ECLIPSE utilizing Gradient Modulated Offset-Independent Adiabaticity (GOIA) Pulses for Highly Selective Human Brain Proton MRSI

Abstract

A multitude of extracranial lipid suppression methods exist for proton MRSI acquisitions. Popular and emerging lipid suppression methods each have their inherent set of advantages and disadvantages related to the achievable level of lipid suppression, RF power deposition, insensitivity to B₁⁺ field and lipid T₁ heterogeneity, brain coverage, spatial selectivity, chemical shift displacement (CSD) errors, and reliability of spectroscopic data spanning the observed 0.9 – 4.7 ppm band. The utility of elliptical localization with pulsed second order fields (ECLIPSE) was previously demonstrated with > 100-fold in extracranial lipid suppression and low power requirements utilizing 3 kHz bandwidth AFP pulses. Like all gradient-based localization methods, ECLIPSE is sensitive to CSD errors, resulting in a modified metabolic profile in edge-of-ROI voxels. In this work, ECLIPSE is extended with 15 kHz bandwidth second-order-gradient-modulated RF pulses based on the gradient offset-independent adiabaticity (GOIA) algorithm to greatly reduce CSD, and improve spatial selectively. An adiabatic double spin-echo ECLIPSE inner volume selection (TE = 45 ms) MRSI method, and an ECLIPSE outer volume suppression (TE = 3.2 ms) FID-MRSI method were implemented. Both GOIA-ECLIPSE MRSI sequences provided artifact free metabolite spectra in vivo, with > 100-fold in lipid suppression and < 2.6 mm in-plane CSD and < 3.3 mm transition width for edge-of-ROI voxels, representing a ~ 5-fold improvement compared to the parent, non-gradient-modulated method. Despite the five-fold larger bandwidth, GOIA-ECLIPSE only required a 1.9-fold increase in RF power. The highly robust lipid suppression combined with low CSD and sharp ROI edge transitions make GOIA-ECLIPSE an attractive alternative to commonly employed lipid suppression methods. Furthermore, the low RF power deposition demonstrates that GOIA-ECLIPSE is very well suited for high-field (≥ 3 T) MRSI applications.

Introduction

Magnetic resonance spectroscopy (MRS) is a powerful technique that can detect levels of neurochemicals in the brain, non-invasively. MR spectroscopic imaging (MRSI) is an extension of MRS with the ability to detect alterations in the neurochemical profile spatially, and has seen applications in a multitude of neurological pathologies^1–5. A major hindrance for reliable whole slice/brain MRSI coverage is, spectral contamination from overwhelming extracranial lipid signals. Existing extracranial lipid suppression techniques can broadly be categorized into three groups, namely traditional MR methods utilizing RF pulses and standard (linear) magnetic field gradients, post processing-based methods, and novel methods utilizing non-standard hardware. Traditional MR methods include cubical inner volume selection (IVS)^6–8, outer volume suppression (OVS) with multi-slice saturation^9–12,T₁-based nulling tailored for extracranial lipids^13–15, and spectrally selective lipid suppression¹⁶. Post-processing based methods include k-space filtering, combined with increased k-space coverage¹⁷, and L2-regularization¹⁸. Non-standard hardware-based methods include the utility of localization using B₁⁺ magnetic field shaping with a parallel transmit array^19,20 at ≥ 7 T, localization using B₀ magnetic field shaping with dedicated crusher coils²¹, and elliptical localization with pulsed second order fields (ECLIPSE^22,23). The interested reader is referred to a consensus paper¹² for a detailed description on the topic.

The ideal lipid suppression method is characterized by 1) effective lipid suppression at short TE, 2) > 90% brain coverage within the slice(s) under investigation, 3) low SAR demands to maintain compatibility with ≥ 7 T applications, 4) minimal chemical shift displacement (CSD) and transition width (TW) effects to maintain a reliable metabolic profile in edge-of-ROI voxels, and 5) have reliable spectroscopic data spanning the entire in vivo observed spectral band (0.9 – 4.7 ppm). All the above outlined extracranial lipid suppression techniques have their own set of strengths and shortcomings^23–25, and approximate the ideal method by varying amounts. Noticeably, most methods give limited lipid suppression and/or limited spatial coverage of the brain.

Recent studies demonstrate ECLIPSE^22,23 to achieve > 100-fold in mean lipid suppression, provide high axial slice coverage, and consume 30% of the power required by standard eight-slice OVS. However, common to all gradient-based localization methods, ECLIPSE is also affected by CSD and TW effects that can lead to perturbation of true metabolite ratios in ROI-edge voxels. The objectives of this work are to extend ECLIPSE methods with second-order-gradient-modulated RF pulses based on the GOIA (gradient-offset-independent adiabaticity) algorithm²⁶ to greatly increase the RF bandwidth and thereby minimize CSD, without a directly proportionate increase in SAR with respect to bandwidth. The performance of GOIA based ECLIPSE is demonstrated for OVS with ultra-short TE (3.2 ms) FID-MRSI, and for inner volume selection (IVS) in medium-TE (45 ms) double-spin-echo MRSI.

Theory

GOIA RF pulses with ECLIPSE

The development of a SAR compatible, high bandwidth adiabatic full passage (AFP) pulse was central to this work to achieve highly selective MRSI acquisitions. An AFP secant pulse with 4^th order modulation (HS4)²⁷ (RF pulse duration, T_p = 6.66 ms and maximum frequency modulation, A = 1.5 kHz, hereafter termed AFP-HS4) was used to derive a GOIA RF pulse (termed GOIA-HS4 hereafter). The starting amplitude modulation, and gradient modulation functions of the GOIA pulse are F₁(τ) = sech(βτ⁴), and F₃(τ) = (1 − GF · sech(βτ²)), respectively, as previously reported²⁶ (for τ = 2t/T_p −1, 0 ≤ t ≤ T_p, and β = 5.3). Here GF = 1 − (1/GA), with the gradient amplification factor denoted as GA. Given F₁(τ) and F₃(τ), the frequency modulation function F₂(τ), was computed numerically in MATLAB by solving the following expression²⁶: $\frac{d{F}_2\left(\tau \right)}{d\tau }-\frac{F_2\left(\tau \right)}{F_3\left(\tau \right)}\cdot \frac{d{F}_3\left(\tau \right)}{d\tau }=\frac{F_1^2\left(\tau \right)}{K}$. Here K is a scaling constant to normalize the frequency sweep of F₂(τ) to span [−1 1], such that A · F₂(τ) spans [−A A].

The GOIA-HS4 pulse used in this study was executed with T_p = 6.66 ms and GA = 5, thus making A = 7.5 kHz and thereby providing a 2A = 15 kHz bandwidth. The final amplitude modulation (AM), frequency modulation (FM), and gradient modulation (GM) functions for AFP-HS4 and GOIA-HS4 are shown in Figure 1. The ROI selected by GOIA-ECLIPSE can be modified by changing the RF pulse FM according to FM = A · F₂(τ) + Δv_RF · F₃(τ), whereby Δv_RF represent the frequency offset needed during the non-gradient-modulated AFP-HS4 RF pulse. Despite the five-fold-increased bandwidth, the GOIA-HS4 requires only 25% more RF amplitude (0.76 kHz versus 0.61 kHz) to achieve 99.5% inversion (labelled B_{1, (99.5%)}). The total time-dependent gradient field during the GOIA pulse G_total(x, y, z, τ) can be expressed as a function of first and second order gradients as follows:

FIGURE 1.

RF pulse components and analytical expressions for the GOIA-HS4 (red line) and AFP-HS4 (dotted blue line) pulses with 15 kHz and 3 kHz bandwidth, respectively. (A) Amplitude modulation function (AM) for the two RF pulses with T_p = 6.66 ms. Illustrated amplitudes correspond to the minimum B₁⁺ amplitude required for 99.5% inversion efficiency (labelled B_{1 (99.5%)}). For a 5-fold increase in bandwidth, the GOIA-HS4 pulse requires a 23% increase in B_{1 (99.5%)}, corresponding to a 51% increase in RF power. (B) Frequency modulation (FM) function and (C) gradient modulation (GM) functions for the GOIA-HS4 and AFP-HS4 pulses. The GM function for the GOIA-HS4 pulse is normalized to the gradient strength required by the AFP-HS4 pulse.

G_total(x, y, z, τ) = GA · F₃(τ)[G_z2 · (2z² − (x² + y²)) + G_X2Y2 · (x² − y²) + G_XY · (xy) + G_X · (x) + G_Y · (y)]. Here G_Z2, G_X2Y2, and G_XY are quadratic magnetic field amplitudes in Hz/mm², and G_X, and G_Y are linear magnetic field amplitudes in Hz/mm.

Chemical shift displacement for first and second order gradients

Figure 2 qualitatively illustrates the CSD for spatial localization with (A) linear and (B) quadratic magnetic field gradients. In both cases the RF pulse bandwidth (BW) and the ROI for on-resonance spins (2r₀) are identical. For spatial localization with linear gradients, the CSD (labelled as Δr) can be quantitatively expressed as

$$\Delta r_{\mathit{\text{linear}}}=\frac{\Delta v}{G_r}=2r_0\frac{\Delta v}{BW}$$ [1]

where Δv is the frequency (or chemical shift) difference between on-resonance (e.g. water), and off-resonance spins (e.g. lipids), and G_r is the linear gradient strength (in Hz/mm). It follows that the CSD, Δr_linear, scales linearly with ROI size, 2r₀.

FIGURE 2.

Comparison of chemical shift displacements (labelled Δr) associated with linear (A) and quadratic (B) gradient fields, given a fixed ROI radius (labelled r₀) of 60 mm is selected with a 3 kHz bandwidth RF pulse. A 60 mm ROI radius is selected as an approximation to the left-right dimension of a typical human brain axial slice. The chemical shift (labelled Δv) between lipid and water is taken to be 600 Hz (representative of lipid-water separation at 4 T). The 1-D ROI’s for water and lipid is illustrated in blue and red vertical dashed lines, respectively in (A-B). The 1-D lipid profile manifests as a translation by Δr with a linear gradient, in contrast to a reduction by Δr in ROI radius as associated with a quadratic gradient. (C) Relationship between $\frac{\Delta v}{BW}\text{vs}\frac{\Delta r_{\mathit{\text{linear}}}}{\Delta r_{\mathit{\text{quadratic}}}}$, as governed by eq [3] illustrates the reduction in CSD with second order gradients relative to first order gradients for a range of chemical shift to RF pulse BW ratios.

For spatial localization with quadratic gradients, the CSD can be quantitatively expressed as

$$\Delta r_{\mathit{\text{quadratic}}}=r_0-\sqrt{r_0^2-\frac{\Delta v}{G_{r^2}}}=r_0-r_0\sqrt{1-\frac{\Delta v}{BW}}$$ [2]

Here $G_{r^2}$ is the quadratic gradient field strength (in Hz/mm²). It follows that the CSD, Δr_quadratic, scales non-linearly with the ROI size, 2r₀. For identical ROIs, the ratio of CSD with linear and quadratic gradients is then given by

$$\frac{\Delta r_{\mathit{\text{linear}}}}{\Delta r_{\mathit{\text{quadratic}}}}=2+2\sqrt{1-\frac{\Delta v}{BW}}$$ [3]

For Δv ≪ BW, the absolute CSD for quadratic gradient localization is circa four times smaller than that of linear gradients. This reduces to a two-fold advantage when Δv approaches BW. Figure 2C displays the non-linear nature of Eq. [3]. For the RF pulse bandwidths 3.0 kHz (for AFP-HS4) and 15.0 kHz (for GOIA-HS4) used in the current work, the CSD advantage for water-lipid spins (Δv ~ 600 Hz at 4 T) are 3.79 and 3.96-fold, respectively, based on Eq. [3]. The absolute CSDs for an elliptical ROI with diameters in the x and y directions of 120 × 150 mm as predicted by Eq. [2] are, 6.3 × 8.0 mm and 1.2 × 1.6 mm for the 3.0 and 15.0 kHz bandwidths, respectively. It must be noted that in practice, the effective RF pulse BW metric in eq. [2] is reduced with quadratic gradient-based localization, due to the finite TW of the RF pulse. If not accounted for, inversion efficiency at the center of the ROI can be compromised due to the inner transition edge of the RF pulse. To maintain 99.5% inversion efficiency throughout the ROI, the effective BW is reduced by ~ 27% and ~ 4% for the AFP-HS4 and GOIA-HS4 pulses, respectively. In this work, the fraction reduction of the effective RF pulse bandwidth was set to 30% for both AFP-HS4 and GOIA-HS4, such that ECLIPSE gradient amplitudes with GOIA-HS4 pulses are 5-fold higher than the gradient amplitudes with AFP-HS4 pulses for equivalent ROI placement.

The CSD analysis with IVS can also be extended for OVS. However, it should be noted that conventional eight-slice OVS executed with 15 kHz GOIA RF pulses is not feasible due to SAR constraints, in vivo. However, since the OVS slice thickness is generally much smaller than for IVS, the absolute CSD for both conventional OVS and ECLIPSE-OVS with high bandwidth RF pulses is largely inconsequential. Furthermore, it should be noted that CSD on the inner edge of an elliptical ROI is slightly larger than that on the outer edge, due to the quadratic spatial dependence of second order magnetic fields.

Methods

MR system

All MR experiments were performed on a 4 T magnet (Magnex Scientific Ltd., Yarnton, UK) interfaced to a Bruker Avance III HD spectrometer running ParaVision 6 (Bruker, Billerica, MA, USA). The system contains actively shielded gradients capable of switching 30 mT/m in 1150 μs, and provides up to third order shimming. A within-brain B₁⁺ (hereafter B₁⁺ will be labeled B₁) optimized, 8-element Tx/Rx volume coil embedded in the ECLIPSE gradient coil was used in a fixed phase configuration. The 8-element Tx/Rx volume coil had a ± 30% and ± 70% B₁ variation within the brain and outside the brain (extracranial region), respectively, based on B₁ maps obtained from volunteers who participated in a previous study²³ and this study. Optimization of the methods in this work, and in vivo RF pulse power calibrations were based on these acquired B₁ maps.

The ECLIPSE gradient system

The ECLIPSE system²² is a home-built, unshielded gradient insert consisting of Z2, X2Y2, and XY second order spherical harmonic magnetic fields with amplitudes of 5.48E-2, 2.58E-2, and 2.76E-2 Hz/mm²/A, respectively. The ECLIPSE gradient fields resulted in minor linear imperfections with a maximum Y gradient of 1.0Hz/mm/A, and minor cross terms between the X2Y2 and XY coils in the order of 2E-3 Hz/mm²/A, which were accounted for in simulation. The three gradient coils are driven with three independent 100 A Techron 7780 current amplifiers (Techron, Elkhart, IN, USA) interfaced to a home-built multi-channel gradient controller²⁸. The minimum rise time of the ECLIPSE magnetic fields is 660 μs, which was increased to 1150 μs to match the system gradient rise time. Temporal alignment of the GOIA gradient waveform F₃(τ) on system gradients and ECLIPSE gradients, were validated to be accurate to within a ± 25 μs using an external oscilloscope. A minimum gradient strength (|G_Z2| − |G_X2Y2|) of 1.4 Hz/mm² was used in the current study (noting that |G_Z2| > |G_X2Y2| is required to generate an elliptical ROI), resulting in a maximum CSD from water-lipids (~ 3.5 ppm or ~ 600 Hz at 4 T) of 2.6 mm along an 85 mm radius ROI. Note that this minimum 1.40 Hz/mm² gradient strength for an 85 mm radius ROI accounts for the 30% reduced effective bandwidth of the RF pulse.

GOIA-ECLIPSE

1-D Bloch simulations of M_z/M₀ following an AFP-HS4 pulse and GOIA-HS4 pulse vs B₁ amplitude, is illustrated in Figure 3A–B for a linear gradient. The inversion pulse TW factor (slice thickness to 0.9 > M_z/M₀ > – 0.9 transition width ratio) for the AFP-HS4 and GOIA-HS4 pulses are 26% and 6%, respectively. Similar to an AFP-HS4 RF pulse, a GOIA-HS4 pulse provides slice-selective excitation (when M_z/M₀ ~ 0) at lower RF amplitudes, and slice-selective inversion above a minimum threshold RF amplitude, while maintaining its excitation/inversion profile with B₁ (Figure 3 A–B). This feature makes the GOIA-HS4 pulse compatible with both ECLIPSE based inner volume selection and outer volume suppression methods, as previously described²³. However, a frequency offset (that manifests from chemical shift) on gradient-modulated AFP pulses results in an asymmetrical inversion profile in addition to a spatial displacement^29,30. Bloch simulations for the AFP-HS4 and GOIA-HS4 pulses in Figure 3A–B were repeated for a 600 Hz offset (lipid-water chemical shift difference at 4 T), as illustrated in Figure 3C–D. The GOIA-HS4 pulse provides a smaller displacement of ~ 1.2 mm (Figure 3D) relative to ~ 6 mm from AFP-HS4, with a notable asymmetrical inversion profile at low RF amplitudes. Steps taken to mitigate effects of the asymmetrical inversion profile for off-resonance spins are detailed in the Results section for the two GOIA-ECLIPSE methods.

FIGURE 3.

Comparison of simulated residual magnetization profiles vs B₁ amplitude for the AFP-HS4 and GOIA-HS4 pulses (A-D), and for the GOIA-ECLIPSE methods (E-H), considering chemical shift related frequency offsets. All simulations are selective for a 30 mm slab with linear gradients. (A-B) Illustrate M_z/M₀ vs B₁ for a 0 Hz offset with near-flat inversion profiles. (C-D) M_z/M₀ profiles for a 600 Hz offset. In (C) an identical inversion profile for the AFP-HS4 pulse is seen relative to (A) with a 6 mm (or 20%) translation, the GOIA pulse in (D) shows a 1.2 mm (or 4%) translation relative to (B), however with an asymmetric inversion profile. (E, G) illustrates M_z/M₀ profiles with RF pulse timings in the GOIA-ECLIPSE OVS method for 0 Hz and 600 Hz offset cases, respectively. (F, H) Illustrates M_xy/M_z profiles for the GOIA-ECLIPSE IVS method for 0 Hz and 600 Hz offset cases.

Figure S1 provides a summary of RF pulse and gradient components required for ECLIPSE based elliptical selection and elliptical ring selection. Switching from an elliptical volume to elliptical ring selection is achieved by shifting the RF pulse frequency offset by - 15 kHz for GOIA-HS4 and - 3 kHz for AFP-HS4 (Figure S1 A,C), respectively; this provides an elliptical ring ROI thickness of ~ 50 mm without having to alter the Z2 and X2Y2 gradient strengths used for elliptical selection. The convention of maintaining identical ECLIPSE gradient fields for a given ROI size was used throughout this work, and results in identical CSD at the ROI edge for both elliptical volume selection and elliptical ring selection.

Switching the unshielded ECLIPSE Z2 gradient coil resulted in eddy currents that were primarily limited to time-varying B₀ field variations; this is likely due to the > 100 mm in radial separation from the system gradient, shim coils, and magnet cryostat²². However, due to the five-fold higher gradient amplitude utilized with GOIA-ECLIPSE relative to regular ECLIPSE, a single empirically determined gradient pre-pulse used previously²³ was found to be insufficient. Therefore, the B₀ modulation following the desired GOIA gradient waveform [F₃(τ)] was experimentally characterized on a water phantom using a slice-selective pulse-acquire sequence. Next, a least-squares minimization was carried out in MATLAB to evaluate placement and amplitudes of identical length and shape, pre and post gradient waveforms (relative to the center gradient waveform of interest) to minimize the B₀ modulation.

GOIA-ECLIPSE outer volume suppression (OVS)

ECLIPSE-OVS is achieved by repeatedly exciting and dephasing an elliptical ring ROI as a multi-pulse sequence with variable powers as previously described²³ with AFP-HS4 pulses. Since each ECLIPSE-OVS saturation pulse uniformly affects the entire in-plane ROI, spatially differential relaxation due to overlapping slices, as associated with conventional OVS, is avoided. Optimization of the GOIA-ECLIPSE outer volume suppression module (GOIA-ECLIPSE OVS) was constrained to four RF pulses with independent amplitudes and delays²³. Specifically, the residual signal (M_z/M₀) was minimized over a B₁ range spanning 0.31 – 1.5 kHz, and a T₁ range spanning 300 – 800 ms. These B₁ and T₁ ranges accommodate the respective RF field variation and heterogeneous T₁ species encountered around the extracranial region. Following optimization of the GOIA-ECLIPSE OVS pulses, a seven-pulse VAPOR-style (variable power RF pulses with optimized relaxation delays³¹) water suppression sequence was re-optimized, considering the optimized OVS pulses. The VAPOR RF pulse nutation angles were limited to α and 1.78α, corresponding to a 5 dB difference in RF power. The water suppression was optimized over a nutation angle range spanning 60° – 120°, and a T₁ range of 500 – 3000 ms; these ranges accommodate the nutation angle range, and the T₁ relaxation time constants of various water compartments encountered within the brain. All OVS pulses were GOIA-HS4 (T_p = 6.66 ms, and BW = 15 kHz) as described above, and water suppression pulses were Gaussian (T_p = 10 ms, and BW = 0.2 kHz). The water-suppressed GOIA-ECLIPSE OVS module was implemented for an ultra-short TE FID-MRSI method (TE = 3.2ms) as illustrated in Figure 4A.

FIGURE 4.

(A) Pulse sequence diagrams for the GOIA-ECLIPSE OVS method with interleaved seven-pulse VAPOR-style water suppression pulses. The water suppression pulses are shaded, and OVS pulses are open. OVS pulses with relative amplitudes 0.64, 0.71, 0.83 and 1.00 are placed 1181, 560, 218, and 50 ms before the slice-selective excitation pulse, respectively. Similarly, the water suppression pulses have relative flip angles of α, α, 1.78α, α, 1.78α, α, and 1.78α, placed 280, 265, 250, 170, 98, 28, and 13 ms before the slice-selective excitation pulse, respectively. The GOIA-ECLIPSE OVS method was implemented for a FID-MRSI acquisition (TE = 3.2 ms). A Shinnar-Le Roux excitation pulse (T_p = 1.35 ms, BW = 4.15 kHz) selects a 10 mm axial slice, and phase encodings are superimposed on the slice rewinder gradient. The collective Z2 and X2Y2 ECLIPSE gradients are labeled as ECL. (B) Simulated residual lipid fraction (|M_z|/M₀) as a function of T₁ and B₁, for the GOIA-ECLIPSE OVS method. Lipid suppression was optimized over a B₁ range spanning 0.31 – 1.5 kHz, and a T₁ range spanning 0.3 – 0.8 s as indicated. (C) Simulated residual water fraction (M_z/M₀) of the optimized water suppression module as a function of T₁ and B₁ (represented as nutation angle α). Water suppression was optimized over a B₁ range spanning 60° – 120°, and a T₁ range spanning 0.5 – 3.0 s as indicated.

GOIA-ECLIPSE inner volume selection (IVS)

The GOIA-ECLIPSE IVS method is based on an adiabatic double-spin-echo²² MRSI sequence with a TE of 45 ms (Figure 5), and provides elliptical localization with B₁ independent lipid suppression (given a minimum B₁ amplitude is met to attain full-intensity adiabatic refocusing within the brain). In addition, this IVS method inherently provides T₁ independent lipid suppression. A seven-pulse VAPOR-style water suppression sequence was re-optimized using Gaussian pulses as described above. Since ECLIPSE gradients are applied after all frequency-selective water suppression RF pulses, the ECLIPSE induced eddy currents are limited to a perturbation of the B₀ field during the acquisition (without higher-order contributions), that does not affect water suppression performance. Therefore, the gradient pre/post pulses utilized with ECLIPSE-OVS were not required, and the acquisition of a water reference scan was sufficient for post-acquisition eddy current correction.

FIGURE 5.

The GOIA-ECLIPSE IVS MRSI method with TE = 45 ms. The excitation pulse (Shinnar-Le Roux, T_p = 1.6 ms, BW = 3.5 kHz) selects a 10 mm axial slice, followed by in-plane ECLIPSE localization with GOIA pulses (T_p = 6.6 ms, BW = 15 kHz). The MRSI phase encodes are superimposed on the final spoiler gradient. Spoiler gradients 5.3 ms in length, with a maximum amplitude of 21 mT/m were placed around the GOIA pulses to cancel out signal from undesired coherence pathways. The adiabatic double spin-echo sequence is preceded by an optimized VAPOR water suppression module.

Human subjects, acquisition parameters, and protocol

Five healthy volunteers (1 female, 4 males) participated in the study to evaluate GOIA-ECLIPSE OVS and IVS methods. All participants provided written informed consent prior to each MRI session, and all procedures were previously approved by the Yale University Institutional Review Board. Multi-slice gradient-echo MRIs were used to plan the MRSI slice location in the z-direction, and a five-slice gradient-echo MRI was used to interactively plan the elliptical ROI. The elliptical ROI can be translated in x, and y directions, along with x, and y radii to best approximate the axial brain shape. High-resolution proton density weighted axial MR images were acquired (TR/TE = 500/6.5 ms, FA = 15°, resolution 1 × 1 × 4 mm³) for anatomical referencing of the MRSI grid. The magnetic field homogeneity over the 10 mm MRSI slice was optimized with first and second-order shim fields using a vendor supplied B₀ field mapping method, resulting in linewidths of ~ 12 Hz over the slice. A reference water scan was acquired for both GOIA-ECLIPSE methods, with water suppression pulses off. An additional water suppressed, non-lipid-suppressed reference MRSI scan was acquired: OVS pulses were turned off for the GOIA-ECLIPSE OVS method, and the ECLIPSE gradients were turned off during the adiabatic double-spin-echo to achieve global in plane localization with the GOIA-ECLIPSE IVS method. No manual RF power adjustments were made for any MRSI sequences following the vendor-supplied, slice-based RF power calibration.

Both GOIA-ECLIPSE MRSI acquisition methods are sampled over the elliptical portion of a 17 × 21 k-space matrix (170 × 210 mm² FOV, 1 mL nominal resolution, TR = 2000 ms, 2048 complex points over 5.0 kHz, leading to an 8.2 min acquisition time). A 10 mm z-slice selective Shinnar-Le Roux³² pulse was used for both GOIA-ECLIPSE IVS (T_p = 1.6 ms, and BW = 3.5 kHz) and OVS (T_p = 1.35 ms, and BW = 4.15 kHz) methods.

ECLIPSE gradient field simulations and data processing

The 1-D Bloch simulation with first order gradient fields illustrated in Figure 3 (A–D) was extended to generate 2-D spatial profiles using calibrated second order ECLIPSE gradient fields as previously described²². The objective of the 2-D Bloch simulations was to evaluate spatial lipid suppression performance of GOIA-ECLIPSE IVS, and to evaluate selectivity metrics (TW and CSD) with 0.05 mm precision in simulation for both methods, that is extremely time-inefficient to achieve experimentally. Optimized inter-pulse timings of both GOIA-ECLIPSE OVS and IVS methods and T₁ = 400 ms was used (considering the dominant lipid species at 4 T²³) when evaluating selectivity metrics. The GOIA-ECLIPSE IVS method was simulated with a 4 × 4 = 16-step phase cycle to mimic spoiling for a 170 mm diameter ROI along the y-axis (similar to Figure S1 A(ii)). Residual M_xy/M₀ metrics were calculated on a 20 mm ring ROI, 3 mm outside of the ROI boundary (the ROI boundary is defined where M_xy/M₀ ~ 0.5) to attain a lipid suppression factor metric for the GOIA-ECLIPSE IVS method. The selectivity metric is calculated to be the ratio between ROI width, and transition width (0.95 > M_xy/M₀ > 0.05). CSD is calculated to be the difference between the ROI radii at 0 Hz offset (analogous to the water ROI in Figure 2B) and 600 Hz offset (analogous to the lipid ROI in Figure 2B), based on the ECLIPSE B₀ fields.

A home-written MATLAB based software toolkit, NMRWizard³³, was used for all data processing. The following steps were employed on all MRSI datasets. 1) The water reference MRSI was used to compute voxel-wise optimal amplitude- and phase-weightings from each RF coil element. The water reference was further used for eddy current corrections with the GOIA-ECLIPSE IVS method. 2) A spatial 2D Fourier transform without apodization was performed, followed by 2 Hz Lorentzian line broadening, Fourier transformation, and zero-order phase correction. 3) A first order phase correction was required on the FID-MRSI spectra due to the 3.2 ms delayed acquisition, 4) NAA, creatine, and choline maps were generated by baseline corrected spectral integration over a 0.2 ppm range centered on the respective metabolite peak from the raw spectral data. Voxel-wise residual lipid fractions were calculated by taking the ratio of the integral over the 0.8 – 1.6 ppm region for the lipid-suppressed, and the water suppressed but non-lipid-suppressed datasets.

Results

Eddy current compensation with pre/post gradient pulses

Eddy current induced B₀ modulations were minimized in simulation when pre/post pulses are applied in succession (with no delays in between) with amplitude values - 109% and - 9% relative to the center gradient pulse (see Figure. 4A). In phantom, the eddy current induced B₀ modulations were minimized 19-fold from 1.8 Hz/A with a decay constant of 70 ms, to < 0.1 Hz/A with a decay constant of 80 ms. No post-processing eddy current corrections were required for MRSI data acquired with the GOIA-ECLIPSE OVS method.

Lipid and water suppression performance in simulation

Figure 4B–C illustrates simulated residual lipid fraction and residual water fraction for the water suppressed GOIA-ECLIPSE OVS method depicted in Figure 4A. The optimized GOIA-ECLIPSE OVS method provides effective lipid suppression, given a minimum B₁ amplitude of 0.31 kHz is met. The max/mean/median/standard deviation of residual magnetization descriptive statistics for the lipid suppression module in the T₁ span of 300 – 800 ms and B₁ span of 0.31 kHz to 1.5 kHz is 0.025/0.004/0.002/0.005, corresponding to a mean lipid suppression factor of 250-fold. The max/mean/median/standard deviation residual magnetization descriptive statistics for the water suppression module optimized over a nutation angle span of 60° to 120° and T₁ span of 500 – 3000 ms is 0.01/0.0013/0.00006/0.002, corresponding to a mean water suppression factor of ~ 770-fold. Note that the displayed T₁ and B₁ span in Figure 4 B–C extends beyond the optimization regime required for this work, to be comparable with simulation results in a previous study²³, and to illustrate compatibility at 3 T and 7 T field strengths.

The effects of chemical shift for GOIA-ECLIPSE OVS and GOIA-ECLIPSE IVS methods in simulation is illustrated in Figure 3E–H. For the GOIA-ECLIPSE OVS method, an asymmetric residual M_z/M₀ profile is present (Figure 3G) relative to the on-resonance case (Figure 3E). However, B₁ independent lipid suppression is achieved in both on-resonance and 600 Hz offset cases given a minimum B₁ amplitude is met (at the cost of a 28% higher B₁ amplitude relative to the on-resonance case in Figure 3E). Alternatively, executing the GOIA-ECLIPSE OVS module on-resonance for lipids at ~ 1.2 ppm is a more power-efficient solution. Therefore, the carrier frequency of the GOIA RF pulses was set to 1.2 ppm for all GOIA-ECLIPSE OVS experiments in this work.

Lipid suppression performance of GOIA-ECLIPSE IVS was evaluated over a 300 – 800 ms T₁ span and a B₁ span of 0 – 1.8 kHz on a 20 mm ring ROI resulting in max/mean/median/std residual magnetization descriptive statistics of 0.0061/0.00018/2×10⁻¹¹/8.4×10⁻⁴, corresponding to a mean lipid suppression factor of ~ 5500-fold. A 20 mm ring ROI was used to attain an aggressive lipid suppression performance metric, since the residual magnetization beyond 20 mm in simulation is impractically small due to improved localization further away from the IVS boundary. While a ~ 5500-fold in mean lipid suppression may not be representative in practice, this statistic demonstrates that the level of lipid suppression with GOIA-ECLIPSE IVS is contingent upon effective dephasing of undesired coherence pathways through improved spoiling. In addition to superior lipid suppression of GOIA-ECLIPSE IVS to GOIA-ECLIPSE OVS, the B₁ independent lipid suppression with the GOIA-ECLIPSE IVS method is evident in both 0 Hz offset (Figure 3F) and 600 Hz offset (Figure 3H) cases (M_xy/M₀ values immediately outside the ROI) due to inner volume selection. However, similar to Figure 3G, an asymmetrical M_xy/M₀ profile is present with the GOIA-ECLIPSE IVS method in Figure 3H due to the 600 Hz offset. Given that the GOIA-ECLIPSE IVS method operates over the entire observed spectral range spanning NAA to water, a 14% increase in B₁ amplitude was employed in vivo.

Localization performance of GOIA-ECLIPSE in simulation and in vitro

Figure 6 compares localization performance (TW and CSD) for GOIA-ECLIPSE IVS in a two-compartment water-lipid phantom with AFP-HS4 and GOIA-HS4 pulses. The CSD metric is defined as the difference between the water and lipid ROIs, and TW is defined as noted earlier, and is illustrated in Figure 6E. CSD along the x-axis (120 mm diameter ROI) was ~ 9.0 mm (0.8 mm in simulation) and ~ 1.7 mm (1.7 mm in simulation) for AFP-HS4 and GOIA-HS4, respectively, demonstrating close agreement between simulation and experimental values. When the Z2 gradient field has a positive amplitude, a radially decreasing gradient field in the X-Y plane is generated, resulting in a smaller ROI for up-field lipid signals relative to water (see Figure 6B, D). This convention was used for all in vivo MRSI acquisitions.

FIGURE 6.

GOIA-ECLIPSE IVS acquisition on a two-compartment water-oil phantom. An elliptical 120 mm × 170 mm diameter ROI has been selected. The second order gradient fields along the x-axis in (A-D) are 0.56 Hz/mm², −0.56 Hz/mm², 2.77 Hz/mm², and −2.77 Hz/mm². A faint larger ring is visible in (B) due to multiple resonances in vegetable oil. (E-F) illustrate 1-D profiles marked with blue and red lines in (A, C), with zoomed in respective panels in green. An ideal profile (with no CSD, and TW) is illustrated with a dashed line for comparison. The TW is calculated in the span 0.95 > |M_xy|/M₀ > 0.05 for a given 1-D profile, and CSD is calculated to be the translation between water and lipid profiles, as illustrated in the zoomed in panel of (E).

Similar to CSD comparisons in Figure 2, TW comparisons were also made in simulation with first and second order gradient fields. Figure S2 illustrates 1-D slice selective inversion profiles for a 170 mm diameter ROI using a GOIA-HS4 pulse with first and second order gradients. The inversion pulse TW (0.9 > M_z/M₀ > – 0.9) is calculated to be 10.9 mm and 2.9 mm for first and second order gradients, demonstrating a 3.75-fold improvement in TW solely due to the utilization of second order fields.

The CSD and TW characteristics were also determined for the GOIA-ECLIPSE OVS method. Since the ECLIPSE gradient strength was set to be identical with both GOIA-ECLIPSE IVS and OVS methods for identical ROI sizes, CSD at the ROI edge (for IVS) and inner ROI edge (for OVS) is identical. The TW, however, can be different between the two methods as localization is achieved with a different number of RF pulses and pulse delays (four-pulse OVS method, vs two-pulse double spin-echo IVS method). Table 1. summarizes the main characteristics of the GOIA-ECLIPSE OVS and IVS methods based on simulation and experimental findings. The maximum absolute CSD and TW metrics scale with ROI size, thus were calculated for a 170 mm ROI diameter in Table 1.

Table. 1.

Comparison of ECLIPSE-OVS and ECLIPSE-IVS characteristics

	ECLIPSE-OVS	ECLIPSE-IVS
Power required relative to one full-inversion (B 1 (99.5%) ) GOIA-HS4 pulse	0.44	2
Minimum peak B1 amplitude required (kHz/μT)	0.31/7.28	0.76/17.9
TR averaged RF power1 in vivo	1.7W	3.1W
Maximum2 CSD (mm)	2.6	2.6
Maximum2 TW (mm)	3.3	3.2
Minimum2 selectivity3	15	53
Minimum TE (ms)	3.24	404
Minimum TR (ms)	15005	20006
Post-processing eddy current correction	Not required	Required with current implementation
Outer volume lipid T1 species effectively suppressed	300 – 800 ms	All
Provides B1 independent lipid suppression	Given a minimum B1 amplitude is met	Yes
Mean lipid suppression factor in simulation 7	~ 250-fold	~ 5500-fold8
Mean lipid suppression factor in vivo	~ 116-fold	~ 204-fold

¹Total power consumed, neglecting contributions from the excitation and water suppression pulses (TR = 2 s, power averaged over 10 s)

²Based on a maximum ROI diameter of 170 mm

³Selectivity is defined as the ROI width to TW ratio

⁴limited by rise time and amplitude of system gradients and T_p of RF pulses used

⁵limited by the OVS module duration

⁶Limited by in-house software monitored time averaged SAR

⁷T₁ and B₁ spans used for the two cases are described in the Results section

⁸Simulation results are based on a 16-step phase cycle to mimic spoiling

Outer volume suppression with FID-MRSI in vivo

Figure 7 summarizes the performance of extracranial lipid suppression for the GOIA-ECLIPSE OVS method in combination with FID-MRSI for three volunteers. The overall residual lipid fraction M_z/M₀ descriptive statistics max/mean/median/standard deviation were 0.068/0.0086/0.0064/0.0077, corresponding to a mean experimental lipid suppression factor of ~ 116-fold. NAA and creatine maps in Figure 7 D–E show no apparent lipid contamination with clear delineations of CSF in the lateral ventricles. Furthermore, minimal attenuation of the metabolic profile along edge-of-brain voxels is observed, demonstrating that GOIA-ECLIPSE OVS provides minimal CSD effects and high spatial selectivity. These lipid suppression metrics are significant, since: 1) short T₁ and T₂ lipid resonances are emphasized at an ultra-short TE and 2) a sparsely sampled k-space acquisition (17 × 21) was used, which provides minimal assistance from the spatial point-spread function to minimize lipid signal leakage into the brain ROI. Voxel-wise spectra also show excellent water suppression across all volunteers (with an overall residual water fraction M_z/M₀ statistics max/mean/median/standard deviation of 0.031/0.0024/0.0016/0.0029), corresponding to a mean water suppression factor of ~ 425-fold. In addition, close observation of spectra in Figure 7 show no apparent eddy current induced B₀ modulations, demonstrating that the gradient pre/post pulse pair was highly effective.

FIGURE 7.

Summary of FID-MRSI acquisitions (TR/TE = 2000 ms/3.2 ms) with the GOIA-ECLIPSE-OVS method from three volunteers on a 10 mm slice with 10 × 10 mm² in-plane resolution. (A) The anatomical image corresponding to the MRSI slice illustrating the ECLIPSE-OVS ROI in blue, and five adjacent MRS voxel locations illustrated in panel (G) for the three volunteers. For volunteers 1 – 3, axial coverage with the ECLIPSE ROI was 89%, 96%, and 97% of the axial brain slice, respectively. (B) The extracranial lipid mask used to evaluate lipid suppression performance as illustrated in (F). (C) Lipid map for the water suppressed MRSI acquisition with ECLIPSE-OVS pulses turned off. (D) NAA and (E) creatine maps obtained from the MRSI acquisition with ECLIPSE-OVS activated. Note that the NAA and creatine maps have been scaled up by a factor of 200 to have comparable intensity to lipid maps in (C). For all three volunteers, artifact free NAA and creatine maps are produced despite the use of a FID-MRSI acquisition that would be highly susceptible to lipid contaminants. In all three volunteers, the lateral ventricular outline is visible to varying levels, based on partial volume effects of the z-slice selection. (F) Residual lipid fraction maps evaluated over the 0.8 – 1.6 ppm spectral region in the extracranial lipid suppressed MRSI, and extracranial lipid unsuppressed MRSI scans. For volunteers 1 – 3, the calculated mean lipid suppression factor is 104-fold, 112-fold, and 131-fold, respectively, and no noticeable hotspots in metabolic maps were present. A maximum residual lipid fraction M_z/M₀ = 0.068 can be seen for one voxel in volunteer 1; however, no noticeable artifacts is registered in the corresponding metabolic maps. (G) First order phase corrected spectra from voxel locations illustrated in (A) for the three volunteers. For all spectra, the first three voxels are within the brain ROI, and voxel 4 is immediately outside the brain ROI. Voxel 3 shows near-full intensity spectra, and a sharp transition with no observable metabolites moving to voxel 4 and no overwhelming lipid resonances. No intensity corrections were made to the data; the hyperintensities on the edge of the brain can be attributed to the receive sensitivity profile of the RF coil.

Inner volume selected MRSI in vivo

Figure 8 summarizes the performance of extracranial lipid suppression for the GOIA-ECLIPSE IVS MRSI method. The metabolic maps in Figure 8 B–D show hypo-intensities consistent with lateral ventricles seen in the anatomical image (Figure 8A). The water-NAA CSD and TW at ROI edge voxels was calculated to be 2.5 mm, and 3.1 mm in the y direction, respectively. The low CSD and TW values with this method also result in near-full intensity brain spectra along the ROI edge, and negligible excitation of extracranial lipid resonances in adjacent voxels (Figure 8B–D).Overall residual lipid fraction M_z/M₀ descriptive statistics max/mean/median/standard deviation were 0.0033/0.0049/0.0031/0.005, corresponding to a mean experimental lipid suppression factor of ~ 204-fold. The highly effective extracranial lipid suppression allows artifact-free macromolecular maps to be generated based on spectral integration covering the 0.8 – 1.6 ppm range without baseline corrections. The ECLIPSE ROI coverage in the experimental slice is 96% of the theoretical maximum axial-brain footprint. Figure S3 is a replication of Figure 8 for a different head shape, illustrating similar results.

FIGURE 8.

Summary of in vivo MRSI (TE/TR = 45 ms/2000 ms) with the GOIA-ECLIPSE IVS method. (A) Anatomical axial slice with ECLIPSE ROI and superimposed voxel locations of spectra. The high level of lipid suppression, minimal CSD effects, and near full-intensity spectra is seen at voxel locations adjacent to the ECLIPSE ROI boundary. (B) The anatomical MRI with the ECLIPSE IVS ROI superimposed. (C-E) Metabolic maps for NAA, creatine, and choline calculated by spectral integration. In the metabolic maps, hypo intensities are seen corresponding to the lateral ventricles in the anatomical image with no observable lipid contamination. (E) Macromolecular maps generated by spectral integration over the 0.8 – 1.6 ppm region conforms to the metabolic map profiles, demonstrating negligible extracranial lipid contamination. Note that the macromolecular maps have been scaled down by a factor of 2 to have comparable intensity to the metabolic maps in (C-E). No intensity corrections were made to the data; the hyperintensities seen on the edge of the brain can be attributed to the receive sensitivity profile of the RF coil.

Discussion

In a previous study we demonstrated the utility of ECLIPSE²² for proton MRSI with > 100-fold in extracranial lipid suppression, while requiring only a fraction of the RF power required by conventional eight-slice OVS²³. Common to all gradient-based localization methods, but especially IVS methods, CSD errors distort the true metabolic profile in edge-of-ROI voxels. The CSD can only be reduced by increasing the RF bandwidth, and gradient-modulation of adiabatic RF pulses has previously been shown to increase the RF bandwidth several-fold without a significant increase in RF power deposition^29,34,35. Here we have demonstrated that GOIA-based ECLIPSE for OVS and IVS can provide effective and robust lipid suppression, while simultaneously reducing the CSD and TW metrics to negligible levels. The low CSD and TW effects at edge-of-ROI voxels facilitated the reliable acquisition of MRSI data for a 10 × 10 mm² in plane resolved MRSI grid, with a near-full intensity metabolic profile for edge-of-ROI voxels void of lipid contamination. The RF power deposition is still well within FDA guidelines and should allow direct translation to 7 T.

In-plane localization performance of the presented GOIA-ECLIPSE methods can be compared against novel selectivity optimized IVS and OVS methods reported in the literature for MRSI. CSD errors of LASER localized cubical IVS-MRSI methods^34,36 utilizing 20 kHz bandwidth GOIA pulses, translate to a ~ 3% fraction relative to the ROI size for water-lipid separation at 4 T. In comparison, the ECLIPSE gradient strength used in this work results in a maximum CSD error fraction of 1.5% (or 0.4% per ppm), which is ~ 5-fold lower than the recommended maximum CSD for 2-D MRSI in a recent consensus paper³⁷. The 3.75-fold lower TW fraction with second order gradients relative to first order gradients with GOIA-HS4 pulse, further demonstrates the benefits of ECLIPSE based localization. The improved in-plane selectivity with GOIA-ECLIPSE IVS is achieved in addition to improved axial brain coverage relative to cubical IVS, while consuming significantly less power; since in-plane localization is achieved with only two GOIA pulses, rather than four with a semi-LASER/LASER approach.

The utility of highly selective pulses for multi-slice OVS MRSI has been reported to minimize CSD and improve spatial selectivity along ROI-edge MRSI voxels^11,38. Based on the RF pulse bandwidth reported in these studies, the gradient amplitudes can be calculated to be 116 Hz/mm¹¹, and 150 Hz/mm³⁸. This would translate to a CSD of ~ 5.2 mm and ~ 4 mm for water-lipid separation at 4 T for an OVS slice thickness of 40 mm. In comparison, the maximum CSD of 2.6 mm over a 50 mm ring ROI obtained with GOIA-ECLIPSE-OVS translates to a 2-fold and 1.6-fold improvement in absolute CSD over the above methods. The TW with highly selective OVS pulses³⁸ is reported to be 1.8 mm for a 30 mm slice thickness (or a 6% fractional TW). In comparison, the maximum TW for GOIA-ECLIPSE OVS was 3.1 mm for a 50 mm thick ring ROI (or a 6.3% fractional TW), demonstrating comparable fractional TW metrics to OVS utilizing highly selective pulses. It must be noted that, unlike AFP pulses used in this work (see Figure 3), the fidelity of excitation/inversion profile of highly selective OVS pulses is achieved over a restricted B₁ range. This restricted B₁ range of very selective pulses, combined with overlapping multi-slice lipid suppression, practically limits B₁ tolerance of multi-slice OVS to 20 – 30 %, and a mean lipid suppression factor of 15 – 25 -fold^10,11,23. In summary, GOIA-ECLIPSE OVS provides reduced CSD and comparable fractional TW metrics relative to selectivity optimized multi-slice OVS methods. This is a highly attractive prospect for GOIA-ECLIPSE OVS, given that it requires low SAR, and provides > 100-fold in mean lipid suppression covering a wide B₁ and T₁ span; a combination of benefits that cannot be achieved simultaneously with multi-slice OVS.

The characteristics of the GOIA-HS4 pulse used in this work (BW = 15 kHz, GA = 5, T_p = 6.66 ms, and B_{1 (99.5%)} = 0.76 kHz) was decided based on a few criteria: 1) the time averaged SAR of the GOIA-ECLIPSE IVS method would comfortably allow a TR = 2 s without triggering the in-house time averaged SAR monitor, 2) the GOIA pulse was developed based on a parent AFP-HS4 pulse with T_p = 6.66 ms used in a previous ECLIPSE study²³, to allow power, localization performance, and lipid suppression comparisons based on gradient modulated and unmodulated pulses, 3) minimize off-resonance asymmetric inversion profile effects with GOIA pulses due to a 600 Hz offset based on simulations findings (as shown in Figure 3), and 4) obtain comparable or improved CSD and TW metrics relative to highly selective cubical inner volume selection and multi-slice outer volume suppression methods described in the literature (compared above).

If the peak B_{1 (99.5%)} of 0.76 kHz (17.9 μT) for ECLIPSE-GOIA IVS is a limitation, a higher order HS modulation GOIA pulse can be implemented. Moving forward, a shorter GOIA pulse with similar bandwidth and B_{1 (99.5%)} metrics would be beneficial (provided that the GM function adheres to the system gradients slew rate) to reduce TE of the GOIA-ECLIPSE IVS method. Short duration and high bandwidth can be realized by increasing the exponent order for the HSn modulation used for HS based GOIA pulses. Benefits and shortcomings of moving towards higher order HSn modulation functions for AFP pulses is well documented^24,26, where the RF pulse bandwidth can be increased without reaching prohibitive peak B_{1 (99.5%)}, however with an increased fractional TW. Specific to gradient modulated AFP pulses, increasing the GA factor can also be used to increase bandwidth, which can further benefit from a reduction in fractional TW; however, at the cost of an accentuated asymmetric inversion character for off-resonance spins (see GOIA-HS4 in Figure 4). The WURST³⁹ modulation based GOIA pulse³⁴ (GOIA-WURST) is shown to be an attractive alternative to HSn modulated GOIA pulses, to attain high bandwidth and reduced susceptibility to off-resonance effects³⁰. A shorter GOIA-WURST pulse (T_p ~ 4 ms, BW ~ 15 kHz, and B_{1 (99.5%)} < 0.9 kHz) will be investigated in future work for GOIA-ECLIPSE.

The minimum TE of 3.2 ms and 40 ms for the FID-MRSI and ECLIPSE-OVS methods can be significantly shortened with a modern clinical MR system and shorter RF pulses. FID-MRSI methods with an effective TE of 0.8 ms have been reported⁴⁰ using a Siemens Prisma 3 T MR system (Siemens, Erlangen, Germany), that can readily be implemented for ECLIPSE-OVS with a shorter duration excitation pulse. MRS and MRSI studies utilizing sLASER^30,37 have also been reported on clinical systems with a minimum TE of ~ 30 ms, thus an ECLIPSE-IVS method on a modern clinical system with a TE < 20 ms should be achievable. The minimum TR of 1500 ms with the ECLIPSE-OVS method and TR of 2000 ms with the ECLIPSE-IVS method limits their use in ultra-short TR (TR = 200 ms), high-resolution MRSI acquisitions²⁵. With the ECLIPSE-OVS method, a reoptimized OVS module constrained to a shorter duration (~ 500 ms) can be realized, where a shorter TR may improve the lipid suppression performance due to partial T₁ saturation of lipids. Minimum TR of the ECLIPE-IVS method is limited by time-averaged SAR, that can be partially mitigated with the use of a less power intensive GOIA based double-spin-echo method. It must be noted that accelerated acquisitions with short TR results in undesired T₁ saturation of metabolites that needs to be addressed for absolute quantification. Therefore, our long-term goal is to operate with a TR ≥ 2000 ms at 4T.

The Z2 gradient amplitude utilization was < 40% of its maximum amplitude throughout this work; thus, can accommodate improved CSD and TW at edge-of-ROI brain voxels if required. In its current configuration, the GOIA-HS4 pulse effective BW can be increased by ~ 25% without a SAR penalty, or any adverse effects. This will only incur a ~ 25% increase in ECLIPSE gradient amplitudes, and provide a proportionate improvement in CSD and TW metrics for both GOIA-ECLISE OVS and IVS methods. In addition, the TW associated with the GOIA-HS4 pulse can be reduced by designing asymmetric gradient modulated RF pulses, using the principles of asymmetric AFP pulse design⁴¹.

Effective lipid suppression with GOIA-ECLIPSE OVS is achieved with a minimum peak B₁ amplitude of 0.31 kHz, as illustrated in Figure 4B. Given that full-inversion (B_{1 (99.5%)}) of one GOIA-HS4 pulse is achieved with a B₁ amplitude of 0.76 kHz, the power required by the overall ECLIPSE-OVS method is 44% of the power required by one GOIA-HS4 pulse at B_{1 (99.5%)}. In contrast, the two GOIA-HS4 pulses in the GOIA-ECLIPSE IVS sequence need to be calibrated at B_{1 (99.5%)} for optimal performance, and requires 4.6x the power required by GOIA-ECLIPSE OVS. The lower power requirement of the GOIA-ECLIPSE OVS method can be attributed to being effective, starting from the non-adiabatic regime. However, considering B₁⁺ inhomogeneity in the extracranial region and within brain, power required by the GOIA-ECLIPSE IVS method was reduced to ~ 1.5x higher than that of the GOIA-ECLIPSE OVS method. TR averaged power consumption of the GOIA-ECLIPSE OVS method was 1.9-fold higher than that of the ECLIPSE-OVS method utilizing AFP-HS4 pulses²³, while providing a ~ 5-fold improvement in CSD for edge-of-ROI voxels.

The GOIA-ECLIPSE IVS method provides lipid suppression unconstrained to T₁ and B₁⁺ heterogeneity in comparison to the GOIA-ECLIPSE OVS method. However, the longer minimum TE (TE > 40 ms) and higher RF power deposition compared to GOIA-ECLIPSE OVS can be a limitation in some applications. The GOIA-ECLIPSE OVS method mitigates these shortcomings by providing effective lipid suppression as demonstrated with FID-MRSI acquisitions. Furthermore, the covered lipid T₁ span makes the GOIA-ECLIPSE OVS method suited to cover T₁ species at 1.5 T¹³, 3 T¹⁵, and 7 T¹⁴, based on reported extracranial lipid T₁ and TI values without any modifications to the sequence. The interleaved water suppression scheme for the GOIA-ECLIPSE OVS method is expected to be sufficient for 3 T and 7 T; however, if improved water suppression at 3 T is required for T₁ species < 1000 ms, the optimized 8-pulse VAPOR module¹² coincidently is compatible with the GOIA-ECLIPSE OVS pulse timings reported in this work. Though the presented GOIA-ECLIPSE OVS method is expected to be effective at 3 T and 7 T, the OVS pulse powers and delays can easily be re-optimized to best fit the field strength of operation, in terms of lipid suppression, B₁ inhomogeneity tolerance, and power requirements. The GOIA-ECLIPSE OVS method provided > 100-fold in lipid suppression in the presence of a ± 70% B₁⁺ heterogeneity, comparable to B₁⁺ heterogeneity encountered around the scalp with novel RF coils developed for ≥ 7 T applications^42–45. As a result, the four-pulse GOIA-ECLIPSE OVS method is expected to provide robust lipid suppression for ≥ 7 T applications, while consuming a fraction of the power needed by conventional OVS²³. The incorporation of pre/post gradient pulses was necessary for the ECLIPSE-OVS method to avoid degraded water suppression performance. The negligible B₀ modulations of the overall GOIA-ECLIPSE OVS method allows edited⁴⁶ MRSI acquisitions as was demonstrated in a pilot study for lactate and β-hydroxybutyrate detection in vivo⁴⁷.

The elliptical ROI generated with ECLIPSE fields is ideally suited for axial brain shapes that are near symmetrical around both x and y axes for maximum brain retention, in comparison to head shapes that deviate in symmetry around the x-axis (Figure S3). However, regardless of head shape, effective lipid suppression is achieved with ECLIPSE, and in most situations an elliptical ROI will provide superior axial coverage relative to cubical IVS, and comparable coverage to multi-slice IVS. In this work, the ECLIPSE magnetic fields were limited to a 10 mm slice in the Z-direction. The spatial coverage along the Z-direction can be extended to a ~ 40 mm slab without significantly compromising axial coverage, thereby providing limited 3D coverage. This is comparable to the superior-inferior coverage achieved by traditional IVS methods^36,47,48, for which the brain curvature limits the extent of the cubical IVS box.

The use of spectral fitting would have been appropriate for the quantification of overlapping resonances instead of a spectral integration approach; however, the scope of this work was on describing and the evaluation of lipid suppression with ECLIPSE utilizing GOIA pulses. In comparison to a spectral fitting method, spectral integration is highly sensitive to lipid contamination and baseline shifts due to insufficient water/lipid suppression; thus, for this work integration serves as an effective means to evaluate lipid suppression performance. Moving forward with ECLIPSE for MRSI applications, a spectral fitting approach will be incorporated for the quantification of overlapping metabolites such as Glx, and myo-inositol. Furthermore, the ability to acquire artifact-free macromolecular maps throughout the axial slice (see Figure 8) with GOIA-ECLIPSE IVS provides avenues to study alterations in the neuro-macromolecular profile as reported in numerous pathologies such as stroke⁴⁹, tumors⁵⁰, multiple sclerosis⁵¹, and with aging⁵².

The ECLIPSE ROI planning is performed during a vendor-supplied B₀ mapping sequence. As such, the ROI planning does not require any additional scan time on the part of the volunteer. The planning of the ECLIPSE ROI typically requires ~ 2 minutes, similar to that of cubical volumes, and could be easily implemented into a clinical setting. The XY gradient can be used to account for head rotations, however was not required in this work by ensuring the volunteer head position was straight. The GOIA pulse power for ECLIPSE-OVS and IVS were fixed based on B₁⁺ maps obtained from all volunteers, and was not altered during sessions. Vendor supplied automated shimming and pulse power calibrations were used, thus the presented ECLIPSE based MRSI methods demonstrate to be robust with minimal operator dependencies. Since the ECLIPSE gradient insert and associated current amplifiers are a separate entity from the 4 T system hardware, the ECLIPSE system can be integrated with clinical MR systems contingent on physical dimensions of the magnet bore.

Attaining an x-fold improvement in CSD requires an x-fold improvement in RF pulse bandwidth, which results in an x-fold increase in RF power (for a given RF pulse type). CSD and TW metrics achieved with second order gradient fields were 3.96-fold and 3.75-fold lower, respectively, compared to first order gradient fields (for IVS with GOIA-HS4 pulses) without a SAR penalty. In essence, a GOIA-ECLIPSE IVS MRSI method can be implemented with a > 4-fold reduction in RF power demands, relative to an equivalent semi-LASER MRSI method with identical maximum ROI dimensions and an x-amount of absolute CSD at edge-of-ROI voxels. As a result, ECLIPSE is a highly attractive prospect for MRSI acquisitions moving to 7 T and beyond, in terms of providing reduced SAR, high localization performance (CSD and TW), and effective lipid suppression. Figure S1 was added primarily as an example of ECLIPSE gradient amplitudes required for a typical brain ROI with IVS and OVS. In this example the amplitudes |G_Z2| and |G_X2Y2| were 1.90 Hz/mm² and 0.42 Hz/mm², with maximum amplitudes of 2.20 Hz/mm² and 0.70 Hz/mm² used in this work, respectively. Note that the scaling constants for the spherical harmonic terms are based on the expressions as defined in the Theory section. The absence of a high amplitude second order gradient-insert from MRI vendors, however is a downside that currently limits the widespread use of ECLIPSE. Extending a higher order shim insert to allow pulsing at sufficient amplitude, may require additional shielding to minimize potential eddy current effects. Shim pre-emphasis has however been demonstrated to be an effective method in mitigating effects of higher-order shim eddy currents⁵³. Fortunately, actively shielded Z2 shim coils are commercially available through third-party vendors, and can thereby open the path towards wide-spread use of ECLIPSE.

Conclusions

Two extracranial lipid suppression methods based on elliptical localization with pulsed second order fields (ECLIPSE) were presented. Both methods were executed with gradient offset-independent adiabaticity (GOIA) pulses to minimize CSD and TW at edge-of-ROI voxels. Both methods provide > 100-fold in lipid suppression and high axial slice brain coverage (> 89%). The GOIA-ECLIPSE OVS method is characterized by low RF power deposition, effective lipid suppression within a range of B₁ and T₁ heterogeneity, and can be combined with ultra-short-TE acquisitions. The GOIA-ECLIPSE IVS method is characterized by B₁ and T₁-independent lipid suppression, however it is limited to medium to long echo-times. The low power requirements in combination with insensitivity to B₁ and T₁ variation encountered in vivo, makes both ECLIPSE-OVS and IVS methods attractive for ≥ 3 T MRSI applications.

Supplementary Material

Supplementary Figure 1

FIGURE S1. Illustrates example ROI’s that can be generated with ECLIPSE gradient fields; namely ECLIPSE based elliptical volume selection, ring selection, and translated ring selection. The ECLIPSE ROI is 130 mm × 165 mm in diameter, which is identical to the ROI used for the volunteer in Figure 8. (A) The two rows illustrate 2-D spatial magnetization plots (M_z/M₀) for AFP-HS4 (top row) and GOIA-HS4 (bottom row) as a result of the amplitude modulation (AM), frequency modulation (FM), and gradient modulation (GM) functions specified in the corresponding columns in (C). Elliptical volume selection due to a 0 Hz offset (i)-(ii), and 600 Hz offset (iii)-(iv) for the two pulse types. The ROI in (iii) is 30% smaller than that of (i); in comparison the ROI in (iv) is 6% smaller than that of (ii). A faint (lighter blue) ring is visible along the outer edge of the ellipse in (iv) as a result of off-resonance effects of the GOIA-HS4 pulse as described in Figure 3. (v)-(viii) illustrate 0 Hz and 600 Hz offset cases similar to (i)-(iv) for ring selection. To select an elliptical ring ROI, an RF-offset is given corresponding to the RF pulse bandwidth; all other parameters remain fixed relative to the elliptical selection case. (ix)-(xii) is identical to the ring selection plots in (v)-(viii) with the addition of translation in X and Y directions with the application of first order gradients G_X and G_Y, as demonstrated in (C). (B) 1-D profiles of the regions marked in (i)-(xii) in (A). The asymmetric inversion profile with the GOIA-HS4 pulse and pronounced horizontal displacement for AFP-HS4 pulse are evident in all three 1-D plots due to the 600 Hz frequency offset. The sharper transition width on the inner edge of the elliptical ring ROI relative to the outer edge ROI is also evident in all plots of (b) and (c).

Supplementary Figure 2

FIGURE S2. 1-D inversion profiles obtained with the 15 kHz bandwidth GOIA-HS4 pulse with first and second order gradients. The first and second order gradient field amplitudes were 88.2 Hz/mm and 1.92 Hz/mm², respectively, selective for a 170 mm length ROI. TW for first and second order localization is 10.9 mm, and 2.9 mm, respectively. The improved inversion profile with second order localization is further emphasized in the green panel spanning - 100 to - 70 mm.

Supplementary Figure 3

FIGURE S3. Summary of in vivo MRSI (TE/TR = 45 ms/2000 ms) with the GOIA-ECLIPSE IVS method on another volunteer, similar to Figure 8. (A) The anatomical axial slice with the IVS ROI superimposed, and voxel locations for illustrated spectra. (B-D) Metabolic maps for NAA, creatine, and choline from spectral integration demonstrate effective extracranial lipid suppression. (E) Macromolecular maps generated by spectral integration over the 0.8 – 1.6 ppm region. Note that the macromolecular maps have been scaled down by a factor of 2 to have comparable intensity to the metabolic maps in (C-E). No intensity corrections were made to the data.

Abbreviations:

AFP

adiabatic full passage

BW

bandwidth

CSD

chemical shift displacement

ECLIPSE

elliptical localization with pulsed second order fields

GA

gradient amplification factor

GOIA

gradient offset-independent adiabaticity

HS

hyperbolic secant

IVS

inner volume selection

OVS

outer volume suppression

ROI

region of interest

TW

transition width