Comparison between quantitative cardiac magnetic resonance perfusion imaging and [15O]H2O positron emission tomography.

Purpose To compare cardiac magnetic resonance imaging (CMR) with [15O]H2O positron emission tomography (PET) for quantification of absolute myocardial blood flow (MBF) and myocardial flow reserve (MFR) in patients with coronary artery disease (CAD). Methods Fifty-nine patients with stable CAD underwent CMR and [15O]H2O PET. The CMR imaging protocol included late gadolinium enhancement to rule out presence of scar tissue and perfusion imaging using a dual sequence, single bolus technique. Absolute MBF was determined for the three main vascular territories at rest and during vasodilator stress. Results CMR measurements of regional stress MBF and MFR showed only moderate correlation to those obtained using PET (r = 0.39; P < 0.001 for stress MBF and r = 0.36; P < 0.001 for MFR). Bland-Altman analysis revealed a significant bias of 0.2 ± 1.0 mL/min/g for stress MBF and − 0.5 ± 1.2 for MFR. CMR-derived stress MBF and MFR demonstrated area under the curves of respectively 0.72 (95% CI: 0.65 to 0.79) and 0.76 (95% CI: 0.69 to 0.83) and had optimal cutoff values of 2.35 mL/min/g and 2.25 for detecting abnormal myocardial perfusion, defined as [15O]H2O PET-derived stress MBF ≤ 2.3 mL/min/g and MFR ≤ 2.5. Using these cutoff values, CMR and PET were concordant in 137 (77%) vascular territories for stress MBF and 135 (80%) vascular territories for MFR. Conclusion CMR measurements of stress MBF and MFR showed modest agreement to those obtained with [15O]H2O PET. Nevertheless, stress MBF and MFR were concordant between CMR and [15O]H2O PET in 77% and 80% of vascular territories, respectively.


Introduction
Cardiac magnetic resonance imaging (CMR) allows for the noninvasive assessment of myocardial perfusion in patients with suspected coronary artery disease (CAD), and its utilization for this task is recommended by contemporary guidelines [1]. In contrast to cardiac radionuclide modalities such as single-photon emission computed tomography and positron emission tomography (PET), considered the mainstay for the noninvasive evaluation of myocardial perfusion, CMR has superior spatial resolution and does not involve exposure to ionizing radiation. Although CMR perfusion images are predominantly assessed through visual analysis in clinical practice, quantification of absolute myocardial blood flow (MBF) and myocardial flow reserve (MFR) using CMR has gained increased interest. Quantification holds several advantages over a visual read. It is less dependent on the skill and experience of the observer and aids in identifying patients at risk for future cardiac events [2]. Most importantly, quantification may have incremental diagnostic value, particularly in the unraveling of homogenously diminished perfusion due to triple vessel or left main disease and subtle regional ischemia that goes undetected in a visual read. Indeed, previous studies have shown the need for quantitative CMR perfusion for improving detection and management of CAD [3][4][5]. Absolute quantification of myocardial perfusion with CMR has been validated ex vivo against microspheres [6,7]. In vivo, [

Study population and design
Sixty patients with stable CAD referred on a clinical basis to the Amsterdam University Medical Centers, location VUmc, were prospectively enrolled. Exclusion criteria were presence of myocardial scar on late gadolinium enhancement, history of coronary artery bypass grafting, acute myocardial infarction, atrial fibrillation, significant valvular disease, heart failure, non-ischemic cardiomyopathy, renal insufficiency (eGFR < 45 mL/min), and contraindications to intravenous adenosine or CMR. All patients underwent [ 15  Patients were instructed to refrain from products containing caffeine or xanthine for 24 h prior to scanning. Images were first obtained during resting conditions and thereafter during vasodilator stress. The PET sequence has been described in detail previously [9]. Briefly, 370 MBq of [ 15 O]H 2 O was injected intravenously as a 5 mL bolus (0.8 mL/s), immediately followed by a 35 mL saline flush (2 mL/s). A dynamic PET emission scan of 6 min was started simultaneously with tracer administration. After a delay of 15 min, an identical PET sequence was performed during continuous infusion of adenosine through a second venous cannula at a dose of 140 μg/kg/min. Adenosine was started 2 min prior to PET scanning to ensure maximal vasodilation. To correct for photon attenuation and scatter, low-dose (10 mA) respirationaveraged computed tomography scans were obtained during normal breathing just before the rest scan and immediately after the stress scan. Post-processing of PET data was done by a single observer (PvD), who was blinded to all clinical and CMR data. Parametric images of rest and stress perfusion were generated in approximately 10 min using an in-house developed software package, Cardiac VUer [10]. Absolute MBF was quantified in mL per minute per g of perfusable tissue.

Cardiac magnetic resonance imaging
Again, patients were instructed to refrain from products containing caffeine or xanthine for 24 h prior to image acquisition. All CMR images were obtained on a 1.5-T whole body MR scanner (Magnetom Avanto, Siemens, Erlangen, Germany). Perfusion imaging was performed using a dual sequence, single bolus technique [11], implemented as a Siemens works in progress software by C. Glielmi. Perfusion images were acquired using an echo-planar imaging sequence in three parallel short-axis slices planned at the basal, mid, and apical levels. To assess the arterial input function, low-resolution turboFLASH images were obtained at the basal level using a sequence optimized for the high gadolinium concentration. Perfusion images were obtained every heartbeat for 50-70 cardiac cycles following intravenous injection of a 0.075 mmol/kg bolus of a gadolinium-based contrast agent (DOTAREM®, Guerbet, Villepinte, France). Patients were asked to hold their breath as long as possible and breathe slowly thereafter. In-plane respiratory motion of the heart was corrected using non-rigid registration [12]. Perfusion images were corrected for surface coil-induced signal inhomogeneities using a separate prescan normalization [13]. Typical in-plane resolution of the myocardial perfusion images was 2.5 × 2.5 mm, with a slice thickness of 10 mm (pre-pulse 90°, repetition time 5.6 ms, echo planar factor 4, echo time 1.1 ms, saturation time 110 ms, flip angle 18°, matrix size 160 × 144, parallel imaging in the temporal direction [TGRAPPA] [14] factor 2). Perfusion imaging was performed first during vasodilator stress, which was induced by continuous infusion of adenosine using the same protocol as applied during PET. Rest perfusion images were obtained 15 min after stress imaging using identical scanning parameters and slice location. Left ventricular cardiac function was assessed in between stress and rest perfusion with steady-state free-precession cine imaging. Late gadolinium enhancement (LGE) was performed 12-15 min after rest perfusion using a 2D segmented inversion-recovery gradient-echo pulse sequence. Analysis of CMR data was done by a single observer (HE), who was blinded to all clinical and PET data. Post-processing of CMR perfusion images was performed in approximately 15 min using dedicated research software (MASS version 2017-Exp, Leiden, the Netherlands). A region of interest was placed in the LV blood pool of the image series obtained for the arterial input function. Care was taken to avoid inclusion of papillary muscles. Endocardial and epicardial contours were drawn manually on a single phase of each slice of the myocardial perfusion images. Subsequently, these contours were propagated to the other phases. Care was taken to avoid inclusion of blood pool or epicardial fat. Rest and stress MBF were quantified in mL per minute per g using Fermi functionconstrained deconvolution, as described previously [15]. Cine and LGE images were analyzed using a commercially available software (QMASS version 7.6, Medis, Leiden, the Netherlands). Left ventricular (LV) end-diastolic volume, endsystolic volume, and ejection fraction were calculated from the cine images. LGE images were visually assessed in order to rule out presence of LV scar tissue.

Data analysis
Perfusion data were analyzed according to the 17-segment model of the American Heart Association (AHA) [16]. The apical cap (segment 17) was excluded from analysis since this segment was not in the imaging planes of the CMR perfusion acquisition. Myocardial segments were also excluded from analysis if either PET or CMR perfusion imaging was of insufficient quality. Global rest and stress MBF were calculated by averaging perfusion over all 16 segments. In addition, myocardial segments were allocated to the three vascular territories (LAD, left anterior descending; LCx, left circumflex artery; and RCA, right coronary artery) as follows: LAD, segments 1,2,7,8,13,14; LCx, segments 5,6,11,12,16; and RCA, segments 3,4,9,10,15. Rest and stress MBF were calculated for each vascular territory by averaging perfusion over the corresponding myocardial segments. Myocardial flow reserve (MFR) was defined as the ratio of stress to rest MBF and was calculated on a global as well as a regional level. Concordance between CMR and PET was assessed on a per-vessel basis. For [ 15 O]H 2 O PET, stress MBF ≤ 2.3 mL/min/g and MFR ≤ 2.5 were considered abnormal according to previously validated cutoff values for diagnosing hemodynamically obstructive CAD (i.e., fractional flow reserve ≤ 0.80) [17]. Receiver operator characteristic (ROC) curve analysis and the Youden index were used to define optimal cutoff values for CMR measurements of stress MBF and MFR (MBF CMR and MFR CMR ).

Statistical analysis
Continuous variables are presented as mean ± standard deviation or median with inter-quartile range. Categorical variables are expressed as frequency with percentage. Pearson's correlation was used to quantify association between continuous variables. Agreement between PET and CMR perfusion was assessed by intraclass correlation coefficients (ICCs) and visually by Bland-Altman analysis. ICCs for absolute agreement of single measures were estimated using a two-way mixed effects model. Paired samples' T-tests were used to compare the means in heart rate and global perfusion measurements between CMR and PET. To account for clustering of multiple vessel measurements per patient, means of regional perfusion indexes were compared using a mixed linear model with a fixed effect for imaging technique and random effects for patient and vessel nested within patient. All statistical tests Stress perfusion CMR images were deemed of insufficient quality in one (2%) patient, which was excluded from analysis. In an additional three (5%) patients, rest perfusion imaging was omitted from the CMR scanning protocol. Baseline characteristics of the final cohort of 59 patients are shown in Table 1. Median time between PET and CMR was 5 [5] days. Table 2 lists data on CMR-derived LV function and volumes. LV ejection fraction was normal, with a mean of 63 ± 5%. Resting heart rate during perfusion imaging did not differ between CMR and PET (63 ± 9 vs. 64 ± 11 bpm; P = 0.52). Heart rate during vasodilator stress (89 ± 14 vs. 87 ± 14 bpm; P = 0.21) and the increment in heart rate (26 ± 12 vs. 22 ± 12 bpm; P = 0.11) was also similar for both techniques, indicating equal hemodynamic response to adenosine. Figure 1 shows a case example of concordance between CMR and PET in a patient with severely impaired stress perfusion and MFR in the vascular territory of the RCA.  2.9 ± 0.8 mL/min/g; P = 0.14). Global MFR was significantly lower for CMR in comparison to PET (2.6 ± 0.7 vs. 3.2 ± 1.0; P < 0.001).

Regional myocardial perfusion
The relationship between CMR and PET measurements of regional myocardial perfusion is shown in Fig. 3. On a per

Discussion
The present study is the largest to date investigating the agreement between CMR and PET measurements of absolute Previouspredominantly PETstudies have shown quantification of MBF to improve both prognostic and diagnostic performance in the management of patients with CAD [2,4,[18][19][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 [ 15 O]H 2 O 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 13 N-ammonia PET in 10 and 4 healthy volunteers, respectively, reported similar results [24,25]. In contrast, Tomiyama et al. studied 10 healthy volunteers with [ 15 O]H 2 O 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

Regional perfusion
Rest MBF (mL/min/g) 1.2 ± 0.3 0.9 ± 0.2 < 0.001 Stress MBF (mL/min/g) 3.1 ± 0.9 2.9 ± 0.8 0.014 MFR 2.7 ± 0.9 3.2 ± 1.1 < 0.001 Data are mean ± standard deviation. CMR ,cardiac magnetic resonance imaging; MBF ,myocardial blood flow; MFR ,myocardial flow reserve; PET ,positron emission tomography (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 noninferior to PET. Engblom Fig. 1 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. Intermethod 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 [ 15 O]H 2 O 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.

Study limitations
The present study lacks invasive confirmation of hemodynamically obstructive CAD. Although [ 15 O]H 2 O PET is considered to be the reference standard for quantification of myocardial perfusion, invasive measurements of fractional flow reserve are the preferred reference for diagnosing hemodynamically obstructive CAD and guiding revascularization. Therefore, our results urge for a new study comparing quantitative PET and CMR head-to-head against invasive measures of physiology. Secondly, resting flow is known to vary according to metabolic demand [37]. Some of the observed discrepancy between CMR and PET is therefore not attributable to differences in methodology but results from physiological fluctuations in rest flow. Finally, the sequence of imaging was similar in all patients as [ 15 Fig. 1 Philips and Biotronik and financial support from the Netherlands Organization for Health Research and Development (grant 9,071,544).

Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of interest.
Ethical approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the principles of the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.
Informed consent Written informed consent was obtained from all individual participants included in the study.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.