In this prospective, single center trial of 50 consecutive patients with suspected CAD, we assessed for the first time, the diagnostic ability of our recently proposed high-resolution, non-contrast coronary CMRA framework [
18] in excluding significant CAD compared against the non-invasive clinical anatomical reference standard of coronary CTA. We applied an intention-to-read analysis approach by including all diagnostic coronary segments regardless of image quality.
We achieved a predictable and clinically feasible average acquisition time of approximately 10 mins, which is comparable to previously published clinical studies despite a significantly higher spatial-resolution (0.9 mm3). Whilst including all coronary segments for analysis, only 24/450 (5%) of all coronary CMRA segments were non-diagnostic, with an overall median image quality score of 4.0 (IQR 3.0–4.0), representing “good to excellent” on the image quality scale.
The sensitivity of our coronary CMRA framework was comparable to previous coronary CMRA studies which had a range of 71–94% per patient, 65–93% per vessel and 46–92% per segment respectively [
10,
11,
14‐
16,
22,
23]. Furthermore, our coronary CMRA approach demonstrated a high specificity and NPV per patient (74% and 100%), per vessel (88% and 97%) and per segment (95% and 99%). In addition, we demonstrated a very high specificity and NPV for excluding LM (98% and 100%), proximal segment (95% and 98%), middle segment (92% and 100%) and distal segment stenoses (97% and 98%), indicating the clinical potential of our proposed coronary CMRA framework in patients with low-intermediate risk of CAD.
To our knowledge, this is the first clinical study to assess the diagnostic performance of a 3D contrast-free coronary CMRA approach using similar patient preparation as coronary CTA that enables a predictable scan time of approximately 10 min for 0.9 mm
3 spatial-resolution. This was achieved by employing a robust motion corrected free-breathing acquisition with 100% respiratory scan efficiency, using image based navigation for 2D translational motion estimation and respiratory data binning combined with 3D non-rigid motion compensated reconstruction based on a 3-4 fold undersampled Cartesian acquisition and a patched based low rank reconstruction. Yang et al [
22] and more recently Sun et al. [
23] utilized a conventional diaphragmatic respiratory navigator in conjunction with a slow infusion of gadobenate dimeglumine to obtain 3D whole-heart coronary CMRA images. Sacrificing spatial-resolution (1.3 mm
3 and 1.3 × 1.3 × 1.8 mm respectively), they achieved an average acquisition time of 9.0-9.5 min (acquisition window of 70–135). However, average respiratory scan efficiency was 35–36%, 9–10% of patients were unable to complete their scan to due to unpredictable scan times and 12–14% of all segments were excluded prior to analysis. Similarly, Kato et al. [
11] combined the same protocol in a non-contrast study (spatial-resolution 1.3 × 1.3 × 1.7 mm), achieving an average acquisition time of 9.5 min (acquisition window of 131 ± 40 ms). However, the respiratory scan efficiency was 37%, 8% of patients were unable to complete the scan, due to unpredictable scan times and 10% of all segments were excluded prior to analysis. To overcome the unpredictability of the conventional diaphragmatic navigator, Piccini et al. [
15] acquired self-navigated 3D whole-heart coronary CMRA with 100% respiratory scan efficiency at 1.15 mm
3 spatial-resolution in a predictable acquisition time of 7.4 min ± 1.9 min. However, this was a contrast-enhanced study and 8% of proximal segments, 16% of middle segments and 44% of distal segments could not be visualized, severely hampering the diagnostic abilities of this framework. He et al. [
14] improved on this by using higher field strength (3T) and spatial-resolution (1 mm
3), achieving a predictable acquisition time of 7.8 ± 0.8 min. However, this was a contrast-enhanced study and they excluded 12% of all segments from analysis due to limited diagnostic quality. Another disadvantage of self-navigation is the difficulty in separating moving (e.g. heart) from static (e.g. chest wall) tissue, which may reduce the motion correction performance [
19]. iNAVs overcome some of the limitations of respiratory self-navigation, e.g. separating moving tissue from static tissue during motion estimation to enable beat-to-beat translational motion correction with or without respiratory gating. We have previously investigated the clinical robustness of the iNAV based coronary CMRA for assessing CAD [
16]. However, this framework utilized contrast-enhancement and fixed 50% respiratory gating. The spatial-resolution was 1.3 mm
3 and the average acquisition time was 7.3 ± 0.5 min (acquisition window of 118 ± 38 ms). However, 9% of all coronary segments were excluded from the analysis due to low spatial-resolution. In contrast, 100% of patients in the present study completed image acquisition and no coronary segments were excluded prior to analysis, despite the significantly higher spatial-resolution. Unique to this framework is the combination of the 3D non-rigid motion correction framework with the golden-step variable density spiral-like Cartesian trajectory which enables further improvement in motion correction and image quality at 100% scan efficiency, particularly at high spatial-resolution. Furthermore, the 3D patch-based reconstruction technique enables sub-1mm spatial-resolution which was not clinically feasible previously.
Study limitations
We observed a significantly lower average HR in patients who underwent a coronary CTA compared to the same patients who subsequently underwent a coronary CMRA scan. Furthermore, only 4% of patients on the day of their coronary CTA scan had a HR of ≥ 70 beats/mins compared with 24% of patients on the day of their coronary CMRA scan (p = 0.008). We also observed a significantly higher coronary CMRA image quality score for the overall 3D whole-heart dataset and RCA in patients with a HR < 70 beats/mins compared with patients with a HR of ≥ 70 beats/mins, although the numbers are limited. This discrepancy could in part be explained by the fact that patients who attended their clinical coronary CTA scan were pre-conditioned with three days of oral metoprolol leading up to the date of their scan. A low HR directly correlates with a longer diastolic resting period of the heart and lower likelihood of cardiac motion artefacts during the coronary CMRA acquisition [
26,
27]. Furthermore, metoprolol suppresses both arrhythmia and ectopic beats [
28], which is crucial for free-breathing 3D whole-heart coronary CMRA acquisition whereby data is acquired over several consecutive heart beats. Additionally, beta blockers have been shown to significantly reduce HR variability in patients undergoing coronary CTA, with significant improvements in image quality and reduction of artefacts [
29].
This study compared the diagnostic performance of coronary CMRA against coronary CTA, without comparison against invasive X-ray coronary angiography. Whilst coronary CTA has an excellent NPV for excluding significant CAD, its specificity is still limited as described in the studies by Liu et al. [
24] and Hamdan et al. [
25]. Furthermore, in our center, patients with very high coronary artery calcium scores (as assessed by the clinical team on the day of the scan) were referred directly for an invasive X-ray coronary angiogram, without performing a contrast enhanced coronary CTA, in order to avoid unnecessary exposure of patients to ionising radiation and iodine contrast agents. Furthermore, the PPV of this coronary CMRA framework was comparatively low compared with previously published studies. This could potentially be explained by the relatively small number of patients in this study, compounded by the low prevalence of severe stenosis in this patient cohort, possible overcautious analysis as well as the absence of invasive X-ray coronary angiography as the arbiter. Therefore, larger multi-center studies should assess the performance of this novel coronary CMRA framework against both coronary CTA and invasive X-ray coronary angiography.
This coronary CMRA framework is currently not applicable to patients who met our exclusion criteria. Future studies should aim to include as many of these patients as possible in order to capture the full spectrum of patients with suspected CAD, e.g. through arrhythmia rejection reconstruction and systolic imaging in atrial fibrillation.