The present study is the largest to date investigating the agreement between CMR and PET measurements of absolute myocardial perfusion. State-of-the-art methods were applied for both CMR and PET perfusion imaging. [15O]H2O, the gold standard for quantification of absolute MBF, was used as tracer for PET, and a dual sequence, single bolus technique optimized for quantification of absolute MBF was used for CMR perfusion imaging. The main finding is that CMR and [15O]H2O PET measurements of stress MBF and MFR showed only modest agreement but were nevertheless concordant in 77% of vascular territories for stress MBF and in 80% of vascular territories for MFR.
Previous – predominantly PET – studies have shown quantification of MBF to improve both prognostic and diagnostic performance in the management of patients with CAD [
2,
4,
18‐
20]. With regard to detection of obstructive CAD, quantitative perfusion measures have been shown to be particularly useful in unmasking balanced ischemia due to three-vessel or left main disease and increase conspicuity of subtle (subendocardial) ischemia [
21]. In addition, absolute stress MBF and MFR may also provide insight in coronary microvascular function [
22]. Although cardiac PET is the commonly used tool for quantitative perfusion imaging, CMR has gained increasing interest for MBF imaging because of its wide availability, high spatial resolution, and non-ionizing nature. In addition, it may also provide information on left ventricular function and viability rendering CMR ideally suited for the noninvasive assessment of CAD.
Previous studies comparing quantitative CMR and PET perfusion have been limited to small numbers of subjects and differ markedly in study population (i.e., patients with CAD vs. healthy volunteers), tracer used for PET quantification, CMR acquisition technique, and CMR field strength. Pärkkä et al. performed CMR and [
15O]H
2O PET in 18 healthy volunteers and reported a significant correlation between CMR and PET measurements of stress MBF (
r = 0.70) and MFR (
r = 0.46), although MFR
CMR was found to be lower than MFR
PET [
23]. Fritz-Hansen et al. and Pack et al., who performed CMR perfusion imaging and
13N-ammonia PET in 10 and 4 healthy volunteers, respectively, reported similar results [
24,
25]. In contrast, Tomiyama et al. studied 10 healthy volunteers with [
15O]H
2O PET and CMR perfusion imaging at 3-T and documented a strong correlation (
r = 0.83) between regional values of MFR
CMR and MFR
PET [
26]. The agreement between quantitative CMR and PET perfusion in patients with CAD was studied by Qayyum and colleagues [
27]. Fourteen patients underwent rubidium-82 PET followed by CMR. Regional MFR, quantified with CMR using a single sequence, single bolus technique, was found to correlate well with PET-derived flow reserve (
r = 0.82). Morton et al. used a dual bolus technique to investigate the agreement between quantitative CMR and PET perfusion in patients with CAD [
28]. CMR measurements of rest and stress MBF showed modest correlation with PET-derived values (
r = 0.32 and
r = 0.37), yet MFR
CMR correlated strongly with MFR
PET (
r = 0.79). Importantly, CMR and PET displayed equal diagnostic performance in a head-to-head comparison against invasive coronary angiography, indicating that although the correlation between CMR and PET in terms of absolute MBF values is modest, diagnostic performance appears to be non-inferior to PET. Engblom et al. and Kunze et al. employed a dual sequence, single bolus technique to quantify absolute MBF with CMR, avoiding multiple contrast bolus injections while preserving accurate caption of the arterial input function [
11,
29,
30]. CMR and
13N-ammonia PET were performed in respectively 21 and 29 patients with stable CAD, and pooled rest and stress measurements of regional MBF were found to correlate strongly between the techniques (
r = 0.83 and
r2 = 0.72). Finally, Kero et al. recently performed CMR and [
15O]H
2O PET in 15 patients with stable CAD using a single sequence, single bolus CMR technique [
31]. Although regional values of stress MBF showed moderate correlation between CMR and PET (
r = 0.69), MFR
CMR correlated poorly with MFR
PET (
r = 0.08). Notwithstanding the interesting results of this study, the small sample size and potential inclusion of patients with myocardial scar may have influenced their findings. In addition, the single sequence single bolus technique used is considered suboptimal for quantification of absolute MBF [
32].
The results of the present study corroborate these prior reports, as we demonstrate only modest correlation between quantitative CMR and PET perfusion measurements. Inter-method reliability between CMR and PET is poor to moderate, as ICC values range from 0.30 to 0.40 with upper bounds of the 95% confidence interval not exceeding 0.60. Although the Bland-Altman plots demonstrate a small mean bias, the limits of agreement are wide, meaning that substantial differences between CMR and PET measurements of stress MBF and MFR are present. It is important to realize however that although [
15O]H
2O PET is not affected by the “roll-off phenomenon,” which occurs with all other myocardial perfusion tracers, the range of perfusion that is clinically important lies apparently beneath this threshold [
33]. This may explain why, despite the modest agreement, stress MBF and MFR are concordant between CMR and PET in the majority of vascular territories. Further support to this hypothesis is provided by a recent meta-analysis reporting a high diagnostic accuracy of quantitative CMR perfusion [
34]. We also observed significantly higher values of rest and stress MBF for CMR compared with PET, which may have resulted from underestimation of the arterial input function with CMR. Although the current dual sequence approach is designed to preserve linearity between gadolinium concentration and signal intensity in the blood pool, saturation effects due to T2* decay still significantly impact the arterial input curve [
35]. Similar to previous reports, we also found that MFR
CMR is lower than MFR
PET, particularly at higher values. The main reason for this lies in the kinetic properties of gadolinium-based contrast agents. The extraction fraction of gadolinium is approximately 0.55 at rest and decreases unpredictably with increasing flow rates [
36]. This results in an underestimation of the tissue response curves at higher flows, subsequently leading to an underestimation of MFR.