Resting myocardial perfusion quantification with CMR arterial spin labeling at 1.5 T and 3.0 T

Background The magnetic resonance technique of arterial spin labeling (ASL) allows myocardial perfusion to be quantified without the use of a contrast agent. This study aimed to use a modified ASL technique and T1 regression algorithm, previously validated in canine models, to calculate myocardial blood flow (MBF) in normal human subjects and to compare the accuracy and repeatability of this calculation at 1.5 T and 3.0 T. A computer simulation was performed and compared with experimental findings. Results Eight subjects were imaged, with scans at 3.0 T showing significantly higher T1 values (P < 0.001) and signal-to-noise ratios (SNR) (P < 0.002) than scans at 1.5 T. The average MBF was found to be 0.990 ± 0.302 mL/g/min at 1.5 T and 1.058 ± 0.187 mL/g/min at 3.0 T. The repeatability at 3.0 T was improved 43% over that at 1.5 T, although no statistically significant difference was found between the two field strengths. In the simulation, the accuracy and the repeatability of the MBF calculations were 61% and 38% higher, respectively, at 3.0 T than at 1.5 T, but no statistically significant differences were observed. There were no significant differences between the myocardial perfusion data sets obtained from the two independent observers. Additionally, there was a trend toward less variation in the perfusion data from the two observers at 3.0 T as compared to 1.5 T. Conclusion This suggests that this ASL technique can be used, preferably at 3.0 T, to quantify myocardial perfusion in humans and with further development could be useful in the clinical setting as an alternative method of perfusion analysis.

pass perfusion cardiovascular magnetic resonance (CMR) would allow efficient measurement of myocardial perfusion, but the necessity of a gadolinium-based contrast agent would limit this technique in terms of repetition in a single CMR examination and use in the renal failure patient. However, arterial spin labeling (ASL), through measurement of myocardial and blood T 1 , is capable of measuring myocardial perfusion multiple times in a single CMR examination [2,5]. ASL uses arterial water as an endogenous tracer and can be implemented through a number of techniques. The method used in this work is based on the flow-sensitive alternating inversion recovery (FAIR) scheme. In FAIR, two sets of inversion recovery (IR) images are acquired: a slice-selective inversion and a nonselective inversion [6]. After the slice-selective inversion, non-inverted proton spins from arterial water flow into the imaging slice and exchange with tissue water, resulting in T 1 shortening in the tissue of interest.
Previously, two studies involving the rat heart in various states (resting, hyperemic, and during post-infarction remodeling) have validated ASL against the previously established technique of myocardial perfusion quantification using microspheres [7,8]. Furthermore, ASL has been validated with both the microsphere technique and firstpass perfusion imaging in a different study using rats [9]. Notably, ASL has been shown to be successful in its application to humans with ischemic heart disease and in the determination of perfusion reserve, a clinically useful metric, in a study that included patients with suspected coronary artery disease [10].
From ASL sequences, myocardial perfusion can be quantified by calculating MBF using the following equation [5]: where λ is the constant blood-tissue coefficient of water, and λ = 0.92 mL/g for blood-perfused myocardial tissue [11]; T 1, GS is the T 1 of the myocardium after the nonselective IR pulse; T 1, Blood is the average T 1 of the left ventricular blood pool; and T 1, SS is the T 1 of the myocardium after the slice-selective IR pulse. This equation assumes identical spatial slice profiles of the slice-selective IR pulse and the slice-selective excitation pulse, a two-compartment model (consisting of the intravascular capillary blood and the extravascular tissue) in which magnetization is homogeneous in each compartment, and rapid exchange of water between the two compartments. The accuracy of the myocardial perfusion rate determined by this equation is highly dependent on both the measured T 1 values and the ASL technique used to obtain them. Recently, we have proposed a modified ASL technique and T 1 regression algorithm that has proved reliable in assessing MBF in canine models [2,12].
In recent years, 3.0 Tesla (T) magnetic resonance imaging systems have been used increasingly in the clinical setting. The primary advantage of 3.0 T over 1.5 T systems is the anticipated, but less than, twofold increase in signal-tonoise ratio (SNR) [13][14][15]. Though it is the intrinsically longer T 1 relaxation times observed at 3.0 T that lead to this lower-than-expected SNR increase in T 1 -weighted imaging, 3.0 T systems still demonstrate a significantly higher SNR than that of 1.5 T, and thus serve as a more reliable platform for the calculation of T 1 [16,17]. Additionally, the intrinsically longer T 1 relaxation times grant 3.0 T greater sensitivity to T 1 changes.
Accordingly, the primary aim of this study was to compare the accuracy and repeatability of myocardial perfusion measurements by the use of a recently modified ASL technique at 3.0 T to those at 1.5 T in resting healthy human subjects [2].

Imaging protocol
A total of 8 subjects (5 male, 3 female, mean age 21.9 ± 2.0 years) were recruited. All subjects had no known history of cardiovascular disease and were in the resting state for both CMR examinations. Each patient underwent two consecutive same-day imaging sessions, one on a 1.5 T whole-body Sonata system (Siemens Medical Solutions, Erlangen, Germany) and one on a 3.0 T whole-body Magnetom Trio system (Siemens Medical Solutions, Erlangen, Germany). Both systems were equipped with a fast gradient system (maximal gradient strength = 40 mT/m; maximal slew rate = 200 mT/m per millisecond). A fourelement phased array coil was used for signal reception and a body coil was used for transmission. For electrocardiogram (ECG) monitoring and pulse sequence triggering, a three-lead patch was attached to the chest of each subject. This allowed for the acquisition of images at middiastole, the time of least cardiac contractile motion. To eliminate respiratory motion artifacts, the subjects were instructed to hold their breath for the duration of each series of acquisitions [18].
A scout image was first obtained to localize a short axis view of the left ventricle (LV) of each subject's heart. Next, a cine acquisition with a segmented steady-state free precession (SSFP) sequence was performed to determine the mid-diastolic period in the cardiac cycle [19]. T 1 -weighted images were then acquired using a recently developed single-shot gradient-echo (GE)-based ASL sequence with an adiabatic hyperbolic secant IR pulse [2]. At both main magnetic field [B 0 ] strengths, the previous ASL sequence was modified by combining nonselective (NonS) and slice-selective (Sel) prepared sequences into one series. This allowed for the reduction of inter-scan motion effects and total scanning time. Based on the Look-Locker scheme, the sequence acquired multiple data points along

MBF T GS T Blood T SS T GS
the magnetization recovery time course after each IR pulse. There was a 3-second idle time between the two IR acquisitions to allow more time for the recovery of magnetizations of GE acquisitions. Therefore, the NonS and Sel T 1 -weighted signals were acquired sequentially. Data acquisition was ECG-triggered ( Figure 1). Each pair of series was repeated three times in order to evaluate the reliability of ASL in perfusion measurement.

Imaging parameters
On the 1. approximately 800 ms, making the total duration of the series, and thus the breath hold duration, 16 to 19 seconds. The first series was followed immediately by a similar second series of acquisitions (data from both parts were combined in the analysis, described below) with a longer initial inversion time (TI). At 1.5 T, the initial inversion time from the first series (initial TI 1 ) ranged from 91 to 99 ms, while that of the second series (initial TI 2 ) ranged from 137 to 155 ms. At 3.0 T, the initial TI 1 ranged from 91 to 122 ms and the initial TI 2 ranged from 137 to 177 ms. The TI was the time from the middle of the IR pulse to the time when the central k-space data of the GE acquisition was acquired. Thus, the initial TI 1 was the minimal TI achieved with the defined image matrix size since k-space data was linearly acquired in each T 1weighted image. Generally, the TI 2 was approximately 40 ms greater than the TI 1 . This ensured that the IR pulse of the second series, together with the following GE acquisition, remained in mid-diastole [2].

Data analysis
Two independent observers drew one region of interest (ROI) in the left ventricular blood pool and one in the entire left ventricular myocardial ring. The use of two independent observers facilitated the determination of the inter-observer variability for the calculated MBF values (note that the T 1 and MBF values presented in the results and the tables represent data from a single observer). The ROI in the left ventricular myocardial ring were drawn in the central 50% of the myocardium, in order to avoid artifacts common to the edges. However, the ROI were not drawn over large full-thickness artifacts that were readily visualized, such as air-solid interface magnetic susceptibility artifacts of the lateral wall or blood-pool and myocardial partial volume effects at the septum [20]. Notably, ROI were not drawn over major coronary artery branches so MBF calculations would presumably include data from the level of the myocardial microvasculature only, necessary for this method to accurately quantify myocardial perfusion. Bulk adjustment of a ROI was performed if necessary to correct for positional changes. In addition to this, each image in a pair of series was screened by one observer for motion of any kind, and the pair was given a motion rating on a scale of zero to five (0 = no motion and 5 = severe motion artifacts). Each score was subsequently classified as respiratory, cardiac, or inter-series related motion.
From the absolute values of the signal intensities measured in each ROI, our previously developed MatLab (The MathWorks Co., Natick, MA)-based nonlinear regression algorithm using real clock time for TI was used to calculate T 1 values [2]. Using these values in Eq. [1], MBF was calculated. The average MBF values, as well as the standard deviations (SD), were calculated from all of the pairs of series from each subject at 1.5 T and 3.0 T, respectively. The repeatability was calculated as the ratio of the average SD to the average MBF. Paired comparison of the two field strengths was performed using a paired two-tailed Student's t test with a P < 0.05 significance level. The strength of agreement between the two sets of observations was determined via MBF comparisons by paired Student's t tests and Bland-Altman tests. To demonstrate the capability of perfusion mapping by this method, a myocardial perfusion map was created pixel-by-pixel using corresponding nonselective and slice-selective T 1 maps ( Figure  2).
Additionally, SNR was calculated for the entire LV myocardial ring (from both slice-selective and nonselective series) and the LV blood pool (from nonselective series only). This was done by dividing the mean signal intensity (of the LV blood pool or the myocardium) by the standard deviation of the signal intensity of extracorporeal air. In both the experimental and the simulation results (see A schematic diagram of the sequence that was used to obtain the data necessary for myocardial perfusion quantification, highlighting ECG triggering and the 3 second idle time between the two IR acquisitions Figure 1 A schematic diagram of the sequence that was used to obtain the data necessary for myocardial perfusion quantification, highlighting ECG triggering and the 3 second idle time between the two IR acquisitions. below), the SNR values from the last image acquired (the image with the longest TI) in each sequence were used.

Error analysis
To evaluate the effect of SNR on the accuracy and precision of the myocardial perfusion measurements using this ASL method, a computer simulation was performed with the T 1 data sets from the averaged values of volunteers (see Results). Using the following equation (Eq. [12] in reference [2]), signal intensity of S N (N is the number of images collected after the IR pulse) can be derived: to compare with the true MBF data (data without noise) for accuracy analysis. For the specific T 1 and SNR data sets of the volunteers at 1.5 T and 3.0 T, these procedures were repeated 3 times to calculate the repeatability and aid in the estimation of the accuracy of the means.

Results
The individual and overall average T 1 values, calculated from the data from one observer, are shown in Table 1.
The average LV blood pool T 1 , nonselective myocardial T 1 , and slice-selective myocardial T 1 values all showed statistically significant increases from 1.5 T to 3.0 T (P < 0.001).
Dynamic normalized (to noise) curves of T 1 recovery are demonstrated in Figure 3. A trend of greater SNR at most time points, as well as larger T 1 values, was evident at 3.0 T as compared to 1.5 T.
The average calculated SNR values from the final acquisition of each sequence are displayed in Figure 4.  Table 2. Higher standard deviations (SD) of MBF values were observed at 1.5 T in subjects 1, 2, 3, and 6. Higher SD of MBF values were observed at 3.0 T in subjects 5, 7, and 8. In subject 4, the SD could not be compared because only one pair of series of images was acquired at 3.0 T. The average SD at 1.5 T was 0.365, representing a repeatability of 0.37, whereas the SD at 3.0 T was 0.221, representing a repeatability of 0.21, a 43% improvement. In subjects 4, 5, and 6, the 1.5 T scans had higher average motion ratings than the 3.0 T scans. In subjects, 1, 2, 3, 7, and 8, the 3.0 T scans had higher average motion ratings than the 1.5 T scans.
The average myocardial perfusion rate for all subjects, calculated from the MBF data from one observer, was found to be 0.990 ± 0.302 mL/g/min at 1.5 T and 1.058 ± 0.187 mL/g/min at 3.0 T. This difference between the two field strengths was not found to be statistically significant (P = NS). Likewise, the average standard deviations of the MBF values (0.365 ± 0.201 mL/g/min at 1.5 T and 0.221 ± 0.128 mL/g/min at 3.0 T) did not demonstrate statistically significant differences between the two field strengths (P = NS). No significant difference was found between the two observers in terms of calculated MBF values at 1.5 T or 3.0 T (P = NS), although a slightly higher average MBF was obtained by the second observer at both B 0 strengths (1.198 ± 0.205 mL/g/min at 1.5 T and 1.211 ± 0.281 mL/ g/min at 3.0 T). Figure 5 shows the Bland-Altman analysis and demonstrates a trend toward less variation in the calculated MBF values by the two observers at 3.0 T as compared to 1.5 T.

Error simulations
The

Discussion
The primary goal of this study was to compare the accuracy and repeatability of myocardial perfusion data obtained at 3.0 T to that at 1.5 T using an ASL technique in normal human subjects. This ASL technique was previously validated at 1.5 T in a canine model [2].
The gold standard in human myocardial perfusion determination is PET, and 13 N-ammonia is considered the tracer of choice, over 15 O-water [21]. In the most recent 13 N-ammonia PET study, Schepis, et al. found the MBF in resting human subjects to be in the range of 0.81 ± 0.28 to 0.84 ± 0.25 mL/g/min, depending on the PET method used [22]. Additionally, Rahimtoola compiled eight 13 Nammonia PET studies from the mid-1990's and found the range of MBF values (± 2 SD) in normal subjects to be approximately 0.40 to 1.50 mL/g/min [4]. In recent years, first-pass perfusion has been the primary CMR method used for the determination of myocardial perfusion.
Using this technique, Hsu, et al. found the perfusion rate in resting human subjects to be 1.02 ± 0.22 mL/g/min [23]. Thus, our perfusion data at both field strengths compare favorably with the data obtained with 13 N-ammonia PET and first-pass CMR.
Secondly, as anticipated, no significant difference was found between the myocardial perfusion data at the two different field strengths or between the perfusion data obtained by the two independent observers. The simulation results suggested the same finding; while the accuracy of perfusion data from the simulation at 3.0 T is somewhat better than that at 1.5 T (as 3.0 T displayed a 61% reduction in error), this difference was not found to be statistically significant. Additionally, less variation was observed in the perfusion data from the two observers at 3.0 T as compared to 1.5 T. Concordantly, the repeatability of MBF measurements at 3.0 T improved to 0.21 from 0.37 at 1.5 T, although these differences were not statistically significant. This is an improvement of approximately 43%; similar to the simulation result of a 38% improvement (P = NS). It should be noted that the absolute values of repeatability are much smaller than the measured ones, suggesting other errors contribute significantly, in addition to the effect of SNR. These errors may include incompletely labeled spins [24], transition time error [25], and imperfect slice profiles of data acquisition RF pulses [26]. Our simplified simulation accounted for none of these, and therefore, further investigations are needed to incorporate the effects of these error sources.

Limitations
The major limitation of the ASL technique remains high sensitivity to motion. The requirement of ECG-gating and breath-holding for the duration of each series is not completely effective in eliminating cardiac and respiratory motion. Additionally, patient movement between the two back-to-back series introduced potential perfusion quantification error associated with inter-acquisition motion since the selective and non-selective scans were performed sequentially. All of these types of motion can lead to blurring and image distortion. In terms of the future application of this technique to patients, these motion types will be significant because patients requiring perfusion analysis will likely suffer from a myriad of comorbidities. It might be unreasonable to request a breath-hold of nearly 20 seconds in some patients, such as those with severe cardiovascular or respiratory disease.
Another limitation is associated with the MBF equation that was used in this study. This equation uses differences in T 1 values between the slice-selective and nonselective acquisitions, rather than differences in signal intensity values, to calculate MBF. Notably, Zhou and van Zijl have theorized that in FAIR techniques, T 1 value differences Perfusion maps at 1.5 T and 3.0 T Figure 2 Perfusion maps at 1.5 T and 3.0 T, demonstrating a more uniform myocardial perfusion distribution at 3.0 T. Each map was created pixel-by-pixel using the corresponding nonselective and slice-selective T 1 maps.
Normalized dynamic curves of T 1 recovery between the slice-selective and nonselective acquisitions may be increased or decreased, depending on the transit time of the "tagged bolus," while the signal intensity difference is always decreased [27]. The unpredictability of the effect of transit time makes it impossible to control for this source of error when T 1 value differences are used to calculate MBF. In addition to this, perfusion quantification using FAIR techniques is susceptible to many sources of error, including macrovascular flow [28][29][30], radiation damping [31], the longer T 1 of arterial blood [28,32,33], and imperfect inversion profiles [34].
There are two main limitations unique to the use of a 3.0 T MRI system for these techniques. First, the specific absorption rate (SAR) quadruples with the doubling of the field strength from 1.5 T to 3.0 T. This often necessitates increased repetition times and, consequently, increased examination durations. Second, magnetic susceptibility artifacts are more prominent at 3.0 T, particularly at the boundary of myocardial tissue and lung air, as well as the boundary of the myocardium and ventricular blood. The ROI was drawn at the center of the myocardium to avoid these effects. Nonetheless, imperfect magnetic shimming might still lead to these artifacts in some large subjects at 3.0 T. This was not often observed with the 1.5 T system.
Finally, this study quantified myocardial perfusion at rest. Such measurement may be helpful to monitor the efficacy of medical therapy in patients with cardiovascular disease. The next step in the study of this technique is to evaluate its efficacy during adenosine-or dipyridamole-induced vasodilatation. This would facilitate the determination of myocardial perfusion reserve, a more important clinical parameter. A study including this parameter as an endpoint would be likely to demonstrate a greater reduction in the magnitude of error from 1.5 T to 3.0 T because the percent error of perfusion reserve is proportional to the percent error of myocardial perfusion during stress. The latter percent error would be greatly reduced during vasodilatation, when perfusion is three-to-four-fold greater than at rest [2].
The average SNR of the LV blood pool (nonselective sequences) and the entire LV myocardial ring (slice-selective and nonselective sequences), calculated from the final acqui-sition of each sequence Figure 4 The average SNR of the LV blood pool (nonselective sequences) and the entire LV myocardial ring (sliceselective and nonselective sequences), calculated from the final acquisition of each sequence. All average SNR demonstrate statistically significant increases from 1.5 T to 3.0 T (P < 0.002). LV BP = Left Ventricular Blood Pool.

Conclusion
This work suggests that CMR ASL can be used to quantify myocardial perfusion at rest in humans with relative accuracy. However, there are several limitations to this technique, to the MBF equation that was used, and to the use of the error simulation to assess accuracy that must temper this suggestion. Furthermore, no significant difference in the use of 1.5 T and 3.0 T systems, in terms of the accuracy and repeatability of myocardial perfusion quantification, was found. However, 3.0 T did demonstrate a significantly higher SNR that yielded slightly improved repeatability, slightly less variation in perfusion data, and more uniform signals in perfusion maps. With continued success, this technique could be applied to patients with significant perfusion defects and eventually become part of the standard clinical work-up of patients with ischemic heart disease.  "Pairs of series" denotes each set of back-to-back series. The motion ratings are on a scale of 0 to 5, with 0 representing no detectable motion and 5 representing severe motion artifacts. Subj = subject; S = interseries motion or motion between the pair of series; R = respiratory motion; C = cardiac motion.
Bland-Altman plots comparing the myocardial perfusion data obtained by the two independent observers (MBF1 and MBF2) at 1.5 T (A) and 3.0 T (B) Figure 5 Bland-Altman plots comparing the myocardial perfusion data obtained by the two independent observers (MBF1 and MBF2) at 1.5 T (A) and 3.0 T (B). The 3.0 T system demonstrated a trend towards reduced variation (mean = -0.153 and 95% confidence interval of -0.772 to 0.467) in MBF measurements by the two observers, as compared to the 1.5 T system (mean = -0.208 and 95% confidence interval of -1.021 to 0.605). Both axes are in units of mL/g/min.
Publish with Bio Med Central and every scientist can read your work free of charge