Evaluation of a Multiple Spin- and Gradient-Echo (SAGE) EPI Acquisition with SENSE Acceleration: Applications for Perfusion Imaging In and Outside the Brain

Abstract

Perfusion-based changes in MR signal intensity can occur in response to the introduction of exogenous contrast agents and endogenous tissue properties (e.g. blood oxygenation). MR measurements aimed at capturing these changes often implement single-shot echo planar imaging (ssEPI). In recent years ssEPI readouts have been combined with parallel imaging (PI) to allow fast dynamic multi-slice imaging as well as the incorporation of multiple echoes. A multiple spin- and gradient-echo (SAGE) EPI acquisition has recently been developed to allow measurement of transverse relaxation rate (R₂ and R₂^*) changes in dynamic susceptibility contrast (DSC)-MRI experiments in the brain. With SAGE EPI, the use of PI can influence image quality, temporal resolution, and achievable echo times. The effect of PI on dynamic SAGE measurements, however, has not been evaluated. In this work, a SAGE EPI acquisition utilizing SENSE PI and partial Fourier (PF) acceleration was developed and evaluated. Voxel-wise measures of R₂ and R₂^* in healthy brain were compared using SAGE EPI and conventional non-EPI multiple echo acquisitions with varying SENSE and PF acceleration. A conservative SENSE factor of 2 with PF factor of 0.73 was found to provide accurate measures of R₂ and R₂^* in white (WM) (r_R2 = [0.55–0.79], r_R2* = [0.47–0.71]) and gray (GM) matter (r_R2 = [0.26–0.59], r_R2* = [0.39–0.74]) across subjects. The combined use of SENSE and PF allowed the first dynamic SAGE EPI measurements in muscle, with a SENSE factor of 3 and PF factor of 0.6 providing reliable relaxation rate estimates when compared to multi-echo methods. Application of the optimized SAGE protocol in DSC-MRI of high-grade glioma patients provided T₁ leakage-corrected estimates of CBV and CBF as well as mean vessel diameter (mVD) and simultaneous measures of DCE-MRI parameters K^trans and v_e. Likewise, application of SAGE in a muscle reperfusion model allowed dynamic measures of R₂′, a parameter that has been shown to correlate with muscle oxy-hemoglobin saturation.

Introduction

MR signal can change rapidly in response to changes in exogenous or endogenous contrast agents (CA) in an imaging voxel. MR methods used to monitor these changes often focus on extracting information about tissue perfusion and/or blood oxygenation. For example, dynamic susceptibility contrast (DSC)-MRI uses measures of the effective transverse relaxation rate to track a CA bolus as it distributes in the vasculature and tissue spaces (e.g. tumor) (1,2). This method aims to characterize first-pass hemodynamics for the purpose of computing measures of blood volume and blood flow, often in brain lesions. In addition to Gd-based contrast enhanced imaging, quantification of MR signal changes due to endogenous contrast (e.g. deoxy-hemoglobin) have been demonstrated in blood oxygen level dependent (BOLD) imaging studies (3). Applications of BOLD imaging include functional MRI in both brain (4) and skeletal muscle (5,6). Regardless of the target tissue or type of CA, these techniques require acquisitions with high temporal resolution and the ability to quantify changes in NMR relaxation rates.

Single-shot echo planar imaging (ssEPI) has been used to capture such signal changes with a temporal resolution less than 2s. Dynamic gradient-echo (GE) acquisitions with ssEPI readouts have been implemented in many MR studies, including perfusion (1,2) and functional MRI (7,8). Though conventional ssEPI imaging can improve temporal resolution, it may also produce geometric distortions and result in longer readout times, leading to T₂^* decay during signal acquisition (9). Multi-shot EPI acquisitions can be used to help alleviate these problems; however, these come at the expense of temporal resolution or slice coverage. A second solution, which has been implemented more recently, is the addition of parallel imaging (PI) (10,11). Combining PI with EPI can be used to maintain high temporal resolution during dynamic imaging while mitigating some of the aforementioned limitations (12–16). PI methods, such as SENSitivity Encoding (SENSE), can be implemented with EPI to shorten echo times (TE), permitting increased slice coverage or the acquisition of additional echoes in a single TR (17).

In recent years, perfusion MRI studies have started to incorporate multi-echo acquisitions (18–20). In the context of DSC-MRI, dual GE acquisitions have been used to correct T₁ leakage effects in tumor tissue and to allow the extraction of T₁-weighted signals for use in dynamic contrast enhanced (DCE)-MRI based pharmacokinetic modeling (21–23). Multi-echo methods that implement a single GE in addition to a spin-echo (SE) acquisition have also been investigated (24–26). More recently, Schmiedeskamp et al. developed a multiple echo spin- and gradient-echo (SAGE) EPI sequence for perfusion measurements in the brain (27,28). The SAGE acquisition permits the simultaneous measurement of R₂ and R₂^* during a dynamic event, such as CA distribution. These measurements can be used to generate traditionally acquired GE-based perfusion parameters such as cerebral blood volume (CBV), blood flow (CBF), and mean transit time (MTT), in addition to their SE-based counterparts. In these multi-echo experiments, simultaneous acquisitions of R₂ and R₂^* provide sensitivity to both macro- and microvasculature and, therefore, allow estimation of relative blood vessel size, a potentially important quantitative MR biomarker in perfusion and functional BOLD imaging (29–31).

Previous studies have investigated the effect of PI on ssEPI image quality and neuronal activation in the case of functional MRI (14–16). Few of these studies, however, included multiple-echo acquisitions and, therefore, have not considered the effect of PI on NMR properties such as R₂ and R₂^*. In previous SAGE EPI experiments (27,28,32), data were only acquired with a single GRAPPA PI factor (Reduction Factor (R) = 3). Though a similar reduction in scan time can be afforded by PI techniques using image-based reconstruction (e.g. SENSE), the optimal (or allowable) amount of image acquisition acceleration, in the context of a SAGE EPI experiment, is unknown. Optimization and implementation of a multi-echo SAGE EPI acquisition requires a trade-off between dynamic scan time, spatial and temporal SNR, and echo time. Depending on voxel size and dynamic repetition time (TR), the use of SENSE with SAGE (33) may be limited by the range of TEs necessary to reliably characterize R₂ and R₂^* in a given tissue. This may be particularly true in tissues outside of the brain (e.g. skeletal muscle) that exhibit higher R₂ and R₂^*. The combination of multiple acceleration methods (e.g. PI plus partial Fourier) may help address this issue by further decreasing SAGE TEs; however, a systematic evaluation is needed.

In this study, a multiple-echo SAGE EPI sequence utilizing SENSE acceleration was developed and optimized in the brain and skeletal muscle. Dynamic relaxation rate measurements were computed from SAGE acquisitions in healthy subjects, while varying SENSE and partial Fourier acceleration factors. These measures were compared, on a voxel-wise basis, to those from conventional non-EPI multiple gradient-echo (MGE) and multiple spin-echo (MSE) MRI methods and were evaluated using the mean percent difference between values, the pooled standard deviation and voxel-wise correlations. Optimized SAGE protocols were applied in DSC-MRI perfusion experiments in high-grade glioma patients as well as measurements of muscle reperfusion in healthy subjects undergoing arterial occlusion and leg contractions.

Methods

SAGE Evaluation and Optimization: Brain and Muscle

For evaluation in the brain, dynamic SAGE EPI data were acquired in healthy subjects (n=4, 28–36 y.o., 2 Males) (under Vanderbilt University Institutional Review Board (IRB) guidelines) at 3T (Achieva, Philips Healthcare, Best, The Netherlands) using a 32 channel head coil for data reception. Scan parameters were: TR = 1.8s, Dynamics = 35; FOV = 240×240mm², Voxel Size = 3.16×3.16×5.0mm³ (reconstructed to a 96×96 square matrix), number of echoes (NE) = 5 (2 GE, 2 asymmetric SE, 1 SE) (27), and slices = 15. The number of dynamics was chosen to provide 30–60s of data, a typical duration for baseline (pre-contrast) measurements in perfusion imaging. Echo times were computed (within the pulse sequence) by minimizing the acquisition time of each EPI readout train and were, therefore, influenced by SENSE acceleration factors (SENSE), partial Fourier acceleration factors (PF), and voxel size. As part of sequence optimization, the SENSE acceleration factor was varied between 1.5 and 4 with and without PF factors of 0.60, 0.73, 0.81, and 1.00. As a prerequisite, all scan parameters were set such that resulting TEs were between ~5ms and 100ms for brain imaging.

The multi-echo SAGE data were fit in a least-squares manner to Eq. (1) (27,28):

$$S(t)=\{\begin{array}{cc}{S}_0·e^{-t·R_2^{\ast }}& 0<t<TE/2\\ \frac{S_0}{\delta }·e^{-TE\left(R_2^{\ast }-R_2\right)}·e^{-t\left(2R_2-R_2^{\ast }\right)}& TE/2<t<TE\end{array}$$ (1)

The four resulting fitted parameters were S₀, R₂, R₂^*, and δ, where δ accounts for possible slice profile mismatch between echoes before and after refocusing (28). To assess fit quality, the root mean squared percent error (RMSPE) was calculated. Due to the importance of dynamic stability in analyses of perfusion and functional MR studies (34), temporal SNR (tSNR) was assessed on a voxel-wise basis over an ROI encompassing a single brain slice for each SAGE echo. Voxel-wise measures of R₂ and R₂^* from the SAGE acquisition (averaged over 25 dynamics) were compared to those from conventional MGE (TR = 1800ms, TE = 5.0ms, echo spacing (ESP) = 5.0ms, NE = 20) and MSE (TR = 6000ms, TE = 5.91ms, ESP = 5.91ms, NE = 16) acquisitions. The group mean and pooled standard deviation (std_p) were computed along with Pearson’s correlation coefficient (r). In addition, the mean percent difference (% diff) in relaxation rates was calculated in a voxel-wise manner over white and gray matter ROIs covering several contiguous slices.

In a similar manner, the SAGE sequence was evaluated and optimized in skeletal muscle. SAGE data was acquired in the legs of healthy subjects (n=4, 29–39 y.o., 3 Females) at 3T (Achieva, Philips Healthcare, Best, The Netherlands) using an 8 channel receive-only knee coil and the body coil for transmission. Scan parameters were: TR = 2.5s, Dynamics = 10, FOV = 180×180mm², Voxel Size = 2.81×2.81×7.6mm³, NE = 5, slices = 1. The SENSE factor was varied between 2 and 4 and the PF factor varied between 0.6 and 0.65, attempting to keep echo times less than 50ms. As with the brain, dynamic stability was assessed by measuring mean tSNR over an ROI encompassing the tibialis anterior (TA). Muscle SAGE data were fitted to Eq. (1) as described above. Resulting measures of SAGE R₂ and R₂^* were compared to those from MGE (TR = 2500ms, TE = 2.5ms, ESP = 2.5ms, NE = 30) and MSE (TR = 2500ms, TE = 10.8ms, ESP = 10.8ms, NE = 8) acquisitions. Due to the homogeneity of muscle tissue over an ROI, the mean percent difference in relaxation rates and the pooled standard deviation were used for comparison in lieu of voxel-wise correlations. For both muscle and brain, the optimal SAGE protocols were selected based on temporal stability (high tSNR) and agreement with conventional multi-echo methods (small mean percent difference, small std_p, and high voxel-wise correlations).

Brain Tumor Perfusion

SAGE data were acquired in high-grade glioma patients (n=3, 40–55 y.o., 3 Males) using the scan parameters listed above and the optimized acceleration scheme (see Results). Measurements were made before, during, and after administration of Gd-DTPA (0.1 mmol/kg, infusion rate of 4ml/s followed by a 20ml saline flush). The scan duration was 7.5 minutes (250 dynamics). The resulting SAGE data were fitted on a voxel-by-voxel basis, first using Eq. (1) to determine an average value of δ over a 60s baseline, and subsequently fixing this parameter when fitting (three-parameter fit to Eq. (1)) the remainder of the dynamic time-course. From this fit, estimates of R₂ and R₂^* were computed with the passage of CA. ΔR₂ and ΔR₂^* were calculated as the difference between the relaxation rates during CA passage and the mean pre-contrast relaxation rate. In addition, a T₁-weighted signal time course was extracted from the first two echoes of the SAGE data (TE₁ and TE₂) by extrapolating to S(TE = 0) (21–23). Multiple flip angle (MFA) data were also acquired (TR = 7.6ms, TE = 4.6ms, FA = 2°-20° in 2° increments) to create a pre-contrast T₁ map, which was combined with the T₁-weighted data to produce ΔR₁ time-courses for each voxel.

DSC-MRI and DCE-MRI Analysis

To compute parametric perfusion maps, an arterial input function (AIF) was extracted from the SAGE GE data, in the region of the medial cerebral artery. An automated selection process based on peak height, bolus arrival time, and correlation between the first two echoes during bolus passage was used to avoid saturation effects in the AIF (35,36). The ΔR₂^* time-course for the AIF was converted to a concentration time-course using a quadratic relationship for CA inside a blood vessel (37). ΔR₂ and ΔR₂^* time-courses in tissue were converted to CA concentration ([CA]) using the CA’s in vivo relaxivity. The effective transverse relaxivity of Gd-DTPA (r₂^*) at 3T was assumed to be 87 mmol⁻¹ ms⁻¹ (38) while the value of r₂ used was 20.4 mmol⁻¹ ms⁻¹ (27). Standard DSC analysis was carried out on the AIF and tissue [CA] time courses (70 dynamics, 30 pre- and 40 post-injection) using a circular deconvolution approach with SVD (1,39). Measures of absolute CBV, CBF, and MTT for both GE and SE data were calculated for the entire tumor volume. Based on the simultaneous acquisition of R₂ and R₂^* (and voxel-wise estimates of tissue diffusion (data not shown)), mean vessel diameter (mVD) was computed using methods described by Kiselev et al. (30). Pharmacokinetic modeling of the computed ΔR₁ time courses (250 dynamics, 30 pre-injection and 220 post-injection) and the AIF (in [CA]) described above was performed using the standard Tofts model in order to estimate voxel-wise measures of K^trans (contrast agent transfer rate constant) and v_e (extracellular-extravascular volume) (40,41).

Muscle Perfusion

Healthy subjects (n=5, 27–37 y.o., 3 Females) were scanned at 3T using an eight channel RF knee coil for signal reception. Specific methods about patient set-up are described elsewhere (5). Briefly, the coil was aligned with the center of the magnet and the subject was positioned supine on the patient bed with a foot stabilized by strapping it into an isometric exercise device. The maximum cross-sectional area of the TA was centered in the coil. For the occlusion experiment, a pneumatic cuff was placed around the subject’s (n=2, 1 Female) thigh and connected to a Hokanson E20 rapid cuff inflator and AG101 air source (Hokanson, Bellevue, WA). Data were acquired for 1 minute prior to cuff occlusion, after which the cuff was rapidly inflated to 248 mmHg. Arterial occlusion continued for a period of 4–5 minutes, after which the cuff was rapidly released and data were acquired for an additional 7–8 minutes. The remaining subjects (n=3, 2 Females) performed a 10s duration maximal voluntary isometric dorsiflexion contraction followed by 4 minutes of rest. Scan parameters for the dynamic SAGE acquisition were the same as in the validation scans with the optimal acceleration scheme (see Results). Dynamic SAGE data were analyzed using a three-parameter fit to Eq. (1)) as described for the brain tumor perfusion experiment. From this fit, measures of R₂ and R₂^* were extracted before, during, and after arterial occlusion and contraction. The post-contraction R₂^* data were smoothed by fitting them to a 6^th order polynomial, similar to the procedure used in (20). The rate of RF-reversible dephasing, R₂′ (= R₂^* – R₂), was also computed as this parameter has been shown to correlate with muscle oxy-hemoglobin saturation (5).

Results

The effect of SENSE PI on SAGE image quality can be seen in Fig. 1. Increasing SENSE acceleration, with the minimum allowable PF factor (PF = 0.6), resulted in PI reconstruction artifacts (Fig. 1a–c). In addition, geometry (g)-factor maps were calculated at each SENSE factor (Fig. 1d–f) in order to assess image noise associated with increased PI. Though the SAGE measurements demonstrated temporal signal stability sufficient for signal modeling in perfusion and functional MRI (Fig. 1g–i), the tSNR decreased by up to 29% in the first two echoes when the SENSE acceleration factor was increased to 4.

Figure 1.

(a–c) Example single-echo (TE₂) images from the SAGE acquisition showing the effect of increased SENSE acceleration on image quality. (d–f) G-factor maps calculated from the SAGE acquisition. (g–i) tSNR at each SAGE echo over an entire slice (mean ± std across subjects).

Validation of the current technique in an example subject is presented in a voxel-wise comparison of relaxation rate estimates with varying image acceleration (Fig. 2). Fig. 2 demonstrates the influence of combining SENSE acceleration and PF on SAGE measures of R₂ and R₂^* in white matter. A wide range of voxel-wise correlations were observed across the different SENSE/PF combinations, with white matter r_R2 = [0.01–0.84] and r_R2* = [0.03–0.74] and gray matter r_R2 = [0.01–0.92] and r_R2* = [0.07–0.77]. Estimates from acquisitions with SENSE ≥3 in combination with PF = 0.6 yielded the lowest correlations with conventional measures. SAGE estimates of R₂ and R₂^* (mean ± std_p across subjects) in white and gray matter ROIs are outlined in Table 1 along with the mean percent difference from a voxel-wise comparison with MGE and MSE measures. Based on high tSNR, small std_p and mean percent difference, and higher correlations (across subjects) with conventional measures (r_R2=[0.55–0.79], r_R2*= [0.47–0.71] in WM and r_R2 = [0.26–0.59], r_R2* = [0.39–0.74] in GM), it was determined that a SENSE factor of 2 with a PF = 0.73 (highlighted in Table 1) provided the most reliable estimates of R₂ and R₂^* from the SAGE acquisitions evaluated. Using this optimal acceleration scheme, fits to the SAGE data, over a single slice, resulted in a RMSPE of 0.12 ± 0.06 across subjects.

Figure 2.

Example voxel-wise plots of R₂ and R₂^* from SAGE and MGE/MSE acquisitions in a white matter ROI at a) SENSE=2, PF=0.6, b) SENSE = 2, PF = 0.73, c) SENSE = 3, PF = 0.6, d) SENSE = 3, PF = 1.0, e) SENSE = 4, PF = 0.6, f) SENSE = 4, PF = 1.0. Regression lines (black line) and correlation coefficients (r) are presented for both R₂ and R₂^* data.

Table 1.

Group analysis of R₂ and R₂^* measures from SAGE acquisitions in brain with various SENSE and partial Fourier acceleration factors.

SENSE/PF	TE (ms)	R2 (s−1)	R2* (s−1)
		mean±stdp (% diff)	mean±stdp (% diff)
White Matter
1.5/0.6	6.6,25,51,69,87	15.46±2.09 (20.38%)	19.85±2.07 (0.43%)
2.0/0.6	5.7,19,41,54,68	16.30±2.19 (26.70%)	20.85±2.92 (5.24%)
2.0/0.73	8.3,24,50,66,82	15.95±1.49 (23.81%)	20.21±2.25 (2.04%)
2.5/0.81	8.8,23,48,62,77	16.48±1.92 (28.03%)	20.44±3.66 (3.14%)
3.0/0.6	5.2,15,33,42,51	16.72±2.20 (30.01%)	20.70±3.31 (4.94%)
3.0/1.0	9.9,25,52,66,81	16.39±1.62 (27.11%)	20.05±2.55 (1.09%)
4.0/0.6	4.8,12,29,37,44	18.26±4.92 (41.95%)	24.93±9.61 (26.14%)
4.0/1.0	9.2,21,45,57,69	16.35±2.93 (26.91%)	21.12±5.29 (6.83%)
MSE/MGE		12.88±0.70	19.83±1.63
Gray Matter
1.5/0.6	6.6,25,51,69,87	14.34±4.90 (42.27%)	19.55±6.15 (−5.63%)
2.0/0.6	5.7,19,41,54,68	15.02±4.98 (49.47%)	20.12±8.94 ((−3.68%)
2.0/0.73	8.3,24,50,66,82	14.04±2.42 (39.40%)	19.82±6.41 ((−4.90%)
2.5/0.81	8.8,23,48,62,77	13.93±2.70 (37.84%)	19.71±7.84 ((−5.80%)
3.0/0.6	5.2,15,33,42,51	14.10±4.24 (39.83%)	17.50±6.80 ((−16.10%)
3.0/1.0	9.9,25,52,66,81	13.45±1.79 (33.13%)	18.58±7.25 ((−10.30%)
4.0/0.6	4.8,12,29,37,44	13.80±5.25 (36.22%)	15.92±9.49 ((−23.22%)
4.0/1.0	9.2,21,45,57,69	13.79±2.16 (37.01%)	18.22±6.52 ((−12.26%)
MSE/MGE		10.20±1.33	20.79±2.53

^*Optimal sequence in highlighted row.

Fig. 3a shows example images at each SAGE echo in muscle. The SAGE decay curve from an ROI encompassing the TA and the corresponding fit can be seen in Fig. 3b. Also shown are the R₂^* (dash-dot) and R₂ (dashed) decay curves. Table 2 shows the effect of varying image acceleration on SAGE measures of R₂ and R₂^* in muscle when compared to MGE and MSE measures. The use of a smaller SENSE factor (e.g. SENSE = 2) resulted in an overestimation in relaxation rates up to 24%, while SENSE > 2 provided more accurate estimates, particularly in R₂^*. Acquisitions with SENSE = 3 and PF = 0.6 as well as SENSE = 4 and PF = 0.65 performed equally as well based on the mean percent difference and std_p reported in Table 2. For the dynamic muscle perfusion studies, a SENSE factor of 3 and PF = 0.6 was selected as the optimal acquisition due to higher mean tSNR at each echo (tSNR_TE1–5 = 423,295,150,114,93).

Figure 3.

a) Example muscle images for each SAGE echo. b) SAGE signal decay data (and fits) for an ROI encompassing the TA. Also shown are portions of the corresponding individual R₂^* (dash-dot line) and R₂ decay curves (dashed line) before and after the refocusing pulse, respectively.

Table 2.

Group analysis of R₂ and R₂^* measures from SAGE acquisitions in muscle with various SENSE and partial Fourier acceleration factors.

SENSE/PF	TE (ms)	R2 (s−1)	R2* (s−1)
		mean±stdp (% diff)	mean±stdp (% diff)
2.0/0.6	5.5,19,37,50,63	39.12±5.72 (23.38%)	42.50±6.80 (11.88%)
3.0/0.6	4.9,14,29,38,47	35.62±5.25 (11.73%)	38.69±6.18 (1.17%)
3.0/0.65	5.5,15,30,39,49	37.10±4.62 (16.69%)	41.15±5.08 (7.81%)
4.0/0.65	5.0,13,25,33,41	33.95±4.78 (6.45%)	39.35±6.18 (2.76%)
MSE/MGE		32.00±1.99	38.28±2.16

^*Optimal sequence in highlighted row.

Fig. 4a shows the SAGE signal decay curves for a region of normal appearing white matter and tumor prior to the administration of CA. Also shown are portions of the corresponding individual R₂^* (dashed lines) and R₂ decay curves (dotted lines) before and after the refocusing pulse. Note that for both tumor and white matter, the GE data points fall directly on the calculated R₂^* decay curve, while the R₂ decay curve intersects the SAGE data points at the SE (TE₅). The resulting SAGE ΔR₂ and ΔR₂^* curves for white matter and tumor ROIs during CA passage are displayed in Fig. 4b and 4c, respectively. Also displayed are ΔR₂ and ΔR₂^* time-courses calculated from single GE (TE₂) and SE (TE₅) data. As expected, only minor differences between SAGE ΔR₂ and ΔR₂^* and single-echo estimates were observed in white matter. The SAGE time-courses in tumor (particularly during washout), however, were quite different when compared to those from single-echo measurements, as the SAGE estimates are intrinsically corrected for T₁ effects due to CA leakage, while enhancing sensitivity to T₂^* effects. Additionally, the contrast-to-noise ratio (CNR), as measured from peak to baseline, in whole tumor across patients was found to be 33.9 ± 18.8 and 10.8 ± 2.1 in SAGE ΔR₂^* and ΔR₂ time-courses, respectively, compared to 50.3 ± 30.3 and 7.2 ± 3.7 in the single-echo estimates.

Figure 4.

a) SAGE signal decay data (and fits) for a region of normal appearing white matter and tumor prior to the administration of contrast agent. Also shown are portions of the corresponding individual R₂^* (dashed line) and R₂ decay curves (dotted line) before and after the refocusing pulse, respectively. b) Resulting ΔR₂ curves from SAGE fits (solid line) and single echo analysis (dashed line) during contrast agent passage in white matter (gray) and tumor (black). c) Resulting ΔR₂^* curves from SAGE fits (solid line) and single echo analysis (dashed line) during contrast agent passage in white matter (gray) and tumor (black).

Fig. 5 demonstrates the utility of SAGE EPI for DSC perfusion imaging. Parametric maps created from ΔR₂^* and ΔR₂ data highlight increases in both CBV and CBF in the enhancing tumor regions. GE CBV, CBF and MTT in whole tumor were 4.07 ± 1.59 ml/100g, 31.53 ± 10.78 ml/100g/min, and 7.75 ± 0.39 s, respectively. SE CBV, CBF and MTT in tumor were 2.97 ± 0.68 ml/100g, 16.95 ± 4.13 ml/100g/min, and 10.95 ± 0.43 s, respectively. Voxel-wise measures of mVD (Fig. 5e) were found to predominantly vary between 2μm–25 μm across the brain. In normal appearing white matter mVD was observed to be 6.6 μm across patients, while whole tumor mVD was 10.7 μm. In addition, voxel-wise T₁-weighted signal time-courses were extrapolated from the GE data, resulting in a mean ΔR₁ time course for the tumor ROI (Fig. 6a). Pharmacokinetic analysis of the ΔR₁ time courses using the standard Tofts model (Fig. 6b, c) resulted in K^trans = 0.16 ± 0.03 min⁻¹ and v_e = 0.45 ± 0.04 in the enhancing tumor volumes across patients.

Figure 5.

a) Post-contrast T₁-weighted anatomical image highlighting a brainstem glioma and an additional enhancing lesion (white arrows). Parametric perfusion maps (CBV, CBF, and MTT) from ΔR₂^* (b–d) and ΔR₂ (f–h) data. e) Map of mean vessel diameter computed from simultaneous measurements of ΔR₂^* and ΔR₂.

Figure 6.

a) Resulting ΔR₁ time course (calculated with the addition of a T₁ map) from an ROI in tumor. The arterial input function acquired for pharmacokinetic modeling is displayed in the inset. DCE-MRI parametric maps of K^trans (b), and v_e (c). Arrows point to regions identified in Fig. 5.

SAGE R₂^*, R₂, and R₂′ time-course data were acquired from the tibialis anterior of one male subject and one female subject before, during, and after 4 or 5 minutes, respectively, of thigh cuff-induced arterial occlusion. Relaxation rate constant time courses included similar features for both subjects: initial rapid decrease with cuff inflation; slow increase during occlusion; rapid decrease to minimum after cuff deflation; slow return to baseline from minimum. Data from the 4 minute occlusion are presented in Fig. 7a. The R₂^*, R₂ and R₂′ baseline 30s prior to cuff inflation for both subjects were 41.4 ± 1.7, 37.5 ± 1.6, and 3.9 ± 0.1 s⁻¹. These values decreased rapidly to 37.9 ± 2.4, 36.0 ± 0.4, and 2.2 ± 2.5 s⁻¹ with cuff inflation, then increased more slowly over the duration of occlusion to 40.7 ± 0.1, 38.2 ± 3.0, and 2.7 ± 2.9 s⁻¹. Following cuff deflation, relaxation rate constants reached minima of 38.2 ± 0.1, 34.7 ± 0.3, and 3.4 ± 0.2 s¹ with times to minimum of 23.8 ± 5.3 s. Return to baseline following occlusion required 445 ± 49.5 and 150 ± 3.5 s for R₂^* and R_2, respectively.

Figure 7.

SAGE R₂^*, R₂, and R₂′ time-courses from the tibialis anterior of one subject a) before, during, and after 4 minutes of cuff-induced arterial occlusion, and from a different subject b) before, during and after a 10 second duration maximal voluntary isometric contractions of the dorsiflexors.

SAGE R₂^*, R₂, and R₂′ time-course data were also acquired from the TA of 3 subjects before, during, and after a 10 second duration maximal voluntary isometric contraction of the dorsiflexors. Example time-course data from a subject are presented in Fig. 7b. R₂^*, R₂ and R₂′ 30 s prior to each contraction were 39.1 ± 6.0, 37.9 ± 5.6, and 1.3 ± 0.5 s⁻¹. Following the contraction, R₂^* and R₂ showed a muscle BOLD related decrease to 37.8 ± 5.7 and 36.1 ± 5.8 s⁻¹ with 20.8 ± 5.2 and 17.5 ± 6.6 s times to minimum, respectively. R₂′ increased followingcontraction to a peak of 2.2 ± 0.4 s⁻¹ with a time to peak of 6.7 ± 2.9 s.

Discussion

This study evaluates a multiple-echo SAGE EPI acquisition with SENSE PI developed for perfusion MRI. Regarding acquisitions in the brain, this analysis demonstrates the potential drawbacks of using large SENSE factors in the context of quantitative imaging with SAGE EPI. Consequently, the use of more conservative PI required the addition of partial Fourier acquisitions to produce TEs suitable for characterizing transverse relaxation rates in brain and muscle. It has been previously shown in the brain that the use of large SENSE factors (≥3) with ssEPI may lead to an increase in image artifact and noise, as observed here, when compared to other PI techniques (11,16). The present study does show that reasonable estimates of R₂ and R₂^* can be made in the brain with SENSE = 3 if PF techniques are excluded, supporting the previous use of GRAPPA (R = 3) with SAGE EPI (27,28,32). It should be noted, however, that these implementations, because they forgo PF imaging, may ultimately be limited to the use of longer TEs (without changing other imaging parameters). This could confound the characterization of tissues with short T₂s, as demonstrated here in muscle, where the use of shorter TEs (with multi-echo imaging) is essential. Therefore, knowledge of the present acceleration techniques’ impact on data reliability and accurate relaxation rate estimates with SAGE EPI is important when considering application across multiple tissue types.

Sequence validation was performed in healthy subjects both in the brain and skeletal muscle. The mean RMSPE for the four-parameter SAGE fit in brain and in the TA ROI were 0.12 and 0.21, respectively, which was significantly less than that observed previously by Schmiedeskamp et al (28). In addition, we observed a mean whole brain tSNR (tSNR > 70 at all echoes) that was larger than previously reported at 3T using SENSE EPI with similar acceleration (15,16). Moderate correlations were found between estimates of R₂ and R₂^* from SAGE acquisitions in the brain when compared to conventional multi-echo techniques. In the optimized acquisition, mean values of R₂^* differed by less than 5% from corresponding MGE measures in the brain, while larger differences were observed in R₂. For muscle, mean values of SAGE R₂ and R₂^* from the optimized acquisition corresponded well (< 12% difference) with estimates from conventional acquisitions (Fig. 3). Despite these differences, the mean transverse relaxation rates estimated from the optimized SAGE acquisitions fell within the range of those previously reported for brain and muscle at 3T (42,43). Aside from known effects of ssEPI imaging, differences in relaxation rate estimates from SAGE EPI may be attributed to slice profile mismatch due to non-ideal RF pulse profiles, as previously reported (28). In this regard, the value of δ in white matter across subjects was 1.09±0.08, which corresponded well to previous SAGE measurements in brain (28). The corresponding measure in muscle was observed to be larger with δ = 1.51±0.40. Though the perfusion experiments implemented a fixed estimate of δ (and therefore a three-parameter fit) throughout the dynamic time course, potential errors associated with excitation and refocusing slice profile mismatch, particularly during CA bolus passage, may still need to be considered.

Brain Tumor Perfusion

The utility of fast multi-echo EPI measurements, as demonstrated with SAGE EPI, is well suited for perfusion imaging, both with and without CA administration. To this end, the work presented here further validates application of SAGE EPI in the evaluation of brain tumors. In DSC-MRI it is known that estimates of CBV, CBF and MTT can be confounded by CA extravasation, as is observed in enhancing tumors (44,45). Because traditional single-echo DSC measures can lead to underestimation of tumor CBV in this scenario, multiple-echo acquisitions have been used in their place to calculate T₁-leakage corrected estimates of ΔR_2*(Fig. 4c) (18,21,46). The range of T₁-corrected CBV, CBF and MTT estimates (Fig. 5) in this study were found to agree with published values in healthy brain and tumor (47,48). The SE-based perfusion maps (Fig. 5f–h) exhibited decreased measures of CBV (27% decrease) and CBF (47% decrease) in tumor, when compared to GE techniques. This is expected, as SE measurements are maximally sensitive to capillary-sized blood vessels. The differences in GE and SE DSC parameters closely reflect those previously observed in leakage corrected SAGE measurements in glioblastoma multiforme patients (32). In addition to traditional DSC parameters, the simultaneous measurement of R₂ and R₂^* with SAGE EPI provides sensitivity to a range of blood vessel sizes. As a byproduct of these simultaneous measurements, information about blood vessel diameter was also ascertained. The resulting estimates of mVD in healthy brain and tumor tissue were in good agreement with those observed in a previous experiment using confocal microscopy (49) and appear to be more similar to histological measurements when compared to previous vessel size imaging studies in humans (30).

In addition to correcting CA leakage effects, multiple-echo DSC acquisitions permit the extraction of T₁-weighted signal time-courses. SAGE EPI data was used to compute ΔR₁ time-courses in order to estimate DCE-MRI parameters K^trans and v_e. These data highlight both perfusion and permeability characteristics of the enhancing lesions. Simple pharmacokinetic modeling was implemented. This approach differed from that previously reported for analyzing T₁-based time-courses acquired with SAGE EPI (32). The previous study by Schmiedeskamp et al. implemented a four-parameter model with abbreviated (< 2 minutes) tissue [CA] time-courses. Though the estimated parameters were used to correct DSC perfusion estimates for T₂^* leakage effects, the complexity of the model may make it more susceptible to noise propagation when compared to the more conventional Tofts’ model. The use of shortened dynamic time-courses may also lead to confounded measures of v_e due to incomplete sampling of tracer washout.

Muscle Perfusion

This study presents the first evaluation and application of dynamic SAGE EPI outside of the head. Overall, the SAGE R₂^* and R₂ responses that we report here agree with reports from the literature when factors such as field strength, the muscle studied, the limited sample size, and pulse sequence are considered. For the occlusion experiments, the 7.4% increase in SAGE R₂^* during occlusion and the relative SAGE R₂^* hyperemic decrease to 90% of the baseline rate constant, after 4–5 minutes of arterial occlusion in the TA, are in very good agreement with studies using a range of cuff durations. Sanchez et al. (50) reported a 7.0% increase in R₂^* of the TA during 3 minutes of cuff occlusion. T₂^* data from Englund et al. (51), when converted to rate constants, showed an 8.7% increase in R₂^* of the soleus during occlusion and a hyperemic decrease to 87% of the baseline signal after 3 minutes of occlusion. Elder et al. (5) reported a 9.8% increase in R₂^* in the TA during occlusion and a decrease to approximately 95% of baseline after 5 minutes of occlusion. In contrast, Toussaint et al. (52) reported a 6.5% decrease in GE (TE = 60 ms) signal intensity after 5 minutes of occlusion in the soleus at 1.5T and Lebon et al. (53) demonstrated a 6% decrease in GE (TE = 15 ms) signal intensity of the gastrocnemius after 7 minutes of occlusion at 3T. Only a 2.2 ± 1.0% decrease in GE (TE = 4.9 ms) signal intensity was observed in the TA during 4 or 5 minutes of occlusion at 3T (data not shown). In addition, our 2.2 s⁻¹ increase in SAGE R₂^* was also less than the ~8 s⁻¹ increase during 3 minutes of occlusion and the 15 s⁻¹ increase in ΔR₂^* observed during 5 minutes of occlusion in the soleus at 3T (TE ≥27.2 ms) by Donahue et al. (26).

The differences in relative signal intensity decrease and increase in R₂^* are likely explained first by differences in the physiological sensitivity of the echo times employed (shorter TE being less sensitive to oxygen saturation (20)), and second, by potential differences in oxygen utilization rate (54) in ischemic muscle between subject populations. Our data agrees, however, with other studies that included similarly short echo times (5,50,51). The 6.1% increase in SAGE R₂ we observed in the TA during occlusion at 3T agrees with the 4.6% increase in R₂ during occlusion that results when the data from Toussaint et al. (52) are adjusted for differences in field strength. This increase, however, is greater than the little to no change in R₂ of the TA reported previously by Elder et al (5). Our 2.2 s⁻¹ increase in R₂ is comparable with the approximately 4 s⁻¹ increase in ΔR₂ of the soleus reported by Donahue et al. (26) when the muscle group and possible differences in training status of the subject populations are considered.

For the muscle contraction experiments, the decrease in SAGE R₂^* of 3.3% following a 10 s isometric contraction is in good agreement with the 4.6% decrease in R₂^* following an 8 s isometric contraction of the same muscle group reported by Damon et al. (20). Meyer et al. (55) reported an average decrease in R₂^* of 0.85 s⁻¹ and a time to peak of the BOLD response of 7.9 s following 1 s isometric contractions of the TA, while we observed a 1.3 s⁻¹ decrease in R₂^* for 10 s contractions and a 20.8 s time to minimum. The signal and time to peak differences are consistent, and the approximately 2 fold difference in the time and magnitude of the response is reasonable considering the potential effect of contraction duration (55) and the effect of differences in physical activity status on post-contraction changes in apparent relaxation rates (54).

The SAGE acquisition, applied in skeletal muscle during arterial occlusion and maximal isometric contractions, allows the simultaneous measurement of R₂^* and R₂ and permits a more direct calculation of R₂′, a parameter that provides a basis for image basedcalculation of muscle oxygen saturation (5). Compared to protocols that separately measure R₂ and R₂^*, SAGE EPI improves experimental efficiency, reduces between-test physiological variability, decreases subject burden, and reduces investigator processing time. These advantages allow many potential applications and would be most useful in the investigation of patient populations such as individuals with diabetes, peripheral vascular disease, and other conditions affecting oxygen delivery or cellular metabolism.

The SAGE EPI acquisition presented here was developed and implemented with relatively minor modifications to an existing SE ssEPI pulse sequence. Furthermore, the current sequence employed vendor-provided RF pulses, PI capabilities, and image reconstruction techniques, all of which were achieved completely on the scanner itself. In addition, SAGE EPI acquisitions with SENSE were successfully made using multiple RF coils (32 channel and 8 channel (data not shown) head coils and 8 channel knee coil). Further optimization and customization of the sequence will be necessary, as translation to higher fields (e.g. 7T) will require custom RF refocusing pulses to mitigate slice profile mismatch, SAR issues, and improve data quality. This will be important in future high-field application of SAGE EPI for functional neuro- and musculoskeletal imaging. Building on the data presented here, larger studies focused on treatment response in high-grade glioma and BOLD imaging of muscle contractions are already in progress.

Conclusion

This work presents a new implementation of a SAGE EPI acquisition using SENSE PI. Voxel-wise estimates of transverse relaxation rates from conventional multi-echo methods were used to validate SAGE EPI measurements using SENSE acceleration with partial Fourier imaging. This combination allowed reliable estimates of relaxation rates with SAGE EPI in tissues with shorter T₂’s due in part to the reduction in achievable echo times. In this regard, we presented the first application of SAGE EPI outside of the head. Simultaneous measurement of R₂^* and R₂ afforded with SAGE provided dynamic tracking of muscle R₂′ during reperfusion. An optimized SAGE EPI acquisition was also used to compute simultaneous perfusion and permeability parameters in high-grade glioma patients. Overall, this study complements the previous development of SAGE EPI while further permitting application of the technique across a wider range of tissue types and MR scanner environments.

Abbreviations

CA

contrast agent

ssEPI

single-shot echo planar imaging

PI

parallel imaging

CBV

cerebral blood volume

CBF

cerebral blood flow

MTT

mean transit time

MGE

multiple gradient-echo

MSE

multiple spin-echo

PF

partial Fourier

RMSPE

root mean squared percent error

TA

tibialis anterior

MFA

multiple flip angle

AIF

arterial input function

mVD

mean vessel diameter