Diagnostic Accuracy of Spiral Whole-Heart Quantitative Adenosine Stress Cardiovascular Magnetic Resonance With Motion Compensated L1-SPIRIT

Abstract

Background:

Variable density spiral (VDS) pulse sequences with motion compensated compressed sensing (MCCS) reconstruction allow for whole-heart quantitative assessment of myocardial perfusion but are not clinically validated.

Purpose:

Assess performance of whole-heart VDS quantitative stress perfusion with MCCS to detect obstructive coronary artery disease (CAD).

Study Type:

Prospective cross sectional.

Population:

Twenty-five patients with chest pain and known or suspected CAD and nine normal subjects.

Field strength/Sequence:

Segmented steady-state free precession (SSFP) sequence, segmented phase sensitive inversion recovery sequence for late gadolinium enhancement (LGE) imaging and VDS sequence at 1.5 T for rest and stress quantitative perfusion at eight short-axis locations.

Assessment:

Stenosis was defined as ≥50% by quantitative coronary angiography (QCA). Visual and quantitative analysis of MRI data was compared to QCA. Quantitative analysis assessed average myocardial perfusion reserve (MPR), average stress myocardial blood flow (MBF), and lowest stress MBF of two contiguous myocardial segments. Ischemic burden was measured visually and quantitatively.

Statistical Tests:

Student’s t-test, McNemar’s test, chi-square statistic, linear mixed-effects model, and area under receiver-operating characteristic curve (ROC).

Results:

Per-patient visual analysis demonstrated a sensitivity of 84% (95% confidence interval [CI], 60%–97%) and specificity of 83% [95% CI, 36%–100%]. There was no significant difference between per-vessel visual and quantitative analysis for sensitivity (69% [95% CI, 51%–84%] vs. 77% [95% CI, 60%–90%], P = 0.39) and specificity (88% [95% CI, 73%–96%] vs. 80% [95% CI, 64%–91%], P = 0.75). Per-vessel quantitative analysis ROC showed no significant difference (P = 0.06) between average MPR (0.68 [95% CI, 0.56–0.81]), average stress MBF (0.74 [95% CI, 0.63–0.86]), and lowest stress MBF (0.79 [95% CI, 0.69–0.90]). Visual and quantitative ischemic burden measurements were comparable (P = 0.85).

Data Conclusion:

Whole-heart VDS stress perfusion demonstrated good diagnostic accuracy and ischemic burden evaluation. No significant difference was seen between visual and quantitative diagnostic performance and ischemic burden measurements.

Evidence Level:

2

Technical Efficacy:

Stage 2

Background

Over 1 million coronary angiograms are performed in the United States annually to evaluate for obstructive coronary artery disease (CAD).¹ While approximately 60% of patients undergo noninvasive evaluation to detect myocardial ischemia prior to coronary angiography, the accuracy of most noninvasive tests for identifying obstructive CAD is modest.² The diagnostic yield for detection of obstructive disease on elective coronary angiography is approximately 40%.³ Cardiac MRI, however, has demonstrated high diagnostic accuracy when compared to coronary angiography with fractional flow reserve (FFR).⁴ Furthermore, the MR-INFORM trial⁵ has shown that patients with stable angina and risk factors for CAD can be managed noninvasively with MRI adenosine stress perfusion imaging rather than invasively with coronary angiography and FFR, leading to a lower rate of revascularization without any difference in major adverse outcomes.

Studies have shown that quantification of myocardial perfusion can confer additional diagnostic and prognostic information, especially in patients with multivessel disease⁶ and microvascular dysfunction.⁷ Positron emission tomography (PET) is the most widely used modality for noninvasive measurements of myocardial blood flow (MBF) and whole-heart quantification but suffers from limited spatial resolution and the need for ionizing radiation.⁸ Fully quantitative MR perfusion imaging has comparable diagnostic performance and perfusion estimates to that of PET,⁹ but whole-heart coverage is not widely available due to limited spatial–temporal resolution and dark-rim artifacts.¹⁰ However, improvements in compressed sensing reconstruction, spatial–temporal acceleration techniques, and efficient k-space trajectories in the past decade have laid the groundwork for whole-heart coverage in first-pass perfusion imaging.¹¹

A previous study demonstrated that a variable density spiral (VDS) pulse sequence with an integrated arterial input function (AIF) could acquire eight short-axis slices with 2 mm in-plane resolution at heart rates up to 125 bpm, providing whole-heart quantitative assessment of perfusion.¹¹ The aim of this study was to assess the clinical performance of this technique for detecting obstructive CAD using both quantitative and visual analysis.

Methods

Study Population

All patients provided written informed consent on a protocol, approved by our local institutional review board. Twenty-seven patients who were scheduled to undergo coronary angiography for evaluation of chest pain with known or suspected CAD were prospectively recruited between September 2014 and November 2018. Patients with prior known history of CAD, myocardial infarction, or percutaneous coronary intervention were eligible for the study. Patients with prior coronary artery bypass, however, were excluded due to the different adenosine stress MRI characteristics seen in this population. Additional exclusion criteria were ejection fraction < 45%, cardiomyopathy, significant valvular disease, pulmonary hypertension, transplant vasculopathy, glomerular filtration rate < 45 mL/min per 1.73 m², or contraindications to perfusion MRI. The latter included implantable cardiac devices, inability to tolerate adenosine due to asthma or severe chronic obstructive pulmonary disease, or history of prior gadolinium contrast agent reaction. All patients had their renal function assessed within 30 days before the MRI study. When feasible, MRI was performed on the morning of the scheduled cardiac catheterization. A detailed history and physical examination was performed by a physician before the MRI study. Nine healthy controls were also included in this study.

MRI Protocol

All stress MRI studies were performed on a 1.5 T scanner (Avanto or Aera, Siemens Medical Systems). Resting heart rate and blood pressure were recorded. The MRI protocol included anatomic imaging using a single-shot steady-state free precession (SSFP) pulse sequence, ventricular function imaging using a segmented cine SSFP pulse sequence, and late gadolinium enhancement (LGE) imaging using a segmented spoiled gradient-echo phase-sensitive inversion recovery (PSIR) pulse sequences following standard methodology.¹² After localization, cine images were acquired in 2, 3, and 4 chamber views with 6 mm slice thickness. After adenosine stress imaging, contiguous short-axis cine images were acquired from base to apex with 8 mm slice thickness. Other sequence parameters included: repetition time (TR) 2.7 msec, echo time (TE) 1.3 msec, flip angle (FA) 73°, field of view (FOV) 300–350 mm, and resolution 1.8 × 1.4 × 8.0 mm. Resting perfusion imaging was performed approximately 10 minutes after stress to allow for contrast washout. Late gadolinium enhancement (LGE) was performed at least 5 minutes following the second gadolinium bolus injection, using a segmented phase sensitive inversion recovery sequence (TR 650 msec, TE 4.2 msec, FA 25°, FOV 300–340 mm, resolution 1.8 × 1.3 × 8 mm) in the same slice location as the cine images.

Perfusion Imaging

Adenosine (Astellas Pharmaceuticals) was infused at 140 mcg/kg/min through a peripheral IV for 4 minutes. A bolus of 0.075 mmol/kg of gadolinium contrast (Magnevist, Bayer Pharmaceuticals) was injected through a second IV in the contralateral arm at 4 mL/sec. Eight short-axis slice locations were imaged per heart beat over 60 heart beats using a saturation recovery (SR) VDS perfusion pulse sequence immediately following R-wave detection. The technical details of the pulse sequence have been described previously.¹¹ Subjects were instructed to hold their breath as long as possible and then to take shallow breaths. Pulse sequence parameters included three interleaves per slice, two interleaved slices per saturation, 5.12 msec readout duration per interleave, TE 1 msec, TR 7 msec, effective repetition time for each slice with interleaved acquisition (TReff) 14 msec, FA 30°, SR time 80 msec (to first radiofrequency pulse of the readout), FOV 340 mm², in-plane resolution 2 mm², and slice thickness 10 mm (Fig. 1). As the spiral readouts from two slice locations were acquired in an interleaved fashion, the data for each slice were acquired with a temporal footprint of 35 msec with a total readout time of 42 msec for the three interleaves from both slices (six total interleaves). Given a minimal overhead for the saturation pulse, the minimum time to acquire two slices including the saturation pulse was 132 msec. AIF images were acquired during the SR time of the first perfusion image with a 2× accelerated single-shot spiral acquisition using a 90° FA, in-plane resolution of 6.95 mm², and SR time of 10 msec. Proton density-weighted (PDW) images were acquired during the first four heart beats for signal modeling. Images were reconstructed using a rigid-motion compensated L1-SPIRIT.^11,13 After a 10 minute washout delay following the stress perfusion acquisition, resting perfusion imaging was performed during a second injection of 0.075 mmol/kg of Magnevist using the same imaging protocol as described above.

FIGURE 1:

Schematic of quantitative whole-heart spiral perfusion pulse sequence. Following the BIR-4 saturation pulse, the arterial input function (AIF) is acquired after a short saturation time (TS) using a single-shot acquisition. Following a fat saturation pulse, two slice locations are sampled in an interleaved fashion. Each slice image is acquired with three interleaves with a repetition time (TR) of 7 msec. Due to slice interleaving, the effective temporal resolution of each slice is 5*TR (35 msec), and the TS times differ by one TR (7 msec).

Perfusion Quantification and Evaluation

The quantitative perfusion and AIF methods have been previously described and validated.^7,14,15 Perfusion images were aligned with nonrigid registration using the ANTs toolbox.¹⁶ Perfusion images and AIF images were normalized by their respective PDW images, and Bloch simulation was used to convert the signal intensity to gadolinium concentration units.¹⁷ For the tissue function quantification, a modification of the equation in Cernicanu and Axel¹⁷ was made to account for the repeated sampling of the center of k-space with spiral acquisition. Quantification of first-pass perfusion was performed on a pixel-wise basis using the constrained Fermi function deconvolution method in a custom-built MATLAB (Mathworks, Natick, Massachusetts) environment.¹⁸ Perfusion was reported on both a per-vessel and per-patient basis. Coronary vessel territories included left anterior descending coronary artery (LAD), right coronary artery (RCA), and left circumflex coronary artery (LCx). Each short-axis slice was divided into six segments with the apical cap slice representing a single segment in the LAD territory. Stress and rest MBF were quantified for each segment. Myocardial perfusion reserve (MPR) was calculated for each segment as the ratio of the stress MBF divided by rest MBF. Average MPR and stress MBF were calculated based on the total number segments on a per-patient and per-vessel basis. The lowest per-vessel stress MBF was calculated based on the lowest average of two contiguous segments within each respective coronary territory. The lowest per-patient stress MBF was calculated from the lowest average of two contiguous segments within any of the three coronary vessel territories.

Visual Stress Perfusion Assessment

Three reviewers (A.A.R. 3 years, M.S. 13 years, and C.M.K. > 20 years) blinded to the coronary angiography data evaluated the perfusion and LGE images to assess for the presence of perfusion abnormalities consistent with myocardial ischemia on a per-patient and per-vessel territory basis. Perfusion studies were rated on scale of 0 to 3 (0 [definitely normal], 1 [probably normal], 2 [probably abnormal], and 3 [definitely abnormal]). Image quality and motion correction were graded on a 5 point scale (1 [poor] to 5 [excellent]). The perfusion images were converted offline to animated gifs and interpreted using ImageJ (http://rsbweb.nih.gov/ij/). In cases of perfusion abnormalities, LGE images were assessed visually to determine whether the region of perfusion abnormality was larger than the area of scar. A positive study was considered one with evidence of a perfusion abnormality in the absence of LGE or with an extent of the perfusion abnormality exceeding the area of LGE, indicating ischemia. Given that the Duke algorithm is only validated for patients without known CAD, and we specifically sought to determine the performance of the stress perfusion component of the study alone, the Duke algorithm was not used in this study.¹⁹

Quantitative Coronary Angiography

Patients underwent coronary angiography using standard techniques.²⁰ The severity of stenosis was assessed by an independent, blinded interventional cardiologist (A.M.T. 16 years) using quantitative coronary angiography (QCA). QCA was performed using automatic edge detection software at an end-diastolic frame based on the visualization of the most severe stenosis with minimal foreshortening or branch overlap. The minimal lumen diameter was recorded in each coronary branch with a reference diameter > 2 mm.

Statistical Analysis

Continuous data were expressed as mean ± SD, and categorical data were expressed as number (percent). Performance of binary classification tests and ischemic burden was expressed as point estimate [95% confidence interval]. Differences in mean stress MBF were assessed using Student’s t-test. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy for the visual and quantitative MRI analysis were calculated based on detection of obstructive CAD, as defined by ≥50% stenosis by QCA. Ninety-five percent confidence intervals (CIs) for diagnostic performance estimates were determined using an exact binomial method based on the F-distribution. For quantitative analysis, optimal cut-off values were determined based on Youden index. Per-vessel sensitivity, specificity, and accuracy were compared using McNemar’s test. Per-vessel quantitative perfusion receiver-operating characteristic (ROC) curves were generated and areas under the curve (AUC) were compared using chi-square statistic. Ischemic burden was measured visually and quantitatively based on the optimal cut-off for average stress MBF. A linear mixed-effects model was used to compare ischemic burden with severity of CAD (single vs. multivessel disease) and analysis method (visual vs. quantitative analysis) as fixed effects and patients as random effects. Statistical analysis was performed using SAS/STAT software, version 9.4, of the SAS System for Windows (SAS Institute Inc., Cary, North Carolina). Statistical significance was determined as P < 0.05. The inter-reader variability for visual analysis was assessed by kappa-statistic. The intra- and inter-reader reliability for quantitative analysis were assessed by Bland–Altman plots using a subset of 10 patients, including both rest and stress MBF. For sensitivity analysis, point estimates were also calculated based on a reference test of ≥70% stenosis by QCA.

Results

Clinical Characteristics

All the patients underwent both adenosine stress and rest imaging without any complications. Of the 27 patients who completed the MRI study, 25 were included in the final analysis. Two patients were excluded due to significant motion artifact from breathing that rendered their studies uninterpretable. Table 1 summarizes the patient characteristics of those included in the study. Eleven patients (44%) had evidence of LGE in an ischemic pattern. A total of 19 patients (76%) demonstrated significant coronary artery stenosis (≥50% luminal diameter reduction in vessels with >2 mm diameter). Ten patients (40%) had single-vessel disease, two (8%) had 2-vessel disease, and seven (28%) had 3-vessel disease. None of the patients had left main disease. In the patients with single-vessel disease, seven were in the LAD territory and three were in the RCA territory. The normal healthy subjects had a mean age of 30 ± 16 with 44% males. All of the patients had normal wall motion, normal stress and resting perfusion imaging, and no LGE.

TABLE 1.

Demographic and Clinical Characteristics

Number of Patients	25 included
Clinical characteristics
Age (years)	63 ± 12
Male (Sex)	21 (84%)
Weight (kg)	89 ± 15
Height (m)	1.75 ± 0.08
Body mass index (kg/m2)	29 ± 4
Risk factors
Hypertension	17 (68%)
Hyperlipidemia	22 (88%)
Diabetes Mellitus	9 (36%)
Prior smoker	10 (40%)
Chest pain	18 (72%)
Dyspnea	12 (48%)
History of CAD	9 (36%)
Prior PTCA	5 (20%)
Prior myocardial infarction	5 (20%)
Medications
Aspirin	18 (72%)
Clopidogrel	2 (8%)
Beta-blocker	14 (56%)
ACE inhibitor/ARB	13 (52%)
Statin	19 (76%)
Other cholesterol medications(s)	2 (8%)
Oral hyperglycemic agent	6 (24%)
Insulin therapy	2 (8%)
Angiography findings
No CAD	3 (12%)
Stenosis < 50%	3 (12%)
Single-vessel	10 (40%)
Two-vessel	2 (8%)
Three-vessel	7 (28%)
Left main	0 (0%)

Values are presented as mean ± SD or n (%).

ACE = angiotensin converting enzyme inhibitors; ARB = angiotensin II receptor blocker; CAD = coronary artery disease; PTCA = percutaneous transluminal coronary angioplasty.

Stress MBF Quantification

Figure 2 shows (a) stress and (b) rest spiral perfusion images from a patient who had obstructive 2-vessel CAD with (c) 70% stenosis of the LCx, and (d) 100% occlusion of the RCA. The patient also had a 40% nonobstructive lesion in the LAD and no evidence of LGE. A visual perfusion defect is seen in the inferior wall, corresponding to the occlusion of the RCA on coronary angiogram. Notably, pixel-wise quantification of perfusion (Fig. 3) demonstrated reduced stress MBF in the LCx and RCA territories consistent with 2-vessel disease. Figure 4 shows (a) stress and (b) rest spiral perfusion images from a patient with obstructive 3-vessel disease. QCA showed 60% stenosis of the LAD, 70% stenosis of the marginal branch from the LCx, and 60% stenosis of the RCA (Fig. 4c,d). Pixel-wise quantification of perfusion (Fig. 5) demonstrated globally reduced stress MBF in all segments consistent with obstructive 3-vessel disease. Refer to Supplemental Material Figs. S1 and S2 for the stress and rest perfusion movies.

FIGURE 2:

Stress and rest perfusion visual analysis in obstructive 2-vessel CAD with an occluded RCA. (a) Stress perfusion images demonstrate a subendocardial perfusion defect (arrows) in the inferior wall, which is not present at (b) rest, correlating with the right coronary artery (RCA) territory. Invasive coronary angiography revealed (c) 70% stenosis in the left circumflex artery (LCx) and (d) 100% occlusion of the RCA (labeled arrows), consistent with 2-vessel obstructive coronary artery disease. There was also a 40% nonobstructive lesion in the left anterior descending artery (LAD). This subject had no evidence of late gadolinium enhancement.

FIGURE 3:

Stress and rest MBF quantification in obstructive 2-vessel CAD with an occluded RCA. Stress perfusion maps (top panel) from the subject in Fig. 2 demonstrate reduced stress myocardial blood flow (MBF) most obvious in the right coronary artery (RCA) territory. Resting perfusion maps are relatively uniform (bottom panel). Bulls-Eye plots of quantitative stress and resting MBF can be seen on the right.

FIGURE 4:

Stress and rest perfusion visual analysis in obstructive 3-vessel CAD. (a) Stress perfusion images demonstrate large subendocardial perfusion defects (arrows), which are not present at (b) rest. Invasive coronary angiography revealed (c) 60% stenosis of left anterior descending artery (LAD), 70% stenosis of the marginal branch (OM) from the left circumflex artery (LCx), and (d) 60% stenosis of the right coronary artery (RCA). Stenotic lesions are shown with labeled arrows. Late gadolinium enhancement was seen in the inferolateral basal and mid segments.

FIGURE 5:

Stress and rest MBF quantification in obstructive 3-vessel CAD. Stress perfusion maps (top panel) from the subject in Fig. 4 demonstrate perfusion defects in the anterior, septal, and inferolateral segments. Resting perfusion maps are relatively uniform (bottom panel). Bulls-Eye plots of quantitative stress and resting myocardial blood flow (MBF) can be seen on the right. Stress MBF is globally depressed.

Figure 6 shows a stepwise decrease in average and lowest overall stress MBF in normal healthy subjects and patients with no CAD, stenosis <50%, single-vessel CAD, 2-vessel CAD, or 3-vessel CAD. Normal healthy subjects and patients with nonobstructive disease had an average stress MBF of 2.74 ± 0.63 mL/g/min and lowest stress MBF of 1.92 ± 0.49 mL/g/min. Patients with obstructive CAD had significantly reduced average stress MBF (1.71 0.55 mL/g/min, P < 0.05) and lowest stress MBF (0.96 0.51 mL/g/min, P < 0.05) when compared to normal healthy subjects and patients with nonobstructive disease.

FIGURE 6:

Average and lowest stress MBF by subject group. Stepwise decrease in average (blue) and lowest (orange) overall stress myocardial blood flow (MBF) in normal healthy subjects and patients with no coronary artery disease (CAD), stenoses < 50%, 1-vessel (1 V) CAD, 2-vessel (2 V) CAD, or 3-vessel (3 V) CAD. A significant difference in average (*) and lowest (†) stress MBF was demonstrated between normal healthy subjects and patients with non-obstructive disease vs. patients with obstructive CAD (P < 0.001). Error bars represent standard deviation.

Diagnostic Performance

Table 2 summarizes the diagnostic performance of the visual and quantitative MRI perfusion methods by patient and vessel. It also includes the optimal cut-offs with ≥50% stenosis by QCA as the reference. Visual analysis demonstrated a point estimate sensitivity, specificity, and accuracy of 69% [95% CI, 51%–84%], 88% [95% CI, 73%–96%], 79% [95% CI, 68%–88%], respectively, on per-vessel basis. The PPV was 83% [95% CI, 64–94%] and the NPV was 76% [95% CI, 61–87%]. There was good inter-reader reliability with a kappa-statistic of 0.65. The average image quality was 4.09 ± 0.86 and the average motion correction score was 4.04 ± 1.04.

TABLE 2.

Diagnostic Performance of Spiral Whole-Heart Adenosine Stress Perfusion

	Cut-offa	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	Accuracy (%)
Visual
Patient	-	84 [60–97]	83 [36–100]	94 [71–100]	63 [24–91]	84 [64–95]
Vessel	-	69 [51–84]	88 [73–96]	83 [64–94]	76 [61–87]	79 [68–88]
Average MPR
Patient	2.38	74 [49–90]	67 [22–96]	88 [62–98]	44 [14–79]	72 [51–88]
Vessel	1.72	49 [31–66]	83 [67–93]	71 [49–87]	65 [50–78]	67 [55–77]
Average Stress MBF
Patient	2.10	84 [60–97]	83 [36–100]	94 [71–100]	63 [24–91]	84 [64–95]
Vessel	1.77	74 [57–88]	70 [53–83]	68 [51–83]	76 [59–88]	72 [60–82]
Lowest Stress MBF
Patient	1.18	79 [54–94]	100 [54–100]	100 [78–100]	60 [26–88]	84 [64–95]
Vessel	1.27	77 [60–90]	80 [64–91]	77 [60–90]	80 [64–91]	79 [68–87]

Diagnostic performance are presented as point estimate [95% confidence interval].

NPV = negative predictive value; MBF = myocardial blood flow; MPR = myocardial perfusion reserve; PPV = positive predictive value.

^aUnitless for MPR and ml/g/min for stress flow.

For the quantitative analysis, lowest stress MBF on a per-vessel basis had the highest point estimate for sensitivity (77% [95% CI, 60–90%]) and accuracy (79% [95% CI, 68–87%]). The specificity was 80% [95% CI, 64–91%]. Comparing per-vessel visual analysis to quantitative analysis by lowest stress MBF, no significant difference was seen for sensitivity (P = 0.61), specificity (P = 0.51), and accuracy (P = 1.00). The ROC curves on a per-vessel basis for the different quantitative perfusion parameters are displayed in Fig. 7. There was no significant difference in AUC (P = 0.06) between average MPR (0.68 [95% CI, 0.56–0.82]), average stress MBF (0.74 [95% CI, 0.63–0.86]), and lowest stress MBF (0.79 [95% CI, 0.69–0.90]). Using the same per-vessel cutoffs for the normal healthy subjects, there was a false positive rate of 11%, 15%, and 7% for the average MPR, average stress MBF, and lowest stress MBF, respectively. The Bland–Altman plots demonstrated good agreement as seen in the Supplementary Figs. S3 and S4. The intra-observer mean difference in MBF was −0.08 mL/g/min and the limits of agreement were −0.48 mL/g/min to 0.33 mL/g/min. The inter-observer mean difference in MBF was 0.03 mL/g/min and the limits of agreement were −0.35 mL/g/min to 0.40 mL/g/min.

FIGURE 7:

Per-vessel area under the receiver-operator characteristic curve. Quantitative perfusion diagnostic performance for myocardial blood flow (MBF) and myocardial perfusion reserve (MPR) based on per-vessel area under the receiver-operator characteristic (ROC) curve. Quantitative perfusion measurements include average stress MBF (blue), lowest stress MBF (red), and average MPR (green).

Using ≥70% stenosis by QCA as the reference, the per-vessel visual analysis showed similar point estimate for sensitivity (76% [95% CI, 53%–92%]), specificity (76% [95% CI, 62%–87%]), and accuracy (79% [95% CI, 68%–87%]). In addition, lowest stress MBF continued to have a high point estimate for sensitivity (86% [95% CI, 64%–97%]), specificity (70% [95% CI, 56%–82%]), and accuracy (75% [95% CI, 63%–84%]).

Ischemic Burden

Ischemic burden measurements in patients with obstructive CAD are shown in Fig. 8. Visual analysis demonstrated an average ischemic burden of 45% [95% CI, 23%–67%] in patients with single-vessel disease and 73% [95% CI, 49%–97%] in patients with multivessel disease. Similarly, quantitative analysis based on average stress MBF showed an average ischemic burden of 44% [95% CI, 15%–73%] in single-vessel disease and 70% [95% CI, 51%–89%] in patients with multivessel disease. Linear mixed model demonstrated a significant difference in ischemic burden between single and multivessel disease (26% [95% CI, 6%–48%], P < 0.05). There was no significant difference in ischemic burden between the visual and quantitative analysis (P = 0.85).

FIGURE 8:

Ischemic burden comparing visual vs. quantitative analysis and single vs. multivessel CAD. Visual and quantitative analysis demonstrated similar ischemic burden measurements and ability to differentiate between single (blue) and multivessel (orange) obstructive coronary artery disease (CAD). Error bars represent standard error.

Discussion

In our study, we clinically evaluated a whole-heart spiral pulse sequence for adenosine stress MRI. Using an accelerated spiral pulse sequence with motion compensated compressed sensing, we achieved high-quality first-pass perfusion images with whole-heart coverage (eight slices) and high in-plane resolution (2 mm) at heart rates up to 125 bpm. Whole-heart spiral perfusion imaging for detection of CAD demonstrated good image quality and accuracy for both visual and quantitative evaluation.

The main advantages of non-Cartesian sampling such as spiral trajectories include high signal-to-noise efficiency, robustness to motion, and isotropic spatial resolution. Furthermore, the reduced motion sensitivity decreases the susceptibility to dark rim artifacts¹⁰ and allows for visualization of subendocardial perfusion defects.²¹ The acquisition efficiency and benign aliasing artifacts also enable higher resolution images for a given acquisition time. With VDS trajectories, a larger extent of k-space with a higher spatial resolution can be sampled in a shorter readout duration.^11,21 As a result, VDS trajectories can achieve higher spatial resolution than a conventional spiral, with increased acquisition efficiency and reduced ringing artifacts.

Our visual and quantitative perfusion results using 2-D whole-heart coverage were comparable to previously published 3-D studies using 3.0 T. A multicenter trial by Manka et al²² prospectively enrolled 155 patients with suspected CAD and scheduled for coronary angiography with FFR. They used FFR < 0.8 and QCA ≥ 50% to define the presence of CAD. Visual per-vessel analysis of 450 territories based on QCA showed a sensitivity, specificity, and diagnostic accuracy of 61.5%, 91.5%, and 79.6%. With FFR, however, visual per-vessel analysis of 387 territories demonstrated a better diagnostic performance with a sensitivity, specificity, and diagnostic accuracy of 73.5%, 91.9%, and 87.1%, respectively. For fully quantitative analysis with 3-D MR perfusion imaging, Motwani et al²³ demonstrated a high per-territory sensitivity, specificity, and AUC of 94%, 95%, and 0.95 for stress perfusion quantification with significant obstruction defined by a QCA > 70%. Similarly in their study, MPR quantification resulted in a lower sensitivity, specificity, and AUC of 85%, 94%, and 0.93, respectively. A meta-analysis²⁴ of fully quantitative perfusion showed an overall per-vessel territory sensitivity, specificity, and AUC of 77%, 86%, and 0.88, respectively, which is more consistent with our results.

Some studies have attempted to compare the standard three-slice model to whole-heart coverage. Using 3-D perfusion MRI at 3.0 T, there was no significant difference in sensitivity, specificity, and diagnostic accuracy between whole-heart coverage and simulated three-slice models.²⁵ In single-photon emission computed tomography myocardial perfusion imaging, three-slice acquisition showed a tendency to miss severe apical defects and underestimate the extent of perfusion defects compared to whole-heart coverage.²⁶ However, there was no difference in diagnostic accuracy.²⁶ In a study by Hamada et al,²⁷ whole-heart 3-D perfusion MRI demonstrated high diagnostic accuracy for significant CAD irrespective of gender. Notably, they showed a high sensitivity of 89% and NPV of 90% for female populations, which are often considered to be an undertreated demographic.²⁷ The disadvantages of current 3-D MR perfusion techniques include the relatively poor in-plane spatial resolution (2.3 × 2.3 mm²) and the long temporal footprint.²⁸ Both 3-D stack of spirals and stack of stars techniques with shorter temporal footprints and higher resolution have been developed but have not been validated clinically.^29,30 There have been no studies directly comparing the diagnostic accuracy of 2-D whole-heart coverage vs. standard three-slice MR perfusion techniques.

Although the benefits of whole-heart coverage remain to be completely realized, the comparable diagnostic performance to standard three-slice perfusion imaging has important implications. Whole-heart coverage would allow simultaneous detection of flow-limiting CAD as well as estimation of prognostically significant myocardial ischemic burden.³¹ A rapid multislice approach eliminates the planning needed for slice selection and reduces the temporal footprint of the images, thus decreasing cardiac motion-induced artifacts. It is possible that in the near future, fully quantitative MRI with whole-heart coverage could serve as a high spatial resolution alternative to PET while avoiding the need for harmful radiation and expensive radiotracers. Ideally, whole-heart coverage will enable cardiologists to identify small areas of ischemia that are contributing to clinically significant symptoms that could otherwise be missed with standard three-slice models. Slice planning is also simplified as the whole heart is covered in all cases. Furthermore, whole-heart coverage may redefine the way we approach perfusion MRI. The current AHA 17-segment model³² has been the standard method for segmenting and reporting MRI results and heavily influences our approach to perfusion analysis. However, variations in coronary anatomy, physiologic differences in rest MBF, and subclinical coronary disease has limited the usefulness of global and segmental MPR.³³ In our analysis, we attempted to identify perfusion defects based on the lowest average stress MBF between two contiguous segments. Using the lowest stress MBF has been reported in the past.³⁴ With whole-heart coverage, however, identifying the region of lowest stress MBF and correlating it with known coronary stenoses should potentially provide a more sensitive and specific method for diagnosing CAD. Combining whole-heart perfusion imaging with magnetic resonance coronary angiography or computed tomography angiography could be a next step for noninvasively evaluating functional and anatomical coronary disease, particularly in light of the results from the MR-INFORM⁵ and ISCHEMIA trials.³⁵

Fully quantitative perfusion analysis could potentially improve visual analysis by detecting subtle perfusion defects and identifying artifacts that contribute to misdiagnoses.¹⁹ Furthermore, quantitative analysis of ischemic burden can confer incremental prognostic value to visual assessment.³⁶ A previous study has shown that only quantitative analysis can accurately identify and differentiate the extent of ischemia.³⁷ However, in our study, we showed that both quantitative and visual analysis methods were able to differentiate between single and multivessel disease based on ischemic burden. This may be due to the higher in-plane spatial resolution and the whole-heart coverage used in our study. We also saw no significant differences in diagnostic performance and ischemic burden measurements between visual and quantitative analysis. There are a number of drawbacks of fully quantitative analysis including the need for manual processing, such as segmentation of the myocardium or selection of the contrast arrival and peak time for the time-signal intensity curves, which can be an operational bottleneck and introduce interobserver variability. Nevertheless, automated analysis techniques offer a solution to alleviate clinical workflow and remove potential human error. One fully automated inline quantification pathway, benchmarked against comprehensive invasive coronary physiology, was able to identify regional obstructive epicardial CAD (FFR ≤ 0.80) as well as distinguish multivessel CAD from coronary microvascular dysfunction (index of microcirculatory resistance ≥ 25).³⁸

Limitations

The sample size in this study was relatively small and there was a high prevalence of obstructive CAD: 76%, compared with 57% in Motwani et al.²³ As a result, we reported per-vessel diagnostic performance and defined positive results based on adenosine-induced perfusion abnormalities that exceeded beyond areas of LGE. In addition, we did not measure transmural perfusion gradients, a technique for which high in-plane resolution acquisitions are ideally suited for. Another limitation was the use of QCA rather than FFR, which is a more comparable reference standard for functionally significant coronary disease. As coronary angiography was performed for routine care, FFR was performed at the discretion of the operator. We also measured ischemic burden but did not specifically investigate the prognostic value given the small sample size and lack of long-term follow-up. Furthermore, spiral pulse sequences possess some drawbacks such as greater sensitivity to gradient hardware fidelity and eddy current effects. The effects of gradient nonlinearity and eddy currents on the spiral trajectory were corrected prior to image reconstruction.³⁹ Spiral sequences are also sensitive to off-resonance effects, which can result in image blurring and signal drop-out. In this study, we utilized spirals with relatively short readout-durations specifically to minimize the effects of off-resonance. Finally, our postprocessing was not fully automated and therefore subject to human error. This method, however, only required the imager to assess and confirm the quality of the fitting on a segmental basis. In future studies, we could implement automated pipelines for quantitative spiral perfusion imaging.⁴⁰

Conclusion

Our initial clinical findings using quantitative whole-heart spiral pulse sequences for adenosine stress MRI demonstrated a promising alternative for full myocardial coverage with high spatial resolution and minimal dark rim artifacts. We demonstrated high diagnostic performance which was comparable to other 3-D and 2-D studies for both visual and quantitative analysis. There were similar ischemic burden measurements between visual and quantitative analysis using this technique, which may be due to the higher spatial resolution and ventricular coverage in comparison with prior techniques.

Supplementary Material

Additional supporting information may be found in the online version of this article