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

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 four-element 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 mid-diastole, 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]. T₁-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 [B₀] 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 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 T₁-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.

On the 1.5 T system, ASL sequence parameters included a TR range of 2.47 to 2.59 ms and a TE range of 1.07 to 1.12 ms. At 3.0 T the TR range was 2.48 to 3.16 ms and the TE range was 1.06 to 1.69 ms. At both B₀ strengths, flip angle = 5°, slice thickness = 8 mm, receive bandwidth = 650 Hz, the number of phase-encoding steps ranged from 64 to 73, and the interpolated matrix size ranged from 128 × 64 to 128 × 73. The first series consisted of 20 to 24 image acquisitions (half obtained with a Sel IR pulse and half obtained with a NonS IR pulse), temporally separated by the RR interval. The average observed RR interval was 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 TI₁) ranged from 91 to 99 ms, while that of the second series (initial TI₂) ranged from 137 to 155 ms. At 3.0 T, the initial TI₁ ranged from 91 to 122 ms and the initial TI₂ 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 TI₁ was the minimal TI achieved with the defined image matrix size since k-space data was linearly acquired in each T₁-weighted image. Generally, the TI₂ was approximately 40 ms greater than the TI₁. This ensured that the IR pulse of the second series, together with the following GE acquisition, remained in mid-diastole [2].

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 T₁ 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 T₁ 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 T₁ 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 below), the SNR values from the last image acquired (the image with the longest TI) in each sequence were used.

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 T₁ 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:

Where A^n/2-1 is defined as M 0 ( 1 − e − T R / T 1 ) 1 − ( cos α e − T R / T 1 ) n / 2 − 1 1 ℮ cos α e − T R / T 1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=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@6066@ and B is equal to cos α e − T R / T 1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGagi4yamMaei4Ba8Maei4CamNaeqySdeMddaqfGaqabeqaleaacqGHsislcqWGubavcqWGsbGucqGGVaWlcqWGubavdaWgaaadbaGaeGymaedabeaaaOqaaiabdwgaLbaaaaa@39CB@, β is the inversion pulse flip angle, α is the RF-train excitation angle, and τ_N is the time interval between the (N - 1)th and Nth measurements. This equation was derived from the Bloch Equation to correlate T₁-weighted signal intensities of dynamic images during T₁ recovery, necessary due to the brevity of the RR interval. Using this equation, the saturation effect induced by the multiple data acquisitions between each cardiac cycle can be accounted for in the T₁ calculation.

Random Gaussian noise was then added to the S_N and SNR was calculated as M 0 sin α N o i s e MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaqcfa4aaSaaaeaacqWGnbqtdaWgaaqaaiabicdaWaqabaGagi4CamNaeiyAaKMaeiOBa4MaeqySdegabaGaemOta4Kaem4Ba8MaemyAaKMaem4CamNaemyzaugaaaaa@3B1B@. Different SNR values were designed and corresponding nonselective and slice-selective myocardial T₁ values were calculated using the nonlinear regression algorithm, as well as non-selective blood T₁ values. MBF values were calculated using Eq. [1] to compare with the true MBF data (data without noise) for accuracy analysis. For the specific T₁ 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.

The relationship of the absolute error of the calculated MBF (in percentages) with the SNR of the images (35.60 at 1.5 T and 79.01 at 3.0 T) was determined with the error simulations. The data were obtained using the T₁ values at 1.5 T. When the SNR is lower than 250 (for all 1.5 T and 3.0 T MRI systems), the error of the MBF is larger than 10% (10–50%). The repeatability of the MBF measurements in this simulation was found to be 0.127 for 1.5 T and 0.078 for 3.0 T (P = NS), whereas the error of the MBF was 13.9 ± 7.1% at 1.5 T and 5.4 ± 4.5% at 3.0 T (P = NS).

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 T₁ 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, T₁ value differences 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 T₁ 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 [28‐30], radiation damping [31], the longer T₁ 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 end-point 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].