Journal of Cerebral Blood Flow & Metabolism (2014) 34, 1111–1116 & 2014 ISCBFM All rights reserved 0271-678X/14 $32.00 www.jcbfm.com
RAPID COMMUNICATION
Snapshot MR technique to measure OEF using rapid frequency mapping Rajiv G Menon1, Edward G Walsh2, Donald B Twieg3, Charles G Cantrell1, Parmede Vakil1, Sumeeth V Jonathan1, Hunt H Batjer4 and Timothy J Carroll1,5 Magnetic resonance (MR)-based oxygen extraction fraction (OEF) measurement techniques that use blood oxygen level-dependent (BOLD)-based approaches require the measurement of the R2’ decay rate and deoxygenated blood volume to derive the local oxygen saturation in vivo. We describe here a novel approach to measure OEF using rapid local frequency mapping. By modeling the MR decay process in the static dephasing regime as two separate dissipative and oscillatory effects, we calculate the OEF from local frequencies measured across the brain by assuming that the biophysical mechanisms causing OEF-related frequency changes can be determined from the oscillatory effects. The Parameter Assessment by Retrieval from Signal Encoding (PARSE) technique was used to acquire the local frequency change maps. The PARSE images were taken on 11 normal volunteers, and 1 patient exhibiting hemodynamic stress. The mean MR-OEF in 11 normal subjects was 36.66±7.82%, in agreement with positron emission tomography (PET) literature. In regions of hemodynamic stress induced by vascular steal, OEF exhibits the predicted focal increases. These preliminary results show that it is possible to measure OEF using a rapid frequency mapping technique. Such a technique has numerous advantages including speed of acquisition, is noninvasive, and has sufficient spatial and temporal resolution. Journal of Cerebral Blood Flow & Metabolism (2014) 34, 1111–1116; doi:10.1038/jcbfm.2014.59; published online 23 April 2014 Keywords: acute stroke imaging; frequency mapping; hemodynamic stress; MR-OEF; oxygen extraction fraction; PARSE
INTRODUCTION Ischemic strokes are the leading cause of long-term disability and the third leading cause of death in the United States,1 representing a significant societal and health-care cost. It has been shown in positron emission tomography (PET) studies that cerebral oxygen extraction fraction (OEF) is an independent predictor of stroke.2–4 The management of stroke in both the acute and chronic phases has benefitted from an improved understanding of the pathophysiology of cerebral ischemia. In the acute setting, the recently reported DEFUSE II (Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution II) trial showed the feasibility of using magnetic resonance imaging (MRI) to perform rapid and reliable triage of patients for thrombolytic therapy.5 In contrast, the MR-RESCUE (Mechanical Retrieval and REcanalization of Stroke Clots Using Embolectomy) clinical trial6 showed inconclusive results for the use of imaging in the acute stroke setting. Thus, the dependence of perfusionweighted/diffusion-weighted imaging mismatch for triage has led to both positive and negative results in large multicenter trials. We believe that current paradigms for identifying tissue at risk using currently available imaging tools would benefit from the inclusion of OEF. In fact, the NIH (National Institutes of Health) and the NINDS (National Institute of Neurological Disorders and Stroke) has recommended researchers ‘conduct poststroke (acute and chronic) imaging studies to understand cerebral hemodynamics, collateral flow, oxygenation, and brain metabolism effects on tissue’.7 There is a critical clinical need for the development of a fast, noninvasive method for direct imaging of tissue oxygenation.
A number of studies have shown that PET imaging using radiolabeled water O15 can be used to measure OEF reliably.8,9 While PET is the gold standard for OEF measurement,8 it has a number of drawbacks that prevent its widespread adoption; including infrastructure requirements (PET scanner, cyclotron in the vicinity for rapid delivery of O15), high cost per scan, the use of ionizing radiation, and limited spatial resolution. The MR-based techniques that have been developed to measure OEF may be divided into three classes. First, BOLD-based approaches are based on the biophysical model of spin dephasing in the presence of inhomogeneities proposed by Yablonskiy and Haacke.10 The technique, including the quantitative BOLD technique, was implemented and improved over the past decade.11–14 A drawback of these techniques is that it requires fitting a large number of parameters or requires the measurement of multiple parameters to derive OEF. Second, using MR-phase information to measure oxygen saturation (SO2) quantification has been shown.15,16 One of the drawbacks of these techniques is that it only provides global measurements and focal changes cannot be measured. A third set of techniques use intravascular T2 to measure oxygenation by using a model that establishes a relation between intravascular T2, deoxyhemoglobin induced signal loss, and hematocrit. Methods developed that use intravascular T2 include TRUST (T2 Relaxation under Spin-Tagging),17 and QUIXOTIC (QUantitative Imaging of eXtraction Of Oxygen and TIssue Consumption).18 The main drawback with these techniques is the difficulty to measure venous blood signal that is free of extravascular contribution.
1 Department of Radiology, Feinberg School of Medicine, Northwestern University, Chicago, Illinois, USA; 2Department of Neuroscience, Brown University, Providence, Rhode Island, USA; 3Deparment of Biomedical Engineering, University of Alabama at Birmingham, Birmingham, Alabama, USA; 4Department of Neurosurgery, UT Southwestern, Dallas, Texas, USA and 5Department of Biomedical Engineering, Northwestern University, Chicago, IL, USA. Correspondence: Professor TJ Carroll, Department of Radiology, Feinberg School of Medicine, Northwestern University Chicago, 737N Michigan Avenue, Suite 1600, Chicago, IL 60611, USA. E-mail: [email protected] Received 17 November 2013; accepted 12 March 2014; published online 23 April 2014
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1112 The techniques described above while promising cannot be performed rapidly enough for translation to the acute stroke setting, where a typical imaging sequence is less than 2 minutes. In an application like stroke, speed of acquisition and noninvasiveness are desirable characteristics. In this paper, we test the hypothesis that we can use a long readout rapid frequency mapping technique to derive OEF. We use a non-Cartesian rosette Parameter Assessment by Retrieval from Signal Encoding (PARSE) acquisition and test it on 11 normal subjects and show proof of concept on a patient with hemodynamic stress. MATERIALS AND METHODS This investigation was fully HIPAA (Health Insurance Portability and Accountability Act) compliant and was approved by Northwestern University’s IRB (Institutional Review Board).
Theory Parameter Assessment by Retrieval from Signal Encoding. Parameter Assessment by Retrieval from Signal Encoding is a multiparametric imaging technique that is similar to the concept of MRI Fingerprinting.19 It extracts relaxation rate (R2*), local frequency change (do), and proton density maps (M0) from a single free induction decay (FID), by encoding the above parameters in sequence and extending the signal readout to 65 ms. It uses a more accurate model of the MR process accounting for the decay and local frequency changes occurring during the acquisition process. A detailed description of the method is reported previously.20,21 The discretized MR signal model is given by Sn ¼
XX
M0 ð~ rÞe ½ðR2 ð~rÞ þ ioð~rÞÞn cnr
ð1Þ
where M0(r) is the initial transverse magnetization at location given by r, ~ R2*(r) is the local decay rate and o(r) is the local frequency cnr ¼ e 2ipkn ~xr is the spatial modulating function, where kn and xr are vectors for k-locations and pixel locations, respectively. To estimate multiple parameters M0(r), R2*(r), and do(r) from a single echo Sn, an inverse estimation technique is employed. Parameter Assessment by Retrieval from Signal Encoding uses a PLCG (Progressive Length Conjugate Gradient) algorithm to solve equation 1. A least squares estimate p is obtained by minimizing the residual between the observed signal Sn and _ pÞ. the model estimates Sn ð~ X _ pÞÞ2 ð2Þ Sð~ pÞ ¼ n ðSn Sn ð~ Oxygen extraction fraction measurement. The MR signal decay in the static dephasing regime, which accounts for the presence of static magnetic field inhomogeneities was described by Yablonskiy10 SðtÞ / expð R2 ðtÞÞ expð idotÞ
ð3Þ
The signal can be modeled as consisting of two parts, a dissipative component and an oscillatory component SðtÞ / expð R2 ðtÞÞ
ð4Þ
SðtÞ / expð idotÞ
ð5Þ
As described by the theory by Yablonskiy and others,10,14,22 the reversible 0 signal relaxation rate R2 is given by 0
R2 ¼ DBVdo
ð6Þ
For TE4tc, where tc is the critical time given by tc ¼ 1/do and deoxygenated blood volume (DBV) is the fraction of the blood volume carrying deoxygenated blood. To calculate the characteristic frequency do, 0 the current quantitative BOLD theories require the estimation of the R2 decay, and knowledge of DBV. In our approach, we directly estimate the oscillatory component, given by equation 5, that is manifested as a local frequency change. These local frequency changes are a surrogate marker for OEF. We ensure that TE is 0 greater than the critical time tc,10 and that the relaxation rate R2 is linearly proportional to the frequency change do, such that equation 5 holds true. We calculate the critical time tc assuming a normal oxygen saturation (Y) of 60%. The calculated tc at 3T is 9.8 ms. Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
The equation that relates OEF to measureable frequency shifts is given by, 4 do ¼ g pDw0 HctOEFB0 3
ð7Þ
where do is the deoxyhemoglobin induced frequency shift; Dw0 is the susceptibility difference between oxygenated and deoxygenated blood. Hct is the hematocrit, and B0 is the magnetic field strength in Tesla. To calculate MR-OEF, the local frequency is directly estimated in each pixel in the PARSE technique. These frequency maps using PARSE are quantitative maps of local off-resonance frequency (do) from the resonant frequency (o0). We use these do maps for the calculation of MR-OEF. The frequency shift that we measure is relative to the resonance frequency of water, which is in the vicinity of a blood vessel with deoxygenated blood to that of free water. A vendor provided advanced shim procedure is used to reduce static field homogeneities. Parameter Assessment by Retrieval from Signal Encoding is well described here.16,20,21,23–25 The rationale behind using the PARSE sequence is to measure OEF-related frequency changes (do) instead of another frequency mapping technique is that if we were to use a Gradient echo frequency map with a long readout time, it would result in signal loss and distortion. Parameter Assessment by Retrieval from Signal Encoding allows us the ability to use a long readout time (65 ms), and iteratively reconstruct the FID with a sparse data set and no distortion. It has high sensitivity to local frequency changes because the PARSE rosette trajectory samples the center of k-space very often. To account for the effect of magnetic field inhomogeneities on the calculated frequency maps, care is taken to use a uniform magnetic field to minimize effects of field inhomogeneities. Bulk magnetic susceptibilities such as air–tissue interfaces present significant challenges to the current technique. Methods to reduce the impact of large inhomogeneities are currently being investigated. Numerical simulations and sensitivity analysis. We designed nominal rosette trajectories and simulated a numerical phantom with input values for offset frequency. The numerical phantom was sampled using the rosette trajectory and the PARSE reconstruction was used to assess the reconstruction of the algorithm. The results from the numerical simulation were used for the sensitivity analysis. Sensitivity analysis was performed to determine the feasibility of using PARSE to measure OEF. From a number of PET studies3,26,27 it has been established that the normal resting brain extracts 40% to 45% of the oxygen delivered to tissue. When vascular insufficiencies develop, the blood flow reduces, the total blood volume increases and consequently the oxygen extraction has to be increased to maintain metabolic demand.2,27,28 We expect that a 4% to 5% increase in OEF would need to be accurately detected to measure mild vascular insufficiencies. This translates to a frequency change (do) of B4 Hz. The OEF increase would be much more (450%) for severe cases,29 a threshold of 13% increase OEF was used for patient selection in the COSS (Carotid Occlusion Surgery Study) study.30 It is known that the offset frequencies in the range shown (10 to 45 Hz) can be estimated using PARSE.23 Also, PARSE is more well suited to estimate frequency as the inherent bias errors and random errors are the smallest for frequency estimates.21 From the sensitivity analysis and supplemental advantages listed above, we conclude that the PARSE technique can be used to measure oxygen extraction.23 Parameter Assessment by Retrieval from Signal Encoding sequence design and optimization. The rosette trajectory was designed and simulated using numerical simulations.21 The rosette trajectory,31 kt ¼ kf cos(o1t)e( io2t), characterized by fast oscillating frequency (o1) and a slow rotating frequency(o2). The o1 frequency was calculated as 3874.8 Hz, and o2 frequency was 1610.8 Hz. Based on constraints of maximum gradient strength, slew rate limits, dwell times, and desired spatial resolution the gradient trajectories were designed. Numerical simulations were performed to test the designed trajectories, to test the reconstruction algorithm, and to aid in the sensitivity analysis. The rosette trajectory as implemented is shown in Figure 1A. The PARSE sequence as shown in Figure 1B is implemented on a Siemens 3T Tim Trio whole body clinical MR scanner (Siemens Medical Systems, Erlangen, Germany). The maximum readout gradient of the sequence was 28.7 mT/m and the peak slew rate was 157.71 mT/m per ms, with a k-space radius of 2.82/cm. Programming of the sequence was performed in the Siemens Integrated Development Environment for Applications (IDEA) programming environment. The implemented PARSE & 2014 ISCBFM
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1113
Figure 1. (A) A rosette trajectory is designed for use in the Parameter Assessment by Retrieval from Signal Encoding (PARSE) sequence. (B) PARSE sequence implemented on the 3T magnetic resonance imaging (MRI) scanner for measurement of local frequency change (do). ADC, analog to digital converter; FID, free induction decay; TR, repetition time.
sequence acquires a single slice in a 5-second acquisition. The FID is read for a 65-ms time window while a 4.9-second delay is set between measurements to ensure that the signal decays before the next acquisition is taken. The TR is 5 seconds, slice thickness is 5 mm, and the in-plane resolution is 3 3 mm2. Calibration. A one-time extensive calibration procedure was performed to determine the actual rosette trajectory generated by the MRI instrument’s gradient system, including eddy current effects. The calibration procedure is necessary to account for drifting of sampled trajectory locations. The calibration would need to be repeated if there are any system or gradient hardware changes. A point phantom was used for the calibration and the method as detailed in Zhang et al22 was used. By changing only one gradient at a time (slice select (Gss) or phase encode (GPE)) and keeping the other gradient constant, signal samples were obtained. The calibration procedure produces a measurement of the k-trajectory, that is, the actual locations in k-space where the data samples are acquired. Test data were taken using a water phantom. A linear curve fitting procedure on the calibration data was performed to measure the implemented rosette trajectory (Figure 1A). Validation of the sequence was performed using a water phantom. The FID data were obtained with the parameters described above. The resulting calibrated trajectory was used for reconstruction of the parameter maps. Data acquisition. After IRB approval, we acquired data on 11 healthy human subjects as controls. In the series of 11 normal volunteers (M/F 6/5, oage4 ¼ 26±10 years) we acquired single slice, 5.0 mm thick, 220 mm 220 mm field of view, 64 64 matrix, resolution ¼ 3 3 5.0 mm3 2D PARSE images. The total sequence time was 5 seconds. A T1weighted MPRAGE acquisition of the same slice was also taken for anatomic reference. To illustrate proof of principle, we acquired a PARSE MR-OEF acquisition on a patient with an arteriovenous malformation (AVM). Quantitative cerebral blood flow (qCBF) and quantitative cerebral blood volume maps were also acquired in the AVM patient using a dynamic susceptibility weighted T1 mapping sequence developed previously32 by our group. Bilateral regions of interest (ROIs) were drawn to calculate the mean blood flows, blood volume, and OEF values. Reconstruction. The PARSE data were exported and reconstructed offline using MATLAB routines. The reconstruction algorithm used an iterative, nonlinear least-squares algorithm to extract M0, R2*, and local offset frequency (do) information from the acquired data. To calculate OEF values, we established a linear relationship (from equation 3 and sensitivity analysis) between frequency change (do) and OEF. In our study, we assumed a constant value of Hematocrit of 18 p.p.m.33 & 2014 ISCBFM
RESULTS Figure 2 shows the results of the numerical simulations. The M0 image (Figure 2A) and the frequency image were estimated. To better visualize the estimated values in the frequency map, a profile through the center (along white line in Figure 2A) is plotted in Figure 2B. The sensitivity analysis is shown in Figure 2C). Table 1 shows a reference PET-OEF values from normal subjects.4,34–36 Figure 3 shows OEF calculated from a representative control subject. An anatomic MR image of the same slice is included for reference (Figure 3A), Figure 3B shows the calculated OEF using PARSE-MR method. The average frequency change (do) across the slice was 17.56±4.12%, and the average OEF across the slice was 36.58±8.58%. The mean slice average across 11 control subjects gave a frequency change of 17.6±3.75%. This corresponded to an average whole-brain slice OEF of 36.66±7.82% across the control subjects. This shows excellent agreement with normative PET literature values (Table 1). Figure 4 shows the results of our study in the AVM patient. Figure 4A is anatomic slice through the AVM location. Figure 4B shows a qCBF map. The ROIs shown in the figure are drawn to compare blood flows, and blood volumes on contralateral sides. The distal side of the brain has a mean CBF value of 48±35 mL/ 100 g per minute and the proximal side is 19.9±13.8 mL/100 g per minute. The calculated CBV for the same ROIs had a twofold increase proximal to the AVM, with distal side 2.10±1.6 mL/100 g and the proximal side 4.8±2.9 mL/100 g. The ROI proximal to the AVM location shows a significantly lower blood flow suggesting a ‘steal effect’.37 Figure 4C shows the calculated OEF map. An ROI drawn proximal to the AVM location shows a high OEF value of 82.08±6.6% (Table 2). The low flow rate and high OEF values calculated in the ROI suggest that the area adjacent to the AVM location is under hemodynamic stress.
DISCUSSION We found that direct measurement of local frequency changes with PARSE agrees with physiologic expectations calculated from an existing model.10 While desirable qualities of a diagnostic technique to measure OEF include a fast, safe, inexpensive, and reliable technique with sufficient spatial and temporal resolution, none of the existing techniques have filled this need for use in the acute stroke imaging setting. The PARSE MR-OEF technique has a Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1114 VALIDATION OF PARSE USING NUMERICAL SIMULATION
B
Estimated frequency profile 20 15
freq (Hz)
C
Sensitivity Analysis
True Values Frequency Change (δω) to be detected
A Estimated Magnitude |M0|
10 5 0
Normal OEF Range
Pathologic OEF
–5 0
50 100 Column Number
Range of OEF (%) that can be measured
Figure 2. Parameter estimates from numerical simulations using Parameter Assessment by Retrieval from Signal Encoding (PARSE). Numerical simulations were performed to test the ability of the CG-based least-squares reconstruction algorithm to accurately measure local frequency. (A) Parameter map of estimated initial transverse magnetization. (B) A profile (along white line in (A)) along the center of the estimated frequency map. Here, the outer annulus was set to 10 Hz, and the center disc to 15 Hz. The rest were left at 0 Hz (on resonance). (C) A sensitivity analysis for oxygen extraction fraction (OEF) measurement shows that to detect an OEF change of 5% the PARSE technique would need to detect a change of 4 Hz or more. The OEF increase would be much more (420%) for severe cases.
Table 1.
Comparison of PET literature values in normal subjects and PARSE MR-OEF
Subject no 1 2 3 4 5 6 7 8 9 10 11 Mean±s.d. across subjects AVM ROI
PARSE MR-OEF (%) (mean±s.d.)
Frequency change (do) (Hz) (mean±s.d.)
41.29±8.79 36.58±8.58 39.77±6.81 28.42±13.44 41.14±8.83 47.94±14.17 48.70±24.03 29.51±8.6 34.73±11.26 27.36±13.82 27.79±14.29
19.83±4.22 17.56±4.12 19.10±3.27 13.65±6.46 19.76±4.24 23.02±6.8 23.39±11.54 14.17±4.12 16.68±5.41 13.14±6.6 13.35±6.86
36.66±7.82
17.6±3.75
82.08±6.6
39.42±3.17
PET literature study 35
Carpenter et al Yamauchi et al27 Diringer et al36 Raichle et al37
PET OEF (%) (mean±s.d.) 35±7% 42±5% 41±6% 40±9%
AVM, arteriovenous malformation; MR, magnetic resonance; OEF, oxygen extraction fraction; PARSE, Parameter Assessment by Retrieval from Signal Encoding; PET, positron emission tomography; ROIs, regions of interest. Normative literature values from a number of PET studies are shown on the right side. The right side shows the mean whole-brain OEF for the five normal subjects and ROI drawn proximal to AVM location.
Figure 3. (A) The anatomic magnetic resonance imaging (MRI) of a normal subject is shown. (B) MR-OEF in a representative normal human brain calculated using Parameter Assessment by Retrieval from Signal Encoding (PARSE). The average whole-brain OEF in this subject was 36.58±8.58%. The mean OEF calculated across 11 subjects was 36.66±7.82%. OEF, oxygen extraction fraction. Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
number of unique advantages over available PET techniques and current MR-OEF measurement techniques. First, MRI is relatively inexpensive and widely available compared with the PET counterparts. Second, speed augurs well for this technique enabling it for potential development in a stroke imaging protocol. Third, this technique is noninvasive and has the advantage of allowing repeated testing of ‘at risk’ patients to monitor their vascular impairment without concerns relating to ionizing radiation dose. The sensitivity of the PARSE technique to local frequency changes is o4 Hz. The sensitivity as obtained in other studies is B4 to 15 Hz.14,38 The PARSE MR-OEF technique assumes a single homogenous lorentzian in each voxel. The reconstruction estimates a single frequency value in each voxel and is susceptible to shifts due to nonlorentzian effects. Such nonlorentzian effects may be a source of error in some individual voxels. We do not expect it to be a significant issue due to the fact that we are looking for large changes in OEF on contralateral sides. This technique achieves a higher spectral resolution due to a high sampling bandwidth in part due to the use of the rosette trajectory that has an increased density of & 2014 ISCBFM
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1115
Figure 4. (A) A large occipital arteriovenous malformation (AVM) induces vascular steal. (B) A quantitative CBF map of the AVM patient is shown. Regions of interest (ROIs) drawn on contralateral sides show reduced blood flow in the region proximal to the AVM location. The presence of steal is confirmed by magnetic resonance (MR) perfusion images which reflect a twofold reduction in perfusion to the adjacent region. (C) The ROI with low CBF also exhibits low CBV and high oxygen extraction fraction (OEF) values, suggesting that the area proximal to the AVM location is under hemodynamic stress. The expected increase in OEF is showed in the proof of principle study.
Table 2.
Results in the patient illustrating hemodynamic stress
qCBF (mL/100 g per minute) qCBV(mL/100 g) OEF (%)
ROI distal to AVM
ROI proximal to AVM
48±35 2.0±1.6 52±5.86
19.9±13.8 4.8±2.9 82.08±6.6
AVM, arteriovenous malformation; OEF, oxygen extraction fraction; qCBF, quantitative cerebral blood flow; qCBV, quantitative cerebral blood volume; ROIs, regions of interest. Comparison of perfusion parameters in distal and proximal ROIs to the AVM location suggest that the area proximal to the AVM is under hemodynamic stress.
sampling near the origin (B1/ms). Since a large proportion of the object’s energy is concentrated in the center suggests that frequent sampling of this energy results in higher effective signal to noise ratio of the sampled echo in the frequency domain. The reconstruction uses an iterative reconstruction technique, which temporally divides data into bins. The least squares algorithm iterates over a bin of data to the desired tolerance before adding the next bin of data. This ensures that the algorithm converges as it works through the data and the cost function in the frequency parameter remains convex. The results from Figure 3 suggest that the normative values obtained using the MR-PARSE technique agree with existing PET literature studies (Table 1). Values from other MR-based OEF studies report a range of similar values. Quantitative BOLD reports it as 47.9±7.2% in humans,14 Christen et al12 with a modified quantitative BOLD approach reports in rats with normoxia as 39.3±5.8%. Jain et al16 using the MR phase-based techniques report values of 64±4%, and intravascular T2-based QUIXOTIC study reported 26±2%18 in humans. An interesting feature to be noted is that the normal subjects show a small elevation of OEF in the occipital region, which is seen in PET studies as well.36 Alteration in perfusion parameters under hemodynamic stress is well documented.2,3,27,29 According to their models a patient in stage II would typically exhibit increased CBV, decreased CBF, and elevated OEF. The results shown in the AVM patient line up with & 2014 ISCBFM
these conditions (increased CBV, decreased CBF, and elevated OEF) suggesting that the area proximal to the AVM is in stage II failure. Thus, the vascular steal occurring proximal to the AVM is resulting in hemodynamic stress and local tissue adjacent to the AVM may be at risk. It is interesting to note that the shunting of oxygenated blood through the high flow AVM do not result in abnormal OEF values over the nidus. This may be because the OEF calculation model assumes a vascular system consisting of randomly oriented infinitely long cylinders where exchange of oxygen takes place with surrounding tissue. Inside the AVM the venous shunting causes the OEF model assumptions to break down and the calculations are no longer valid within the AVM, hence we ignore the OEF values within the AVM. The strong R/L asymmetries in this illustrative case are similar to published PET study results showing R/L asymmetry in patients with carotid occlusion3 As shown in Table 2 and Figure 4, the proximal and distal sides to the AVM show marked differences in perfusion parameters, including OEF values. The PARSE MR-OEF technique is not without limitations. Background field inhomogeneities have the potential to cause local frequency shifts, which may result in artifactually high OEF values. To mitigate this issue, the vendor’s advanced Shimming technique was used to ensure that inhomogeneities are minimized. A potential confounder may be the focal deposition of Fe (e.g., hemosiderin from hemorrhage). In these cases, MR-OEF images must be reviewed along with standard GRE images which are reference standard image for the detection of hemorrhage. It is anticipated that the proposed MR-OEF approach here will have limited use in the evaluation of complex changes such as neovascularity in brain tumors. The MR-OEF has the potential to be complementary addition as part of a comprehensive analysis of cerebral physiology in neurovascular disease, which includes perfusion, anatomy, and vascular imaging. While it is beyond the scope of this implementation, it is possible to use spin-echo PARSE39 to use R2’ information as a validation technique. The PARSE technique that is implemented uses a conjugate gradient-based least squares minimization to estimate multiple encoded parameters from the FID signal. Currently, this is a computationally intensive technique. The goal of the study was not, however, to optimize the reconstruction time. Several strategies may be employed to reduce reconstruction times to make it feasible in a clinical setting. Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1116 Parallelizing the reconstruction code, using GPU-based processors, and algorithmic improvements can potentially reduce the reconstruction time by a few orders of magnitude. This technique assumes a constant value of hematocrit. Future versions of this experiment may get this information during routine blood draws. In Figure 4C, a sinus susceptibility effect is seen to reflect in the OEF map near the orbitofrontal cortex. This is caused by bulk magnetic susceptibilities at air–tissue interfaces. It is challenging to remove these effects with our current implementation. Among the normal subjects, there were five subjects with below normal OEF values, which may have been a result from artifacts caused by large susceptibility gradients from the ear canals. A limitation of this technique is the inability to account for large susceptibility gradients caused by sinuses and ear-holes. An improved cardiac gated acquisition method used to remove static field inhomogeneities is being investigated. In conclusion, we have showed a rapid frequency mapping technique to calculate OEF using the PARSE MRI technique. Its primary advantages including speed, wide availability, noninvasiveness, and cost effectiveness make it suitable for use in an application like acute stroke imaging. Our preliminary results suggest strong agreement with PET and other MR-OEF studies. Further development and validation of this technique would be required. DISCLOSURE/CONFLICT OF INTEREST The authors declare no conflict of interest.
ACKNOWLEDGMENTS The authors would like to acknowledge the following funding sources for this research: NIH/NHLBI R01 HL088437; NIH/NINDS R01 NS049395-01A2; NIH/NIBIB T32 EB005170.
REFERENCES 1 Lloyd-Jones D, Adams R, Carnethon M, De Simone G, Ferguson TB, Flegal K et al. Heart disease and stroke statistics--2009 update: a report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Circulation 2009; 119: 480–486. 2 Derdeyn CP, Grubb Jr. RL, Powers WJ. Cerebral hemodynamic impairment: methods of measurement and association with stroke risk. Neurology 1999; 53: 251–259. 3 Derdeyn CP, Yundt KD, Videen TO, Carpenter DA, Grubb Jr. RL, Powers WJ. Increased oxygen extraction fraction is associated with prior ischemic events in patients with carotid occlusion. Stroke 1998; 29: 754–758. 4 Yamauchi H, Fukuyama H, Nagahama Y, Nabatame H, Ueno M, Nishizawa S et al. Significance of increased oxygen extraction fraction in five-year prognosis of major cerebral arterial occlusive diseases. J Nucl Med 1999; 40: 1992–1998. 5 Lansberg MG, Lee J, Christensen S, Straka M, De Silva DA, Mlynash M et al. RAPID automated patient selection for reperfusion therapy: a pooled analysis of the Echoplanar Imaging Thrombolytic Evaluation Trial (EPITHET) and the Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution (DEFUSE) Study. Stroke 2011; 42: 1608–1614. 6 Kidwell CS, Jahan R, Gornbein J, Alger JR, Nenov V, Ajani Z et al. A trial of imaging selection and endovascular treatment for ischemic stroke. N Engl J Med 2013; 368: 914–923. 7 Moskowitz MA, Grotta JC, Koroshetz WJ. The NINDS Stroke Progress Review Group final analysis and recommendations. Stroke 2013; 44: 2343–2350. 8 Mintun MA, Raichle ME, Martin WR, Herscovitch P. Brain oxygen utilization measured with O-15 radiotracers and positron emission tomography. J Nucl Med 1984; 25: 177–187. 9 Videen TO, Perlmutter JS, Herscovitch P, Raichle ME. Brain blood volume, flow, and oxygen utilization measured with 15O radiotracers and positron emission tomography: revised metabolic computations. J Cereb Blood Flow Metab 1987; 7: 513–516. 10 Yablonskiy DA, Haacke EM. Theory of NMR signal behavior in magnetically inhomogeneous tissues: the static dephasing regime. Magn Reson Med 1994; 32: 749–763. 11 An H, Lin W. Quantitative measurements of cerebral blood oxygen saturation using magnetic resonance imaging. J Cereb Blood Flow Metab 2000; 20: 1225–1236.
Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
12 Christen T, Lemasson B, Pannetier N, Farion R, Segebarth C, Remy C et al. Evaluation of a quantitative blood oxygenation level-dependent (qBOLD) approach to map local blood oxygen saturation. NMR Biomed 2011; 24: 393–403. 13 Fujita N, Shinohara M, Tanaka H, Yutani K, Nakamura H, Murase K. Quantitative mapping of cerebral deoxyhemoglobin content using MR imaging. NeuroImage 2003; 20: 2071–2083. 14 He X, Yablonskiy DA. Quantitative BOLD: mapping of human cerebral deoxygenated blood volume and oxygen extraction fraction: default state. Magn Reson Med 2007; 57: 115–126. 15 Fernandez-Seara MA, Techawiboonwong A, Detre JA, Wehrli FW. MR susceptometry for measuring global brain oxygen extraction. Magn Reson Med 2006; 55: 967–973. 16 Jain V, Langham MC, Wehrli FW. MRI estimation of global brain oxygen consumption rate. J Cereb Blood Flow Metab 2010; 30: 1598–1607. 17 Lu H, Ge Y. Quantitative evaluation of oxygenation in venous vessels using T2-Relaxation-Under-Spin-Tagging MRI. Magn Reson Med 2008; 60: 357–363. 18 Bolar DS, Rosen BR, Sorensen AG, Adalsteinsson E. QUantitative Imaging of eXtraction of oxygen and TIssue consumption (QUIXOTIC) using venular-targeted velocity-selective spin labeling. Magn Reson Med 2011; 66: 1550–1562. 19 Ma D, Gulani V, Seiberlich N, Liu K, Sunshine JL, Duerk JL et al. Magnetic resonance fingerprinting. Nature 2013; 495: 187–192. 20 Twieg DB. Parsing local signal evolution directly from a single-shot MRI signal: a new approach for fMRI. Magn Reson Med 2003; 50: 1043–1052. 21 Twieg DB, Reeves SJ. Basic properties of SS-PARSE parameter estimates. IEEE Trans Med Imaging 2010; 29: 1156–1172. 22 He X, Zhu M, Yablonskiy DA. Validation of oxygen extraction fraction measurement by qBOLD technique. Magn Reson Med 2008; 60: 882–888. 23 Menon RG, Walsh EG, Twieg DB, Sutton BP, Carroll TJ. A novel pulse sequence to measure oxygen extraction fraction in the brain using parameter assessment using retrieval from signal encoding (PARSE) techniques. Proc Intl Soc Magn Reson Med, Melbourne, Australia, 2012. 24 Zuo J, Bolding M, Twieg DB. Validation of V-SS-PARSE for single-shot flow measurement. Magn Reson Imaging 2007; 25: 335–340. 25 Zuo J, Walsh EG, Deutsch G, Twieg DB. Rapid mapping of flow velocity using a new PARSE method. Magn Reson Med 2006; 55: 147–152. 26 Yamauchi H, Fukuyama H, Nagahama Y, Nabatame H, Nakamura K, Yamamoto Y et al. Evidence of misery perfusion and risk for recurrent stroke in major cerebral arterial occlusive diseases from PET. J Neurol Neurosurg Psychiatry 1996; 61: 18–25. 27 Derdeyn CP, Videen TO, Yundt KD, Fritsch SM, Carpenter DA, Grubb RL et al. Variability of cerebral blood volume and oxygen extraction: stages of cerebral haemodynamic impairment revisited. Brain 2002; 125(Pt 3): 595–607. 28 Grubb Jr. RL, Derdeyn CP, Fritsch SM, Carpenter DA, Yundt KD, Videen TO et al. Importance of hemodynamic factors in the prognosis of symptomatic carotid occlusion. JAMA 1998; 280: 1055–1060. 29 Nemoto EM, Yonas H, Chang Y. Stages and thresholds of hemodynamic failure. Stroke 2003; 34: 2–3. 30 Powers WJ, Clarke WR, Grubb Jr. RL, Videen TO, Adams Jr. HP, Derdeyn CP. Extracranial-intracranial bypass surgery for stroke prevention in hemodynamic cerebral ischemia: the Carotid Occlusion Surgery Study randomized trial. JAMA 2011; 306: 1983–1992. 31 Noll DC. Multishot rosette trajectories for spectrally selective MR imaging. IEEE Trans Med Imaging 1997; 16: 372–377. 32 Srour JM, Shin W, Shah S, Sen A, Carroll TJ. SCALE-PWI: a pulse sequence for absolute quantitative cerebral perfusion imaging. J Cereb Blood Flow Metab 2011; 31: 1272–1282. 33 Weisskoff RM, Kiihne S. MRI susceptometry: image-based measurement of absolute susceptibility of MR contrast agents and human blood. Magn Reson Med 1992; 24: 375–383. 34 Carpenter DA, Grubb Jr. RL, Tempel LW, Powers WJ. Cerebral oxygen metabolism after aneurysmal subarachnoid hemorrhage. J Cereb Blood Flow Metab 1991; 11: 837–844. 35 Diringer MN, Yundt K, Videen TO, Adams RE, Zazulia AR, Deibert E et al. No reduction in cerebral metabolism as a result of early moderate hyperventilation following severe traumatic brain injury. J Neurosurg 2000; 92: 7–13. 36 Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman GL. A default mode of brain function. Proc Natl Acad Sci USA 2001; 98: 676–682. 37 Poublanc J, Han JS, Mandell DM, Conklin J, Stainsby JA, Fisher JA et al. Vascular steal explains early paradoxical blood oxygen level-dependent cerebrovascular response in brain regions with delayed arterial transit times. Cerebrovasc Dis Extra 2013; 3: 55–64. 38 Sutton BP, Noll DC, Fessler JA. Dynamic field map estimation using a spiral-in/ spiral-out acquisition. Magn Reson Med 2004; 51: 1194–1204. 39 Li N, Bolding M, Twieg DB. Spin-echo SS-PARSE: a PARSE MRI method to estimate frequency, R(2) and R(2)(’) in a single shot. Magn Reson Imaging 2010; 28: 1270–1282.
& 2014 ISCBFM
RAPID COMMUNICATION
Snapshot MR technique to measure OEF using rapid frequency mapping Rajiv G Menon1, Edward G Walsh2, Donald B Twieg3, Charles G Cantrell1, Parmede Vakil1, Sumeeth V Jonathan1, Hunt H Batjer4 and Timothy J Carroll1,5 Magnetic resonance (MR)-based oxygen extraction fraction (OEF) measurement techniques that use blood oxygen level-dependent (BOLD)-based approaches require the measurement of the R2’ decay rate and deoxygenated blood volume to derive the local oxygen saturation in vivo. We describe here a novel approach to measure OEF using rapid local frequency mapping. By modeling the MR decay process in the static dephasing regime as two separate dissipative and oscillatory effects, we calculate the OEF from local frequencies measured across the brain by assuming that the biophysical mechanisms causing OEF-related frequency changes can be determined from the oscillatory effects. The Parameter Assessment by Retrieval from Signal Encoding (PARSE) technique was used to acquire the local frequency change maps. The PARSE images were taken on 11 normal volunteers, and 1 patient exhibiting hemodynamic stress. The mean MR-OEF in 11 normal subjects was 36.66±7.82%, in agreement with positron emission tomography (PET) literature. In regions of hemodynamic stress induced by vascular steal, OEF exhibits the predicted focal increases. These preliminary results show that it is possible to measure OEF using a rapid frequency mapping technique. Such a technique has numerous advantages including speed of acquisition, is noninvasive, and has sufficient spatial and temporal resolution. Journal of Cerebral Blood Flow & Metabolism (2014) 34, 1111–1116; doi:10.1038/jcbfm.2014.59; published online 23 April 2014 Keywords: acute stroke imaging; frequency mapping; hemodynamic stress; MR-OEF; oxygen extraction fraction; PARSE
INTRODUCTION Ischemic strokes are the leading cause of long-term disability and the third leading cause of death in the United States,1 representing a significant societal and health-care cost. It has been shown in positron emission tomography (PET) studies that cerebral oxygen extraction fraction (OEF) is an independent predictor of stroke.2–4 The management of stroke in both the acute and chronic phases has benefitted from an improved understanding of the pathophysiology of cerebral ischemia. In the acute setting, the recently reported DEFUSE II (Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution II) trial showed the feasibility of using magnetic resonance imaging (MRI) to perform rapid and reliable triage of patients for thrombolytic therapy.5 In contrast, the MR-RESCUE (Mechanical Retrieval and REcanalization of Stroke Clots Using Embolectomy) clinical trial6 showed inconclusive results for the use of imaging in the acute stroke setting. Thus, the dependence of perfusionweighted/diffusion-weighted imaging mismatch for triage has led to both positive and negative results in large multicenter trials. We believe that current paradigms for identifying tissue at risk using currently available imaging tools would benefit from the inclusion of OEF. In fact, the NIH (National Institutes of Health) and the NINDS (National Institute of Neurological Disorders and Stroke) has recommended researchers ‘conduct poststroke (acute and chronic) imaging studies to understand cerebral hemodynamics, collateral flow, oxygenation, and brain metabolism effects on tissue’.7 There is a critical clinical need for the development of a fast, noninvasive method for direct imaging of tissue oxygenation.
A number of studies have shown that PET imaging using radiolabeled water O15 can be used to measure OEF reliably.8,9 While PET is the gold standard for OEF measurement,8 it has a number of drawbacks that prevent its widespread adoption; including infrastructure requirements (PET scanner, cyclotron in the vicinity for rapid delivery of O15), high cost per scan, the use of ionizing radiation, and limited spatial resolution. The MR-based techniques that have been developed to measure OEF may be divided into three classes. First, BOLD-based approaches are based on the biophysical model of spin dephasing in the presence of inhomogeneities proposed by Yablonskiy and Haacke.10 The technique, including the quantitative BOLD technique, was implemented and improved over the past decade.11–14 A drawback of these techniques is that it requires fitting a large number of parameters or requires the measurement of multiple parameters to derive OEF. Second, using MR-phase information to measure oxygen saturation (SO2) quantification has been shown.15,16 One of the drawbacks of these techniques is that it only provides global measurements and focal changes cannot be measured. A third set of techniques use intravascular T2 to measure oxygenation by using a model that establishes a relation between intravascular T2, deoxyhemoglobin induced signal loss, and hematocrit. Methods developed that use intravascular T2 include TRUST (T2 Relaxation under Spin-Tagging),17 and QUIXOTIC (QUantitative Imaging of eXtraction Of Oxygen and TIssue Consumption).18 The main drawback with these techniques is the difficulty to measure venous blood signal that is free of extravascular contribution.
1 Department of Radiology, Feinberg School of Medicine, Northwestern University, Chicago, Illinois, USA; 2Department of Neuroscience, Brown University, Providence, Rhode Island, USA; 3Deparment of Biomedical Engineering, University of Alabama at Birmingham, Birmingham, Alabama, USA; 4Department of Neurosurgery, UT Southwestern, Dallas, Texas, USA and 5Department of Biomedical Engineering, Northwestern University, Chicago, IL, USA. Correspondence: Professor TJ Carroll, Department of Radiology, Feinberg School of Medicine, Northwestern University Chicago, 737N Michigan Avenue, Suite 1600, Chicago, IL 60611, USA. E-mail: [email protected] Received 17 November 2013; accepted 12 March 2014; published online 23 April 2014
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1112 The techniques described above while promising cannot be performed rapidly enough for translation to the acute stroke setting, where a typical imaging sequence is less than 2 minutes. In an application like stroke, speed of acquisition and noninvasiveness are desirable characteristics. In this paper, we test the hypothesis that we can use a long readout rapid frequency mapping technique to derive OEF. We use a non-Cartesian rosette Parameter Assessment by Retrieval from Signal Encoding (PARSE) acquisition and test it on 11 normal subjects and show proof of concept on a patient with hemodynamic stress. MATERIALS AND METHODS This investigation was fully HIPAA (Health Insurance Portability and Accountability Act) compliant and was approved by Northwestern University’s IRB (Institutional Review Board).
Theory Parameter Assessment by Retrieval from Signal Encoding. Parameter Assessment by Retrieval from Signal Encoding is a multiparametric imaging technique that is similar to the concept of MRI Fingerprinting.19 It extracts relaxation rate (R2*), local frequency change (do), and proton density maps (M0) from a single free induction decay (FID), by encoding the above parameters in sequence and extending the signal readout to 65 ms. It uses a more accurate model of the MR process accounting for the decay and local frequency changes occurring during the acquisition process. A detailed description of the method is reported previously.20,21 The discretized MR signal model is given by Sn ¼
XX
M0 ð~ rÞe ½ðR2 ð~rÞ þ ioð~rÞÞn cnr
ð1Þ
where M0(r) is the initial transverse magnetization at location given by r, ~ R2*(r) is the local decay rate and o(r) is the local frequency cnr ¼ e 2ipkn ~xr is the spatial modulating function, where kn and xr are vectors for k-locations and pixel locations, respectively. To estimate multiple parameters M0(r), R2*(r), and do(r) from a single echo Sn, an inverse estimation technique is employed. Parameter Assessment by Retrieval from Signal Encoding uses a PLCG (Progressive Length Conjugate Gradient) algorithm to solve equation 1. A least squares estimate p is obtained by minimizing the residual between the observed signal Sn and _ pÞ. the model estimates Sn ð~ X _ pÞÞ2 ð2Þ Sð~ pÞ ¼ n ðSn Sn ð~ Oxygen extraction fraction measurement. The MR signal decay in the static dephasing regime, which accounts for the presence of static magnetic field inhomogeneities was described by Yablonskiy10 SðtÞ / expð R2 ðtÞÞ expð idotÞ
ð3Þ
The signal can be modeled as consisting of two parts, a dissipative component and an oscillatory component SðtÞ / expð R2 ðtÞÞ
ð4Þ
SðtÞ / expð idotÞ
ð5Þ
As described by the theory by Yablonskiy and others,10,14,22 the reversible 0 signal relaxation rate R2 is given by 0
R2 ¼ DBVdo
ð6Þ
For TE4tc, where tc is the critical time given by tc ¼ 1/do and deoxygenated blood volume (DBV) is the fraction of the blood volume carrying deoxygenated blood. To calculate the characteristic frequency do, 0 the current quantitative BOLD theories require the estimation of the R2 decay, and knowledge of DBV. In our approach, we directly estimate the oscillatory component, given by equation 5, that is manifested as a local frequency change. These local frequency changes are a surrogate marker for OEF. We ensure that TE is 0 greater than the critical time tc,10 and that the relaxation rate R2 is linearly proportional to the frequency change do, such that equation 5 holds true. We calculate the critical time tc assuming a normal oxygen saturation (Y) of 60%. The calculated tc at 3T is 9.8 ms. Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
The equation that relates OEF to measureable frequency shifts is given by, 4 do ¼ g pDw0 HctOEFB0 3
ð7Þ
where do is the deoxyhemoglobin induced frequency shift; Dw0 is the susceptibility difference between oxygenated and deoxygenated blood. Hct is the hematocrit, and B0 is the magnetic field strength in Tesla. To calculate MR-OEF, the local frequency is directly estimated in each pixel in the PARSE technique. These frequency maps using PARSE are quantitative maps of local off-resonance frequency (do) from the resonant frequency (o0). We use these do maps for the calculation of MR-OEF. The frequency shift that we measure is relative to the resonance frequency of water, which is in the vicinity of a blood vessel with deoxygenated blood to that of free water. A vendor provided advanced shim procedure is used to reduce static field homogeneities. Parameter Assessment by Retrieval from Signal Encoding is well described here.16,20,21,23–25 The rationale behind using the PARSE sequence is to measure OEF-related frequency changes (do) instead of another frequency mapping technique is that if we were to use a Gradient echo frequency map with a long readout time, it would result in signal loss and distortion. Parameter Assessment by Retrieval from Signal Encoding allows us the ability to use a long readout time (65 ms), and iteratively reconstruct the FID with a sparse data set and no distortion. It has high sensitivity to local frequency changes because the PARSE rosette trajectory samples the center of k-space very often. To account for the effect of magnetic field inhomogeneities on the calculated frequency maps, care is taken to use a uniform magnetic field to minimize effects of field inhomogeneities. Bulk magnetic susceptibilities such as air–tissue interfaces present significant challenges to the current technique. Methods to reduce the impact of large inhomogeneities are currently being investigated. Numerical simulations and sensitivity analysis. We designed nominal rosette trajectories and simulated a numerical phantom with input values for offset frequency. The numerical phantom was sampled using the rosette trajectory and the PARSE reconstruction was used to assess the reconstruction of the algorithm. The results from the numerical simulation were used for the sensitivity analysis. Sensitivity analysis was performed to determine the feasibility of using PARSE to measure OEF. From a number of PET studies3,26,27 it has been established that the normal resting brain extracts 40% to 45% of the oxygen delivered to tissue. When vascular insufficiencies develop, the blood flow reduces, the total blood volume increases and consequently the oxygen extraction has to be increased to maintain metabolic demand.2,27,28 We expect that a 4% to 5% increase in OEF would need to be accurately detected to measure mild vascular insufficiencies. This translates to a frequency change (do) of B4 Hz. The OEF increase would be much more (450%) for severe cases,29 a threshold of 13% increase OEF was used for patient selection in the COSS (Carotid Occlusion Surgery Study) study.30 It is known that the offset frequencies in the range shown (10 to 45 Hz) can be estimated using PARSE.23 Also, PARSE is more well suited to estimate frequency as the inherent bias errors and random errors are the smallest for frequency estimates.21 From the sensitivity analysis and supplemental advantages listed above, we conclude that the PARSE technique can be used to measure oxygen extraction.23 Parameter Assessment by Retrieval from Signal Encoding sequence design and optimization. The rosette trajectory was designed and simulated using numerical simulations.21 The rosette trajectory,31 kt ¼ kf cos(o1t)e( io2t), characterized by fast oscillating frequency (o1) and a slow rotating frequency(o2). The o1 frequency was calculated as 3874.8 Hz, and o2 frequency was 1610.8 Hz. Based on constraints of maximum gradient strength, slew rate limits, dwell times, and desired spatial resolution the gradient trajectories were designed. Numerical simulations were performed to test the designed trajectories, to test the reconstruction algorithm, and to aid in the sensitivity analysis. The rosette trajectory as implemented is shown in Figure 1A. The PARSE sequence as shown in Figure 1B is implemented on a Siemens 3T Tim Trio whole body clinical MR scanner (Siemens Medical Systems, Erlangen, Germany). The maximum readout gradient of the sequence was 28.7 mT/m and the peak slew rate was 157.71 mT/m per ms, with a k-space radius of 2.82/cm. Programming of the sequence was performed in the Siemens Integrated Development Environment for Applications (IDEA) programming environment. The implemented PARSE & 2014 ISCBFM
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1113
Figure 1. (A) A rosette trajectory is designed for use in the Parameter Assessment by Retrieval from Signal Encoding (PARSE) sequence. (B) PARSE sequence implemented on the 3T magnetic resonance imaging (MRI) scanner for measurement of local frequency change (do). ADC, analog to digital converter; FID, free induction decay; TR, repetition time.
sequence acquires a single slice in a 5-second acquisition. The FID is read for a 65-ms time window while a 4.9-second delay is set between measurements to ensure that the signal decays before the next acquisition is taken. The TR is 5 seconds, slice thickness is 5 mm, and the in-plane resolution is 3 3 mm2. Calibration. A one-time extensive calibration procedure was performed to determine the actual rosette trajectory generated by the MRI instrument’s gradient system, including eddy current effects. The calibration procedure is necessary to account for drifting of sampled trajectory locations. The calibration would need to be repeated if there are any system or gradient hardware changes. A point phantom was used for the calibration and the method as detailed in Zhang et al22 was used. By changing only one gradient at a time (slice select (Gss) or phase encode (GPE)) and keeping the other gradient constant, signal samples were obtained. The calibration procedure produces a measurement of the k-trajectory, that is, the actual locations in k-space where the data samples are acquired. Test data were taken using a water phantom. A linear curve fitting procedure on the calibration data was performed to measure the implemented rosette trajectory (Figure 1A). Validation of the sequence was performed using a water phantom. The FID data were obtained with the parameters described above. The resulting calibrated trajectory was used for reconstruction of the parameter maps. Data acquisition. After IRB approval, we acquired data on 11 healthy human subjects as controls. In the series of 11 normal volunteers (M/F 6/5, oage4 ¼ 26±10 years) we acquired single slice, 5.0 mm thick, 220 mm 220 mm field of view, 64 64 matrix, resolution ¼ 3 3 5.0 mm3 2D PARSE images. The total sequence time was 5 seconds. A T1weighted MPRAGE acquisition of the same slice was also taken for anatomic reference. To illustrate proof of principle, we acquired a PARSE MR-OEF acquisition on a patient with an arteriovenous malformation (AVM). Quantitative cerebral blood flow (qCBF) and quantitative cerebral blood volume maps were also acquired in the AVM patient using a dynamic susceptibility weighted T1 mapping sequence developed previously32 by our group. Bilateral regions of interest (ROIs) were drawn to calculate the mean blood flows, blood volume, and OEF values. Reconstruction. The PARSE data were exported and reconstructed offline using MATLAB routines. The reconstruction algorithm used an iterative, nonlinear least-squares algorithm to extract M0, R2*, and local offset frequency (do) information from the acquired data. To calculate OEF values, we established a linear relationship (from equation 3 and sensitivity analysis) between frequency change (do) and OEF. In our study, we assumed a constant value of Hematocrit of 18 p.p.m.33 & 2014 ISCBFM
RESULTS Figure 2 shows the results of the numerical simulations. The M0 image (Figure 2A) and the frequency image were estimated. To better visualize the estimated values in the frequency map, a profile through the center (along white line in Figure 2A) is plotted in Figure 2B. The sensitivity analysis is shown in Figure 2C). Table 1 shows a reference PET-OEF values from normal subjects.4,34–36 Figure 3 shows OEF calculated from a representative control subject. An anatomic MR image of the same slice is included for reference (Figure 3A), Figure 3B shows the calculated OEF using PARSE-MR method. The average frequency change (do) across the slice was 17.56±4.12%, and the average OEF across the slice was 36.58±8.58%. The mean slice average across 11 control subjects gave a frequency change of 17.6±3.75%. This corresponded to an average whole-brain slice OEF of 36.66±7.82% across the control subjects. This shows excellent agreement with normative PET literature values (Table 1). Figure 4 shows the results of our study in the AVM patient. Figure 4A is anatomic slice through the AVM location. Figure 4B shows a qCBF map. The ROIs shown in the figure are drawn to compare blood flows, and blood volumes on contralateral sides. The distal side of the brain has a mean CBF value of 48±35 mL/ 100 g per minute and the proximal side is 19.9±13.8 mL/100 g per minute. The calculated CBV for the same ROIs had a twofold increase proximal to the AVM, with distal side 2.10±1.6 mL/100 g and the proximal side 4.8±2.9 mL/100 g. The ROI proximal to the AVM location shows a significantly lower blood flow suggesting a ‘steal effect’.37 Figure 4C shows the calculated OEF map. An ROI drawn proximal to the AVM location shows a high OEF value of 82.08±6.6% (Table 2). The low flow rate and high OEF values calculated in the ROI suggest that the area adjacent to the AVM location is under hemodynamic stress.
DISCUSSION We found that direct measurement of local frequency changes with PARSE agrees with physiologic expectations calculated from an existing model.10 While desirable qualities of a diagnostic technique to measure OEF include a fast, safe, inexpensive, and reliable technique with sufficient spatial and temporal resolution, none of the existing techniques have filled this need for use in the acute stroke imaging setting. The PARSE MR-OEF technique has a Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1114 VALIDATION OF PARSE USING NUMERICAL SIMULATION
B
Estimated frequency profile 20 15
freq (Hz)
C
Sensitivity Analysis
True Values Frequency Change (δω) to be detected
A Estimated Magnitude |M0|
10 5 0
Normal OEF Range
Pathologic OEF
–5 0
50 100 Column Number
Range of OEF (%) that can be measured
Figure 2. Parameter estimates from numerical simulations using Parameter Assessment by Retrieval from Signal Encoding (PARSE). Numerical simulations were performed to test the ability of the CG-based least-squares reconstruction algorithm to accurately measure local frequency. (A) Parameter map of estimated initial transverse magnetization. (B) A profile (along white line in (A)) along the center of the estimated frequency map. Here, the outer annulus was set to 10 Hz, and the center disc to 15 Hz. The rest were left at 0 Hz (on resonance). (C) A sensitivity analysis for oxygen extraction fraction (OEF) measurement shows that to detect an OEF change of 5% the PARSE technique would need to detect a change of 4 Hz or more. The OEF increase would be much more (420%) for severe cases.
Table 1.
Comparison of PET literature values in normal subjects and PARSE MR-OEF
Subject no 1 2 3 4 5 6 7 8 9 10 11 Mean±s.d. across subjects AVM ROI
PARSE MR-OEF (%) (mean±s.d.)
Frequency change (do) (Hz) (mean±s.d.)
41.29±8.79 36.58±8.58 39.77±6.81 28.42±13.44 41.14±8.83 47.94±14.17 48.70±24.03 29.51±8.6 34.73±11.26 27.36±13.82 27.79±14.29
19.83±4.22 17.56±4.12 19.10±3.27 13.65±6.46 19.76±4.24 23.02±6.8 23.39±11.54 14.17±4.12 16.68±5.41 13.14±6.6 13.35±6.86
36.66±7.82
17.6±3.75
82.08±6.6
39.42±3.17
PET literature study 35
Carpenter et al Yamauchi et al27 Diringer et al36 Raichle et al37
PET OEF (%) (mean±s.d.) 35±7% 42±5% 41±6% 40±9%
AVM, arteriovenous malformation; MR, magnetic resonance; OEF, oxygen extraction fraction; PARSE, Parameter Assessment by Retrieval from Signal Encoding; PET, positron emission tomography; ROIs, regions of interest. Normative literature values from a number of PET studies are shown on the right side. The right side shows the mean whole-brain OEF for the five normal subjects and ROI drawn proximal to AVM location.
Figure 3. (A) The anatomic magnetic resonance imaging (MRI) of a normal subject is shown. (B) MR-OEF in a representative normal human brain calculated using Parameter Assessment by Retrieval from Signal Encoding (PARSE). The average whole-brain OEF in this subject was 36.58±8.58%. The mean OEF calculated across 11 subjects was 36.66±7.82%. OEF, oxygen extraction fraction. Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
number of unique advantages over available PET techniques and current MR-OEF measurement techniques. First, MRI is relatively inexpensive and widely available compared with the PET counterparts. Second, speed augurs well for this technique enabling it for potential development in a stroke imaging protocol. Third, this technique is noninvasive and has the advantage of allowing repeated testing of ‘at risk’ patients to monitor their vascular impairment without concerns relating to ionizing radiation dose. The sensitivity of the PARSE technique to local frequency changes is o4 Hz. The sensitivity as obtained in other studies is B4 to 15 Hz.14,38 The PARSE MR-OEF technique assumes a single homogenous lorentzian in each voxel. The reconstruction estimates a single frequency value in each voxel and is susceptible to shifts due to nonlorentzian effects. Such nonlorentzian effects may be a source of error in some individual voxels. We do not expect it to be a significant issue due to the fact that we are looking for large changes in OEF on contralateral sides. This technique achieves a higher spectral resolution due to a high sampling bandwidth in part due to the use of the rosette trajectory that has an increased density of & 2014 ISCBFM
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1115
Figure 4. (A) A large occipital arteriovenous malformation (AVM) induces vascular steal. (B) A quantitative CBF map of the AVM patient is shown. Regions of interest (ROIs) drawn on contralateral sides show reduced blood flow in the region proximal to the AVM location. The presence of steal is confirmed by magnetic resonance (MR) perfusion images which reflect a twofold reduction in perfusion to the adjacent region. (C) The ROI with low CBF also exhibits low CBV and high oxygen extraction fraction (OEF) values, suggesting that the area proximal to the AVM location is under hemodynamic stress. The expected increase in OEF is showed in the proof of principle study.
Table 2.
Results in the patient illustrating hemodynamic stress
qCBF (mL/100 g per minute) qCBV(mL/100 g) OEF (%)
ROI distal to AVM
ROI proximal to AVM
48±35 2.0±1.6 52±5.86
19.9±13.8 4.8±2.9 82.08±6.6
AVM, arteriovenous malformation; OEF, oxygen extraction fraction; qCBF, quantitative cerebral blood flow; qCBV, quantitative cerebral blood volume; ROIs, regions of interest. Comparison of perfusion parameters in distal and proximal ROIs to the AVM location suggest that the area proximal to the AVM is under hemodynamic stress.
sampling near the origin (B1/ms). Since a large proportion of the object’s energy is concentrated in the center suggests that frequent sampling of this energy results in higher effective signal to noise ratio of the sampled echo in the frequency domain. The reconstruction uses an iterative reconstruction technique, which temporally divides data into bins. The least squares algorithm iterates over a bin of data to the desired tolerance before adding the next bin of data. This ensures that the algorithm converges as it works through the data and the cost function in the frequency parameter remains convex. The results from Figure 3 suggest that the normative values obtained using the MR-PARSE technique agree with existing PET literature studies (Table 1). Values from other MR-based OEF studies report a range of similar values. Quantitative BOLD reports it as 47.9±7.2% in humans,14 Christen et al12 with a modified quantitative BOLD approach reports in rats with normoxia as 39.3±5.8%. Jain et al16 using the MR phase-based techniques report values of 64±4%, and intravascular T2-based QUIXOTIC study reported 26±2%18 in humans. An interesting feature to be noted is that the normal subjects show a small elevation of OEF in the occipital region, which is seen in PET studies as well.36 Alteration in perfusion parameters under hemodynamic stress is well documented.2,3,27,29 According to their models a patient in stage II would typically exhibit increased CBV, decreased CBF, and elevated OEF. The results shown in the AVM patient line up with & 2014 ISCBFM
these conditions (increased CBV, decreased CBF, and elevated OEF) suggesting that the area proximal to the AVM is in stage II failure. Thus, the vascular steal occurring proximal to the AVM is resulting in hemodynamic stress and local tissue adjacent to the AVM may be at risk. It is interesting to note that the shunting of oxygenated blood through the high flow AVM do not result in abnormal OEF values over the nidus. This may be because the OEF calculation model assumes a vascular system consisting of randomly oriented infinitely long cylinders where exchange of oxygen takes place with surrounding tissue. Inside the AVM the venous shunting causes the OEF model assumptions to break down and the calculations are no longer valid within the AVM, hence we ignore the OEF values within the AVM. The strong R/L asymmetries in this illustrative case are similar to published PET study results showing R/L asymmetry in patients with carotid occlusion3 As shown in Table 2 and Figure 4, the proximal and distal sides to the AVM show marked differences in perfusion parameters, including OEF values. The PARSE MR-OEF technique is not without limitations. Background field inhomogeneities have the potential to cause local frequency shifts, which may result in artifactually high OEF values. To mitigate this issue, the vendor’s advanced Shimming technique was used to ensure that inhomogeneities are minimized. A potential confounder may be the focal deposition of Fe (e.g., hemosiderin from hemorrhage). In these cases, MR-OEF images must be reviewed along with standard GRE images which are reference standard image for the detection of hemorrhage. It is anticipated that the proposed MR-OEF approach here will have limited use in the evaluation of complex changes such as neovascularity in brain tumors. The MR-OEF has the potential to be complementary addition as part of a comprehensive analysis of cerebral physiology in neurovascular disease, which includes perfusion, anatomy, and vascular imaging. While it is beyond the scope of this implementation, it is possible to use spin-echo PARSE39 to use R2’ information as a validation technique. The PARSE technique that is implemented uses a conjugate gradient-based least squares minimization to estimate multiple encoded parameters from the FID signal. Currently, this is a computationally intensive technique. The goal of the study was not, however, to optimize the reconstruction time. Several strategies may be employed to reduce reconstruction times to make it feasible in a clinical setting. Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
Snapshot MR technique measure OEF using rapid frequency mapping RG Menon et al
1116 Parallelizing the reconstruction code, using GPU-based processors, and algorithmic improvements can potentially reduce the reconstruction time by a few orders of magnitude. This technique assumes a constant value of hematocrit. Future versions of this experiment may get this information during routine blood draws. In Figure 4C, a sinus susceptibility effect is seen to reflect in the OEF map near the orbitofrontal cortex. This is caused by bulk magnetic susceptibilities at air–tissue interfaces. It is challenging to remove these effects with our current implementation. Among the normal subjects, there were five subjects with below normal OEF values, which may have been a result from artifacts caused by large susceptibility gradients from the ear canals. A limitation of this technique is the inability to account for large susceptibility gradients caused by sinuses and ear-holes. An improved cardiac gated acquisition method used to remove static field inhomogeneities is being investigated. In conclusion, we have showed a rapid frequency mapping technique to calculate OEF using the PARSE MRI technique. Its primary advantages including speed, wide availability, noninvasiveness, and cost effectiveness make it suitable for use in an application like acute stroke imaging. Our preliminary results suggest strong agreement with PET and other MR-OEF studies. Further development and validation of this technique would be required. DISCLOSURE/CONFLICT OF INTEREST The authors declare no conflict of interest.
ACKNOWLEDGMENTS The authors would like to acknowledge the following funding sources for this research: NIH/NHLBI R01 HL088437; NIH/NINDS R01 NS049395-01A2; NIH/NIBIB T32 EB005170.
REFERENCES 1 Lloyd-Jones D, Adams R, Carnethon M, De Simone G, Ferguson TB, Flegal K et al. Heart disease and stroke statistics--2009 update: a report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Circulation 2009; 119: 480–486. 2 Derdeyn CP, Grubb Jr. RL, Powers WJ. Cerebral hemodynamic impairment: methods of measurement and association with stroke risk. Neurology 1999; 53: 251–259. 3 Derdeyn CP, Yundt KD, Videen TO, Carpenter DA, Grubb Jr. RL, Powers WJ. Increased oxygen extraction fraction is associated with prior ischemic events in patients with carotid occlusion. Stroke 1998; 29: 754–758. 4 Yamauchi H, Fukuyama H, Nagahama Y, Nabatame H, Ueno M, Nishizawa S et al. Significance of increased oxygen extraction fraction in five-year prognosis of major cerebral arterial occlusive diseases. J Nucl Med 1999; 40: 1992–1998. 5 Lansberg MG, Lee J, Christensen S, Straka M, De Silva DA, Mlynash M et al. RAPID automated patient selection for reperfusion therapy: a pooled analysis of the Echoplanar Imaging Thrombolytic Evaluation Trial (EPITHET) and the Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution (DEFUSE) Study. Stroke 2011; 42: 1608–1614. 6 Kidwell CS, Jahan R, Gornbein J, Alger JR, Nenov V, Ajani Z et al. A trial of imaging selection and endovascular treatment for ischemic stroke. N Engl J Med 2013; 368: 914–923. 7 Moskowitz MA, Grotta JC, Koroshetz WJ. The NINDS Stroke Progress Review Group final analysis and recommendations. Stroke 2013; 44: 2343–2350. 8 Mintun MA, Raichle ME, Martin WR, Herscovitch P. Brain oxygen utilization measured with O-15 radiotracers and positron emission tomography. J Nucl Med 1984; 25: 177–187. 9 Videen TO, Perlmutter JS, Herscovitch P, Raichle ME. Brain blood volume, flow, and oxygen utilization measured with 15O radiotracers and positron emission tomography: revised metabolic computations. J Cereb Blood Flow Metab 1987; 7: 513–516. 10 Yablonskiy DA, Haacke EM. Theory of NMR signal behavior in magnetically inhomogeneous tissues: the static dephasing regime. Magn Reson Med 1994; 32: 749–763. 11 An H, Lin W. Quantitative measurements of cerebral blood oxygen saturation using magnetic resonance imaging. J Cereb Blood Flow Metab 2000; 20: 1225–1236.
Journal of Cerebral Blood Flow & Metabolism (2014), 1111 – 1116
12 Christen T, Lemasson B, Pannetier N, Farion R, Segebarth C, Remy C et al. Evaluation of a quantitative blood oxygenation level-dependent (qBOLD) approach to map local blood oxygen saturation. NMR Biomed 2011; 24: 393–403. 13 Fujita N, Shinohara M, Tanaka H, Yutani K, Nakamura H, Murase K. Quantitative mapping of cerebral deoxyhemoglobin content using MR imaging. NeuroImage 2003; 20: 2071–2083. 14 He X, Yablonskiy DA. Quantitative BOLD: mapping of human cerebral deoxygenated blood volume and oxygen extraction fraction: default state. Magn Reson Med 2007; 57: 115–126. 15 Fernandez-Seara MA, Techawiboonwong A, Detre JA, Wehrli FW. MR susceptometry for measuring global brain oxygen extraction. Magn Reson Med 2006; 55: 967–973. 16 Jain V, Langham MC, Wehrli FW. MRI estimation of global brain oxygen consumption rate. J Cereb Blood Flow Metab 2010; 30: 1598–1607. 17 Lu H, Ge Y. Quantitative evaluation of oxygenation in venous vessels using T2-Relaxation-Under-Spin-Tagging MRI. Magn Reson Med 2008; 60: 357–363. 18 Bolar DS, Rosen BR, Sorensen AG, Adalsteinsson E. QUantitative Imaging of eXtraction of oxygen and TIssue consumption (QUIXOTIC) using venular-targeted velocity-selective spin labeling. Magn Reson Med 2011; 66: 1550–1562. 19 Ma D, Gulani V, Seiberlich N, Liu K, Sunshine JL, Duerk JL et al. Magnetic resonance fingerprinting. Nature 2013; 495: 187–192. 20 Twieg DB. Parsing local signal evolution directly from a single-shot MRI signal: a new approach for fMRI. Magn Reson Med 2003; 50: 1043–1052. 21 Twieg DB, Reeves SJ. Basic properties of SS-PARSE parameter estimates. IEEE Trans Med Imaging 2010; 29: 1156–1172. 22 He X, Zhu M, Yablonskiy DA. Validation of oxygen extraction fraction measurement by qBOLD technique. Magn Reson Med 2008; 60: 882–888. 23 Menon RG, Walsh EG, Twieg DB, Sutton BP, Carroll TJ. A novel pulse sequence to measure oxygen extraction fraction in the brain using parameter assessment using retrieval from signal encoding (PARSE) techniques. Proc Intl Soc Magn Reson Med, Melbourne, Australia, 2012. 24 Zuo J, Bolding M, Twieg DB. Validation of V-SS-PARSE for single-shot flow measurement. Magn Reson Imaging 2007; 25: 335–340. 25 Zuo J, Walsh EG, Deutsch G, Twieg DB. Rapid mapping of flow velocity using a new PARSE method. Magn Reson Med 2006; 55: 147–152. 26 Yamauchi H, Fukuyama H, Nagahama Y, Nabatame H, Nakamura K, Yamamoto Y et al. Evidence of misery perfusion and risk for recurrent stroke in major cerebral arterial occlusive diseases from PET. J Neurol Neurosurg Psychiatry 1996; 61: 18–25. 27 Derdeyn CP, Videen TO, Yundt KD, Fritsch SM, Carpenter DA, Grubb RL et al. Variability of cerebral blood volume and oxygen extraction: stages of cerebral haemodynamic impairment revisited. Brain 2002; 125(Pt 3): 595–607. 28 Grubb Jr. RL, Derdeyn CP, Fritsch SM, Carpenter DA, Yundt KD, Videen TO et al. Importance of hemodynamic factors in the prognosis of symptomatic carotid occlusion. JAMA 1998; 280: 1055–1060. 29 Nemoto EM, Yonas H, Chang Y. Stages and thresholds of hemodynamic failure. Stroke 2003; 34: 2–3. 30 Powers WJ, Clarke WR, Grubb Jr. RL, Videen TO, Adams Jr. HP, Derdeyn CP. Extracranial-intracranial bypass surgery for stroke prevention in hemodynamic cerebral ischemia: the Carotid Occlusion Surgery Study randomized trial. JAMA 2011; 306: 1983–1992. 31 Noll DC. Multishot rosette trajectories for spectrally selective MR imaging. IEEE Trans Med Imaging 1997; 16: 372–377. 32 Srour JM, Shin W, Shah S, Sen A, Carroll TJ. SCALE-PWI: a pulse sequence for absolute quantitative cerebral perfusion imaging. J Cereb Blood Flow Metab 2011; 31: 1272–1282. 33 Weisskoff RM, Kiihne S. MRI susceptometry: image-based measurement of absolute susceptibility of MR contrast agents and human blood. Magn Reson Med 1992; 24: 375–383. 34 Carpenter DA, Grubb Jr. RL, Tempel LW, Powers WJ. Cerebral oxygen metabolism after aneurysmal subarachnoid hemorrhage. J Cereb Blood Flow Metab 1991; 11: 837–844. 35 Diringer MN, Yundt K, Videen TO, Adams RE, Zazulia AR, Deibert E et al. No reduction in cerebral metabolism as a result of early moderate hyperventilation following severe traumatic brain injury. J Neurosurg 2000; 92: 7–13. 36 Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman GL. A default mode of brain function. Proc Natl Acad Sci USA 2001; 98: 676–682. 37 Poublanc J, Han JS, Mandell DM, Conklin J, Stainsby JA, Fisher JA et al. Vascular steal explains early paradoxical blood oxygen level-dependent cerebrovascular response in brain regions with delayed arterial transit times. Cerebrovasc Dis Extra 2013; 3: 55–64. 38 Sutton BP, Noll DC, Fessler JA. Dynamic field map estimation using a spiral-in/ spiral-out acquisition. Magn Reson Med 2004; 51: 1194–1204. 39 Li N, Bolding M, Twieg DB. Spin-echo SS-PARSE: a PARSE MRI method to estimate frequency, R(2) and R(2)(’) in a single shot. Magn Reson Imaging 2010; 28: 1270–1282.
& 2014 ISCBFM
