Study Population
Anterior AMI patients were scanned prospectively with 4D flow MRI as an adjunct to their clinical CMR exams performed in 2016-2017. These patients were compared against historical controls from a previously reported, 2010-2011 study [11]. This study was approved by the Institutional Review Board and was compliant with the Health Insurance Portability and Accountability Act. Written informed consent was obtained from all subjects. Inclusion criteria for anterior AMI patients were hospitalization and revascularization for AMI with the left anterior descending artery or left main coronary artery identified as the culprit vessel by coronary angiography, and a clinically ordered CMR exam. Exclusion criteria were contraindications to MRI or gadolinium-based contrast agents. Twelve patients with anterior AMI were recruited. Exclusion criteria for control subjects were standard contraindications to MRI and to gadolinium-based contrast agents, high cardiovascular risk factors (body mass index > 30, history of smoking, diabetes, or hypertension), and drugs affecting cardiovascular function. Data from 19 control subjects were included in the analysis, for a total of 31 scans analyzed.
MRI Acquisition
CMR examinations in controls were performed in a 3.0T scanner (MR750, GE Healthcare, Waukesha, WI). CMR examinations in AMI subjects were acquired on 1.5T (MR450w or HDxt, GE Healthcare, Waukesha, WI; N=9) and 3.0T (MR750 or MR750w, GE Healthcare, Waukesha, WI; N=3) scanners. The choice of field strength was based on the clinical availability of the scanners. The CMR protocol included a short-axis bSSFP cine acquisition to segment the LV cavity & compute LV strain, 4D flow imaging for velocity mapping, and short-axis late gadolinium enhancement (LGE; only in AMI patents) imaging to measure infarction size. 4D flow data was acquired with PC VIPR, a three-dimensional radially-undersampled, three-directionally velocity encoded technique [12, 13], with a scan duration of 9-14 minutes including respiratory gating efficiency. Each set of 4D flow data was reconstructed in two ways: a high-resolution image set was reconstructed using a gridding technique and used to measure through-plane flow and intraventricular KE, and a low-resolution, low-noise image set was reconstructed using compressed sensing with a spatial-wavelet-transform L1-norm penalty (λ=0.01) in order to separate LV flow into different compartments by tracking flow pathlines. This separate reconstruction was used for pathline tracking because this method is sensitive to noise due to compounding errors in pathline integration. Intravenous contrast was administered to all subjects prior to 4D flow imaging. AMI patients received 0.15 mmol/kg of gadobenate dimeglumine (Multihance; Bracco, Milan, IT) and controls received 0.03 mmol/kg of gadofosveset trisodium (Ablavar; Lantheus, Billerica, MA, USA). Table 1 shows the MRI acquisition parameters.
Image Analysis
The LV cavity was segmented at each time frame on short-axis bSSFP images (including outflow tract, excluding papillary muscles) using the software Segment (Medviso, http://segment.heiberg.se; v2.0 R5399) [14]. LV end diastolic volume (EDV), end systolic volume (ESV), stroke volume (SV), and ejection fraction, were calculated from the endocardial borders. Global longitudinal strain (GLS) was computed from long-axis bSSFP images, and regional radial and circumferential strain were computed from short-axis bSSFP images using one slice each in the base, mid-ventricle, and apex in Segment using feature tracking [15].
4D flow data background phase errors were corrected by fitting a 3rd order polynomial to static tissue phase. Subsequently, 4D flow data was registered to the short-axis bSSFP dataset using the Advanced Normalization Toolkit [16, 17]. Proper registration was confirmed by overlaying the images in ITK-SNAP [18]. Each slice in the LV segmentation was assigned to one of three equal-length segments divided along the LV long axis: base, mid-ventricle, and apex (Figure 1). Through-plane flow was computed for each slice and time point by multiplying the average through-plane velocity component in the LV with the cross-sectional area of the LV. Through-plane flow in each LV region was computed as the average through-plane flow for each slice in that region. Peak systolic and diastolic flows were defined for each region as maximum positive and negative flows, with the positive through-plane direction running from the apex to the base.
Average KE was computed by summing the KE contributions for all voxels in the LV and averaging over time: (see Equation 1 in the Supplementary Files)
Where Nt is the number of cardiac time frames, ρblood is the density of blood (1.06 g/cm3), Vvox is the voxel volume, and vvox is the velocity magnitude. KEavg was then indexed to EDV (KEiEDV) as in Garg et al [10]. KE was not computed on a regional scale because the squared velocity term results in a measurement dominated by high-velocity voxels, which is therefore highly sensitive to noise in slow-flow regions (such as the LV apex) with a full-ventricle high velocity encoding imaging approach.
The distribution of different LV flow components was determined in all subjects using the method of Eriksson et al [19]. Blood pathlines were emitted from the LV blood volume and traced forwards and backwards in time from end diastole until end systole, thus including the entire cardiac cycle. Pathlines were computed by integrating the velocity field using a 4th order Range-Kutta numerical integration through time. Pathline location was used to separate the pathlines into four different components of flow: Direct Flow (blood that enters the LV during diastole and leaves the LV during systole in the analyzed heartbeat), Retained Inflow (blood that enters the LV during diastole but does not leave during systole in the analyzed heartbeat), Delayed Ejection Flow (blood that starts and resides inside the LV during diastole and leaves during systole), and Residual Volume (blood that resides within the LV for at least two cardiac cycles). Pathlines passing through the ventricle wall (either entering or leaving the LV through the mid-ventricle or apical regions) were excluded from analysis. The fraction of EDV containing pathlines from each compartment was computed for all subjects.
All flow computations were performed in Matlab (R2018a, The Mathworks Inc., Natick, Massachusetts, USA). All bSSFP and 4D flow images were analyzed by PAC (3 years of CMR analysis experience). Kim’s method [20] was used on LGE images to compute infarct size as follows: each segment in the 17-segment AHA myocardial model was scored for infarction transmurality using a 5-point scale (0=no infarction, 1=0%-25%, 2=25%-50%, 3=50%-75%, 4=>75% transmurality), the scores were averaged, and the result was divided by four. The images were scored by consensus of two radiologists with expertise in cardiothoracic imaging (reader 1, 17 years of experience; reader 2, 5 years of experience). Disagreements were handled by consulting a third radiologist. The LGE readers were blinded to the results of the flow analysis and vice-versa.
Statistical Analysis
Demographic and traditional CMR measures (LV function and strain) are presented as mean ± standard deviation and were compared between AMI patients and controls using independent sample t-tests. The proportion of male subjects was compared between groups using a chi-squared test. The 11 LV flow parameters (peak systolic and diastolic flow in each region, global KEiEDV, and the percentage of LV flow in each of the 4 compartments) are not assumed to be normally distributed and are presented as median ± inter-quartile range (IQR). Multivariable linear regression was used to test if there were differences in the 11 LV flow parameters between anterior AMI patients and controls, while adjusting for age, sex and heart rate differences between the groups as covariates. For each LV flow parameter, a multivariable linear regression model was fitted with LV flow parameter as outcome, group of subjects (AMI patient or control) as predictor variable, and age, sex, and heart rate as covariates in the model. Model diagnostics were performed and no serious violations of the assumptions of the linear models were found. Regional LV through-plane flow parameters were correlated with heart rate and traditional CMR measures (infarct size, SV, CO, EF, EDV, GLS, and regional radial and circumferential strain) using Spearman’s rank correlation test. Regional through-plane flow and regional strain were compared on a region-by-region basis (i.e. flow in the LV base was compared with strain in the LV base). Multivariable regression modeling was performed in R version 3.5.2. All other analysis was performed in MATLAB. A significance level of 0.05 was used for all tests.