Stress myocardial perfusion with qualitative magnetic resonance and quantitative dynamic computed tomography: comparison of diagnostic performance and incremental value over coronary computed tomography angiography

Abstract

Aims

Assessment of haemodynamically significant coronary artery disease (CAD) using cardiovascular magnetic resonance (CMR) imaging perfusion or dynamic stress myocardial perfusion imaging by computed tomography (CT perfusion) may aid patient selection for invasive coronary angiography (ICA). We evaluated the diagnostic performance and incremental value of qualitative CMR perfusion and quantitative CT perfusion complementary to cardiac computed tomography angiography (CCTA) for the diagnosis of haemodynamically significant CAD using fractional flow reserve (FFR) and quantitative coronary angiography (QCA) as reference standard.

Methods and results

CCTA, qualitative visual CMR perfusion, visual CT perfusion, and quantitative relative myocardial blood flow (CT-MBF) were performed in patients with stable angina pectoris. FFR was measured in coronary vessels with stenosis visually estimated between 30% and 90% diameter reduction on ICA. Haemodynamically significant CAD was defined as FFR <0.80, or QCA ≥80% in those cases where FFR could not be performed. A total of 218 vessels from 93 patients were assessed. An optimal cut-off of 0.72 for relative CT-MBF was determined. The diagnostic performances (area under the receiver-operating characteristics curves, 95% CI) of visual CMR perfusion (0.84, 0.77–0.90) and relative CT-MBF (0.86, 0.81–0.92) were comparable and outperformed visual CT perfusion (0.64, 0.57–0.71). In combination with CCTA ≥50%, CCTA + visual CMR perfusion (0.91, 0.86–0.96), CCTA + relative CT-MBF (0.92, 0.88–0.96), and CCTA + visual CT perfusion (0.82, 0.75–0.90) improved discrimination compared with CCTA alone (all P < 0.05).

Conclusion

Visual CMR perfusion and relative CT-MBF outperformed visual CT perfusion and provided incremental discrimination compared with CCTA alone for the diagnosis of haemodynamically significant CAD.

Introduction

For the identification and management of patients with stable angina pectoris, current guidelines recommend non-invasive testing using anatomical or functional imaging prior to invasive assessment.¹ Cardiac computed tomography angiography (CCTA) is an excellent anatomical imaging modality for accurately ruling out coronary artery disease (CAD). One limitation of CCTA, however, is that when this test is used as a first-line modality, anatomical findings of coronary luminal stenoses provided by CCTA are not always informative of the haemodynamic significance of CAD, especially in the case of intermediate coronary stenoses.

Cardiovascular magnetic resonance (CMR) perfusion imaging has established high diagnostic performance in determining visually assessed myocardial ischaemia in clinical practice and is radiation free.^2,3 In recent years, new CT-based techniques for the functional assessment of CAD have emerged. CT-derived fractional flow reserve (FFR) applies advanced computational modelling and/or machine learning techniques and does not require additional radiation or contrast injection. CT-derived FFR, however, is not ubiquitous and cannot always replace myocardial physiologic or functional testing. Dynamic stress computed tomography myocardial perfusion (CT perfusion) allows the derivation of myocardial blood flow (MBF), which can be assessed visually from colour-coded polar maps and can be quantified. Animal⁴ and patient studies^5,6 have demonstrated the feasibility of this technique. Large-scale application is, however, currently limited by radiation exposure and iodinated contrast requirement. If technical developments will enable improved, lower dose CT perfusion in the near future, this technique may become a practical tool to complement CCTA in selected circumstances and care settings, for instance, where levels of CMR availability are suboptimal or where patients cannot tolerate CMR.

This study aimed to evaluate the diagnostic performance of CMR perfusion alongside CT perfusion for the diagnosis of haemodynamically significant CAD after first-line CCTA. We reported FFR and quantitative coronary angiography (QCA) as invasive reference standard.

Methods

Study population

This prospective observational study included patients presenting with stable chest pain who were clinically referred for invasive coronary angiography (ICA) between 2014 and 2016. One to four weeks before ICA, patients underwent CCTA and adenosine-stress dynamic CT perfusion, followed by adenosine-stress CMR perfusion on the same day.

Study exclusion criteria were acute coronary syndrome, previous percutaneous or surgical coronary revascularization, severely impaired left ventricular ejection fraction (≤35%), estimated glomerular filtration rate <60 mL/min, and documented or suspected allergy to contrast and contraindications to adenosine infusion.

All patients gave written informed consent. This study was approved by the local ethics committee and complies with the second Declaration of Helsinki.

An overview of the image analysis techniques is given in Figure 1A–D.

CT acquisition protocol

A second-generation dual-source CT scanner (Somatom Definition Flash, Siemens, Forchheim, Germany) was used. The CT protocol included a prospectively electrocardiogram-triggered CCTA followed by a dynamic stress CT perfusion using a prospectively electrocardiogram-triggered axial shuttle scan mode with the table alternating between two positions for a total scan coverage of 72 mm. The dynamic stress CT perfusion dataset consisted of 13–14 volumes of the left ventricle acquired over 30 s. Adenosine was infused intravenously at a rate of 140 µg/mL/min for at least 3 min prior to scanning. A volume of 60 mL of iodinated contrast agent (Omnipaque 300, GE Healthcare, Chalfont St. Giles, UK) was injected intravenously using a 15-mL test bolus to time image acquisition, with an injection rate of 7.5 mL/s. CT perfusion imaging required a second 60 mL contrast bolus injection that was additional to that for CCTA. The total contrast volume received was 135 mL. The detailed scan protocol is given in Supplementary data online, Figure S1.

CMR acquisition protocol

A 1.5-T scanner (Achieva CV, Philips Healthcare, Best, the Netherlands) with a cardiac 32-channel phase array coil was used. Adenosine-stress CMR perfusion was undertaken using a single-shot balanced steady-state free precession sequence combined with parallel imaging (sensitivity encoding) in three 10-mm thick short-axis slices, with an intravenous bolus of 0.05 mmol/kg of gadoteric acid (Dotarem, Guerbet, USA) injected at 4 mL/s and followed by 30 mL of saline with the same injection rate. The adenosine administration protocol was the same as for CT. Rest perfusion imaging was performed 10–15 min after the stress acquisition with a further injection of 0.05 mmol/kg of gadoteric acid. For late gadolinium enhancement (LGE) imaging, a T1-weighted segmented inversion-recovery gradient echo pulse sequence was used. The left ventricle was imaged in 8 mm thick two-, three-, and four-chamber views, and with continuous short-axis views from base to apex with a 2-mm gap. The detailed scan protocol is given in Supplementary data online, Figure S1.

CT post-processing and image analysis

Haemodynamic response to adenosine was considered adequate if either  ≥ 10 bpm heart rate increase or ≥10 mmHg systolic blood pressure decrease was registered. Commercially available software (syngo 3D, MMW, Siemens, Erlangen, Germany) was used to analyse CCTA datasets and stenosis severity was reported as no stenosis, <25%, 25–49%, 50–69%, 70–99%, and occlusion.⁷

CT perfusion datasets were post-processed using commercial software (Volume Perfusion CT Body, Siemens) by an experienced operator. First, the left ventricle was segmented by drawing a volume of interest (VOI). The arterial input function (AIF) was sampled by placing regions of interest (ROIs) on the descending aorta in the cranial and caudal image stacks. Time-attenuation curves (TACs) were created for each myocardial volumetric image element (voxel). Dedicated parametric deconvolution based on a two-compartment model of intra and extravascular space was applied to fit the TACs. MBF (mL/100 mL/min) was calculated as:$MBF=maximum slope of the fit curve/maximum AIF6$

Prototype software (Cardiac Functional Analysis Protocol Build Data; Siemens) was used to generate polar maps representing the MBF distribution within the subendocardial layer of the left ventricular myocardium.⁸ The following two variables were generated:

1. Visual CT perfusion: MBF polar maps based on the 17-segment American Heart Association (AHA) model were inspected visually using a fixed colour scale. According to this scale, the colours violet, blue, green, yellow, and red were used to display values from 0 to 200 mL/100 mL/min. Areas coded as violet or blue were considered hypoperfused.

2. Relative CT-MBF: VOIs of at least 0.5 cm³ were manually drawn on the perfusion polar maps, guided by the colour-coded scale, to sample absolute values of MBF within each vascular territory. Relative CT-MBF was calculated as the ratio between the sampled blood flow and the value corresponding to the 75th percentile of the segmental MBF distribution.⁹

CMR image analysis

The presence of the splenic switch off phenomenon was assessed in each patient as a marker of adequate adenosine stress.¹⁰ This method was not suitable for CT as the spleen was not included in the scan range. Patients with inadequate haemodynamic response were excluded. A visual assessment of CMR perfusion was performed as follows:

Visual CMR perfusion: hyperintense territories observed on LGE suggesting myocardial infarction were excluded from the perfusion analysis. Both stress and rest perfusion images were evaluated. Hypoperfused areas were considered indicative of myocardial ischaemia when present on stress images for >6 heartbeats but not detectable on rest images. Hypoperfused areas were reported according to the 17-segment AHA model, excluding the apical segment. Equivocal hypoperfused areas were considered positive in the final analysis.

Matching coronary anatomy and vascular myocardial territories

To ensure accurate matching of between coronary vessels and myocardial territories, patient specific coronary anatomy on CCTA (right, left, or balanced dominance, and length of the left anterior descending artery) was used to decide which vessel (right coronary artery, left coronary artery, or both) supplied the inferior and inferoseptal segments in CT and CMR perfusion datasets, following an approach outlined by Cerci et al.¹¹

Invasive coronary angiography

ICA was performed according to local clinical standards. FFR was measured in coronary lesions with a visually assessed diameter narrowing between 30% and 90% using a sensor-tipped 0.014-inch guidewire (Pressure Wire, Radi Medical Systems, Uppsala, Sweden) placed just distal to the lesion. FFR was calculated as the ratio of mean distal pressure measured by the pressure wire divided by the mean proximal pressure measured by the guiding catheter during rest and during maximal hyperaemia induced by continuous intravenous infusion of adenosine (140 µg/kg/min for a minimum of 2 min).

ICA images were analysed offline on multiple projections by a single observer (7 years of experience) blinded to CCTA, CT, and CMR perfusion results. The most severely diseased segment in each coronary vessel was identified to derive the percentage diameter stenosis using validated software (QAngio^® XA, 7.3, Medis, Leiden, the Netherlands).

Reference standard for haemodynamically significant lesions

Vessels with FFR ≤0.80 were considered haemodynamically significant. If FFR could not be measured due to logistics, QCA was utilized and lesions were classified as: (i) lesions with ≥80% diameter reduction on QCA: haemodynamically significant and (ii) lesions with <30% diameter reduction on QCA: haemodynamically insignificant. This was based on the observation that a QCA of 80% is likely to correspond to a 90% visual stenosis (oculostenotic reflex).¹² Lesions with a QCA between 30% and 80% and not interrogated with FFR were excluded.

Statistical analysis

Continuous variables are presented as means ± SD or medians with interquartile ranges (IQR). Categorical variables are shown as frequencies and percentages. The Mann–Whitney U test adjusted for clustering using the Rosner–Glynn–Lee method was used to compare continuous non-normally distributed variables.¹³

CCTA was dichotomized using ≥50% and ≥70% stenosis thresholds. The Youden index from area under the receiver-operating characteristics curve (AUC) analysis was used to define a cut-off for relative CT-MBF. Logit-binomial regression analyses [using generalized estimating equations (GEEs) with an exchangeable working correlation matrix to account for clustering] were calculated to identify predictors of haemodynamically significant CAD.

Sensitivities, specificities, positive predictive values and negative predictive values were calculated using GEEs for the dichotomized variables, for combined evaluations and for a sequential approach where only perfusion assessments from vessels with obstructive CAD by CCTA (≥50%/≥70%) were assessed. Here, positive matched functional evaluations of included vessels were considered positive. Sensitivities and specificities were compared using the Obuchowski χ² test.¹⁴ AUCs were corrected for clustering and compared using the method described by Obuchowski.¹⁵

Interobserver reproducibility of absolute CT-MBF was evaluated in 72 randomly selected myocardial territories in 30 patients using the Bland–Altman method and reported as mean bias and 95% limits of agreement.¹⁶ Normality of the mean difference was tested. Furthermore, the coefficient of variation (CV) was calculated.

Two-sided P-values < 0.05 were considered statistically significant. Data were analysed using STATA Statistical Software release 15 (StataCorp LP, College Station, TX, USA), IBM SPSS Statistics Version 22.0 (IBM Corp., Armonk, NY, USA), R version 3.5.1, and R Foundation for Statistical Computing (Vienna, Austria).¹⁷

Results

Study population

A total of 218 coronary vessels and corresponding myocardial territories from 93 patients were available (Figure 2, study flowchart). Thirty-six vessels with a 30–80% diameter reduction on QCA were not investigated with FFR and were excluded. A total of 79 of the 218 vessels (36%) in 49/93 (53%) patients were directly assessed with FFR. A further 13 vascular territories were LGE positive and were excluded. For CCTA and CT perfusion, the median (interquartile range) dose length products were 235 (120–386) and 733 (624–795) mGy × cm, respectively. Using a k-factor of 0.026,¹⁸ these values correspond to effective doses of 6 and 19 mSv, respectively. Baseline characteristics are reported in Table1.

Relationship between coronary stenosis severity by CCTA and QCA with FFR

The relationship between the coronary stenosis severity by CCTA and QCA with FFR is illustrated in Figure 3. Using CCTA ≥50%, CCTA misclassified 29/79 (37%) coronary lesions as haemodynamically significant and 8/79 (10%) lesions as haemodynamically insignificant. QCA misclassified 12/79 (15%) lesions as haemodynamically significant, all in the 50–69% range, and 10/79 (13%) lesions as haemodynamically insignificant.

CMR perfusion

Visual CMR perfusion predicted haemodynamically significant CAD (Supplementary data online, Table S1). The diagnostic performance of visual CMR perfusion is given in Table 2 and Supplementary data online, Table S2. CCTA ≥50% + visual CMR perfusion provided incremental discrimination compared with CCTA ≥50% alone (Figure 4A–C).

CT perfusion

Median (IQR) absolute and relative CT-MBF in ischaemic vessel territories was lower than in myocardial territories supplied by unobstructed vessels: 95.77 (81.02–116.19) vs. 163.92 (130.72–191.72) mL/100 mL/min (P < 0.001) and 0.62 (0.49–0.69) vs. 0.92 (0.80–0.99) (P < 0.001), respectively.

Dichotomization of relative CT-MBF resulted in a Youden index-derived cut-off of 0.72 with 84% sensitivity and 89% specificity for the detection of haemodynamically significant CAD. Territorial odds ratios for both visual CT perfusion and relative CT-MBF ≤0.72 were highly significant for the detection of haemodynamically significant CAD (Supplementary data online, Table S1).

The diagnostic performance of visual CT perfusion and relative CT-MBF is given in Table 2 and Supplementary data online, Table S2. Compared with CCTA ≥50%, a 70% positivity threshold increased specificity at the expense of decreased sensitivity. Additionally, CCTA ≥50% + relative CT-MBF provided incremental discrimination over CCTA alone when using both ordinal/continuous and dichotomous variables (Figure 4A–C).

CMR perfusion vs. CT perfusion

AUCs (95% CI) for CCTA ≥50%, visual CMR perfusion, visual CT perfusion, and relative CT-MBF ≤0.72 were 0.79 (0.72–0.85), 0.84 (0.77–0.90), 0.64 (0.57–0.71), and 0.86 (0.81–0.92), respectively. Relative CT-MBF and visual CMR perfusion showed comparable discrimination for the detection of haemodynamically significant CAD (P = 0.514) and both outperformed visual CT perfusion (both P < 0.001). The discrimination of all CT perfusion parameters to diagnose haemodynamically significant CAD is given in Supplementary data online, Table S3.

Sensitivity and specificity of each modality are given in Table 2. When considering the sequential model, CCTA ≥50% + visual CMR perfusion and CCTA ≥50% + relative CT-MBF had better sensitivity compared with CCTA ≥50% + visual CT perfusion (all P < 0.01). Sensitivities for CCTA ≥50% + visual CMR perfusion and CCTA ≥50% + relative CT-MBF were similar (P = 0.743). Regarding specificity, there were no differences between CCTA + visual CMR perfusion, CCTA + relative CT-MBF and CCTA + visual CT perfusion in the sequential model (all P ≥ 0.105).

Reproducibility

The interobserver reproducibility of absolute CT-MBF had a mean bias and 95% limits of agreement of 2.21 ± 27.16 mL/100 mL/min and a CV of 10% demonstrating good reproducibility (Figure 5).

Discussion

The main findings of this study were that visual CMR perfusion and relative CT-MBF had similar diagnostic value and were superior to visual CT perfusion. Dynamic CT perfusion obtained a diagnostic performance similar to CMR perfusion when it was assessed quantitatively. Combined CCTA and perfusion imaging by CMR or CT increased diagnostic accuracy compared with CCTA alone.

A diagnostic strategy starting with CCTA followed by CMR perfusion in patients with stable chest pain was evaluated by Groothuis et al.¹⁹ In this study, we used established CMR perfusion alongside more recently introduced CT perfusion imaging. Given that it is unlikely that all patients will undergo both an anatomical and perfusion assessment, we simulated a diagnostic workup where CCTA was used as a first-line examination followed by a functional examination in the case of a positive CCTA result. This could be cost-effective and radiation-saving, as proposed by guidelines (level IB).^1,20 In our study, only 62 patients underwent perfusion assessments (compared with 93 patients) when using a 50% CCTA threshold.

This serial approach was also tested in a randomized trial by Lubbers et al.²¹ In this trial, the rate of downstream invasive procedures with an indication for revascularization in the CT arm was 88% compared with 50% in the standard approach arm. This raises the issue of how to best interpret and report CCTA to maximize diagnostic yield and patient benefit without triggering excessive rates of downstream testing. In the present study, a CCTA positivity threshold of 70% increased specificity at the expense of sensitivity. While CCTA ≥70% was specific for flow limiting CCTA, haemodynamically significant lesions with lower anatomical severity may be overlooked. These findings were in accordance with CAD-RADS guidelines,⁷ which suggested referral to ICA in case of CCTA ≥70% and to functional testing in intermediate (50–69%) lesions.

Arguably, the anatomical severity of CCTA stenosis in a coronary vessel may not be the only or the most important piece of information to guide subsequent management. As emerged from the ISCHAEMIA trial,²² severe extent of coronary disease was associated with mortality and myocardial infarction, while severe perfusion deficits on stress testing were associated with myocardial infarction. In both groups of patients, this trial found a similar overall lack of benefit for invasive vs. conservative therapy. In light of these findings, revascularization for stable CAD should be weighted in the context of a patient’s angina burden and effectiveness of background medical management, provided that optimal revascularization can be achieved with a very low rate of procedural complications. Anatomical and functional imaging may be used to support these clinical decisions. For the purpose of this diagnostic performance study, the reported perfusion imaging performance metrics were based on standalone and standardized landmarks of anatomical coronary lesion severity.

We found better sensitivities of visual CMR-MPI compared with the Dan-NICAD trial.²³ One difference between this study and the Dan-NICAD trial resided in the prevalence of CAD in the study population. This was more than two-fold higher in the present study (49%) compared with the Dan-NICAD trial (21%).

A few clinical and imaging features may affect the diagnostic performance of CCTA, CT, and CMR perfusion. On CCTA, calcified plaque and the presence of stents impede the correct assessment of luminal narrowing due to the blooming effect. Similarly, high body mass index, poor heart rate control, arrhythmias, and breathing artefacts all result in reduced CT image quality with detrimental impact on diagnostic performance. In perfusion imaging, blunt or suboptimal hyperaemic response may hinder the ability to detect reversible ischaemia.

CT perfusion involves exposure to radiation and contrast agent. Although efforts towards dose reduction and quality standardization of CT perfusion are ongoing with encouraging results,^24–30 if a patient undergoes CCTA and CT perfusion, radiation exposure and contrast volume will be increased. In patients with impaired kidney function, a double contrast injection may be of concern. Arguably, CCTA followed by perfusion CMR, an imaging modality free from ionizing radiation, may be a safer option in these circumstances. CT-derived FFR may replace perfusion imaging in a number of situations. It makes sense to consider these technologies side-by-side to exploit the full diagnostic potential of CCTA. CT-derived FFR, however, is not equivalent to physiologic testing and may not be available in cases with suboptimal image quality, small coronary vessel anatomy, previous coronary revascularization, or other clinical or logistic circumstances.

Better knowledge of the physiological range and variability of blood flow across strata of the population, with validated normative values, is a further requirement prior to robust implementation of CT perfusion. The absolute range of MBF values from CT appears to be underestimated compared with other imaging modalities such as positron emission tomography, probably due to limitations in sampling frequency and mathematical modelling. Its discriminatory ability to identify functionally CAD, however, appears very encouraging.³¹

Last but not least, despite contemporary guidance, it could be argued that a strategy based on first-line CCTA may be suboptimal or questionable in certain clinical circumstances. Patients who are very likely to have CAD, with extremely calcified coronary arteries, with complex previous revascularization or patients who cannot receive potentially nephrotoxic contrast agent may not be ideal candidates for CCTA. Young patients, especially women, could benefit from a test free from radiation, such as perfusion CMR or stress echocardiography.

Limitations

Absolute quantification of CMR perfusion is likely to outperform the parameters of visual CMR investigated in the present work, as shown in previous research.^32–36 In this work, however, it was not possible to provide absolute quantification of CMR perfusion due to a single-bolus, single-sequence approach used.

FFR was not performed in all vessels. Angiographically near-normal vessels or vessels with severe stenosis were not eligible for FFR. QCA was used to predict the flow-limiting potential of these lesions. This approach was unlikely to have resulted in functional misclassification of severe anatomic lesions and represents routine clinical practice.¹² Lesions with a QCA between 30% and 80% and not interrogated with FFR were excluded (n = 36). FFR was performed at the operator’s discretion and lesion stenosis severity during ICA was judged based on visual analysis. Discrepancies between visual assessment and QCA (performed post-ICA) are to be expected and, in addition to possible logistic reasons, explain the exclusion of these 36 vessels.

FFR is a pressure measurement restricted to the epicardial vessels while myocardial perfusion is a functional measure of both narrowing in epicardial coronary vessels and the endothelial response of the microvasculature. In the present study, the significance of microvascular dysfunction was reduced by excluding patients with previous myocardial infarction and severely impaired left ventricular ejection fraction as well as excluding myocardial territories that exhibited LGE on CMR.

With regards to the relative CT-MBF cutoff used in this study, the study sample size did not allow for a derivation and validation population and further studies are, therefore, needed for external validation.

In this exploratory study, power calculations were not performed. The findings from the present study may, however, be used to power future prospective studies.

Finally, generalizability of findings to other populations may be reduced as 80% of the study subjects were male and all were clinically assessed as requiring invasive investigation.

Conclusions

In patients presenting with stable angina pectoris, anatomical assessments of stenosis severity by CCTA and dynamic stress perfusion by CMR and CT performed well in predicting haemodynamically significant CAD. Visual CMR perfusion and relative CT-MBF outperformed visual CT perfusion.

Supplementary data

Supplementary data are available at European Heart Journal - Cardiovascular Imaging online.

Funding

M.C.K. was supported by The Danish Heart Foundation (grant number 16-R107-A6719-22959), Eva og Henry Frænkels Mindefond and Snedkermester Sophus Jacobsen og hustru Astrid Jacobsens Fond. A.R. has received a training grant awarded by the European Society of Cardiology. F.P. has received grant support from Siemens Healthineers. This work forms part of the translational research portfolio of the NIHR Barts BRC, which was supported and funded by the NIHR. The funders provided support in the form of salaries for authors but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conflict of interest: M.C.K. has received lecturing fees from Toshiba Medical Systems. F.P. received research funding from Siemens Healthineers and speaker’s honoraria from Bracco. S.E.P. provides consultancy to and has options for shares for Circle Cardiovascular Imaging, Inc., Calgary, Canada. K.N. reports unrestricted institutional research support from Siemens Healthineers, HeartFlow Inc., Bayer, and GE Healthcare. F.B. has received institutional research grants from Siemens Healthineers and Bayer Healthcare and speakers honoraria from Siemens Healthineers, Bayer Healthcare, and Bracco. All other authors report no support from any organization for the submitted work; no financial relationships with any organizations that might have an interest in the submitted work in the previous two years; and no other relationships or activities that could appear to have influenced the submitted work.

References