Dynamic pressure–volume loop analysis by simultaneous real-time cardiovascular magnetic resonance and left heart catheterization

Animal experiments were approved by the local Institutional Animal Care and Use Committee and conducted per contemporary guidelines from the National Institutes of Health. We prospectively performed 16 dynamic PV loop experiments in female juvenile Yorkshire swine. Three porcine models were studied: naïve (n = 7, 50 ± 6 kg), chronic afterload increase by aortic banding (n = 6, 46 ± 7 kg), and ischemic cardiomyopathy (n = 3, 83 ± 9 kg). Induction of a chronic afterload increase was created by inserting a transcatheter external aortic band [18]. In short, pericardial access was obtained through percutaneous subxiphoid access. A guidewire loop was formed around the ascending aorta within the pericardial sac using deflectable catheters, which was subsequently exchanged for a suture band secured in position with a Rumel-style tourniquet and subxiphoid pocket [19]. Ischemic cardiomyopathy was induced as a model of impaired systolic function induced by multi-vessel transcoronary ethanol chemoablation of the mid left anterior descending coronary artery and an obtuse marginal branch. The chronic afterload increase model pigs were examined 30 days after aortic band insertion, and cardiomyopathy pigs were examined 154 (142–161) days after chemoablation.

Animals were maintained under general anesthesia with mechanical ventilation and isoflurane inhalation. Percutaneous femoral arterial and venous access was obtained. Under X-ray, a fluid-filled dual-lumen catheter (Langston, Teleflex, Morrisville, North Carolina, USA) was placed in the LV to allow simultaneous recording of LV and aortic pressures, and a non-metallic polymer based 24 mm air filled balloon catheter (Atlas, Bard, New Providence, New Jersey, USA) was positioned within the right atrium (RA). Animals were transferred to the adjacent CMR scanner for further real-time CMR-guided catheterization [20, 21]. Preload reduction was obtained under CMR guidance by fully inflating and withdrawing the balloon to occlude the IVC-RA junction. Intravenous heparin to achieve an activated clotting time > 250 s and amiodarone were administered.

Imaging was performed using a contemporary 0.55 T CMR system (prototype MAGNETOM Aera, Siemens Healthineers, Erlangen, Germany) [22]. Real-time long-axis 4-chamber cardiac CMR cine images were acquired during end-expiratory breath-hold using a Cartesian balanced steady-state-free-precession sequence (TE/TR/θ = 1.4 ms/3.3 ms/80°, 2.3 × 2.3 × 8 mm reconstructed resolution, 360 × 270 mm FOV, acceleration rate 3, 76 ms temporal resolution, 326 timeframes). The IVC-occlusion was performed ~ 5 s after the start of imaging, and deflated after an additional 15–20 s.

A custom, open source, hemodynamic recording system (PRiME, Physiological Recording in the CMR Environment) was used to record LV pressures [17]. PRiME allows synchronized recording of a six lead ECG, two invasive blood pressures, and CMR gradient waveforms at a sampling frequency of 1 kHz, and supports cardiac triggering from either the ECG or pressure signal. The recorded gradient activity provided a temporal fiducial to ensure that only physiological signals recorded simultaneously with the real-time CMR acquisition were sampled for the PV signal pairing.

An automated signal processing pipeline that derives PV loops, contractility and compliance in six steps was implemented (Fig. 1):

The pipeline was implemented in MATLAB (Mathworks, Natick, Massachusetts, USA), allows manual corrections of end-diastolic and end-systolic detections, and is available as open source (https://​github.​com/​NHLBI-MR/​dynamic_​pv_​loops).

To validate volumes calculated by real-time long-axis CMR, breath-held retrospectively cardiac gated balanced steady state free precession short-axis-stack cine images (TE/TR/θ = 1.79 ms/4.4 ms/78°, 0.7 × 0.7 × 8 mm reconstructed resolution, 360 × 270 mm FOV, 16 slices covering the left ventricle (LV), bandwidth 305 Hz/Pixel, 34 ms temporal resolution, 30 timeframes) were acquired in all animals, without balloon inflation. Short-axis LV segmentations at end-diastole and end-systole were performed using the freely available software Segment (v3.1 R8154, Medviso, Lund, Sweden) [23]. The same segmentations were used for the conductance catheter calibrations. EDV and ESV LV volumes measured with real-time long-axis CMR and with a conductance catheter, from the first recorded heartbeat prior to balloon inflation, were compared to short-axis-stack volumes.

Statistical analysis was performed using GraphPad Prism 9 (GraphPad Software, Inc, La Jolla, CA, USA). Continuous variables were reported as mean ± standard deviation (SD). Contractility and compliance between naïve and diseased animal models were compared using one-way ANOVA. Agreement between measured long-axis and short-axis derived LV volume were described with Bland–Altman analysis and intraclass correlation coefficient (ICC), where levels of agreement was defined as poor (0.00–0.30), weak (0.31–0.50), moderate (0.51–0.70), strong (0.71–0.90), and excellent (0.91–1.00) [24]. Pearson R were reported and statistical significance was considered for p < 0.05. Differences between parameters derived from PV loops measured in the CMR environment and with a conductance PV loop catheter was assessed with a paired t-test.

LV volumes derived from real-time long-axis imaging and centerline rotation across all animal models had an excellent agreement with short-axis-stack imaging at both end-diastole (ICC = 0.97, R = 0.98, p < 0.001, bias 7.6 ± 9.2% or 8.2 ± 14.5 ml) and end-systole (ICC = 0.99, R = 0.99, p < 0.001, bias − 4.2 ± 8.6% or − 2.8 ± 6.5 ml). Figure 3 provides correlation plots and Bland–Altman comparison of the two measurement methods and Fig. 4B provides an example comparison of the cardiac volume over time in a naïve animal. Volumetric biases across the different animal models at end-diastole and end-systole were respectively 8.4 ± 6.5 ml and − 2.9 ± 5.3 ml in the naïve animals, 9.8 ± 10.0 ml and − 3.3 ± 7.9 ml in the aortic banding animals, and 0.7 ± 8.4 ml and 4.7 ± 34.4 ml in the cardiomyopathy animals. Cardiomyopathy animals were larger (83 ± 9 kg) than naïve (50 ± 6 kg) and aortic banding animals (46 ± 7 kg), and therefore had larger ventricles, stroke volume, and cardiac output compared to the other animal models. The low LV ejection fraction (LVEF) in the cardiomyopathy animals was expected (27 ± 4.5%). However, both naïve and aortic banding model pigs used in this study had mildly reduced LVEF at baseline (39 ± 12% and 36 ± 10%, respectively). This could be due to an increased sensitivity to amiodarone, which we use as a routine pre-medication in the catheter laboratory, in this cohort.

Quantified contractility and compliance are shown in Fig. 5. We detected differences in contractility and compliance between naïve and cardiomyopathy models, but not between naïve and banded models.

The cardiomyopathy animal model developed a dilated ventricle (short-axis EDV 277 ± 36 ml, ejection fraction 27.0 ± 4.5%), with lower contractility (0.2 ± 0.1 mmHg/ml vs 0.6 ± 0.2 mmHg/ml, p < 0.05) and increased compliance (12.0 ± 2.1 ml/mmHg vs 4.9 ± 1.1 ml/mmHg, p < 0.01) compared to naïve animals. We measured no difference between aortic banding model and naïve animals in contractility (0.5 ± 0.1 mmHg/ml vs 0.6 ± 0.2 mmHg/ml, p = 0.82) and compliance (6.8 ± 3.6 ml/mmHg vs 4.9 ± 1.1 ml/mmHg, p = 0.32), which is explained by the 10 ± 6 mmHg pressure gradient across the aortic band, that was considered modest and not clinically significant [25].

An example of single-beat LV pressures, volumes, and PV loops measured in the CMR environment and with a conductance catheter is shown in Fig. 4, measured before IVC occlusion in a naïve animal. The example highlights strengths and shortcomings of both the CMR and catheter methods. The conductance catheter pressure measurement was steadier than the fluid-filled catheter pressure, as expected. Fluid-filled catheters are inherent sensitive to damping, which in this example was apparent both peak systole and early diastole, and is likely ascribed to an air bubble in the fluid column. Long-axis real-time CMR volume measurements were on the other hand comparable to the reference standard short-axis-stack volumes, whereas the conductance PV loop catheter in this example measured higher volumes throughout out the entire cardiac cycle, despite being calibrated with short-axis cine volumes. Differences in pressure and volume measurements were also perceptible in the combined PV loop representation, resulting in different positioning of end-systolic and end-diastolic detections in the PV plane. These different positions will impact the derivation of the ESPVR and EDPVR lines, and in turn, the quantification of contractility and compliance.

The impact on quantitative parameters is illustrated in Fig. 6, showing correlation plots and Bland–Altman comparison of contractility and compliance derived from measurements by CMR and a conductance PV loop catheter. There was a weak to moderate agreement and large bias for both quantitative parameters (contractility ICC = 0.56, bias − 51 ± 73%, compliance ICC = 0.44, bias − 16 ± 66%), but no statistically significant differences were found in contractility (0.44 ± 0.33 mmHg/ml vs 0.76 ± 0.51 mmHg/ml, p = 0.31) or compliance (8.4 ± 4.5 ml/mmHg vs 15.0 ± 15.0 ml/mmHg, p = 0.35) quantified from dynamic PV loop measurements in the CMR environment vs with a conductance PV loop catheter. Real-time long-axis volumetric measurements and pressure measurements from the CMR environment are compared to conductance PV loop catheter measurements in Table 2, disclosing differences in ESV, ejection fraction, cardiac output, and peak LV pressure between the methods. Volumes measured at baseline with a conductance catheter and short-axis-stack volumes did not differ significantly at end-diastole (EDV: 220 ± 94 ml vs 205 ± 102 ml, p = 0.08, bias 11 ± 13% or 15 ± 15 ml), or end-systole (ESV: 143 ± 73 ml vs 145 ± 82 ml, p = 0.84, bias 4 ± 13% or 2 ± 16 ml).

In this study, we approximated the ESPVR and EDPVR to linear relationships and derived contractility and compliance from their respective slopes. The ESPVR is approximately linear within the physiological range of end-systolic pressures and volumes, whereas the EDPVR is exponential [26]. A linear approximation of the EDPVR has however been shown to suffice as a quantitative metric of compliance characterizing the overall diastolic function [27], as the slope of and exponential varies greatly depending on loading conditions.

Our method was able to characterize both a decreased contractility and increased compliance in animals with cardiomyopathy compared to naïve animals. These results are consistent with changes seen in dilated cardiomyopathy in humans [26]. Conversely, a decrease in compliance due to an increased afterload was expected but not observed in the animals with aortic banding. This negative result is explained by the clinically non-significant pressure gradient that developed across the banding. Future studies using another animal model with a significantly increased afterload, as well as evaluation in other diseases, are warranted to further evaluate the efficacy of the proposed method.

Intracardiac pressures can only be measured invasively, often using either a fluid-filled catheter transducer or a catheter equipped with a solid-state pressure sensor. Solid-state pressure sensors are positioned on the catheter and therefore measure pressure at the site of interest, a technique typically used in conductance PV loop catheters. Unfortunately, these catheters are not CMR-compatible. Fluid-filled catheters offer the advantage of being CMR compatible, but measure pressures that have been translated through the fluid column to a transducer located outside of the body. Fluid-filled catheters are therefore inherently susceptible to damping and latency, specifically at peak systole and early diastole, especially if there is an air bubble in the catheter [28]. In this study, we observed damping in some experiments performed with fluid-filled catheters in the CMR environment, but not in the conductance catheter experiments. We also measured higher peak systolic LV pressures with the fluid-filled catheter compared to conductance catheters, which may be explained by underdamping in the fluid-filled pressure transducer [28]. However, this damping did not nominally impact pressures at end-systole and end-diastole, meaning that the pressures at the sampled points used to derive ESPVR and EDPVR, and in turn, contractility and compliance, should not be affected by this limitation. Other hemodynamic parameters that may be of interest might be impacted, for example stroke work which is measured as the area inside the PV loop. A catheter with a solid-state pressure sensor would thus theoretically be preferable to the fluid-filled catheter for this application, if CMR compatible.

Another challenge of the CMR environment is the RF-induced interference on physiological signal recordings. In this study, this was evaded by filtering out CMR-induced noise using the custom open source PRiME system [17]. The PRiME system was also used to ensure that only physiological signals recorded simultaneously to the real-time CMR sequence acquisition were sampled for the PV signal pairing. The simultaneous and synchronized ECG measurements also allowed detection of end-diastole from the straightforwardly detected peak R-wave rather than from the LV pressure signal.

Short-axis-stack cine CMR is considered reference standard for measuring intracardiac volumes and is widely used in the clinical setting [16], but typically requires several minutes of image acquisition to produce a cine image of one heartbeat. In order to capture the transient LV volumes during IVC occlusion, we opted to measure volume from a real-time long-axis 4-chamber cine by performing center line rotation on 2D LV segmentations. These long-axis derived volumes were comparable to short-axis-stack volumes without a need of volumetric calibration, with a clinically acceptable bias of < 10 ml which held across the three animal models [29]. Center line rotation does however approximate the LV to a perfect circular shape around the rotated center line axis. This approximation was sufficient in our study, but may introduce bias in subjects with a more irregularly shaped ventricular cavities or with regional wall motion abnormalities not depicted in the long-axis view. Conductance PV loop catheters assume a cylindrical LV geometry and provide relative volume estimates using measures of the time-varying electrical conductance induced by electrodes on the catheter, using an approximation of the myocardial parallel conductance derived from the hypertonic saline calibration procedure [30, 31]. A secondary calibration is then required for absolute volume estimations, which in this study was performed using ventricular volumes measured by short-axis CMR. We found that these calibrated conductance PV loop catheter volumes were not comparable to end-diastolic short-axis-stack CMR with bias > 10 ml, confirming previous findings by Jacoby et al. [8]. The volumetric discrepancy impacted computation of both contractility and compliance, resulting in a weak to moderate agreement between conductance catheter and CMR-derived measurements of these parameters. Quantitative values of contractility and compliance derived from dynamic PV loops measured in the CMR environment and conductance catheter are therefore not directly comparable. Given the superiority of volumetric measurements in CMR, this technique may provide the more accurate value.

The temporal resolution of the conductance catheters is significantly higher than CMR (4 ms vs 76 ms in this study). Conductance catheters are therefore better at characterizing the short isovolumic phases of the cardiac cycle (IVCT ~ 50 ms, IVRT ~ 80 ms) [32]. Witschey et al. previously found that an CMR temporal resolution of < 100 ms is sufficient to derive reliable PV loops, and suggested a real-time sequence implementation with 95.2 ms temporal resolution [13]. The proposed method in this study provides a real-time temporal resolution of 76 ms. Nevertheless, a sequence with higher temporal resolution, for example using spiral or radial sampling [33, 34], would be desirable to better characterize the volumetric variations over the cardiac cycle, especially at higher heart rates where 76 ms would only capture a few timeframes per heartbeat. In addition to offering reliable volumetric measurements, other advantages related to CMR-guided PV loops include the lack of ionizing radiation compared to X-ray fluoroscopy which is used for conductance PV loop catheterization, and that the procedure can be combined with a comprehensive CMR examination for evaluation of for example tissue characterization and other volumetric assessments.

The contemporary 0.55 T CMR system used in this study is well suited for CMR-guided interventional procedures due to the reduced RF-induced heating of metallic devices, which is quadratically related to field strength [22]. Device heating was, however, not measured in this study as we used a polymer based catheter without metallic-braiding. Cardiac catheterizations under CMR guidance have also been performed at 1.5 T [10, 20, 35], and we anticipate that the proposed dynamic PV loop recording would translate to higher field strengths.

This study had some notable limitations, including a relatively small sample size of 16 pigs. Additionally, the different sampling frequencies of the measured signals warranted the need to down-sample pressure and up-sample volume signals. The lower temporal resolution in the CMR measurements compared to the pressure measurement might impact the short isovolumic phases of the cardiac cycle in the derived PV loop representations. Another limitation for both approaches we used, is that the approximation of the ESPVR to a load independent linear relationship assumes an unchanged myocardial inotropy [36]. The physiological response to a preload alteration challenge includes the activation of the sympathetic autonomic nervous system, leading to a change in inotropic conditions. To avoid fitting the ESPVR to end-systolic points at altered inotropic states, we limited the included number of heartbeats after IVC occlusion to ten, while also monitoring for a significant beat-to-beat increase in heart rate which in this study was low (0.77%). Including fewer heartbeats would result in a poor linear fit. Hence, when selecting the number of points used for fitting of the ESPVR line, there was a compromise between including enough data to derive a reliable linear fit while avoiding including too many points at an altered inotropic state. Furthermore, volumetric estimations from long-axis images may be adversely affected by asymmetric ventricles, regional wall-motion abnormalities, and through-plane motion in the case of inconsistent breath holds. Further evaluation of this volumetric estimation is warranted in larger numbers of subjects with ischemic cardiomyopathy. Future work could also explore a cine or a real-time 3D cine sequence.