Author + information
- Received February 3, 2009
- Revision received May 29, 2009
- Accepted June 3, 2009
- Published online September 1, 2009.
- Timothy F. Christian, MD⁎,⁎ (, )
- Stephen P. Bell, BS⁎,
- Lawrence Whitesell, BS† and
- Michael Jerosch-Herold, PhD‡
Reprint requests and correspondence:
Dr. Timothy F. Christian, Professor of Medicine, Director of Cardiac Imaging, MCHV McClure 1060, University of Vermont, Burlington, Vermont 05495
Objectives The aim of this study was to determine the accuracy of cardiac magnetic resonance (CMR) first pass (FP) perfusion measures of absolute myocardial blood flow (MBF) with a 3.0-T magnet and compare these measures with FP perfusion at 1.5-T with absolute MBF by labeled microspheres as the gold standard.
Background First-pass magnetic resonance (MR) myocardial perfusion imaging can quantify MBF, but images are of low signal at conventional magnetic field strength due to the need for rapid acquisition.
Methods A pig model was used to alter MBF in a coronary artery during FP CMR (intracoronary adenosine followed by ischemia). This produces an active zone with a range of MBF and a control zone. Microspheres were injected into the left atrium with concurrent reference sampling. FP MR perfusion imaging was performed at 1.5-T (n = 9) or 3.0-T (n = 8) with a saturation-recovery gradient echo sequence in short-axis slices during a bolus injection of 0.025 mmol/kg gadolinium–diethylenetriamine pentaacetic acid. Fermi function deconvolution was performed on active and control region of interest from short-axis slices with an arterial input function derived from the left ventricular cavity. These MR values of MBF were matched to microsphere values obtained from short-axis slices at pathology.
Results Occlusion MBF was 0.21 ± 0.26 ml/min/g, adenosine MBF was 2.28 ± 0.99 ml/min/g, and control zone MBF was 0.70 ± 0.22 ml/min/g. The correlation of MR FP CMR with microsphere was close for both field strengths: 3.0-T, r = 0.98, p < 0.0001 and 1.5-T, r = 0.95, p < 0.0001. The 95% confidence limits of agreement were slightly narrower at 3.0-T (3.0-T = 0.49 ml/min/g, 1.5-T = 0.68 ml/min/g, p < 0.05). The FP CMR image characteristics were better at 3.0-T (noise and contrast enhancement were both superior at 3.0-T). In myocardial zones where MBF <0.50 ml/min/g, the correlation with microspheres was closer at 3.0-T (r = 0.55 at 1.5-T, r = 0.85 at 3.0-T).
Conclusions Absolute MBF by FP perfusion imaging is accurate at both 1.5- and 3.0-T. Signal quality is better at 3.0-T, which might confer a benefit for estimating MBF in ischemic zones.
Higher field magnetic resonance (MR) systems are becoming more widely used for all medical applications. These systems provide an increase in signal-to-noise ratio (SNR) and prolonged time relaxation constant (T1) proton recovery (1–4). For neurologic, orthopedic, and many other static types of imaging, this improvement in image physics comes with little downside. However, cardiac imaging has not benefited as broadly, due to an increase in artifacts from its dynamic nature. We have previously reported improved SNR and contrast-to-noise ratio (CNR) for myocardial perfusion imaging in human volunteers at 3.0-T compared with 1.5-T and improved persistence of myocardial tagging at the higher field strength (5,6). Others have found that sequences commonly used at 1.5-T to define ventricular function are significantly degraded at higher field strength from banding artifacts and radiofrequency (RF) pulse inhomogeneity (7–9) and pose energy absorption issues (10). For institutions considering upgrades in MR imaging, this dilemma of higher field strength is very real.
Myocardial perfusion imaging by cardiac magnetic resonance (CMR) can be accomplished with first pass (FP) imaging techniques and is clinically combined with pharmacologic stress to detect coronary artery disease with good accuracy and prognostic value (11–13). First pass imaging can be quantified to estimate absolute myocardial blood flow (MBF) that can be beneficial in measuring serial changes from therapeutic interventions, determining the adequacy of pharmacologic hyperemia, and identifying balanced perfusion abnormalities. Because of the need to image a bolus of contrast as it transits the myocardium, speed in imaging is of paramount importance. To image multiple short-axis left ventricular slices during FP, SNR is sacrificed in favor of temporal resolution. Consequently, MR FP perfusion imaging is among the most signal-starved sequences performed clinically. This becomes even more important in hypoperfused myocardium, where the signal is diminished on a physiologic basis.
The purpose of this study is 2-fold: 1) to determine the accuracy of FP CMR-derived estimates of absolute MBF measures at 3.0-T and compare these measures with absolute MBF determined by left atrial microsphere injection as the gold standard in an animal model of dynamic MBF; and 2) to compare these estimates with those derived at 1.5-T in the same model.
The protocol was reviewed and approved by the institutional animal care and use committee at the University of Vermont. The animal model used was an anesthetized sus scrofa swine preparation of variable MBF. A total of 20 pigs were used for this study. Two pigs died suddenly after surgery, and 1 pig had unacceptable MR images due to persistent arrhythmia, leaving 17 pigs in the study cohort. Pigs were anesthetized with intramuscular ketamine and atropine, and anesthesia was maintained with 1% to 2% inhaled isoflurane during the entire procedure. Ventilation was accomplished with a mechanical ventilator and endotracheal intubation. A mid-line sternotomy was performed, and the proximal portion of the left anterior descending coronary artery (LAD) was dissected free. A hydraulic occluder was placed around the LAD that could be inflated to a specific pressure to cause luminal narrowing of a variable degree. Distal to this, a 24-gauge catheter was placed into the LAD for regional infusion of adenosine. A silastic catheter was placed through a purse-string incision into the left atrial appendage for microsphere injection. A femoral vein and artery were cannulated for contrast injection and reference arterial blood flow sampling, respectively. The chest was loosely closed with wide sutures in order not to pinch-off the catheters and transported to the MR scanner facility.
Animals were placed in either a Philips Achieva 3.0-T magnet (Philips Healthcare, Andover, Massachusetts) (n = 8) or a GE Signa 1.5-T magnet (GE Medical Systems, Milwaukee, Wisconsin) (n = 9), and localization scans were performed. For regional hyperemia experiments, adenosine was infused at 80 μg/kg/min intracoronary through the LAD catheter. Perfusion imaging was performed during the infusion as was left atrial injection of fluorescent microspheres with concurrent arterial reference sampling. The total duration of adenosine infusion was 6 to 7 min (3 min to reach equilibrium, 1 min for the FP study, and 2-min reference sampling after microsphere injection). The coronary ischemia experiments were performed approximately 30 min later to allow return to baseline physiology and blood pool washout of contrast. A coronary hydraulic occluder was inflated to produce partial coronary occlusion with perfusion imaging, and microsphere injection was repeated. After a 30-min recovery, the process was repeated with full coronary occlusion. After completion of the total occlusion scan, the animal was killed with an overdose of intravenous pentobarbital. The heart was later removed and sectioned into 4-mm short-axis slices. These were grouped in pairs to produce 8-mm slices that corresponded to the MR FP perfusion short-axis slices with anatomical landmarks (right ventricle, papillary muscles) and sectioned radially into 8 segments for MBF analysis (1.6 ± 0.6 g). Each segment was weighed and sent for microsphere content (IMT Labs, Irvine, California). MBF was calculated by: where counts = fluorescent microsphere content and the reference arterial sampling rate is 7.5 ml/min.
Perfusion imaging was performed with a bolus injection of 0.025 mmol/kg of gadolinium (Gd)-diethylenetriamine pentaacetic acid (DTPA) at 5 ml/s through a power injector. We chose this dose to be in the linear range of Gd-DTPA concentration and signal intensity (SI) within the blood pool. Both systems used the same type of perfusion sequence: saturation-recovery gradient echo imaging with an echo-train readout. The field of view was 30 cm2, matrix was 1.4 × 1.4 mm, and short-axis slice thickness was 8 mm. For 1.5-T imaging, the repetition time = 6.3 ms, echo time = 1.3 ms, flip angle = 20°, and the echo train = 4. For 3.0-T imaging, repetition time = 1.7 ms, echo time = 1.3 ms, flip angle = 20°, and the echo train = 1. During bolus transit, 3 to 4 short-axis images were acquired per 1 R-R interval gated to the electrocardiographic signal of the animal over 40 to 60 heart beats.
Perfusion images were analyzed offline with an interactive data language-based software program (Cine Tool GEMS, Milwaukee, Wisconsin). Time intensity curves (TIC) were generated from both the left ventricular blood pool to obtain the arterial input function to the coronary circulation and from the myocardial tissue. These were 1.0 to 1.5 cm2 in area to coincide with volume of the pathologic sections. Myocardial TIC regions were registered to coincide with the pathologic MBF analysis: hyperemic zones where the MBF values were highest, occlusion zones where they were lowest (together referred to as intervention zones), and control zones where MBF was unaltered. Two intervention zones (apical and mid-ventricle) and 1 control zone (mid-ventricle) were analyzed for a total of 3 data points/perfusion analysis. This approach minimizes partial volume effects related to arbitrary segment boundaries and statistical inflation by multiple segment analysis (14).
The concept of deconvolution of arterial input function with a shaped function to fit the tissue enhancement curve was employed as first described by Axel (15) and adapted to CMR by Jerosch-Herold et al. (16) and previously described in a similar animal model (17,18). A Fermi function is used to deconvolve the arterial input function to fit the myocardial TIC in a region of interest. Perfusion scans were acquired for 40 R-R intervals, and the entire myocardial TIC was used to generate the fit with the arterial input function. The resultant amplitude of the Fermi function after the fit reflects absolute MBF (15,16). An interactive data language-based quantitative software program was created to automate these calculations, but they could be adjusted to improve the fit (17).
Signal noise was calculated for each study by measuring the SD of the mean SI to a region of interest outside the chest cavity. This was done to separate the impact of noise from tissue SI. The contrast enhancement ratio was calculated by: (peak enhancement − pre-contrast) SI/pre-contrast SI, during FP perfusion.
Data are presented as mean ± SD. Unpaired t tests were used to compare continuous variables by grouping variable. Paired t tests were used to compare variables with paired measures. Analysis of variance was used when more than 2 variables were being compared simultaneously with post hoc comparisons performed with the Fisher least significant difference test. Simple linear regression analysis was used to compare MR perfusion estimates with MBF by microspheres, and Bland Altman plots were constructed from these. There was no adjustment used for multiple data points (n = 3) from a single experiment.
There were 9 pigs imaged on the 1.5-T system, and 8 pigs imaged on the 3.0-T system. Myocardial blood flow ranged from 0.001 to 4.34 ml/min/g with a significant difference (p < 0.0001) in MBF by intervention (in ml/min/g): severe ischemia = 0.09 ± 0.09 ml/min/g, moderate ischemia = 0.45 ± 0.26 ml/min/g, control = 0.70 ± 0.21 ml/min/g, and adenosine = 2.28 ± 0.99 ml/min/g (p < 0.0001 for all groups simultaneously). The heart rate was 92 ± 17 (ischemia scans = 90 ± 16, adenosine scans = 95 ± 19, p = NS).
There were marked differences in image characteristics between the 2 groups as expected. The SD of the noise was significantly reduced at higher field strength: 1.5-T = 6.54 ± 0.86, 3.0-T = 1.23 ± 0.92, p < 0.0001. The contrast enhancement ratio was approximately double at 3.0-T compared with 1.5-T, despite a trend for lower MBF in the control zone for 3.0-T animals: 1.5-T = 0.31 ± 0.13, 3.0-T = 0.68 ± 0.32, p < 0.001.
Consequently, perfusion imaging at 3.0-T produces significantly better contrast for a given degree of MBF with a sharp reduction in background noise. Examples of perfusion images and TIC at peak enhancement by field strength are shown in Figure 1.
The correlation between microsphere-derived measures of absolute MBF and CMR FP deconvolution estimates of MBF were close for both systems, with correlation coefficients exceeding 0.90 (Fig. 2). Both comparisons demonstrated significant correlations between microsphere and MR-derived measures of absolute MBF in the subset of ischemic zones (Fig. 3). However, the correlation coefficient was higher for 3.0-T imaging.
Bland-Altman plots of the overall comparison revealed that the agreement was better for imaging at 3.0-T (Fig. 4), with significantly narrower confidence limits (0.68 ml/min/g at 1.5-T vs. 0.49 ml/min/g at 3.0-T, p < 0.05). The difference between MR estimates of absolute MBF with absolute MBF by microspheres is shown in Figure 5. The MR values were significantly closer to absolute MBF at 3.0-T.
The MR perfusion imaging first-pass sequence provides the least signal of any of the standard cardiac sequences, because of the need for dynamic imaging. Because myocardial contrast enhancement is transient and interpretation is dependent on the properties of the bolus transit, temporal resolution must be optimized, particularly to image multiple slices/R-R interval. Consequently, sampling is brief for each image, and myocardial signal is markedly reduced. Increasing Gd-DTPA concentration is an option, but quickly results in nonlinearity between tracer concentration and tracer signal that can lead to errors in interpretation (19). Another mechanism whereby SI can be increased is by using a higher-strength magnet.
The theoretical physical explanation for increasing signal strength with magnetic field strength (Bo) is well understood and predicts a linear relationship between Bo and SNR. When Bo is within the range of 0.001- and 5.0-T, the SNR is predicted to double for every doubling of Bo. Although the measured increases in SNR can be less than predicted by theory, increasing Bo is an attractive method to improve signal (20). An important advantage of increased field strength is the effect on T1 relaxation after contrast administration. The T1 increase for all tissues at 3.0-T compared with 1.5-T (21), whereas the acceleration upon proton recovery in a magnetic field after an RF pulse for Gd-based contrast agents shows a relatively small decrease between 1.5- and 3.0-T. A longer T1 at baseline at 3.0-T provides a wider dynamic range for contrast-induced T1 changes, compared with 1.5-T. For cardiac imaging sequences, this difference can be exploited to further increase the contrast-to-noise differences between tissues with differential perfusion rates.
The tradeoff for higher field strengths are an increase in chemical shift, T2*, and banding artifacts (7–9,21). Particularly significant is the effect on Bo homogeneity and RF (B1) penetration (7–10). A number of corrections for such confounders have been proposed, but clearly cardiac imaging at 3.0-T is a much more complicated process than 1.5-T (22,23).
The results of the present study demonstrate that accurate quantitation of MBF is possible, despite the increase in potential physical distortions described in the preceding text, with the limitation that we did not image the same animal on both systems. The correlation with microsphere-derived measures of absolute MBF were at least as close as values obtained at 1.5-T. This held for a broad range of physiologic blood flow rates, and there was a trend toward improved correlation in ischemic zones where myocardial signal is most reduced. Myocardial enhancement was approximately doubled at 3.0-T imaging with a fixed concentration of Gd-based contrast, and noise was markedly reduced, which might have accounted for the apparent improvement at low-flow states. Quantification of ischemic flow has particular significance for therapies altered at improving MBF.
These findings support an overall benefit seen for higher field strengths for myocardial FP perfusion imaging. Several clinical and simulation studies have confirmed an overall physical imaging advantage at 3.0-T for FP perfusion (24,25). This has translated to improved myocardial TIC (26) and a slight advantage for the diagnosis of coronary artery disease in patients, when combined with pharmacologic stress (24). Perfusion imaging can be based on steady-state free precession imaging (27) but, as previously discussed, might produce unacceptable artifacts at 3.0-T from banding and RF inhomogeneity. High-field myocardial perfusion can be performed with parallel imaging techniques developed at 1.5-T to speed acquisition (2). This can expand myocardial coverage during an FP exam but with some sacrifice in signal.
We originally designed the study so that each animal was imaged at both 1.5- and 3.0-T. It became apparent that resting MBF dropped significantly and the hyperemic response was markedly attenuated for the second study, independent of magnet strength. This was likely due to the prolonged anesthesia in an open chest model necessary to do 2 complete MR setups in different suites and the repeated accumulation of microspheres in the coronary circulation. Consequently, it was necessary to switch from a “paired” comparison to a “group” comparison. Because animal weight, body habitus, and MBF values were quite similar, we felt this was an acceptable compromise.
The magnets used were from 2 separate manufacturers, and therefore the perfusion sequences, gradients, and RF pulses were not exactly matched. They were, however, identical in terms of the basic principles of imaging (i.e., they were both saturation recovery prepared gradient echo sequences with identical contrast concentration and injection rate). The difference in echo train length by system might have contributed to the slight overestimation of MBF at 1.5-T. Multiple echo readout might reduce apparent signal concentration (but improve temporal resolution), impacting the amplitude of the arterial input function, consequently generating slightly higher estimates of MBF compared with a single readout.
The analysis was confined to 3 myocardial segments/animal: 2 in the intervention zone, and 1 in a remote control zone. The statistical power would have been markedly increased had we analyzed all segments that were sectioned at pathology (24 total). However, this would falsely increase the power of the comparisons without really adding truly original data (14). Consequently, the more rigid standard for comparison was applied.
The estimation of MBF from the measured contrast enhancement is not based on a physiological model of the blood-tissue exchange, as for example embodied by a 2-compartment model (28,29), but rather on the relationship between tissue impulse response amplitude and blood flow, originally established by Zierler (30). The Fermi function represents here a general parametrized model of a tissue impulse response.
The tissue impulse response can be thought of as the contrast enhancement for the theoretical limit when the contrast injection is an infinitesimally short contrast bolus. The MBF corresponds to the amplitude of the best-fit impulse response, with adjustment for the temporal resolution of the measured TIC. The basis of the study was that the improvement in physical imaging characteristics would generate truer TIC curves, consequently improving the performance of the model. In this framework, Figure 1 is important in the interpretation of the results, because it shows clearer definition of the TIC at 3.0-T. This improvement translated to closer estimates in low MBF zones and less difference with absolute values in all zones.
This work was supported in full by a Grant-in-Aid from the American Heart Association (to Dr. Christian).
- Abbreviations and Acronyms
- magnetic field strength
- cardiac magnetic resonance
- contrast-to-noise ratio
- diethylenetriamine pentaacetic acid
- first pass
- left anterior descending coronary artery
- myocardial blood flow
- magnetic resonance
- signal intensity
- signal-to-noise ratio
- time intensity curve
- Received February 3, 2009.
- Revision received May 29, 2009.
- Accepted June 3, 2009.
- American College of Cardiology Foundation
- Jahnke C.,
- Nagel E.,
- Gebker R.,
- et al.
- Christian T.F.
- Cheng A.S.,
- Pegg T.J.,
- Karamitsos T.D.,
- et al.
- Jerosch-Herold M.,
- Wilke N.,
- Wang Y.,
- et al.
- Bassingthwaighte J.B.,
- Goresky C.A.