Development of a spiral spin- and gradient-echo (spiral-SAGE) approach for improved multi-parametric dynamic contrast neuroimaging

Abstract

Purpose:

The purpose of this study was to develop a spiral-based combined spin- and gradient-echo (spiral-SAGE) method for simultaneous dynamic contrast-enhanced (DCE-MRI) and dynamic susceptibility contrast MRI (DSC-MRI).

Methods:

Using this sequence, we obtained gradient-echo TEs of 1.69 and 26 ms, a SE TE of 87.72 ms, with a TR of 1663 ms. Using an iterative SENSE reconstruction followed by deblurring, spiral-induced image artifacts were minimized. Healthy volunteer images are shown to demonstrate image quality using the optimized reconstruction, as well as for comparison with EPI-based SAGE. A bioreactor phantom was used to compare dynamic-contrast time courses with both spiral-SAGE and EPI-SAGE. A proof-of-concept cohort of patients with brain tumors shows the range of hemodynamic maps available using spiral-SAGE.

Results:

Comparison of spiral-SAGE images with conventional EPI-SAGE images illustrates substantial reductions of image distortion and artifactual image intensity variations. Bioreactor phantom data show similar dynamic contrast time courses between standard EPI-SAGE and spiral-SAGE for the second and third echoes, whereas first-echo data show improvements in quantifying T₁ changes with shorter echo times. In a cohort of patients with brain tumors, spiral-SAGE-based perfusion and permeability maps are shown with comparison with the standard single-echo EPI perfusion map.

Conclusion:

Spiral-SAGE provides a substantial improvement for the assessment of perfusion and permeability by mitigating artifacts typically encountered with EPI and by providing a shorter echo time for improved characterization of permeability. Spiral-SAGE enables quantification of perfusion, permeability, and vessel architectural parameters, as demonstrated in brain tumors.

1 |. INTRODUCTION

Intracranial neoplasms are associated with a multitude of hemodynamic changes, including increased cerebral blood flow (CBF) and volume (CBV), altered vessel size and vascular architecture, and increased permeability of blood vessels. ¹ Dynamic susceptibility contrast (DSC-) and dynamic contrast-enhanced (DCE-) MRI methods are often applied either alone or in tandem to assess complementary vascular characteristics in these patients, specifically perfusion and permeability metrics, respectively. Standard DSC methods generally leverage gradient-echo (GRE) EPI readouts to provide high ${\mathrm{T}}_2^{\mathrm{*}}$ sensitivity, yielding measures of $\mathrm{C}\mathrm{B}\mathrm{V}$ and $\mathrm{C}\mathrm{B}\mathrm{F}$ that reflect vessels of all sizes. DSC methods that include spin-echo (SE) EPI readouts, which are differentially sensitive to capillary-sized vessels,^2,3 can inform on microvascular CBV and CBF. Because of this differing vascular sensitivity, methods that combine both GRE and SE signals can quantify parameters related to vessel size^4–6 and architecture.^7,8 Additionally, DSC methods that include multiple GREs permit the simultaneous estimation of both DSC perfusion and DCE permeability metrics.^9–11 The inclusion of multiple echoes provides high relative CBV (rCBV) accuracy, with the potential to further optimize pulse-sequence parameters for higher T₁ sensitivity.¹² The advent of multiecho and multicontrast methods, which include both SE and (multiple) GRE (SAGE) acquisitions,^13–17 permits the estimation of total and microvascular perfusion, vascular architecture, and permeability, thus providing a comprehensive hemodynamic assessment.

Prior implementations of the SAGE sequence leveraged multiple EPI acquisitions, which provide high temporal resolution (<2 s) with moderate spatial resolution (2–3 mm in-plane, 4–5 mm through-plane).^13–17 However, EPI-based readouts are associated with image distortion, signal pileups, and susceptibility-induced signal dropout near air–tissue interfaces. In addition, as the readout trajectory starts at the edge of k-space, the first EPI-SAGE TE is typically longer (>5 ms) than recommended for DCE-MRI,¹⁸ which reduces T₁ sensitivity and may negatively bias the resulting permeability metrics.¹⁹ Spiral-based readouts circumvent both of these issues, mitigating image distortion via spiral deblurring and providing shorter first TEs with a spiral-out trajectory.¹¹ Other possible advantages of spiral acquisitions include decreased motion sensitivity, shorter readouts based on more efficient use of the gradient hardware, and better conditioning for high undersampling in combination with compressed sensing reconstruction.^20–22 Spiral perfusion imaging with consecutive (GREs) echoes (spiral perfusion imaging with consecutive echoes [SPICE]¹¹ enables simultaneous DSC- and DCE-MRI analysis, while removing the confounding relaxation effects that reduce the reliability of perfusion and permeability metrics. Although this method enables quantification of total perfusion and permeability, the lack of a SE precludes analysis of microvascular perfusion and vessel architectural (VAI) parameters. Notably, the VAI framework provides insight into the relationship between vascular volume and flow, vessel size, and oxygenation across the vascular network.^7,8

The purpose of this study was to develop a spiral-based SAGE method that includes dual GRE and SE readouts, consistent with the simplified EPI-SAGE method previously developed.^16,23 We implemented dual spiral-out trajectories for the GRE acquisitions and a spiral-in trajectory for the SE acquisition, thus maximizing temporal efficiency. An advanced spiral-SAGE reconstruction scheme was developed and optimized, and spiral-SAGE images were compared with standard EPI-based SAGE images at baseline in a healthy volunteer to assess image quality. Subsequently, a bioreactor phantom was used to compare spiral-SAGE with EPI-SAGE with consecutive injections of gadolinium (Gd)-based contrast agent. Finally, spiral-SAGE data were acquired in a cohort of patients with brain tumors, which were then compared with standard single-echo DSC-MRI. The comprehensive assessment of brain tumor hemodynamics and vascular architectural parameters enabled by spiral-SAGE is also shown.

2 |. METHODS

2.1 |. Spiral-SAGE acquisition and reconstruction

The spiral-SAGE sequence shown in Figure 1 was developed on the Philips 3T platform. The GRE trajectories are both spiral-out, whereas the SE trajectory is spiral-in. As such, the center of k-space occurs at the beginning of the acquisition window for both gradient echoes and at the end of the acquisition window for the SE. The spiral readout length $\left(\tau \right)$ was 15.1 ms, and variable density undersampling was employed with the center 10% of k-space fully sampled and a maximum undersampling factor of 2.5. The spiral-SAGE sequence provided the following imaging parameters: TEs = 1.69,26.00, and 87.72 ms, TR = 1663 ms, FOV = 240 × 240 mm², 15 slices, and voxel size = 3.158 × 3.158 × 5 mm³. Because of slightly differing hardware configurations between scanners, spiral-SAGE data were acquired in the clinic with a shorter readout length (14.6 ms), which resulted in small variations in TEs (2.00, 21.00, and 80.00) and TR (1497 ms). All pulse sequence parameters are shown in Table 1. (The compiled spiral-SAGE patch is available upon request [Philips R5.3]).

FIGURE 1.

Spiral-based combined spin and gradient echo (spiral-SAGE) (top) and EPI-SAGE (bottom) pulse sequences. The spiral trajectories are spiral-out for both gradient echoes and spiral-in for the SE. The spiral-SAGE sequence provides a shorter first TE with increased T₁-weighting for dynamic contrast-enhanced MRI (DCE-MRI), whereas the second and SE TEs can be optimized for ${\mathrm{T}}_2^{\mathrm{*}}$ and ${\mathrm{T}}_2$ contrast, respectively

TABLE 1.

Comparison of pulse-sequence parameters for EPI-SAGE versus spiral-SAGE

Parameters	EPI-SAGE	Spiral-SAGE	Spiral-SAGE-clinic
FOV	240 × 240 mm2	240 × 240 mm2	240 × 240 mm2
Voxel size, mm2	3.158 × 3.158	3.158 × 3.158	3.158 × 3.158
Slice thickness	5	5	5
Number of slices	15	15	15
TEs	8.0 ms	1.7 ms	2.0 ms
	26.0 ms	26.0 ms	21.0 ms
	79.6 ms	87.8 ms	80.0 ms
TR	1800 ms	1663 ms	1497 ms
Flip angle (EXC-RFC)	90–180	90–180	90–180
Readout length $\left(\tau \right)$	17.3 ms	15.1 ms	14.4 ms
Sense acceleration	2	-	-
Partial Fourier undersampling	0.74	-	-
Variable density undersampling		Center 10% fully sampled; Max 2.5x undersampling	Center 10% fully sampled; Max 2.5x undersampling

Note: The first set of spiral-SAGE parameters were used for the bioreactor and healthy volunteer studies, whereas the spiral-SAGE–clinic version was used for patient studies.

Abbreviations: SAGE, spin and gradient echo; EXC, excitation; RFC, refocusing.

Spiral data were reconstructed off-line in a prototyping environment called graphical programming interface (GPI)²⁴ with nodes for gridding and reconstruction,²⁵ deblurring²⁶ and dynamic ${\mathrm{B}}_0$ drift correction,²⁷ as shown in Supporting Information Figure S1. SENSE reconstruction can be used to combine undersampled data by leveraging the local sensitivities of multichannel-phased array coils. The drawbacks of SENSE include SNR losses as a function of both total acceleration factor and regional coil sensitivities. A previously described iterative SENSE algorithm^25,28 reduces the effect of local coil sensitivities by modifying the aliasing point-spread function, thus reducing the geometrical variations in SNR (also known as g-factors). Gridding reconstruction from the undersampled spiral-SAGE acquisition was performed using this iterative SENSE algorithm,²⁸ with sampling density weights calculated a priori from the spiral k-space coordinates.^25,29 Receiver coil-sensitivity maps were generated on the scanner by the vendor’s software. Further improvement of this algorithm is obtained by Tikhonov regularization based on a tissue probability map determined from the coil-sensitivity prescan, where the regularization coefficient was empirically determined in a previous study.²⁵

Although spiral trajectories offer high readout efficiency, spiral acquisitions are sensitive to off-resonance effects, which manifest as blurring caused by both chemical shift and inhomogeneous ${\mathrm{B}}_0$. With the acquisition of a separate ${\mathrm{B}}_0$ map, accurate deblurring can be performed iteratively using a convolution-based conjugate-gradient method in the image domain.^26,30,31 This approach accounts for both phase accumulation and blurring simultaneously and was previously shown to perform particularly well in regions with large ${\mathrm{B}}_0$ variations.²⁶ Following SENSE reconstruction, spiral-SAGE images were deblurred using this deblurring algorithm and a ${\mathrm{B}}_0$ map generated from a separate Dixon acquisition.²⁶ A convergence threshold of 5% was used to reduce the computational time.

A related issue is that ${\mathrm{B}}_0$ maps acquired before the dynamic scan may not accurately account for the temporal evolution of ${\mathrm{B}}_0$. Unfortunately, even small errors in the ${\mathrm{B}}_0$ map can yield suboptimal deblurring. Previous studies have shown that ${\mathrm{B}}_0$ drift is on the order of 2.2 Hz/min for a high gradient load²⁷; as such, ${\mathrm{B}}_0$ drift correction is particularly important for long scan durations (5 min herein) and heavy gradient load applications. In this study, residual blurring artifacts persisted in the deblurred images at later time points because of ${\mathrm{B}}_0$ drift. To compensate for this effect over the long perfusion scan times, dynamic ${\mathrm{B}}_0$ drift correction²⁷ was performed by interpolation between the first and last images of the dynamic time series. More specifically, the ${\mathrm{B}}_0$ maps were temporally demodulated using ${\mathrm{B}}_{0corr}\left(\mathrm{t}\right)={\mathrm{B}}_{0dynamic-1}+\mathrm{\Delta }{\mathrm{B}}_{0drift}\left(\mathrm{t}\right)$, where $\mathrm{\Delta }{\mathrm{B}}_{0drift}\left(\mathrm{t}\right)$ represents the dynamic changes in ${\mathrm{B}}_0$ caused by drift. These dynamic drift-corrected ${\mathrm{B}}_0$ maps were then input into the conjugate-gradient deblurring technique described above. The critical nodes of this network were previously implemented and can be found in Anderson III et al,²⁵ Wang et al,²⁶ and Ooi et al.²⁷

Data were acquired at 3T in healthy volunteers (non-contrast only), a bioreactor perfusion phantom, and patients with brain tumors $(\mathrm{n}=8)$. This study was approved by the Barrow Neurological Institution; healthy subjects provided written informed consent. Patient scans were completed with a waiver of consent as part of their standard-of-care examination. For the patients with brain tumors (six male, two female), the mean age ± SD was $58.5\pm 9.8$ years. Tumor types included glioblastoma multiforme (grade IV; $\mathrm{n}=4$), anaplastic astrocytoma (grade III; $\mathrm{n}=2$), oligodendroglioma (grade II; $\mathrm{n}=1)$, and a secondary metastatic brain tumor $(\mathrm{n}=1)$.

Acquisition parameters for spiral-SAGE are listed above (where patient data were acquired with the clinic version of spiral-SAGE), and parameters for EPI-SAGE were TE = 7.96, 26.02, and 79.61 ms; TR = 1800 ms; FOV = 240 × 240 mm²; 15 slices; and voxel size = 3.158 × 3.158 × 5 mm³. For the bioreactor, deionized water was pumped through the bioreactor using a flow rate of 700 mL/min with a Cole-Palmer Masterflex peristaltic pump (model EW-77921-75); in addition, to ensure a continuous flow, a Cole-Palmer pulse dampener (model EW-07596-20) was used. Using a power injector, consecutive injections of Gd-based contrast agent (4 mL of Gadavist, diluted to 0.25M) were injected at 3.5 mL/s, followed by saline flush. In patients $(\mathrm{n}=8)$, spiral-SAGE data were acquired during a preload injection of Gadavist (standard dose: 0.1mmol/kg) using the optimized spiral-SAGE-clinic parameters (Table 1; 220 dynamics, 5.5-min acquisition time). Six minutes after the preload injection, the main perfusion injection (Gadavist, standard dose: 0.1mmol/kg) was injected during a standard single-echo DSC-MRI acquisition (TE=30 ms,TR=1.4 s, flip angle $\left[\mathrm{F}\mathrm{A}\right]={75}^{\circ },1.75\times 1.75\mathrm{m}\mathrm{m}$ in-plane, 5-mm-slice thickness, 100 dynamics, 2.3-min acquisition time).

2.2 |. Data analysis

Spiral-SAGE data were reconstructed in GPI using the optimized algorithm, as described above, before perfusion analysis. Perfusion analysis was performed using in-house MATLAB software (MathWorks). For SAGE (spiral and EPI), GRE $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{SE}\mathrm{\Delta }{\mathrm{R}}_2$ were calculated using the simplified SAGE method^16,23 in Equations (1) and (2), respectively:

$$\mathrm{\Delta }{R}_{2,GRE}^{\mathrm{*}}(t)=\frac{1}{\left(T{E}_2-T{E}_1\right)}\left(\mathrm{l}\mathrm{n}\left(\frac{S_{TE1}(t)}{S_{TE1}\left(0\right)}\right)-\mathrm{l}\mathrm{n}\left(\frac{S_{TE2}(t)}{S_{TE2}\left(0\right)}\right)\right)$$ (1)

$$\mathrm{\Delta }{R}_{2,SE}(t)=\frac{1}{T{E}_5}\left(\mathrm{l}\mathrm{n}\left(\frac{S_{TE=0}(t)}{S_{TE=0}\left(0\right)}\right)-\mathrm{l}\mathrm{n}\left(\frac{S_{TE,SE}(t)}{S_{TE,SE}\left(0\right)}\right)\right)$$ (2)

where $S(t)$ and $S\left(0\right)$ denote the dynamic signal and prebolus (baseline) signals, respectively, for each echo time, and $S_{TE=0}$ is the signal extrapolated to $\mathrm{T}\mathrm{E}=0$ using the dual-GRE signals, as follows:

$$S_{TE=0}=S_{TE1}{\left(\frac{S_{TE1}}{S_{TE2}}\right)}^{TE1/(TE2-TE1)}$$ (3)

These dynamic time ceourses represent the separate contributions from ${\mathrm{T}}_2^{\mathrm{*}},{\mathrm{T}}_2$, and ${\mathrm{T}}_1$, respectively.

For the patients with brain tumors, the single-echo $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ was calculated using Equation (4):

$$\mathrm{\Delta }{R}_2^{\mathrm{*}}(t)=\frac{-\mathrm{l}\mathrm{n}\left(\frac{S(t)}{S\left(0\right)}\right)}{TE}$$ (4)

In these patients, leakage correction was performed on both the spiral-SAGE and standard DSC-EPI data using the standard Boxerman-Schmainda-Weisskoff (BSW) method,³² modified to account for both ${\mathrm{T}}_1$ and ${\mathrm{T}}_2^{\mathrm{*}}$ leakage effects.³³ The $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$-based arterial input function (AIF) was determined using previously published automated methods with specific multiecho³⁴ and single-echo³⁵ criteria; subsequently, the AIF and tissue $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ were converted to concentration using quadratic and linear relaxivity relationships, respectively.^36,37 CBV was determined using integration of the dynamic $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ curve up to 2 min (up to 60 s each of baseline and postinjection), and negative values of $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ were not included in the integration. CBV was normalized to normal-appearing white matter (NAWM), yielding rCBV. Cerebral blood flow (CBF) was determined from the maximum of the impulse response function obtained from circular singular value decomposition (cSVD) of the input AIF with the tissue $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$,³⁸ using an adaptive threshold.³⁹ CBF was then calibrated to a mean NAWM CBF of 22 mL/100 g tissue/min.¹¹

For DCE-MRI, a precontrast ${\mathrm{T}}_1$ map was acquired using a variable FA approach (10 FA equally distributed between 2° and 20°; $\mathrm{T}\mathrm{R}=7.590\mathrm{m}\mathrm{s}$). Combining the precontrast ${\mathrm{T}}_1$ map $\left({\mathrm{T}}_{\mathrm{1,0}}\right)$ and the signal extrapolated to $\mathrm{T}\mathrm{E}=0$ [Equation (3)], dynamic $R_1$ maps were obtained:

$$R_1(t)=\frac{1}{TR}\left(\mathrm{l}\mathrm{n}\left(\frac{S_{TE=0}\left(0\right)}{(1-m)\cdot \left(\frac{S_{TE=0}\left(0\right)}{(1-m)}-S_{TE=0}(t)\right)}\right)\right)$$ (5)

where $m=\mathrm{e}\mathrm{x}\mathrm{p}\left(-TR/T_{1,0}\right)$. DCE analysis⁴⁰ was then performed using the extended Toft’s model⁴¹ to extract the volume-transfer coefficient $K^{trans}$, the extravascular volume fraction $v_e$, and the vascular volume fraction $v_p$. Measures of vessel size index (VSI) were obtained from a linear fit of $\mathrm{\Delta }{\mathrm{R}}_2^{3/2}$ versus $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ during bolus passage, with a scaling factor that includes ADC and blood volume fraction (from GRE rCBV).⁴⁵ ADC maps were computed from diffusion data acquired in three orthogonal directions using two b-values (0 and 1000 s/mm²). VAI parameters included vortex area and vortex direction from temporal analysis of $\mathrm{\Delta }{\mathrm{R}}_2$ vs $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ during bolus passage, as previously described.^7,8

2.3 |. Statistical analysis

For the bioreactor data, regions of interest (ROIs) covering the entire bioreactor were drawn separately for the spiral- and EPI-SAGE data because of slight shifts in the location between injections (the bioreactor was removed to fully flush out all residual contrast agent between injections). $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ between spiral- and EPI-SAGE were compared across all voxels. For $\mathrm{\Delta }{\mathrm{R}}_1$, only voxels with purely real values across all dynamics were used; that is, complex-valued voxels, which are thought to be associated with concentrations above the detectable limit,¹⁹ were removed from both spiral and EPI-SAGE $\mathrm{\Delta }{\mathrm{R}}_1$. For the patient data, ROIs for tumors were drawn by the first author (AMS), who has nine years of experience. ROIs for NAWM were segmented using a ${\mathrm{T}}_1$-weighted anatomical image with the Functional Magnetic Resonance Imaging of the Brain Software Library (FSL) fast automated segmentation tool (FAST).⁴² These images were coregistered to the perfusion maps using FSL-FLIRT.⁴²

Correlation analyses were performed in MATLAB using linear regression between standard EPI-$\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and spiral-$\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ rCBV maps. In tumor ROIs, concordance correlation coefficient (CCC) and Pearson’s correlation coefficient $\left(r\right)$ were used to assess agreement and linearity, respectively. Correlation coefficients were interpreted as weak (<0.39), moderate (0.40 to 0.59), strong (0.60 to 0.79), and very strong (>0.80).

3 |. RESULTS

The spiral SAGE data were reconstructed using the post-processing pipeline illustrated in Supporting Information Figure S1. On a computer with a 3.1-GHz quad-core processor with 16 GB of RAM, the reconstruction time for a single dynamic was 20.15±10.28 s; thus, reconstruction of all 200 dynamics required approximately 67 min. The computational time for each step of the pipeline is also provided in Supporting Information Figure S1. The effect of parallel imaging reconstruction using SENSE²⁸ and spiral deblurring using the conjugate-gradient (CG)method²⁶ are shown in Figure 2 for each echo of the spiral-SAGE acquisition. Initially, undersampling artifacts overwhelm the images, which are largely mitigated using a SENSE reconstruction node. Blurring artifacts similarly degrade the image quality but are minimized using CG deblurring. The effect of dynamic ${\mathrm{B}}_0$ corrections²⁷ is shown in Figure 3 for each echo of the spiral-SAGE acquisition, where the importance of dynamic ${\mathrm{B}}_0$ correction can be appreciated by comparing the first and last dynamic. Further quantification of these improvements in SNR and deblurring can be found in Supporting Information Figure S2.

FIGURE 2.

Representative example of SENSE reconstruction and deblurring on each spiral-based combined spin and gradient echo (spiral-SAGE), shown for the first dynamic of the time course. Compared with the initial reconstruction (top row), the addition of SENSE (middle row) and conjugate-gradient (CG) deblurring (bottom row) in the reconstruction pipeline yields substantially reduced artifacts across the images

FIGURE 3.

Representative example comparing the first and last dynamic with dynamic ${\mathrm{B}}_0$ correction for each spiral-based combined spin and gradient echo (spiral-SAGE), where the acquisition lasts for 5 min. Compared with reconstruction without ${\mathrm{B}}_0$ correction (top row), dynamic ${\mathrm{B}}_0$ correction (bottom row) shows substantially reduced artifacts across the dynamic timeframe, as seen in the last dynamic image for each spiral-SAGE. CG, conjugate-gradient

The SAGE images for all three echoes using spiral and EPI readouts are shown in Figure 4 for two slices in a healthy volunteer. The inferior slice (Figure 4A) shows signal voids and pileups on the echo images for EPI, particularly near air-tissue interfaces, as shown by the white arrows. The corresponding ${\mathrm{R}}_2^{\mathrm{*}},{\mathrm{R}}_2$, and ${\mathrm{S}}_{\mathrm{T}\mathrm{E}=0}$ maps in Figure 5 show that these artifacts lead to spatially inconsistent values, particularly in the left temporal lobe. These effects and associated artifacts are largely mitigated with spiral trajectories. The superior slice (Figures 4 B and 5 B) is also impacted by image distortion and intensity variations with EPI, particularly in the anterior region, which are improved by the use of spiral readouts.

FIGURE 4.

Example images for EPI-spin and gradient echo (SAGE) (top) and spiral-SAGE (bottom) for two slices (A and B), along with a corresponding ${\mathrm{T}}_1$-weighted (T1w) image. The image distortion and intensity variations associated with EPI readouts are abated using spiral readouts

FIGURE 5.

Calculated images for signal extrapolated to $\mathrm{T}\mathrm{E}=0\left({\mathrm{S}}_{\mathrm{T}\mathrm{E}=0}\right),{\mathrm{R}}_2^{\mathrm{*}}$, and ${\mathrm{R}}_2$ for slices A and B using EPI-spin and gradient echo (SAGE) (top) and spiral-SAGE (bottom), corresponding to the echo images from Figure 4. Both ${\mathrm{S}}_{\mathrm{T}\mathrm{E}=0}$ and ${\mathrm{R}}_2$ are more spatially consistent using spiral-SAGE compared with EPI-SAGE. A.U., arbitrary units

The bioreactor phantom is shown in Figure 6A,B, where the latter shows a cross-sectional slice with a representative ROI shown in green. Dynamic SAGE data using spiral and EPI readouts in this bioreactor phantom with consecutive Gd-based contrast injections are shown in Figure 6C–F. Spiral and EPI readouts produced similar dynamic time courses between injections across all echoes (Figure 6C), whereas $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ (Figure 6D,E, respectively) were consistent between spiral and EPI readouts. A double-peak in the bioreactor data can be seen in Figure 6C–F, most notably in Figure 6D; this likely occurred because of signal saturation, which would confound hemodynamic analysis but is not typically observed at the doses used in vivo. Spiral $\mathrm{\Delta }{\mathrm{R}}_1$ (Figure 6F) has higher ${\mathrm{T}}_1$ sensitivity, leading to higher measurable contrast agent concentrations. More specifically, the maximum observed contrast agent concentrations are 1.1 and 0.41 with spiral and EPI, respectively.

FIGURE 6.

(A) Photograph of the bioreactor phantom. (B) Cross-sectional image of the bioreactor phantom, where a representative ROI is shown in green. (C) Dynamic signal time courses in the bioreactor using spiral-spin and gradient echo (SAGE) and EPI-SAGE (solid and dotted lines, respectively), where TE1 and TE2 reflect the gradient-echo signals and SE is the SE signal. (D-F) Dynamic $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}},\mathrm{\Delta }{\mathrm{R}}_2$, and $\mathrm{\Delta }{\mathrm{R}}_1$, respectively, for spiral-SAGE (purple) and EPI-SAGE (green). The signals $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ time courses are similar between spiral- and EPI-SAGE, whereas the higher $\mathrm{\Delta }{\mathrm{R}}_1$ for spiral-SAGE reflects higher ${\mathrm{T}}_1$-weighting

Dynamic DSC time-courses in a patient with a high-grade glioma are shown in Figure 7. ${\mathrm{T}}_1$ leakage effects are evident in all spiral echoes (tumor ROI); most notably, no ${\mathrm{T}}_2^{\mathrm{*}}$-induced signal drop is observed in the first echo because of the short TE. The subsequent $\mathrm{\Delta }{\mathrm{R}}_1$ and $\mathrm{\Delta }{\mathrm{R}}_2$ curves represent readouts that are unavailable using a standard single-echo acquisition, reflecting permeability and microvascular perfusion, respectively. The $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ curves show high similarity between spiral and EPI acquisitions in tumor, NAWM, and the AIF.

FIGURE 7.

Dynamic time courses for spiral-SAGE (first injection) and standard EPI (single-echo, second injection) in a patient with high-grade glioma. (Top) Spiral-SAGE echo 1–3 (red) and EPI single-echo (blue) are shown in tumor and normal-appearing white matter (NAWM) ROIs (solid and dotted lines, respectively). The vertical dotted lines for the spiral echoes indicate the analogous timespan for the EPI acquisition, which was only acquired for 150 s. (Bottom) Dynamic spiral $\mathrm{\Delta }{\mathrm{R}}_1$, spiral and EPI $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$, spiral $\mathrm{\Delta }{\mathrm{R}}_2$ are shown in tumor and NAWM ROIs, as well as the corresponding spiral and EPI $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$-based AIFs. $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ curves are corrected for ${\mathrm{T}}_1$ leakage using the simplified SAGE approach,^16,23 followed by leakage correction for residual ${\mathrm{T}}_2^{\mathrm{*}}$ effects using the Boxerman-Schmainda-Weisskoff method⁴⁶

Integration of the $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ produces rCBV maps, where the GRE $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and standard single-echo $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ are both sensitive to vessels of all sizes, as shown in Figure 8. Both spiral-based and standard rCBV maps show similar features, including a rim of elevated rCBV (indicated by arrow) adjacent to a previous tumor resection. The voxel-wise comparison between rCBV maps shows a strong linear relationship $(\mathrm{R}=0.74;\mathrm{C}\mathrm{C}\mathrm{C}=0.72$), with near-unity slope (0.95). Comparisons of spiral-$\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}\mathrm{r}\mathrm{C}\mathrm{B}\mathrm{V}$, and standard single-echo rCBV across all subjects $(\mathrm{n}=8$) show a similar trend: a strong linear relationship $(\mathrm{R}=0.77$ and $\mathrm{C}\mathrm{C}\mathrm{C}=0.73)$ and near-unity slope (0.99). Comparisons of spiral-SAGE CBF and standard single-echo CBF show similar hemodynamic characteristics in this tumor and across the brain. A strong linear relationship ($\mathrm{R}=0.74$ voxel-wise; 0.85 group-wise) was observed between spiral and standard CBF, though with lower CCCs (0.50 and 0.74 for voxel-wise and group-wise, respectively). Across most subjects, the mean spiral CBF was higher than the standard CBF.

FIGURE 8.

Comparison of spiral-spin and gradient echo (SAGE) and EPI $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$-based relative cerebral blood volume (rCBV) and cerebral blood flow (CBF). (Top) ${\mathrm{T}}_1$-weighted $\left({\mathrm{T}}_1\mathrm{w}\right.$) postcontrast image (tumor outlined in magenta and indicated by the arrow), spiral-based global rCBV, standard single-echo rCBV, spiral-based global CBF, and standard single-echo CBF. (Bottom) Voxel-wise comparison of rCBV and CBF values between spiral and EPI in the tumor ROI, along with the group-wise comparison of mean tumor rCBV and CBF values across all subjects ($\mathrm{n}=8$). The $△{\mathrm{R}}_2^{\mathrm{*}}$-based rCBV and CBF are similar between spiral-SAGE and EPI-SAGE both within a single subject and across subjects

Beyond measures of global rCBV, Figure 9 shows a representative example of the wide array of vascular parameters that can be obtained using spiral-SAGE; more specifically, global (from $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$) and microvascular (from $\mathrm{\Delta }{\mathrm{R}}_2$) measures of rCBV and CBF, as well as $K^{trans}$, VSI, VAI_area, and VAI_direction, can all be quantified from a single injection and single scan. In the tumor ROI, outlined in magenta, elevated rCBV and CBF can be discerned, as well as higher permeability $\left(K^{trans}\right)$ and vessel size (VSI). VAI_area, a parameter that may relate to oxygen extraction, is isointense with surrounding tissue, whereas VAI_direction shows a mix of clockwise (red) and counterclockwise (blue) vortex directions. The summary of these parameters across all subjects is shown in Supporting Information Figure S3.

FIGURE 9.

Spiral-spin– and gradient–echo (SAGE) maps in an enhancing brain tumor (glioblastoma, outlined in magenta), showing the range of parameters available using spiral-SAGE. A ${\mathrm{T}}_1$-weighted $\left({\mathrm{T}}_1\mathrm{w}\right)$ postcontrast image is shown, along with spiral-SAGE-based maps of global and microvascular relative cerebral blood volume (rCBV) and cerebral blood flow (CBF), permeability $\left(K^{trans}\right)$, and vessel architectural parameters including vessel size (VSI), vortex area VAI_area, and vortex direction VAI_direction

4 |. DISCUSSION

DSC perfusion is increasingly used in the management of patients with brain tumors, covering all aspects of clinical care from diagnosis and monitoring to assessing treatment. However, standard DSC methods yield a single-echo, and thus only a limited set of hemodynamic metrics can be quantified. In principle, more information can be gleaned about the tumor vasculature by acquiring additional scans using either DCE to quantify blood-brain barrier permeability or SE DSC to assess complementary microvascular hemodynamic metrics. Unfortunately, these approaches increase both contrast agent dose and scan time. More advanced methods leverage multiple GREs or combine GRE with SE readouts, and the most comprehensive of these is the combined SAGE method. Previous studies have investigated SAGE perfusion in brain tumors,^13,14,16,23 ischemia,^15,17 multiple sclerosis,⁴³ and Alzheimer disease⁴⁴ suggesting the improved informational content of SAGE-based approaches may find widespread applications across neuropathologies. However, EPI-SAGE suffers from signal distortion associated with EPI readouts and poor ${\mathrm{T}}_1$ sensitivity for DCE analysis. To overcome these drawbacks, we have implemented a spiral-based SAGE approach, which improves signal quality and provides higher ${\mathrm{T}}_1$ sensitivity.

Image artifacts are a major confounding factor in brain tumor hemodynamic assessment; more specifically, susceptibility artifacts may be associated with certain tumor locations (eg, near the skull base, near bone/air/soft tissue interfaces, and near metal calvarial fixation devices), tumor resection cavities, and shunts, which cause large susceptibility effects. For EPI readouts, these effects lead to signal pileup artifacts; additionally, they can manifest as geometric distortion in the phase-encode direction. These artifacts can be seen in the baseline images from a healthy volunteer, and they persist in the resulting ${\mathrm{R}}_2^{\mathrm{*}}$ and ${\mathrm{R}}_2$ maps. In contrast, more consistent ${\mathrm{R}}_2^{\mathrm{*}}$ and ${\mathrm{R}}_2$ images were observed for spiral-SAGE. Although more advanced postprocessing pipelines for EPI-SAGE could include distortion correction, as well as deblurring and drift correction, these have not been widely implemented for EPI-SAGE to date. The implementation of such steps could reduce the impact of many of these artifacts and yield more spatially consistent ${\mathrm{R}}_2^{\mathrm{*}}$ and ${\mathrm{R}}_2$ images for EPI-SAGE. The disadvantages of spiral-based readouts include more complicated reconstruction and spiral blurring artifacts that require the acquisition of ${\mathrm{B}}_0$ maps and a deblurring step in the reconstruction. In this study, reconstruction for spiral-SAGE was performed offline using GPI, but future implementations could leverage inline spiral reconstruction. Although reconstruction includes nodes to correct for spiral blurring, residual artifacts in the spiral data may be the result of the confounding effects of aliasing and spiral blurring on the distribution of artifacts. Future studies could optimize spiral-SAGE parameters to minimize these confounding aliasing and blurring effects.

In terms of comparability between spiral-SAGE and EPI-SAGE, we observed similar signal characteristics in healthy volunteers and a bioreactor phantom. In particular for the bioreactor, the ${\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ curves were highly similar between approaches, although there was a clear benefit for $\mathrm{\Delta }{\mathrm{R}}_1$ in terms of higher ${\mathrm{T}}_1$ sensitivity. In patients with brain tumors, the resulting rCBV maps for spiral- ${\mathrm{R}}_2^{\mathrm{*}}$ were similar to those obtained using standard (single-) GRE EPI readouts, both on a voxel-wise level and across subjects. This is similar to the findings of Schmainda et al, who showed comparable standardized rCBV measures between dual-echo spiral acquisition and a single-echo EPI acquisition across both high- and low-grade brain tumors.⁴⁵ Higher voxel-wise CBF was observed using spiral-SAGE relative to the standard single-echo CBF, though high linear correlation and CCC were observed across all subjects. Similar to EPI-SAGE, ${\mathrm{T}}_1$ leakage effects are inherently removed from $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ measures using spiral-SAGE; thus, the use of multidose protocols to minimize ${\mathrm{T}}_1$ leakage effects are obsolete with this approach. Remaining ${\mathrm{T}}_2^{\mathrm{*}}$ leakage effects can be effectively removed using leakage correction postprocessing methods,^46,47 leading to more accurate global and microvascular rCBV maps. ${\mathrm{T}}_1$ leakage effects form the basis of DCE-MRI, which can be leveraged using dual-echo approaches. However, it is worth noting that the ${\mathrm{T}}_1$ sensitivity of virtually all dual-echo DSC-based methods developed to date is still far below that achieved with standard DCE-based methods because of the use of longer TRs.

Approaches that combine both DSC- and DCE-MRI must balance competing requirements: DCE-MRI acquisitions are typically acquired with short TE and TR to fully maximize ${\mathrm{T}}_1$ sensitivity at the cost of temporal resolution; DSC-MRI acquisitions are typically acquired with moderate TEs to optimize ${\mathrm{T}}_2^{\mathrm{*}}$ sensitivity and higher temporal resolution to capture the rapid pharmacodynamics of intravascular bolus passage. Dual-echo DCE-MRI approaches have been proposed to correct for residual ${\mathrm{T}}_2^{\mathrm{*}}$ effects that can confound pharmacokinetic analysis, but these approaches typically lack the ${\mathrm{T}}_2^{\mathrm{*}}$ sensitivity for DSC-MRI. Additionally, EPI-SAGE is largely optimized for DSC-MRI at the expense of ${\mathrm{T}}_1$ sensitivity based on the longer TEs and TRs. Reduced ${\mathrm{T}}_1$ sensitivity may negatively bias the resulting DCE parameters, in particular for $K^{trans}$.^19,48,49 Dual-echo spiral approaches, including both SPICE¹¹ and spiral-SAGE, enable a short TE (<2 ms) via spiral-out trajectories, thus providing a better balance between rCBV and $K^{trans}$ accuracy. The first TE signal can either be used outright for DCE analysis, or the GRE signals can be combined to extrapolate to $\mathrm{T}\mathrm{E}=0$. Although the latter approach may yield lower SNR caused by echo combinations⁵⁰ Paulson et al compared both approaches and found appreciable ${\mathrm{T}}_2^{\mathrm{*}}$ effects near vessels, suggesting a benefit from removal of ${\mathrm{T}}_2^{\mathrm{*}}$ effects using extrapolation to $\mathrm{T}\mathrm{E}=0.$¹¹ Although the shorter TE improves ${\mathrm{T}}_1$ sensitivity, spiral-based approaches have long TRs relative to single-line DCE-MRI acquisitions, which ultimately limits the ${\mathrm{T}}_1$ sensitivity achievable. Although future studies should directly compare DCE metrics between standard DCE-MRI approaches and our spiral-SAGE approach, the study design herein precluded such a direct comparison.

The SAGE approach, including both EPI and spiral readouts, extends dual-GRE DSC-MRI to include a SE acquisition. The inclusion of both ${\mathrm{T}}_2^{\mathrm{*}}$ and ${\mathrm{T}}_2$ contrast enables the measurement of vessel architectural parameters caused by the differing vessel size sensitivities. Increased vessel diameter is a hallmark of the rapid tumor angiogenesis associated with high-grade brain tumors, and quantitative MRI-based vessel-size measurements have been validated against histological vessel-size measurements.⁵¹ Recently, other vascular parameters have been revealed using the VAI paradigm, including the relationship between perfusion, vessel size, and oxygenation within the vascular network. These parameters may reveal key information in the context of tumor grading,⁵² recurrence,⁵³ and response to therapy.⁷ Using SAGE, these parameters can be unlocked and combined with complementary parameters related to perfusion and permeability. For VAI, spiral-SAGE may have additional advantages over EPI-SAGE because of the improved time efficiency of spiral-based readouts.

SAGE-based approaches must further balance slice coverage with the longer TRs required for inclusion of a SE acquisition; in this study, we achieved 75 mm of brain coverage through the use of the maximum recommended slice thickness (5 mm, per Boxerman et al⁵⁴). Simultaneous multislice approaches⁵⁵ could be implemented in the future to achieve higher through-plane resolution without loss of slice coverage.

In this study, we observed elevated postbolus $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ and $\mathrm{\Delta }{\mathrm{R}}_2$ for spiral-SAGE, which were evident in both tumor and NAWM voxels. Previous studies have similarly observed this effect, and it is generally attributed to contrast agent recirculation and/or residual susceptibility effects caused by contrast agent leakage.^11,56 Previous studies have proposed a two-step pharmacokinetic model (modeled as $\mathrm{\gamma }$ variate fits over the first pass) to account for these effects,^11,56 but as the elevated postbolus curves are not entirely attributable to leakage effects (that is, they are also present in NAWM), there is a lack of consensus for their biophysical origin or their impact on the resultant perfusion parameters. As the focus of this study was to establish the spiral-SAGE sequence, these effects were minimized by short integration limits and normalization to NAWM, which is consistent with virtually all other dual-echo studies. Future studies should investigate the impact of residual postbolus effects in multiecho perfusion metrics, including approaches to correct for these effects.

There are several limitations to this study. First, we were unable to directly compare EPI-SAGE and spiral-SAGE metrics in a patient cohort. This study was completed in the context of our clinical standard of care, where the first injection is used clinically to give the preload dose (and thus provides the opportunity to track this first dose), and the second injection is the clinically acquired standard perfusion. As such, we leveraged a bioreactor phantom to directly compare dynamic contrast injections using both EPI- and spiral-SAGE, which showed excellent comparability. The clinical standard of care did enable direct comparisons with EPI- $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}\mathrm{r}\mathrm{C}\mathrm{B}\mathrm{V}$ and CBF maps, thus validating spiral- $\mathrm{\Delta }{\mathrm{R}}_2^{\mathrm{*}}$ perfusion parameters; however, no direct comparison with DCE-MRI was performed to validate the resulting $K^{trans}$ metrics. Future studies should prospectively compare DCE metrics between standard and spiral-SAGE approaches. In terms of the patient population, the cohort scanned here all exhibited low to moderate rCBV; thus, we were unable to determine rCBV accuracy with spiral-SAGE across a wide range of hemodynamic properties. However, our previous studies, and those of others, have shown excellent agreement between dual-echo approaches and single-echo approaches.^12,45,57 Finally, although Cartesian readouts continue to dominate DSC- and DCE-MRI, increasing availability of both spiral acquisition and analysis methods should further enable advanced multiecho and multicontrast combined approaches. Although the spiral-SAGE parameters were chosen to enable comparisons with EPI-SAGE, spiral readouts provide a significant opportunity for optimization of image parameters for DSC and DCE-MRI. Overall, this approach improves quantification of DCE-MRI parameters, while providing robust DSC-MRI readouts across the full scale of hemodynamic properties.

5 |. CONCLUSIONS

Multiecho acquisitions permit the measurement of multiple distinct hemodynamic parameters in a single acquisition, which is highly advantageous for reducing scan times and contrast agent dose in neuroimaging. The use of spiral readouts for dynamic contrast imaging has several advantages over EPI readouts, including increased time efficiency, shorter first echo times, reduced motion sensitivity, and improved image quality. An advanced spiral-based SAGE method was developed and implemented, which provides substantial flexibility to optimize contrast for both DSC- and DCE-MRI. In the context of brain tumors, we show the wide range of hemodynamic parameters that can be obtained, including standard perfusion metrics, plus microvascular perfusion, DCE, and VAI parameters. The combination of these metrics represents the most complete physiological assessment of brain vasculature to date.