Synopsis

Keywords: Heart, Modelling

Clinical pressure-volume data are often subjected to measurement noise. In the present work we explore the potential of a biomechanical model of the heart as a data-filtering tool. We use cardiovascular MRI volume data either simultaneously or sequentially acquired with catherization. We use time-varying elastance metrics for a quantitative comparison of data- and model-derived pressure-volume loops (PV loops). Our results demonstrate that the biomechanical model provides high quality PV loops that could be used to replace noisy data signals. In addition, this work contributes to the increased understanding of cardiovascular MRI results.

Introduction

Therapy planning may benefit from the assessment of myocardial energetics using ventricular pressure-volume loops (PV loops) ¹. PV loops can be reconstructed from interventional cardiovascular magnetic resonance imaging (MRI) (simultaneous acquisition of catheter pressure and MRI). Clinical pressure-volume (P-V) data are often subjected to measurement noise, e.g. temporal dyssynchrony, non-physiological over-/undershoot in pressure measurements (Figure 2). Our previous work demonstrated that a biomechanical model could be deployed to synchronize clinical P-V data in time ². This work aims to further explore the potential of the model to filter measurement noise in clinical data. We aim to quantitatively compare data- and model-derived P-V waveforms based on the time-varying elastance (TVE) metrics. The data-derived P-V waveforms will use acquired data directly from cardiovascular MRI and catheter-based measurements.

Methods

Data: The study included 15 P-V datasets of patients with single-ventricle physiology undergoing interventional cardiac MRI within the pre-Fontan assessment, and 17 left and 20 right ventricular P-V datasets of patients with repaired tetralogy of Fallot (rTOF). In the Fontan cohort, P-V data were acquired simultaneously. A highly accelerated prospectively ECG-triggered cine bSSFP sequence was used (kt-SENSE factor 6, partial Fourier 0.625, slice thickness 10 mm, spatial resolution 2.4). In the rTOF cohort, retrospective ECG-gated cine bSSFP sequence was used (SENSE=2, spatial resolution 2x2x10 mm). Catheterization was sequentially performed during percutaneous pulmonary valve replacement therapy. The target temporal resolution was ~30 ms in all acquisitions. The motion tracking algorithm ³ was used to extract time-vs.-volume plots from cine MRI.
Biophysical model of the heart: The model represented the mechanical (contractile and viscoelastic) behavior of a single ventricular cavity with spherical shape ^4,5. Cavity mass and radius were prescribed from the patient’s MRI data. Mechanical laws were calibrated so that simulated P-V waveforms corresponded to the maximum and minimum P-V traces for a given patient. The circulation system was represented by a Windkessel model. It was coupled to the heart cavity via a system of diodes representing the ventricular outflow valves (Figure 1). Calibration of the model provided patient-specific PV loops and associated mechanical quantities of the heart and circulation – myocardial contractility and stiffness, distal resistance of the circulation.
Quantitative assessment of pressure-volume data: TVE, i.e. E(t)=P(t)/V(t), maximum value of TVE (E_max)⁶ and maximum time derivative of ventricular pressure (max(dP/dt)) were computed from data- and model-derived P-V waveforms (Figure 2). TVE theory conceptually links ventricular isovolumetric and end of ejection phases via ⁷:

$$max(dP/dt)/EDV=E_max/t_max * k,$$

(1)where EDV denotes end-diastolic volume, tmax is the time of the occurrence of E_max, and k is a constant. Eq. (1) and linear regression analysis were used to quantitatively compare data- and model-derived P-V waveforms. In addition, Bland-Altman plots and Wilcoxon signed-rank tests at p<0.05 were used to evaluate the difference between data- and model-derived E_max/t_max and max(dP/dt).

Results

All data – and model-derived linear regressions were significant with p<0.01. Model-derived max(dP/dt) vs. model-derived E_max/t_max linear regressions showed the highest R² values in all groups (Figure 3). Model-derived max(dP/dt) vs. data-derived E_max/t_max showed stronger linear regressions than data-derived max(dP/dt) vs. data-derived E_max/t_max in all patient groups. Bland-Altman plot statistics demonstrated that mean bias ± standard deviation (SD) for data- vs. model-derived max(dP/dt) were -272 ± 168 mmHg/s, -72 ± 215 mmHg/s, and -40 ± 118 mmHg for Fontan, rTOF-LV and rTOF-RV patients, respectively (Figure 4 and Table 1).

Discussion

This study demonstrated that both data – and model-derived P-V traces produced statistically significant relationships between ventricular isovolumetric and end of ejection phases – in accordance with the prediction of TVE theory. However, model-derived P-V traces provided quantitatively stronger relationships between max(dP/dt) and E_max/t_max in all patient groups. Also, model-derived max(dP/dt) was more accurately reflective of data-derived E_max/t_max than data-derived max(dP/dt). Our results suggest that such a biomechanical model could be used as data-filtering tool. The calibration of the model requires discrete patient data inputs (i.e. maximum and minimum P-V values). Therefore, if clinically acquired continuous data signal is too noisy or not available at all, this biomechanical model could provide high quality patient-specific PV loops for functional evaluation of cardiac health.

Conclusion

This model of the heart can assist in clinical data-processing by complementing noisy P-V data thus improving an objective assessment of clinical P-V loops. This study contributes to the increased understanding of cardiovascular MRI results and their links to cardiac (patho)physiology.

Acknowledgements

The study was supported by the Inria-UTSW Associated Team TOFMOD and the W. B. & Ellen Gordon Stuart Trust & The Communities Foundation of Texas.