Forty-seven patients with suspected CAD but without apparent coronary stenosis on CCTA were recruited for NCMRA examination at Fuyang People’s Hospital from September 2021 to March 2022. The exclusion criteria were severe arrhythmia, stenosis or occlusion, and MRI-related contraindications, including pacemaker implantation and claustrophobia. Our local ethics institutional review board approved this prospective study, and all participants provided informed consent. The flowchart of the study participants is shown in Fig. 1.

Flowchart of the study participants. CAD, coronary artery disease; CCTA, coronary CT angiography; CMRA, coronary MR angiography

The CCTA imaging was performed via prospective electrocardiographic (ECG) gating using 256-slice multi-detector CT (Brilliance iCT; Philips Healthcare, OH, USA). The CT data acquisition was initiated under the full inspiration of 6 s after a predetermined signal attenuation threshold of 180 Hounsfield units was attained. About 60–70 mL of contrast media (Iodixanol 320; Jiangsu Hengrui Pharmaceuticals, Lianyungang, Jiangsu, China), followed by 20 mL of saline, was intravenously injected into the antecubital vein at a flow rate of 5–6 mL/s. No beta-blockers, nitroglycerin, or contrast agents were used. The imaging parameters were as follows: tube potential, 120 kVp; effective tube current–time product, 50 mA; detector configuration, 32 × 0.625 mm2; rotation time, 330 ms; field of view, 250 mm; reconstructed slice thickness, 2.5 mm; and an increment of 2.5 mm.

All MRI examinations were performed using a clinical 3 T MR scanner (Ingenia CX; Philips Healthcare, Amsterdam, the Netherlands) with a 32-channel body phased-array surface coil. No beta-blockers, nitroglycerin, and contrast agents were used. After a free-breathing, four-chamber cine imaging was acquired, the optimal data acquisition window was determined by the minimal motion phase of the right coronary ostium. A 3D whole-heart turbo-field gradient-echo sequence was used with ECG-gating and diaphragm navigator-gating for NCMRA data acquisition.

All patients underwent four different scanning sequences in a random order: the conventional SENSE sequence (denoted as SENSE), and the other three sequences using CS technology with acceleration factors 4, 5, and 6 (indicated as CS4, CS5, and CS6). The detailed imaging parameters considered in this study are listed in Table 1. The SENSE sequence was considered the control sequence, and CCTA images were taken as the gold standard.

All original-source CT and MR images were transferred to the IntelliSpace Portal, Version 7.0 (Philips Healthcare) for curved planar reconstruction (CPR) and vessel length measurement (Fig. 2). The visible vessel lengths measured using NCMRA sequences were compared with those measured using CCTA. Two radiologists with 8 years (Y.Z.) and 3 years (X.H.) of experience in CMRA performed CPR blindly on each set of CCTA and NCMRA images. Then, the visualized vessel lengths of the right coronary artery (RCA), left anterior descending artery (LAD; including the left main artery), and left circumflex artery (LCX) were obtained from the CPR images semiautomatically in the workstation.

Sample coronary artery CPR and visible length measurement using IntelliSpace Portal workstation. CPR, curved planar reconstruction

The intra- and inter-observer reliabilities of visible lengths were assessed in 50 coronary arteries randomly selected using Bland–Altman plots. The intra-observer reliability was derived from repeated measurement by one radiologist (Y.Z.) after at least 1 week of blinding to the previous results. The inter-observer reliability was independently assessed by both radiologists (Y.Z. and X.H.).

According to the 15-segment American Heart Association classification [28], we evaluated three segments (proximal, middle, and distal) of RCA, left main artery (LM), two segments (proximal and middle) of LAD, and two segments (proximal and distal) of LCX (Fig. 3).

Eight coronary artery segments evaluated in this study

Two other experienced radiologists (B.P. and Y.Y.) with more than 5 years of experience in CMRA, who were blinded to the CCTA results, determined the subjective image scores and CR independently.

A four-point subjective score was used to assess the quality of the NCMRA image: 4, excellent (vessels that were well depicted with sharply defined borders); 3, good (vessels that were adequately visualized with only mildly blurred boundaries); 2, fair (coronary vessels that were visible, but with low confidence in the diagnosis due to moderately blurred borders); and 1, poor (coronary vessels that were barely seen or obscured by noise) [29] (Fig. 4).

Four-point subjective score for the qualitative image analysis

It would be inaccurate to use the classic measurement approaches for calculating the SNR and CNR because the iterative reconstruction of CS could result in an artificial reduction in noise in MR images [30]. Instead, the CRs among the eight coronary artery segments and myocardium were calculated as \(} = \frac}}} - \mu_}}} } \right|}}}}}^ + \sigma_}}}^ } }}\), where μ is the mean signal intensity of the corresponding tissue and σ is the variance of the related tissue [31]. The regions of interest (ROIs) were placed on the same slices of the four NCMRA source images for measuring the signal intensity of the arteries for each segment, while ROIs of the myocardium were drawn on the left ventricular septum.

The intra- and inter-observer reliabilities of the image scores and CR were assessed in 50 segments randomly selected using the Kappa test and Bland–Altman plots, respectively. The intra-observer reliability was derived from repeated measurement by one radiologist (Y.Z.) after at least 1 week of blinding to the previous results. Two radiologists independently assessed the inter-observer reliability (B.P. and Y.Y.).

The data from participants with successful NCMRA and CCTA were statistically analyzed using SPSS Statistics 26.0 (SPSS, Inc., IL, USA). The Shapiro–Wilk test was used to assess the normal distribution of the continuous data. The quantitative variables were reported as mean ± standard deviation if normally distributed and as a median or interquartile range in case of non-normal distribution, and categorical variables were expressed as numbers (percentage).

The differences between the scan times of NCMRA sequences were assessed using the Friedman nonparametric statistical test. The linear regression analyses with the coefficient of determination (R2) were performed to determine the correlation between the visible vessel lengths of four NCMRA and CCTA sequences. A further agreement was tested via the Bland–Altman analysis (mean difference and upper and lower limits of agreement). The Friedman tests were used to compare image quality scores and CR among these sequences. Multiple comparisons were performed between three CS factor sequences and the SENSE sequences using Wilcoxon signed-rank tests. A p value < 0.05 indicated a statistically significant difference.