Study population
From April 2018 to June 2020, 30 healthy controls and 33 patients with LV diastolic dysfunction confirmed by echocardiography were retrospectively enrolled[16]. The inclusion criteria for the healthy controls were as follows: normal blood pressure (systolic blood pressure < 140mm Hg, diastolic blood pressure < 90mm Hg); regular sinus rhythm; normal echocardiography findings; and no history of cardiovascular disease. The 33 patients were composed of ischemic cardiomyopathy (n=3), dilated cardiomyopathy (n=15), hypertrophic cardiomyopathy (n=5), ventricular arrhythmia (n=4), myocarditis (n=4), cardiac amyloidosis (n=1), and constrictive pericarditis (n=1). The exclusion criteria were common contraindications to CMR and poor image quality. All participants underwent a CMR study which included conventional segmented cine (the gold standard) and CS cine sequences. All subjects did not take extra medicine for the CMR examination. The protocol and the use of data were approved by the local ethics committee. Written informed consent for CMR examination was obtained by all the study patients.
Image acquisition
All CMR scans were performed on a 3T system (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany). Standard long-axis views (2-, 3- and 4-chamber) and a short-axis stack of cine images were acquired using retrospective electrocardiogram (ECG)-gated balanced steady-state free precession conventional cine imaging and an adaptive prospective ECG-triggered CS sequence, both during inspiratory breath-hold. The total scan time for each cine CMR was obtained. CS sequence used incoherent undersampling on the Cartesian k-space by implementing a pseudorandom variable-density readout pattern. Image reconstruction was performed with a nonlinear iterative reconstruction with k-t regularization [17]. Spatial resolution (1.8*1.6 mm) and slice orientations were the same for the two cine imaging methods. Detailed imaging parameters are shown in Table 1. The “real” number of cardiac phases depends on the patient’s HR during CS cine. To facilitate post-processing of the data, all cine images were set to 25 calculated cardiac phases calculated retrospectively using the “real” obtained cardiac phases.
Image analysis
All image quality assessments and LA parameters were analyzed independently by two radiologists (2 years and 5 years of CMR experience). When it was difficult to reach consistent image quality, it was evaluated by a third radiologists (10 years of CMR experience). LA volume, ejection fraction (EF), and strain were quantified by using the commercial software package (CVI 4.2 v. 5.0, Circle Cardiovascular Imaging, Calgary, Canada) [12, 18]. Pulmonary veins and LA appendages were excluded during the quantification of LA parameters.
The image quality of both cine imaging methods was evaluated according to the following image features: the clarity of the endocardial and epicardial border, the contrast of the myocardium and blood pool, the visualization of papillary muscle and valves, and the degree of artifacts [19]. The qualitative analysis of image quality used a 4-point scale: 1 = poor quality (moderate artifacts with unclear image features), 2 = fair quality (mild artifacts with slight blurring of endocardial borders), 3 = good quality (minimal artifacts but not affecting the views of the remaining features), 4 = excellent quality (clear endocardial and epicardial borders, clear contrast of the myocardium and blood pool, clear papillary muscles and valves, and no artifacts). Images with a score ≥ 2 were considered acceptable for further measurements of LA parameters.
The LA endocardial border was manually drawn in 2- and 4-chamber views at three phases of the cardiac cycle: LA minimum volume (LAVmin) at the first phase after mitral valve (MV) closing during LV systole, LA maximum volume (LAVmax) at the last phase before MV opening during LV early diastole, and LA pre-contraction volume (LAVpre) at the last phase before the second diastolic opening of the MV during LV late diastole. For the biplane area-length method, LA volumes were calculated using the following equation: LA volume (LAV) = (0.85 * Area2ch * Area4ch) / Lmin [20], and indexed to body surface area (BSA).
Lmin is equivalent to the shorter long-axis length of the LA either in 2-chamber or 4-chamber view [20]. LA EFs included LA total EF (EFtotal), passive ejection fraction (EFpassive), and active ejection fraction (EFbooster), and these were calculated using the following equations:
EFtotal = (LAVmax - LAVmin) / LAVmax * 100%
EFpassive = (LAVmax - LAVpre) / LAVmax * 100%
EFbooster = (LAVpre - LAVmin) / LAVpre * 100%
LA endocardial and epicardial borders were manually traced in 2- and 4-chamber views at LV end-systole, automatically propagating the borders over the whole cardiac cycle. Automatic tracking was applied to ensure the accuracy of propagation and manually corrected if necessary. LA global radial and longitudinal strain curves were automatically generated by the software (Supplementary Figure). Global radial strain and global longitudinal strain contained total strain (Ers, Els, respectively), passive strain (Ere, Ele, respectively), and active strain (Era, Ela, respectively), respectively, which correspond to the LA reservoir function, conduit function, and booster function.
Statistical analysis
Continuous data were expressed as mean ± standard deviation (SD) or median and interquartile range (IQR). The Wilcoxon matched-pairs signed-rank test was used to compare image quality and LA parameters between segmented and CS cines. The correlation of all LA parameters between the two cine images was evaluated using intraclass coefficient (ICC) analysis with two-way random effects, and corresponding 95% confident interval (CI). The Bland-Altman analysis was chosen to evaluate the agreement of LA EFpassive, Ere, and Ele between both cine methods. The Mann-Whitney U test was applied to compare LA parameter differences between the two methods in healthy controls and patients with LV diastolic dysfunction. The diagnostic performance of EFpassive, Ere, and Ele in terms of sensitivity, specificity, negative predictive values, and positive predictive values (NPV and PPV, respectively), and accuracy to distinguish between patients and healthy controls were assessed using receiver-operating-characteristic (ROC) curves, and cut-off values were developed. The diagnostic accuracies of LA EFpassive, Ere, and Ele between the segmented and CS cine images were compared with a Chi-squared test. Inter- (two readers) and intra-observer (two readings) reproducibility were also assessed using ICC, corresponding 95% CI and the coefficient of variation (CoV). ICCs greater than 0.90, 0.75, and 0.50 were considered excellent, good, and moderate, respectively [21]. The CoV was calculated as the SD of the differences between repeated measurements normalized to the mean. A value of p < 0.05 was regarded as statistically significant. Statistical analysis was performed using SPSS (IBM SPSS Statistics 23 for Windows).