 patient (NA not acquired). *This shim was measured in four subjects. # This shim was measured in two subjects. § This shim was measured in one subject. www.nature.com/scientificreports/ The in-vivo 2HG results are in line with results from the calibration phantom that showed the highest correlation between the measured and ground-truth 2HG concentrations, and from simulations which predicted that precision and accuracy of 2HG quantification is dependent on the linewidth. Our 3D MRSI sequence has three advantages over previous 2D MRSI 47 used for 2HG detection at 7 T: (1) compared to PRESS used in Ref. 47 , our ASE excitation has 14-times less chemical shift displacement error for slab localization with a sharp profile, and also compensates for B 1 transmit inhomogeneity; (2) our efficient spectral-spatial encoding using spiral out-in allowed us to obtain the highest spatial resolution so far reported for 2HG imaging at 7 T and in a shorter acquisition time; (3) our whole-brain slab coverage and high resolution metabolic imaging makes possible to probe the full spatial extent and heterogeneity of tumor metabolism, which complements better anatomical imaging. Compared to semi-adiabatic single voxel excitation shown at 7 T for 2HG detection 46 , our ASE uses fewer pulses reducing specific absorption rate and echo time to allow faster repetition times and increased SNR. Our study has some limitations. We limited the AC/DC shimming to a brain slab of 70 mm thickness because there is less gain in B 0 homogeneity for thicker slabs. Our AC/DC coil used a design that was optimized for detection of RF signal, rather than to generate local B 0 field patterns, but improved versions may be designed in the future that are optimized for thicker brain slabs. Our study was also limited by a small sample size of four subjects and further validation of this methodology may be needed in larger studies. However, we obtained reproducible and consistent improvements in all subjects and the calibration phantom.
In summary, AC/DC shimming significantly improves B 0 homogeneity at 7 T, enabling MRSI to take full advantage of higher SNR available at ultra-high field. Hence, AC/DC shimming provides better 3D MRSI data quality and metabolite quantification for imaging human brain metabolism at 7 T. This methodology may increase the number of successful imaging investigations and reduce false results. Hence, AC/DC can increase the throughput of 7 T without the need for repeated scans after failed examinations, which may save costs in clinical operation and increase patient comfort. Robustness is further enhanced by the use of a navigator for real-time motion correction and frequency update. This robust performance can facilitate the clinical translation of ultra-high field metabolic imaging in patients with brain diseases. Imaging metabolism brings more specificity to molecular mechanisms of disease in patients, which could help disambiguate the confounding effects 61 of changes associated with water in anatomical MR imaging, and better guide patient management 41 . AC/DC shimming hardware and software. We used recently-developed integrated RF-receive and multi-coil B0 shim coil array hardware for dynamic shimming. The array consists of a 31-channel AC/DC coil array patterned on a close-fitting 3D-printed polycarbonate helmet (3D printer used Fortus 360mc, Stratasys, Rehovot, Israel) (Fig. 9A). The AC/DC coil had 31 RF-receive channels and 31 B 0 -shim channels. In 25 of the loops, B 0 shim capability was added by creating a DC current path using twisted pair wires and inductive chokes to bridge the DC into the RF circuit. Six of the loops on the inferior-posterior part of the helmet deemed less important for B 0 shimming were retained as RF receive-only loop. Six four-turn B 0 -shim-only loop coils were placed over the face to target prefrontal cortex building on prior local shim coil work 62 from other groups. This AC/DC design has been shown in simulations 9 to provide comparable B 0 shim performance to a 4 th -order spherical harmonics shim insert. Thus this hardware setup preserves the receive sensitivity of a close-fitting brain RF receive array, while also providing local B 0 field control capability using the same array-both of these features are critical for obtaining high-quality MRSI data. A home-built detunable quadrature birdcage coil was used for RF transmit.

Our methodology was developed and implemented on a whole-body 7 T Magnetom MR scanner (Siemens
Coil loops with a diameter of 9.5 cm made of AWG16 solid wire are arrayed in a hexagonal-pentagonal pattern, with critical overlap to decouple neighboring elements. Other details of the RF coil are given in 26,63 . The AC/DC coil shim currents were driven by a bank of digitally-programmable low-voltage amplifiers that provide DC current up to ± 2.5 A per channel and allow very fast switching in less than 1 ms between different B 0 field patterns 28 . The output stage devices are mounted to heat sinks with in-laid piping for water cooling. RF excitation was achieved using a detunable quadrature birdcage coil. Dielectric pads were added around the sides of the subject's head to improve the transmit B 1+ homogeneity and achieve more uniform MRSI data quality across the whole brain 64 . The AC/DC shimming was superimposed on the 2SH shimming produced by the scanner hardware. In addition to different hardware combinations, several shimming algorithms were investigated based on the target shim volume, which included only the brain or was a rectangular "box" that included all the head compartments (brain, bone, and scalp) within the MRSI slab. In total five shimming combinations were studied: (1) 2SH box , (2) 2SH brain , (3) 2SH box + AC/DC brain , (4) 2SH brain + AC/DC brain , and (5) (2SH brain + AC/DC brain ) joint . The 2SH box is the shimming method provided by the manufacturer to which all users have access. The 2SH brain was previously reported 20 to improve 2SH box , and here we needed to investigate whether AC/DC shimming can provide futher improvement. The joint optimization of 2SH and AC/DC shimming (2SH brain + AC/DC brain ) joint is expected to make the best use of the orthogonal spatial basis functions of all shimming channels. For calculating the 2SH box shim we employed the vendor supplied software using three shim-iterations with the dual echo steady state (DESS) sequence.
For calibrating the AC/DC shim coils, a vendor-provided two-echo (∆TE = 1.02 ms) gradient echo (GRE) fieldmapping sequence was used to measure the B 0 -fields caused by each individual AC/DC shim coil in a large phantom that completely fills the coil. The same B 0 mapping sequence was also used to calibrate the scanner www.nature.com/scientificreports/ 2SH shim coils for the 2SH brain shimming. During the subject measurements B 0 field maps were measured using the same sequence. All field maps were acquired with field-of-view (FOV) of 224 × 224 × 200 mm 3 , matrix size 112 × 112 × 100, and 2 × 2 × 2 mm 3 isotropic voxel, acquisition time 1:55 min:s. The phase difference GRE image was spatially unwrapped with FSL PRELUDE 65 , converted to a B 0 -fieldmap, and transfered to an offline computer where the optimal shim currents for the 31 shim channels were computed in under one minute using Matlab's "quadprog" optimization function (The MathWorks, Natick, MA). A single shim-iteration was performed as follows. Shim currents were calculated using a least squares penalty on ΔB 0 with the goal of minimizing the standard deviation of the ΔB 0 over the brain slab of interest that had 70 mm thickness in all subjects. The brain was masked using the FSL Brain Extraction Tool from the magnitude image of the first GRE echo. The objective function is a quadratic program with linear inequality constraints enforcing maximum current per channel (2 A) and total current to the array (25 A). Because of low currents, there is minimal heating of the loops that does not require water cooling of the coil. A similar optimization was performed for 2SH brain using the current limitations and field calibration of the scanner 1st and 2nd order shim coils. Because the AC/DC shim is optimized over the brain the B 0 field can vary rapidly outside the brain in the subcutaneous fat, and this can degrade the lipid suppression employed by our sequence. To avoid this problem, a trigger pulse from the sequence is used to switch off the AC/DC shim immediately prior to lipid suppression. A second trigger pulse is used to switch the optimal AC/DC shim back on immediately prior to the water suppression pulses. Because of the low inductance of the coils and rapid switching capability of the control electronics, the fields can be quickly updated during the acquisition in this manner without introducing MRSI artifacts.
In summary, our proposed shim methodology consists of four components: (1) The scanner 2SH-shim, which provides the "baseline" shim, (2) The AC/DC coil, which adds localized ΔB 0 fields, (3) the possibility to dynamically switch those AC/DC fields within each TR, allowing separate optimization of the shim for metabolite detection and lipid suppression, (4) our shim software, which more readily than the scanner's shim software allows shimming only of the brain instead of the whole head.
MRSI acquisition and processing. The whole-brain 3D MRSI sequence (Fig. 9B) consisted of five modules: (1) adiabatic spin-echo (ASE) slab excitation 66 using an AHP4 excitation pulse and two GOIA-W(16,4) refocusing pulses; (2) fat suppression with inversion recovery using an asymmetric adiabatic hypergeometric (HGSB) pulse 67 and 270 ms inversion time; (3) four-pulse WET module 5 of 160 ms total duration which was optimized for water suppression at 7 T, and played out within the inversion time; (4) a stack of spiral out-in trajectories 49,50 designed for the 7 T-SC72CD gradient; (5) volumetric EPI navigator which was interleaved each TR for real-time motion correction, fieldmapping and frequency drift correction 52,53,68 .
ASE excitation used an AHP4 69 pulse of 4 ms duration, 5 kHz bandwidth, 12 μT B 1+ amplitude, and two GOIA-W(16,4) 70 refocusing pulses of 5 ms duration, 20 kHz bandwidth, 14 μT B 1+ amplitude (20% above adiabatic threshold). The ASE sequence is a double refocussing sequence similar to PRESS, and the echo time was optimized for 2HG detection at 7 T using TE1/TE2 = 58/20 ms (TE = 78 ms) as proposed for PRESS 45 sequence. It was verified by simulations that similar to PRESS the ASE with TE1/TE2 = 58/20 ms provides a negative peak for 2HG at 2.25 ppm (Fig. 1). However, the ASE sequence has much smaller chemical shift displacement error (1.5% for 1 ppm at 7 T) compared to PRESS 47 (21% for 1 ppm at 7 T), and also compensates for B 1+ transmit inhomogeneity. The slice profile of GOIA pulses used in our MRSI acquisition were simulated for ± 1 ppm chemical shift offset (Fig. 9B), and indicate minimal (2.5%) distortion of the flat top of the passband over 2 ppm chemical shift range. Narrow transition band (10% of the passband) and no out of band exciation are also evident.
The HGSB pulse had 30 ms duration (A = 3.2842; B = 0.1751; C = -1.7; D = 1.4231; Ω = 9.1809), 12 μT B 1+ amplitude, 2 kHz inversion band, and a transition band of 90 Hz that was centered at 1.6 ppm, providing full inversion below 1.4 ppm and no inversion above 1.8 ppm, hence it suppressed the main lipid peaks at 1.2 and 0.9 ppm while preserving the metabolite SNR. Simulation of the HGSB inversion profile is shown in Fig. 9B. Triggers placed before and after the HGSB pulse were used to switch off/on the AC/DC shimming. WET module used Gauss pulses of 150 Hz bandwidth, flip angles of 83.6°, 99.7°, 74.7°, 160°, and 40 ms interpulse delays. The spiral out-in trajectories were designed for human 7 T MRSI with a spectral window of 2.7 kHz, FOV of 220 × 220 mm 2 , and matrix of 44 × 44, which required a maximum gradient amplitude G max = 14.19 mT/m, and maximum slew rate S max = 158.89 mT/m/s. The spiral out-in design eliminates rewinders used for spiral-out, which increase sampling efficiency and the SNR of MRSI 49,50 .
In addition to MRSI metabolite data, water unsuppressed MRSI data (matrix of 22 × 22 × 8; acquisition time of 4:19 min:s) were acquired for coil combination and phasing of metabolite spectra. Coil combination of metabolite data was performed using S/N 2 weighting 71,72 of the individual channel data based on the water data. After coil combination, the reconstruction of the non-cartesian sampled data was performed via the nonuniform discrete Fourier transform (NUDFT) 73 , followed by removal of residual lipid signal with the L1 penalty 74  MR spectra were phase/frequency corrected and fitted with LCModel 77 between 1.8 and 4.2 ppm, with a basis-set simulated in GAMMA 78 using the same pulses and gradient modulation as played by the scanner and including 20 metabolites: D-2-hydroxyglutarate (2HG), aspartate (Asp), creatine (Cr), gamma-aminobutyric acid (GABA), glutamate (Glu), glutamine (Gln), glutathione (GSH), glycine (Gly), glycerophosphocholine (GPC), glycerophosphoethanolamine (GPE), myo-inositol (Ins), lactate (Lac), N-acetyl-aspartate (NAA), N-acetylaspartyl glutamate (NAAG), phosphocholine (PCh), phosphocreatine (PCr), phosphorylethanolamine (PE), scyllo-Inositol (Scy), serine (Ser), and taurine (Tau). Total NAA (tNAA) is reported as the sum contribution of NAA and NAAG, total choline (tCho) as the sum of GPC and PCh, and total creatine (tCr) as the sum of Cr and PCr. Note that macromolecule (MM) signals at 2.25 ppm and 4 ppm overlapping 2HG have a T2 relaxation (T2 ~ 20 ms 79,80 ) approximately five times shorter than T2 relaxation of 2HG at 7 T (T2 ~ 100 ms, assumed similar to glutamate 81 ), hence the MM signals will decay 20 times more than 2HG for TE = 78 ms minimizing the potential quantification bias of 2HG due to MM. Metabolic maps and quality parametric maps for SNR, CRLB and linewidth (FWHM, full width half maximum) are produced for each MRSI data set as estimated by the LCModel fitting routine. Linewidth less than 0.1 ppm and Cramer-Rao lower bound (CRLB) less than 20%, and SNR > 10 were used to determine acceptable goodness of fit for the metabolites quantification.
Human subjects. Three healthy subjects (2 females and 1 male, ages 28-33 years) and one mutant IDH1 glioma patient (female, age 42 years) were measured to test our methodology. All experiments and methods were carried out in accordance with relevant guidelines and regulations. The study had ethical approval from the Massachusetts General Hospital Ethics Committee and adhered to the Helsinki Declaration and in accordance to the US government guidelines. Informed consent was obtained from each subject using a study protocol that was approved by the institutional review board (IRB). The mutant IDH glioma patient had previous brain tumor surgery and the IDH1-mutational status was established by immunohistochemistry (IHC) analysis using an antihuman R132H antibody (DIANOVA 83 ), and also subsequently confirmed by genetic sequencing (SNaPshot 84 ).
All five shimming conditions could not be tested in all four subjects because some subjects could not tolerate the entire two hour duration of the full imaging protocol. The 3D MRSI data were measured as following: (a) 2SH box and 2SH box + AC/DC brain in all four subjects; (b) 2SH box , 2SH brain , 2SH box + AC/DC brain , 2SH brain + AC/ DC brain , (2SH brain + AC/DC brain ) joint in one healthy volunteer; (c) 2SH box , 2SH brain , 2SH box + AC/DC brain , 2SH brain + AC/DC brain in the patient. We chose to measure the 2SH box and 2SH box + AC/DC brain in all subjects because the comparison of these two shimming conditions is the most relevant: (a) 2SH box is the vendor shimming tool that is used by all users, and (b) 2SH box + AC/DC brain represents the most fair comparison to measure the improvements of AC/DC shimming in addition to the manufacturer shimming. In two subjects (one healthy and one patient) we investigated additional improvements that would be possible when a better shim focused only on the brain (2SH brain ) could be performed with the scanner hardware. For the glioma patient the tumor contrast-to-noise ratio (CNR) in metabolic images was calculated using the following definition

Phantom measurements.
A structural-metabolic phantom (Fig. 2) was custom made with six tubes (25 mm diameter) placed symmetrically inside of a larger cylindrical container (110 mm diameter). The six tubes contained buffered (pH = 7) solution of brain metabolites and 2HG as following: (a) the D-2HG concentration was chosen to be different across the six compartments, respectively 0, 1, 2, 3, 4, and 5 mM, while (b) the same background of brain metabolites was used in all compartments, assuming a tumor metabolic profile with 6 mM of NAA, 8 mM of glutamate, 1 mM of GABA, 4 mM of creatine, 5 mM of choline, 8 mM of myoinositol, and 4 mM of lactate. The tubes were doped with magnevist 1 ml/l to shorten T 1 , while outside the tubes a buffer aqueous solution without metabolites and magnevist was used to fill the cylinder. All five shim conditions 2SH box , 2SH brain , 2SH box + AC/DC brain , 2SH brain + AC/DC brain , (2SH brain + AC/DC brain ) joint were used to test 3D MRSI metabolic imaging against ground truth.
Simulations. In order to evaluate the effect of linewidth on 2HG quantification, we performed quantum mechanics simulations (GAMMA 78 ) of brain tumor spectra for different spectral linewidths assuming the ASE pulse sequence parameters used in vivo. Synthetic tumor spectra were obtained by combining simulated spectra of 14 brain metabolites and 2HG. The concentration of 2HG was set to 5 mM while for the other 14 metabolites the concentrations were: 4 mM glutamate, 7 mM glutamine, 1 mM GABA, 0.5 mM glutathione, 2 mM glycine, 8 mM myo-ionsitol, 3 mM lactate, 5 mM NAA, 2 mM NAAG, 2 mM PC, 2 mM GPC, 3 mM Cr, 3 mM PCr, 2 mM taurine. T 2 relaxation times of 132 ms for NAA, 152 ms for phosphocholine and glycerol-phosphocholine, 95 ms for Cr and PCr, and 93 ms for all other metabolites including 2HG were considered 81 .
To mimic the effects of B 0 inhomogeneity, we applied line broadening in the range of 3-60 Hz (0.01-0.2 ppm at 7 T) with 1 Hz (0.0033 ppm) step size. 10% white noise was added after line broadening in all simulations, Scientific RepoRtS | (2020) 10:15029 | https://doi.org/10.1038/s41598-020-71623-5 www.nature.com/scientificreports/ yielding SNR of 50 for the narrowest linewidth and SNR of 10 for the widest linewidth. The range of SNR and linewidth in simulations covered the range measured in the in vivo spectra. The simulated spectra were fitted with LCModel 77 equivalently to experimental spectra.
Statistical analysis. Statistical analysis was performed using GraphPad Prism (GraphPad Software, Inc. V4.03, CA, USA). Mean differences were compared using the non-parametric Mann-Whitney test with the threshold for statistical significance defined as P < 0.05. Data are presented as mean ± standard deviation (SD). Pearson correlation was performed for phantom measurements.