FULL PAPER Magnetic Resonance in Medicine 74:990–998 (2015)

Myocardial Perfusion Assessment in Humans Using Steady-Pulsed Arterial Spin Labeling Thibaut Capron,1 Thomas Troalen,1 Benjamin Robert,2 Alexis Jacquier,1 Monique Bernard,1 and Frank Kober1* Purpose: Although arterial spin labeling (ASL) has become a routinely performed method in the rodent heart, its application to the human heart remains challenged by low tissue blood flow and cardiac and respiratory motion. We hypothesized that an alternative steady-pulsed ASL (spASL) method would provide more efficient perfusion signal averaging by driving the tissue magnetization into a perfusion-dependent steady state. Methods: We evaluated the feasibility of spASL in the human heart by combining pulsed labeling in the aortic root with a balanced steady state free precession sequence. The spASL scheme was applied to 13 subjects under free breathing. Breathing motion was addressed using retrospective image exclusion based on a contour-based cross-correlation algorithm. Results: The measured signal with spASL was due to labeled blood. We found that the perfusion signal was larger than that obtained with the earlier flow-sensitive alternating inversion recovery (FAIR) method. Averaged myocardial blood flow (MBF) over four myocardial regions was 1.28 6 0.36 mLg1min1. Conclusion: spASL was able to quantify MBF in healthy subjects under free breathing. Because quantification with ASL is more direct than with first-pass perfusion MRI, it appears particularly suited for pathologies with diffuse microvascular alterations, MBF reserve, and follow-up studies. Magn Reson C 2014 Wiley Periodicals, Inc. Med 74:990–998, 2015. V Key words: myocardial perfusion; arterial spin labeling; steady state; blood flow; steady-pulsed; cine-ASL

INTRODUCTION Arterial spin labeling (ASL) (1,2) appears as a powerful, direct and fully noninvasive alternative to first-pass techniques for assessing myocardial perfusion in humans. Beyond human brain studies, it has become a routinely performed method in the rodent (3–10), but its applica-

tion remains challenging in the human myocardium. Lower tissue blood flow compared with rodents, motion constraints, and physiological noise are major difficulties in a reliable assessment of human myocardial perfusion by ASL. Despite these difficulties, several groups have performed ASL successfully in the human heart (11–15) and have shown its ability to assess myocardial perfusion reserve (16). In these experiments, sensitivity and robustness of the perfusion assessment were limited by both a weak signal inherent to the ASL methods (about 2%) and by low acquisition efficiency due to the alternation of pulse and relaxation periods in the acquisition scheme. As a result, these methods remain subject to physiological noise (14). Sensitivity improvement and the reduction of physiological noise are therefore still of major interest for assessing myocardial ASL in humans. In this study, we propose a steady-pulsed ASL (spASL) approach for improving sensitivity in human heart applications. Steady-pulsed labeling is expected to be more efficient than pulsed ASL in a similar way as continuous ASL while remaining compatible with cardiac motion, the relatively complex orientation of vessels and the pulsatile blood flow in the heart. This labeling approach has been recently set up in animal studies (17,18) under the name cine-ASL, and its superiority in terms of acquisition efficiency compared with LookLocker flow-sensitive alternating inversion recovery (FAIR) gradient echo has been demonstrated. Here, the technique was adapted to measure myocardial blood flow (MBF) in the human heart with a different readout strategy. The technique was implemented for freebreathing acquisition along with a simple contour correlation-based image selection algorithm, and its performance was assessed in 13 subjects. The proposed steady-pulsed approach was compared in terms of signal and perfusion quantification with the earlier used FAIR– balanced steady state free precession (bSSFP) technique.

1

, UMR 7339, CNRS, CRMBM (Centre de Re sonance Aix-Marseille Universite tique Biologique et Me dicale), 13385, Marseille, France. Magne 2 Siemens Healthcare France SAS, Saint-Denis, France. Grant sponsor: Centre National de la Recherche Scientifique (CNRS); Grant , Agence Nationumber: UMR 7339; Grant sponsor: Aix-Marseille Universite nale de la Recherche (QASAREM); Grant number: BLAN-2008-058 and PNR-IMA0706. Thomas Troalen received Ph.D. grant support from Siemens Healthcare France.  UMR *Correspondence to: Frank Kober, Ph.D., Aix-Marseille Universite sonance Magne tique Biologique et Me dicale, CNRS no. 7339, Centre de Re  de Me decine, 27 Bd Jean Moulin, 13385 Marseille Cedex 5, Faculte France. E-mail: [email protected] Received 6 August 2014; revised 4 September 2014; accepted 7 September 2014 DOI 10.1002/mrm.25479 Published online 26 September 2014 in Wiley Online Library (wileyonlinelibrary.com). C 2014 Wiley Periodicals, Inc. V

METHODS Sequence Design The steady-pulsed labeling principle was implemented in a standard bSSFP (true-FISP) sequence using Siemens IDEA VB17 software. An electrocardiogram (ECG)-triggered balanced bSSFP acquisition during diastole (TbSSFP ¼ 334 ms; repetition time (TR) ¼ 3.22 ms; echo time (TE) ¼ 1.61 ms; flip angle ¼ 50 ; slice thickness ¼ 10 mm) was combined with radiofrequency (RF)-pulse labeling applied in end-systole (Sech 15 ms) yielding one image at each cardiac cycle (Fig. 1). The block duration

990

Human MBF Using Steady-Pulsed Arterial Spin Labeling

FIG. 1. Left: position of the Tag and Control slabs in symmetry to the imaging slice, chosen from an LV outflow tract image of the heart. Right: sequence scheme showing triggered bSSFP acquisition during diastole combined with RF-pulse labeling applied in end systole at each cardiac cycle. A total of 128 Tag and 128 Control short-axis images were acquired sequentially under free breathing.

TbSSFP as defined here did not include the bSSFP state initialization pulses and dummy scans, whose duration was Tinit ¼ 20 TR. The labeling slab of 60-mm thickness was placed in the basal heart for the Tag scans so as to cover the aortic root. The gap between labeling slab and image slice was approximately 4 cm but was slightly dependent on the cardiac volume. Due to the high blood velocity in the main coronary arteries, the slight distance variations should only weakly contribute to the arterial transit time (ATT). Control scans were acquired with the inversion slab positioned symmetrically to the imaging slice in order to compensate for magnetization transfer effects, similarly to the EPISTAR technique (19). A series of 128 Tag images followed by a series of 128 Control images (short axis view; field of view [FOV] ¼ 244  300 mm2; matrix size ¼ 104  128) was acquired to maintain the respective Tag and Control steady states while the subject was freely breathing. Rectangular FOV was used to keep the acquisition block as short as possible. The total duration of the acquisition was about 4 min. bSSFP acquisition modules were chosen for their robustness to flow artifacts and their higher signal-tonoise ratio in the cardiovascular context. A 10-tip linear ramp and 10 additional dummy scans were used for bSSFP steady state initialization before each image. The image acquisition time TbSSFP ¼ 334 ms was always much smaller than the respiratory period Tresp  3500 ms. The position of the myocardium, however, was different in successive images due to the breathing motion. This sequence allowed for driving the tissue magnetization into a perfusion dependent steady state while maintaining compatibility with human heart constraints. In contrast to the cine-ASL sequence, however, the labeling pulse was played out prior to and not within each acquisition block in order to avoid signal disturbances and oscillations due to bSSFP steady state interruptions. Protocol A total of 13 healthy subjects were included in the study (men, n ¼ 8; women, n ¼ 5; mean age 6 standard deviation [SD], 29.5 6 6 y). All subjects gave written and informed consent to participate in the study. The protocol was approved by the institutional ethics committee.

991

Data were acquired on a Siemens Verio 3T clinical MRI scanner using a 32-channel phased array receiver coil. After standard adjustments and field map-based shimming, a cine-MRI scan was performed in left ventricular outflow tract orientation to visualize the aortic valve. This cine series was used to choose timing and location of the labeling pulse in the steady-pulsed sequence. The labeling slab was positioned in the aortic root, and the labeling pulse was played out at the instance of aortic valve closure measured for each subject. A first set of data was obtained with the steady-pulsed sequence described above, setting the slab thickness to 60 mm. In order to prove that the signal indeed results from capillary blood flow and not from residual motion of any kind (respiratory, cardiac) during acquisition, a second steady-pulsed scan was performed with a slab thickness reduced to its minimum (3 mm) so that the resulting blood labeling was negligible. The slab thickness was minimized rather than the pulse removed, since this way all sequence properties inducing magnetization transfer remained unchanged. Repeatability was investigated with a third scan, setting back the slab thickness to 60 mm. In order to evaluate the magnetization behavior at the onset of the sequence during the transient phase, all subjects were submitted to an additional spASL breath-hold acquisition allowing for a series of eight Tag/Control pairs, and the group standard deviations during breath-hold were compared with the free breathing situation over a similar duration. A comparison with a FAIR-bSSFP method as published by Zun et al. (14) was performed. We implemented this sequence with FAIR labeling (20) (nonselective/slice-selective Sech pulses) in diastole and bSSFP acquisition in the following diastole such that the inversion time (TI) matched the duration of the cardiac cycle. The sequence parameters—including FOV, slice thickness, matrix size, TR, and flip angle—were identical to those used with the spASL technique. As described by Zun et al., images were acquired during six breath-holds followed by an additional short breath-hold for baseline image acquisition. Because spASL perfusion quantification depends on T1, a T1 map was acquired for each subject using a sequence provided by the MRI manufacturer based on modified Look-Locker inversion recovery (MOLLI) (21). Heart rate and breath rate were continuously monitored during the experiment. In order to assess the MBF response to modified physiologic conditions, two subjects were submitted to a cold pressor test (CPT) by immersing the subject’s left hand into ice water during acquisition of an additional spASL scan (4 min). CPT has been shown to increase MBF using positron emission tomography (22) or MR coronary sinus flow measurement (23). Blood pressure was measured by a cuff at calf level before and during acquisition. Postprocessing and Quantification spASL In previous work (17), the signal behavior has been theoretically described for a similar steady-pulsed experimental scheme. The model was based on the fast exchange assumption in the mouse heart, which can be

992

Capron et al.

considered valid for human myocardium as well. The main difference arises from the fact that in this work, bSSFP readouts were used instead of FLASH, improving both readout signal-to-noise ratio and robustness to flow artifacts in the blood chambers in humans. With bSSFP sequences and assuming zero dephasing between alternating excitation pulses, the steady state signal intensity Sss for a single image is given by (24,25) Sss ¼ M0

1  ðe

eTR=2T2 ð1  eTR=T1 Þsina ;  eTR=T2 Þcosa  eTR=T1 eTR=T2

TR=T1

where M0 is the equilibrium magnetization, a is the flipangle, TR is the repetition time, T1 is the longitudinal relaxation time, T2 is the transverse relaxation time, and TE ¼ TR/2 is the echo time. The transient behavior of the bSSFP signal is characterized by a monoexponential decay whose rate is governed by the apparent longitudinal relaxation time T1* T1 ¼



1 1 1 cos2 ða=2Þ þ sin2 ða=2Þ : T1 T2

Theoretical models for ASL rely on longitudinal magnetization. Because we did not use flip-back pulses at the end of the bSSFP pulse train, the longitudinal projection of the steady-state signal is: z MSS ¼

Sss : tanða=2Þ

In earlier work (17), taking advantage of the monoexponential decay of the signal, the following relation between perfusion and longitudinal magnetization difference using the spASL scheme was obtained with FLASH readout: FLASH C1 ¼

2b 1þ

z lMSS fM0 T1

blood

b¼ð1  1=2eðRRTIÞ=T1

:

Here, f is the blood flow, k is the blood-tissue partition coefficient for water, and b is the effective inversion efficiency. Using bSSFP acquisition, a simplification occurs if the condition TR  T1,2 is fulfilled (see Appendix for calculation details): z MSS M0 sin a ¼ T1 2 T1 tan ða=2Þ

and therewith bSSFP ¼ C1

2b 1þ

1 l sin a 2 f T1 tan ða=2Þ

;

[1]

which was used for quantification in this study. The contrast C1 was experimentally obtained from the measurement of the Control (M1c) and Tag (M1t) steady state values as C1 ¼

ery delay (RD) is set between each acquisition block. As a result of ECG-triggering, RD ¼ RR  Tinit TbSSFP for each cardiac cycle. We note, however, that no flip-back pulse was used at the end of the bSSFP pulse train, leading to a flip-angle dependent partial saturation, which is not optimal in this implementation. If flip-back pulses were used, the equation for MBF quantification would have to be modified accordingly. The calculation of MBF in this work accounts for the currently present partial saturation as long as the RR variations remain small compared with T1. Therefore, heart rate variations had a small impact on the signal stability in our experiment. For estimating b, we used the blood signal in the left ventricle (LV) as a starting point and the following considerations. Ventricular blood is ejected in systole and flows through the coronaries to the capillary bed in the following diastolic phase. The initial state of blood magnetization before entering the coronaries can therefore be estimated using the LV blood signal on the ASL raw images. In our series, the relative amount of remaining blood magnetization in the LV was Mblood,Tag/ Mblood,Ctrl ¼ 15.5%, and we therefore assumed saturated LV blood for the Tag image series. On its way from the LV to the capillary bed, LV blood first undergoes a period of relaxation with T1blood. The duration of this period is taken as the delay between the image k-space center and the following inversion pulse in the next cardiac cycle (duration RR – TI). Blood is then inverted again by the inversion slab in the aortic root and again relaxes during an ATT before reaching the capillary bed. Using a singleexponential Bloch-model to describe relaxation during each of these two consecutive periods of durations (RRTI) and ATT, the magnetization at the entry of the capillary system can therefore be estimated by two Bloch relaxation terms separated by an inversion leading to

c t M1  M1 : c M1

Within the proposed experimental scheme, image acquisition occurs during the time TbSSFP ¼ 334 ms, and a recov-

blood

ÞeATT=T1

;

[2]

where TI is the inversion time (as defined in Figure 1). For a FAIR experiment with 3- to 4-cm-thick inversion slices, ATT was estimated by Wang et al. (26), who found values around 400 ms. We assumed that similar ATT values can be considered in the spASL experiment as well, since the distance traveled from the aortic root to the imaging slice is only slightly longer than that in FAIR, and the velocity in the coronary vessels is high. T1 blood was set to 1.69 s as measured in the MOLLI T1 experiment. Using Equation 2 with these values, we obtained b values around 0.5. For calculating MBF, Tag/Control difference images were required. These difference images can in general be calculated using any image pair from the series, since all images are obtained in the same stationary state. The total number of possible Tag/Control combinations before discarding was therefore 128  128, which implies that images are used multiple times for pair formation. However, only a subset of image pairs were selected using the following procedure. The magnitude image series obtained with the spASL sequence was first filtered by global intensity, such as to exclude images acquired during the transient regime after start of the series or after switch from Control to Tag scan. This filter

Human MBF Using Steady-Pulsed Arterial Spin Labeling

993

FIG. 2. Series of four Tag (top row) and Control (bottom row) images showing respiratory motion and motion handling. A lateral myocardial region was defined (dashed lines in the fourth column), and Control/Tag pairs were chosen based on their contour cross-correlation coefficient within the chosen region. For illustration, the top right image shows the best correlated pair and the bottom right image shows the worst correlated case. Gray pixels represent discordance between contours, whereas black and white pixels are displayed when Control and Tag contours are superimposed.

also excluded images acquired after occasionally missing ECG trigger pulses that led to signal increase. This resulted in keeping about 80% of the acquired data. Since the acquisition was done during free breathing, the myocardium underwent position changes throughout the series (Fig. 2). In order to assess regional perfusion only when the myocardium was at equal positions, the following contour-based selection algorithm was applied. A Canny-Deriche edge-enhancement filter (the default parameters of the IDL implementation were used: upper threshold, 0.8; lower threshold, 0.4; sigma, 0.6) was applied to each image. A region covering the analyzed myocardial segment was manually drawn, and within

this region, the cross-correlation coefficient of contour image pairs was calculated for every possible Control and Tag combination. The contour-based selection and signal evaluation were performed in each region of interest (ROI). A cross-correlation coefficient acceptance level was set such that the 30% best-correlated pairs of the scans were retained. However, the measured signal showed only slight changes when the acceptance level was varied between 10% and 50%. In order to remove potential bias related to the very high bulk-flow blood signal in the LV cavity, the signal difference maps were masked with an intensity threshold. The threshold was symmetric to zero and set

FIG. 3. a: Typical signal map obtained in a representative subject. b: Signal obtained in a manually motion-followed myocardial region during free breathing Tag and Control scans. The outlying points represent images acquired after a missed ECG trigger, which were excluded in the first post processing step. The standard deviation over the entire series was 2% during the Tag scan and 4% during the control scan. c: Myocardial signal evolution during steady state initialization. The data were experimentally obtained for the first eight successive Tag and Control scans during the breath-hold experiment (top) and during free breathing (bottom). In the stationary regime, the group standard deviations of signal over time were comparable between breath-hold and free breathing acquisitions and found to be around 8%.

994

Capron et al.

manually and individually until the endocardial rim introduced by limited spatial resolution and partial volume disappeared. Perfusion was assessed from the signal averaged from all unmasked remaining pixels in the ROI. After removal of the endocardial rim, about 35–50 pixels were retained for each region, and there were three transmural pixel ranges available for quantification. The group average myocardial T1 measured with the MOLLI sequence was used for all calculations, as group dispersion was low. FAIR-bSSFP For the FAIR-bSSFP sequence, signal was quantified as described by Zun et al. (14). In order to allow for comparison with the steady-pulsed method, the relative FAIR-bSSFP signal was defined as SFAIR ¼ (C – T)/B, where C, T, and B refer to the mean signal in the ROI in Control (slice-selective), Tag (nonselective), and baseline images. C, T, and B were obtained as average in regions that were manually segmented in the same four locations assessed with spASL for each breath-hold, averaged over multiple breath-holds. MBF was then obtained using (14) MBF ¼

SFAIR 2 RR eRR=T1

;

[3]

where RR is the interval between heartbeats. RESULTS Selection Procedure Figure 2 shows a series of images approximately covering a complete respiratory cycle, with Tag (top) and Control (bottom) labeling. The image selection process is illustrated on the right side. The contour cross-correlation is calculated for each pair in a manually defined ROI shown in the last images. Among all Control/Tag pairs, the top right image shows the best cross-correlation, whereas the bottom right shows the worst case for illustration. Gray pixels indicate discordance between Control and Tag contours, whereas white and black pixels show well-overlaying Control and Tag contours.

nal difference measured for eight successive Tag and Control scans in all subjects is shown in Figure 3c under breath-hold and free breathing. The stationary magnetization difference, which is related to perfusion by Equation 1, was reached after about 5 to 6 heartbeats, and the signal stability was comparable between breath-hold and free breathing acquisitions (group dispersion, 8%). The group averages of signal difference values are shown in Figure 4 for the Tag experiment (labeling slab, 60 mm) along with the “No Tag” experiment (labeling slab, 3 mm). The chest muscle can be assumed to be perfused with unlabeled blood, since the time for blood to travel from the heart to the chest muscle is long compared with T1. In two subjects, insufficient stability of the signal over time was found in the “No Tag” scan. These subjects were excluded from the analysis. For comparison, Figure 4 also shows the FAIR signal difference measured in the group. MBF calculated from the steady-pulsed ASL signal using Equation 1 as well as that obtained with FAIR is shown in Figure 5 for four segments of myocardium. Global MBF was obtained by averaging the values from the four segments. For comparison, perfusion procedure was also applied to the signal obtained in the chest muscle with FAIR and spASL techniques. From the MOLLI sequence at 3T, the group mean 6 SD T1 values in myocardium, chest muscle, and blood were T1myoc ¼ 1.16 6 0.03 s, T1muscle ¼ 1.06 6 0.05 s, and T1blood ¼ 1.69 6 0.04 s, respectively. MBF averaged over all myocardial regions was MBFspASL ¼ 1.28 6 0.36 mLg1min1. Using the FAIR technique, MBF was calculated using Equation 3, with T1blood ¼ 1.69 s and RR measured for each subject. The FAIR data yielded a group average MBFFAIR ¼ 1.32 6 0.47 mLg1min1. Individual results are summarized in Table 1. Measurement Reproducibility and Stress Response Intrasubject repeatability was evaluated from the two measurements performed at about 4 min interval during the protocol. Results are depicted as a Bland-Altman plot in Figure 6a. Because the placement of the different

Signal Measurement The signal difference map calculated from selected images of a freely breathing subject is shown in Figure 3a. For this map, the selection was based on contour correlation within a region covering the entire heart. Figure 3b shows the signal obtained in a manually motion-followed myocardial region during free breathing Tag and Control scans. The outlying points represent images acquired after a missed ECG trigger, which were excluded in the first postprocessing step. Once these points were excluded, the standard deviation over the entire series was 2% during the Tag scan and 4% during the control scan. Note that these data points were obtained without applying masks and likely contain contributions from blood in the LV cavity. To show the initialization of the steady state under free breathing compared with breath-hold, the average myocardial sig-

FIG. 4. Mean contrast signal in % as defined in the text, obtained in myocardium (dark gray bars) and chest muscle (light gray bars). Signal was measured for the “Tag”, “No Tag,” and “FAIR” experiments as described in the text. Error bars show group standard deviation (n ¼ 11).

Human MBF Using Steady-Pulsed Arterial Spin Labeling

995

FIG. 5. Mean MBF in mLg1min1 obtained in four regions of myocardium, chest muscle, as well as global MBF, comparing spASL (dark gray bars) and FAIR (light gray bars). Error bars show the group SD (n ¼ 13).

regions of interest and image thresholding might vary from one observer to another, interobserver variability was evaluated by comparing MBF values obtained from two different readers following the procedure described in the Methods section. Results are shown in the BlandAltman plot in Figure 6a. Results of the stress test performed in two subjects are shown in Figure 6b. Hemodynamic response to CPT was characterized by a significant blood pressure increase and stable heart rate. MBF was increased for both subjects (8% and 56%). Although one subject showed a minor increase, MBF variation was larger than the intrasubject repetition standard deviation in both cases. The parameters of this experiment are summarized in Table 2. Heart rate varied by less than 5% in a single subject along the 4-min acquisition at both rest and stress states. The standard deviation of RR interval variability was estimated at 28 ms in the group at rest. During the stress test, blood pressure was measured three times, yielding systolic and diastolic pressure dispersions of 7 mm Hg and 4 mm Hg, respectively.

DISCUSSION To improve the sensitivity of myocardial ASL, we implemented an alternative steady-pulsed labeling method for acquiring a series of perfusion-weighted cardiac images under free breathing. The results obtained in 13 healthy subjects showed that the technique yielded a higher intrinsic signal than the FAIR method. Although the postprocessing procedure can still be improved, the method offers efficient averaging capability and good reproducibility. Signal increase during CPT could also be shown. Free-breathing acquisition is an important advantage compared with breath-hold techniques when targeting pathological populations. Concerning the selection of free-breathing images, we note that from a perfusionweighting point of view, all steady state image pairs contain the same information. Any pair can therefore be formed within the series. Out of the 1282 possible combinations, only 30% were typically accepted by the algorithm. The method can thus be improved regarding respiratory motion handling. The simple selection

Table 1 Individual Characteristics and Results of MBF Measurements Using FAIR and the Two spASL Measurements in the Study Group

Subject 1 2 3 4 5 6 7 8 9 10 11 12 13 Mean Group SD a

Age (y)

Sex

28 26 25 44 29 29 24 36 20 30 28 32c 33c — —

M F M M F F F M M F M M M — —

HR (min1)

BR (min1)

MBF, spASL-1 (mLg1min1)

MBF, FAIR (mLg1min1)

MBF, spASL-2 (mLg1min1)

FAIR, SDrep, (mLg1min1)a

Number of original pairsb

64 56 65 67 69 76 70 54 89 81 80 65 66 69 10

17 13 16 14 18 18 17 16 17 16 14 18 19 16 2

1.54 1.44 1.07 0.92 1.80 1.25 1.58 0.30 1.29 0.94 1.62 0.94 1.37 1.24 0.40

2.33 1.06 1.47 1.45 1.11 1.94 1.44 0.73 1.57 0.71 0.85 0.98 1.5 1.32 0.47

1.37 1.39 1.65 1.11 1.77 1.19 1.71 0.57 1.91 1.06 1.32 0.93 1.21 1.32 0.37

1.23 0.74 0.40 0.59 0.91 2.83 0.68 0.65 0.62 5.43 1.32 0.67 1.08 1.75 1.41

23 23 29 26 22 30 27 14 27 24 23 34 30 26 5

SDrep for FAIR represents the SD across the six different repetitions. HR: heart rate, BR: breath rate. Effective number of image pairs contributing to the averaged final blood flow; can be seen as the number of averages. This number was approximated by the square root of the total number of included pairs. The number of pairs was six for FAIR. c Subject is undergoing CPT. b

996

Capron et al.

FIG. 6. a: Bland-Altman plot of individual MBF measurements illustrating interobserver (dark gray diamonds, n ¼ 13) and intrasubject (gray squares, n ¼ 13) repeatability. b: MBF increase in response to CPT for two subjects. Error bars show the SD of repeated measurements at rest.

algorithm used here (Fig. 2) should in the future be replaced by a retrospective or even prospective motion correction such that a greater percentage of the acquired data can be used for averaging. However, retrospective techniques cannot assess potential bias due to throughplane motion. Prospective motion correction (“motion tracking”) would have the additional advantage of addressing these issues. Also, adaptive reconstruction algorithms such as GRICS (generalized reconstruction by inversion of coupled systems) (27) should enable the use of all images for signal averaging, thereby improving sensitivity. The signal difference within image pairs was related to perfusion by an equation derived from a theoretical model developed previously (17). Using bSSFP sequences with short TE, the quantification simplifies to Equation 1. Because no tip-back pulses were used at the end of each bSSFP block, the amplitude of the contrast signal is not optimal and may be improved in future work. The flip angle was, however, chosen such that Sss was close to the maximum. Spatial homogeneity for a may also lead to potential errors in perfusion quantification, especially at 3T. In our setting, a variability of 10% of the flip-angle value yields less than 5% variation of the final perfusion value. The transient signal behavior reported in Figure 3 arises from the successive alternation of labeling acquisition and relaxation periods. Such a steady-pulsed scheme was modeled in previous theoretical work and showed that the transient regime is mainly characterized by TR and RD. In our setting, we used TbSSFP ¼ 334 ms and RD  470 ms in group average and we observed signal convergence to the steady state after about 5 to 6 heart beats, in accordance with theoretical predictions. Nevertheless, the contrast itself is independent of RD, so that arrangement of the steady-pulsed scheme only

affects the signal convergence, as illustrated in this experiment. As presented in Figure 4, spASL consistently provided signal in the myocardium, whereas insensitive regions (e.g., chest muscle) showed no measurable signal, although a tendency to a positive offset existed. Also, when neutralizing the labeling, good signal nulling was obtained. In comparison with the earlier employed FAIR-bSSFP technique, a larger signal was found with the spASL method. Due to the perfusiondependent steady state feature, spASL gives intrinsically larger signal than the single-TI FAIR method. When comparing the results between FAIR and spASL, one has to keep in mind that the sensitivity advantage of spASL is also owing to the large number of acquired signals, since it was applied in a free-breathing implementation, whereas the labeling efficiency including arterial transit times should be significantly better with FAIR. Since relatively long relaxation periods are required for FAIR-bSSFP, it does not acquire a sufficient number of images per time for allowing retrospective image exclusion. Global MBF obtained with both methods was in good agreement. A paired Student t test performed on global MBF yielded P > 0.76, although the regional values had a rather large distribution. Also, both methods showed higher values in the septum. Because the septum is bordered on both sides by chamber blood, the higher MBF values are likely due to residual influence of LV and RV chamber blood signal that was not completely masked. However, because this variation was small, it shows that masking was relatively efficient in our experiment. Table 2 also shows some particularly strong-differing values between FAIR and spASL. This might be caused by breathing motion that influences the data differently for spASL and FAIR. In some cases, breathing patterns

Table 2 CPT Parameters in Two Subjects Rest Subject no. 12 13

CPT

HR (min1)

MBF (mLg1min1)

Blood pressure, systolic/diastolic (mm Hg)

HR (min1)

MBF (mLg1min1)

Blood pressure, systolic/diastolic (mm Hg)

DMBF (%)

DMBF/ DRep

65 66

0.94 1.29

117/57 105/56

64 69

1.01 2.01

154/81 139/68

8 56

13.0 4.7

Blood pressure was measured at calf level during acquisitions. MBF was assessed using the spASL technique. DMBF/DRep represents the ratio of MBF variation between rest and stress DMBF over the MBF SD between the two measurements at rest DRep. HR: heart rate.

Human MBF Using Steady-Pulsed Arterial Spin Labeling

on the one hand and stability of the cardiac position under breath-hold on the other hand might have caused a particular difference in response. Note, however, that the FAIR reproducibility SD cannot be compared directly with the spASL variation, since the number of measurements was six for FAIR and only two for spASL. As a limitation, there is a remaining uncertainty in quantification, and a potential loss of sensitivity related to the labeling efficiency. The labeling slab chosen was rather large to ensure compatibility with a free-breathing acquisition. This led to multiple labeling of blood before entering the left ventricle and labeling in the left atrium in particular. As discussed in previous studies on the mouse heart (18), multiple fast labeling likely leads to a saturation that is b ¼ 0.5. In humans, however, the delay between two labeling events is larger, and labeling efficiency should be reconsidered. Based on our observations of blood signal in the LV blood chamber throughout the Tag series, the assumption of nulled blood magnetization described in the Methods section was fulfilled for all subjects. When using Equation 3 for estimation, the error in b—even with heart rate or ATT variations of 20%—should thus remain within 10%. This introduces errors and a potential bias in perfusion quantification, which varies linearly with b (i.e., of the order of 10%). The use of a two-dimensional spatially selective labeling pulse as proposed by Botnar et al. (28) in a different context is likely to improve the labeling efficiency and to remove uncertainties regarding b. On the other hand, the relatively large slab thickness ensured good robustness to breathing motion. As a future improvement, real-time slice-tracking might be employed for improving the robustness of the steady state regime against respiratory motion. Indeed, through-plane motion during respiration leads to partial loss of steady state and associated signal transients that might affect acquisition, and this was not taken into account in the current version of the sequence. Figure 3 shows, however, good stability of the myocardial signal under free breathing during the steady state. Also, signal transients due to respiratory motion should affect Tag and Control scans in a similar way so that the influence on perfusion quantification remains limited. A second step of improvement will be the use of flip-back pulses at the end of each bSSFP readout, which should improve robustness against heart rate variations and would attenuate the flip angle dependency of the perfusion quantification. With the spASL technique, perfusion values were found with a good intrasubject and interobserver repeatability. The upper masks were indeed necessary to delimit ROIs in a way that ensures that all pixels lie entirely within the myocardium and so as to exclude pixels from the myocardial border. Despite the mask, MBF values in the septum were higher for both techniques indicating nonnegligible partial volume effects. Future implementations should therefore use higher spatial resolutions and potentially reduced slice thicknesses. To ensure minimal operator bias, however, a more uniform and robust postprocessing procedure will be desirable in the future. The sequence was also able to report perfusion changes induced by the CPT performed in two subjects.

997

In one subject, the perfusion increase was small but highly significant with respect to the intrasubject variation. A large interindividual variability in response to CPT was also observed in other studies (22,23), and a larger number of subjects will ultimately be required for a proper comparison. The spatial resolution used in this study was higher than that used in previous studies (14), yet further resolution improvements would be desirable to increase the number of pixels available for processing on the relatively thin myocardium. However, no lengthening of the acquisition window within the cardiac cycle should occur to avoid blurring by contractile motion. On the other hand, segmented acquisition across several cardiac cycles cannot be done under free breathing without advanced reconstruction. Future versions of this sequence will include k-space acceleration by parallel imaging, which will allow a resolution increase within the given constraints, at the expense of SNR. For improving robustness against long-term motion and heart rate variations, one might also consider to acquire several Tag and Control blocks in an interleaved fashion. We note, however, that this would introduce a higher number of discarded scans, since the transitory regime between Tag and Control blocks cannot be taken into account for signal measurement. Because of the requirement for a symmetric Control/ Tag slab configuration, steady-pulsed ASL is a singleslice method. Alternative labeling schemes such as TILT (transfer insensitive labeling technique) (29) may allow for multislice spASL in the future, although the bSSFP image acquisition time still has to remain within the diastolic duration. In conclusion, this study reports the application of a steady-pulsed ASL sequence for the quantitative assessment of MBF in humans. Although we believe that a detailed comparison of performance with FAIR would require a dedicated investigation, higher signal and equivalent performance compared with FAIR was found in a group of healthy subjects at 3T. Overall MBF values were comparable between spASL and FAIR. spASL appears to be a good candidate for non–contrast perfusion assessment, although several sequence and postprocessing details leaving room for technical improvements were identified. The main motivation for developing ASL techniques for cardiac perfusion assessment relies in its repeatability and in the absence of contrast agents. They appear particularly suited for studying pathologies with diffuse microvascular alterations longitudinally and in combination with cardiac stress tests. APPENDIX Here we present the details for the simplification occurring using spASL with bSSFP readout. The ratio to be evaluated can be written as z MSS eTR=2T2 ð1  eTR=T1 Þ 1 sina : ¼ M0 T1 1  ðeTR=T1  eTR=T2 Þcosa  eTR=T1 eTR=T2 T1 tanða=2Þ

When TR  T1,2,

998

Capron et al.

eTR=T1;2  1  TR=T1;2 ; which yields z MSS M0 1 sina    ¼ T1 T1 T T1 1 tanða=2Þ þ 1 þ cosa 1  T2

T2

when neglecting second-order terms in TR/T1,2. As presented in the text, T1 ¼



1 1 1 cos2 ða=2Þ þ sin2 ða=2Þ ; T1 T2

and replacing cosa ¼ cos2 ða=2Þ  sin2 ða=2Þ leads to z MSS M0 cos2 ða=2ÞT2 þ M0 sin2 ða=2ÞT1 sina h i    ¼ : T1 T1 T2 1 þ TT12 þ 1  TT12 cos2 ða=2Þ  sin2 ða=2Þ tanða=2Þ

This equation simplifies to z MSS M0 cos2 ða=2ÞT2 þ M0 sin2 ða=2ÞT1 sin a h i ;  ¼ T1 T1 T2 2cos2 ða=2Þ þ 2 TT12 sin2 ða=2Þ tan ða=2Þ

and then z MSS M0 sina : ¼ T1 2T1 tanða=2Þ

This simplification therefore holds for any value of flip angle a or T2, provided the condition TR  T1,2 is fulfilled. REFERENCES 1. Williams DS, Detre JA, Leigh JS, Koretsky AP. Magnetic resonance imaging of perfusion using spin inversion of arterial water. Proc Natl Acad Sci U S A 1992;89:212–216. 2. Detre JA, Leigh JS, Williams DS, Koretsky AP. Perfusion imaging. Magn Reson Med 1992;23:37–45. 3. Belle V, Kahler E, Waller C, Rommel E, Voll S, Hiller K, Bauer W, Haase A. In vivo quantitative mapping of cardiac perfusion in rats using a noninvasive MR spin-labeling method. J Magn Reson Imaging 1998;8:1240–1245. 4. Waller C, Kahler E, Hiller K-H, Hu K, Nahrendorf M, Voll S, Haase A, Ertl G, Bauer WR. Myocardial perfusion and intracapillary blood volume in rats at rest and with coronary dilatation: MR imaging in vivo with use of a spin-labeling technique. Radiology 2000;215:189– 197. 5. Vandsburger MH, Janiczek RL, Xu Y, French BA, Meyer CH, Kramer CM, Epstein FH. Improved arterial spin labeling after myocardial infarction in mice using cardiac and respiratory gated look-locker imaging with fuzzy C-means clustering. Magn Reson Med 2010;63: 648–657. 6. Iltis I, Kober F, Dalmasso C, Lan C, Cozzone PJ, Bernard M. In vivo assessment of myocardial blood flow in rat heart using magnetic resonance imaging: effect of anesthesia. J Magn Reson Imaging 2005;22: 242–247. 7. Kober F, Iltis I, Izquierdo M, Desrois M, Ibarrola D, Cozzone PJ, Bernard M. High-resolution myocardial perfusion mapping in small animals in vivo by spin-labeling gradient-echo imaging. Magn Reson Med 2004;51:62–67. 8. Jacquier A, Kober F, Bun S, Giorgi R, Cozzone PJ, Bernard M. Quantification of myocardial blood flow and flow reserve in rats using arterial spin labeling MRI: comparison with a fluorescent microsphere technique. NMR Biomed 2011;24:1047–1053. 9. Campbell-Washburn AE, Price AN, Wells JA, Thomas DL, Ordidge RJ, Lythgoe MF. Cardiac arterial spin labeling using segmented ECG-

gated Look-Locker FAIR: variability and repeatability in preclinical studies. Magn Reson Med 2013;69:238–247. 10. Kampf T, Helluy X, Gutjahr FT, Winter P, Meyer CB, Jakob PM, Bauer WR, Ziener CH. Myocardial perfusion quantification using the T1-based FAIR-ASL method: the influence of heart anatomy, cardiopulmonary blood flow and Look–Locker readout. Magn Reson Med 2014;71:1784–1797. 11. Poncelet BP, Koelling TM, Schmidt CJ, Kwong KK, Reese TG, Ledden P, Kantor HL, Brady TJ, Weisskoff RM. Measurement of human myocardial perfusion by double-gated flow alternating inversion recovery EPI. Magn Reson Med 1999;41:510–519. 12. Wacker CM, Fidler F, Dueren C, Hirn S, Jakob PM, Ertl G, Haase A, Bauer WR. Quantitative assessment of myocardial perfusion with a spin-labeling technique: preliminary results in patients with coronary artery disease. J Magn Reson Imaging 2003;18:555–560. 13. Northrup BE, McCommis KS, Zhang H, Ray S, Woodard PK, Gropler RJ, Zheng J. Resting myocardial perfusion quantification with CMR arterial spin labeling at 1.5 T and 3.0 T. J Cardiovasc Magn Reson 2008;10:53. 14. Zun Z, Wong EC, Nayak KS. Assessment of myocardial blood flow (MBF) in humans using arterial spin labeling (ASL): feasibility and noise analysis. Magn Reson Med 2009;62:975–983. 15. Do HP, Jao TR, Nayak KS. Myocardial arterial spin labeling perfusion imaging with improved sensitivity. J Cardiovasc Magn Reson 2014; 16:15. 16. Zun Z, Varadarajan P, Pai RG, Wong EC, Nayak KS. Arterial spin labeled CMR detects clinically relevant increase in myocardial blood flow with vasodilation. JACC Cardiovasc Imaging 2011;4:1253–1261. 17. Capron T, Troalen T, Cozzone PJ, Bernard M, Kober F. Cine-ASL: a steady-pulsed arterial spin labeling method for myocardial perfusion mapping in mice. Part II. Theoretical model and sensitivity optimization. Magn Reson Med 2013;70:1399–1408. 18. Troalen T, Capron T, Cozzone PJ, Bernard M, Kober F. Cine-ASL: a steady-pulsed arterial spin labeling method for myocardial perfusion mapping in mice. Part I. Experimental study. Magn Reson Med 2013; 70:1389–1398. 19. Edelman RR, Siewert B, Darby DG, Thangaraj V, Nobre AC, Mesulam MM, Warach S. Qualitative mapping of cerebral blood flow and functional localization with echo-planar MR imaging and signal targeting with alternating radio frequency. Radiology 1994;192:513–520. 20. Kim SG. Quantification of relative cerebral blood flow change by flow-sensitive alternating inversion recovery (FAIR) technique: application to functional mapping. Magn Reson Med 1995;34:293–301. 21. Messroghli DR, Radjenovic A, Kozerke S, Higgins DM, Sivananthan MU, Ridgway JP. Modified Look-Locker inversion recovery (MOLLI) for high-resolution T1 mapping of the heart. Magn Reson Med 2004; 52:141–146. 22. Schindler T, Zhang X-L, Prior J, Cadenas J, Dahlbom M, Sayre J, Schelbert H. Assessment of intra- and interobserver reproducibility of rest and cold pressor test-stimulated myocardial blood flow with 13N-ammonia and PET. Eur J Nucl Med Mol Imaging 2007;34:1178– 1188. 23. Moro P-J, Flavian A, Jacquier A, Kober F, Quilici J, Gaborit B, Bonnet J-L, Moulin G, Cozzone PJ, Bernard M. Gender differences in response to cold pressor test assessed with velocity-encoded cardiovascular magnetic resonance of the coronary sinus. J Cardiovasc Magn Reson 2011;13:54. 24. Scheffler K. On the transient phase of balanced bSSFP sequences. Magn Reson Med 2003;49:781–783. 25. Schmitt P, Griswold MA, Jakob PM, Kotas M, Gulani V, Flentje M, Haase A. Inversion recovery TrueFISP: quantification of T(1), T(2), and spin density. Magn Reson Med 2004;51:661–667. 26. Wang DJ, Bi X, Avants BB, Meng T, Zuehlsdorff S, Detre JA. Estimation of perfusion and arterial transit time in myocardium using freebreathing myocardial arterial spin labeling with navigator-echo. Magn Reson Med 2010;64:1289–1295. 27. Odille F, Vuissoz P-A, Marie P-Y, Felblinger J. Generalized reconstruction by inversion of coupled systems (GRICS) applied to freebreathing MRI. Magn Reson Med 2008;60:146–157. 28. Botnar RM, Kim WY, Bornert P, Stuber M, Spuentrup E, Manning WJ. 3D coronary vessel wall imaging utilizing a local inversion technique with spiral image acquisition. Magn Reson Med 2001;46:848–854. 29. Pruessmann KP, Golay X, Stuber M, Scheidegger MB, Boesiger P. RF pulse concatenation for spatially selective inversion. J Magn Reson 2000;146:58–65.