Exercise-induced irregular right heart flow dynamics in adolescents and young adults born preterm

This study was approved by the local Institutional Review Board and is compliant with the Health Insurance Portability and Accountability Act. Eleven preterm young adults (27 ± 1 years; 5 male,) and 17 preterm adolescents (13 ± 1 years; 6 male) were recruited. These subjects were drawn from the Newborn Lung Project (NLP) cohort [18] at the University of Wisconsin-Madison. The NLP cohort included individuals born prior to 32 weeks of gestation at a birthweight below 1500 g. Invitations to participate in this study were emailed to cohort members or their guardians with up-to-date contact information and residence in Wisconsin. Approximately 10% of those contacted responded and participated in phone interviews which provided further details on the study and determined if the individual met any specific exclusion criteria. Exclusion criteria included any history of cardiovascular or cardiopulmonary disease, exclusive of neonatal comorbidities such as bronchopulmonary dysplasia or resolved pulmonary hypertension, any physical limitations inhibiting exercise capabilities, and any other significant health problems that required daily medication use. Qualified individuals were given the chance to accept or decline participation in this study at this point. Eleven young adults (26 ± 1 years; 8 male) and eighteen adolescents (13 ± 1 years; 8 male) born at term were also recruited to serve as age-matched healthy controls. All subjects were screened with a Global Physical Activity Questionnaire (GPAQ) [19] or a Physical Activity Questionnaire for older Children (PAQ-C) [20] to ensure there were similar levels of physical activity between the preterm groups and the age-matched control groups.

All imaging was performed on a clinical 3T CMR scanner (Discovery 750, General Electric Healthcare, Waukesha, Wisconsin, USA) with an 8-channel cardiac coil between March 2016 and November 2017. 4D flow acquisitions were performed at rest and during exercise with a radially-undersampled, 5-point encoded PC VIPR [21, 22] sequence with the following parameters: TR/TE = 6.2/2.0 ms, flip angle = 10°, velocity encoding (venc) = 200 cm/s, field of view = 32 × 32 × 32 cm, acquired spatial resolution = 1.25 mm isotropic, scan duration = 9.25 min. The imaging volume covered the entire upper torso, including the heart and great vessels. Acquisitions were free-breathing with retrospective electrocardiogram (ECG) gating and respiratory gating from an abdominal belt. Average heart rate during imaging was calculated from the ECG gating files recorded on the scanner.

As a result of increased motion during exercise and reduced reliability of ECG leads with perspiration, there was an increase in missed ECG triggers for many subjects during exercise. Unreliable gating profiles resulted in inaccurate retrospective temporal binning of projections, leading to reconstructed flow waveforms with erroneous increases in diastolic volumetric blood flow rates. To counter this, a novel ECG gating correction algorithm was applied to compromised gating files [23]. Correction was performed if more than 5% of the acquired projections were acquired at a cardiac time longer than the median RR interval length (i.e., outside of the expected cardiac cycle length). The details of this algorithm are presented in an additional file (see Additional file 1).

Images were reconstructed offline and only used data acquired during expiration identified through adaptive thresholding of the respiratory waveform at 50% efficiency [24]. The number of reconstructed cardiac phases was kept to fifteen to compensate for minor signal-to-noise (SNR) losses expected with exercise [15] with an increased number of projections per phase.

The exercise power corresponding to each subject’s VO_2max (maximal oxygen uptake) was established on an upright bicycle ergometer in an exercise laboratory as previously described [16]. For exercise imaging, each subject exercised in a supine position in the scanner bore with a commercial CMR-compatible recumbent stepper (Cardio Stepper Module, Ergospect GmbH, Innsbruck, Austria). An example of the setup is shown in Fig. 1. Subjects were verbally coached to step at a cadence of 60 steps/min at 70% of the power corresponding to their VO_2max for the duration of imaging. Exercise imaging began after three minutes of stepping to allow subjects to achieve a steady-state heart rate. The stepper automatically adjusted the stepping resistance in response to changes in stepping cadence to maintain the target power setting.

Background phase correction for the velocity maps was performed on all datasets by fitting a 3^rd order polynomial to phase in stationary tissue, defined by semi-automatically thresholding time-averaged magnitude and velocity data [25]. To assess right heart flow, time-averaged complex difference 4D flow datasets were processed using MIMICS (version 17.0, Materialize, Leuven, Belgium) to segment the heart and great vessels. Two-dimensional measurement planes were placed orthogonal to the main pulmonary artery in Ensight (Version 10.0, Ansys, Canonsburg, Pennsylvania, USA). A custom MATLAB (The Mathworks Inc., Natick, Massachusetts, USA) tool [26] was used to identify the vessel walls and integrate velocity across the vessel lumen to quantify blood flow over the cardiac cycle. Stroke volume was calculated by integrating flow through the MPA over the reconstructed cardiac cycle. Cardiac output was calculated by multiplying stroke volume (SV) by heart rate (HR). SV and cardiac output were both indexed by subject body surface area.

To quantify ventricular KE, the RV was segmented from time-averaged magnitude 4D flow images with MIMICS. A MATLAB script applied this segmentation mask to the time-resolved velocity images to extract the velocity magnitude for each voxel included in the RV mask. Ventricular KE was calculated using the following formula:where ρ = 1060 kg/m³ is the assumed density of blood, V = 1.95 mm³ is the voxel volume, and v_i is the velocity magnitude for voxel i. RV energy “efficiency” was defined as the SV normalized by ventricular KE, η \(=\frac{SV}{KE}\). To further characterize the behavior of the intra-ventricular velocity fields, vorticity was calculated for each voxel in the segmented RV by taking the curl of the velocity field. Total vorticity was determined by adding all voxels included in the mask. KE, η, and vorticity were calculated for all cardiac phases. Peak systolic and diastolic values for these parameters were defined as the local maximum values on the time-resolved curves during the cardiac phases associated with systole and diastole, respectively. These measurements were indexed by mask size.

To lend further context to the quantitative vorticity measurements, streamline and pathline visualizations were generated in Ensight to qualitatively assess differences in intra-ventricular flow. The streamlines and pathlines were generated from a plane bisecting the tricuspid valve, RV apex, and pulmonary valve to characterize the main flow patterns through the ventricle. The flow dynamics of subjects with abnormal vorticity measurements were inspected to identify differences in flow patterns compared to representative controls. Special attention was given to the structure of the diastolic filling vortex, which has been shown to be important to efficient RV function [27].

All measurements are presented as the mean of the samples plus/minus standard deviation. To further enhance the statistical power and allow for age-adjusted comparisons, the two age groups were combined, and multivariate linear regression was performed with age and birth status as independent variables. Term/preterm status was represented as a dummy variable. Cardiac index, SV index, systolic KE, diastolic KE, systolic η, diastolic η, systolic vorticity, and diastolic vorticity were included in the model as dependent variables. Significant differences between groups was determined from the t-statistic of a hypothesis test with the null hypothesis that the birth status coefficient was equal to zero. A significance threshold of α = 0.05 was used for all tests. p-values were uncorrected for multiple comparisons. While this confers an increased risk for type I error, it reduces type II error, which we deemed was appropriate in this attempt to characterize subtle differences in flow dynamics between two overtly healthy cohorts.

The approach used to quantify KE and vorticity in this study has some inherent limitations. RV volumes were segmented from time-averaged 4D flow magnitude images. This approach suffers from reduced myocardium-to-blood-pool contrast and lack of dynamic information when compared to cine bSSFP-based segmentation [36]. KE measurements using time-averaged 4D flow magnitude segmentations have been shown to suffer from reduced repeatability compared to flow measurements in the MPA [15]. While resting bSSFP images could be registered to resting 4D flow data to improve ventricle segmentation, this approach was not feasible during exercise due to the incompatibility of the breath-hold requirement for bSSFP imaging during exercise. In addition to segmentation challenges, the high VENC employed in this study was selected to characterize high velocities in the MPA and was not optimized for slower ventricular flow. This likely contributed to increased noise in KE and vorticity measurements. Dual-VENC approaches [37] may allow for optimized velocity encoding in the great vessels and cardiac chambers but would almost double scan times. Finally, while the decision to reconstruct 15 cardiac phases helped offset SNR loss during exercise, it increased the apparent temporal resolution, reducing sensitivity to the early diastolic motion important for properly characterizing diastolic dysfunction.