Preparation
In this study, seven hearts were retrieved from pigs slaughtered for human consumption (Dutch Landrace hybrids, approximately 110 kg live weight). The protocols at the slaughterhouse and during the experiment were in agreement with European Community regulations 1069/2009 and 142/2011 regarding the use of slaughterhouse byproducts for research and were approved by the associated legal authorities of animal welfare. 4D flow MRI was applied in five pig hearts with native valves. In two additional experiments described in Additional file 3, the native values of two other hearts were replaced in the preparation phase of the experiment with TAVR valves by a specialist (A.d.W.) with 12 years of experience. All experimental and post-processing steps were the same as for hearts with native values. The prosthetic valves available were the CoreValve (29 mm, Medtronic Inc., Minneapolis, MN, USA) and Edwards SAPIENS XT (26 mm, Edwards Lifesciences LLC, Irvine, CA, USA). The prosthetic valve sizes were not selected according to the pigs’ annulus size, which is why the Edwards SAPIENS XT valve was additionally stabilised by a purse string suture.
Each slaughterhouse pig heart was harvested and arrested with crystalloid cardioplegic solution after a very short warm ischaemic time. Approximately 10 L of blood was collected from the same animal and heparinised. The heart was transported in an iced cardioplegic solution and connected to the MRI-compatible isolated beating pig heart platform [19] (PhysioHeartTM platform, LifeTec Group BV, Eindhoven, The Netherlands) on average 4 h after death. During preparation, the pericardial sac was discarded and the right pulmonary veins, vena cava inferior, vena cava superior, and the left azygos vein were tied off. A 27-mm cannula was inserted into the opened left pulmonary vein and secured with a purse string suture. The connecting tubes were compliant for blood flow and connected to preload simulating pulmonary resistances. A cannula with an internal diameter of 24 mm was inserted into the ascending aorta and fixed downstream to the aortic valve.
The aortic tube drained in an afterload simulating body resistance (systolic/diastolic pressure approximately 120/80 mmHg). A CO of approximately 4.5 L/min was maintained during the experiments to match human conditions. The CO was real time controlled via a flow sensor and regulated by adapting the preload and afterload resistances. A 17-mm cannula was inserted in the pulmonary artery, which was connected directly to a reservoir. The balanced blood was warmed and oxygenated by a heart lung machine outside the MRI room. The blood returned as a nutrient-rich, warm, oxygenated solution of 5% CO2 and 25% O2 and with a temperature of 38 °C. A schematic overview of the connections of pre- and afterload is shown in Fig. 1a.
Reperfusion
To resuscitate the heart, a Langendorff perfusion [20] mode was created by cross clamping preload and afterload systems (Fig. 1b). Via a side port in the aortic cannula, blood was pumped retrograde in the aorta at approximately 80 mmHg closing off the aortic valve forcing flow into the coronary system. After perfusion of the myocardium, the deoxygenated blood drained in the coronary sinus, the right atrium, and the right ventricle. Through the pulmonary artery, blood was pumped back to the reservoir. The heart was left in Langendorff perfusion to recover for about 30 min until a steady state (in terms of a physiological colour and temperature, a stable sinus rhythm, a constant coronary flow, and constant pressures) was obtained (Additional file 1: Video S1).
Hereafter, the platform was switched to the working mode (Fig. 1c): Langendorff perfusion was stopped and the aorta and left pulmonary vein (preload) were opened, resulting in blood actively pumped by the LV with sufficient preload leading to a physiological CO. Coronary filling was created by LV function and aortic pressure only. To ensure a stable heart rate (HR) during MRI scanning, a pacemaker was attached to the heart. The paced HR was approximately 5–10 beats per min (bpm) over the irregular HR of the resuscitated heart. As soon as the heart was beating with a stable rhythm (approximately 20 min after switching from the Langendorff mode to the working mode), the setup was inserted into the MRI scanner (Fig. 1e) and the acquisitions were started.
4D flow MRI
All acquisitions were performed with a 3-T scanner (Ingenia, Philips Healthcare, Best, The Netherlands) using retrospectively triggered 4D flow MRI. Two medium flex coils (diameter 10 cm) were connected to the isolated beating pig heart platform below and above the heart (Fig. 1c). Electrocardiography sensors were attached via copper wires on the hearts’ surface and were used as a cardiac trigger signal. Twenty-four cardiac frames at a temporal resolution of 21 ms were acquired covering the cardiac cycle.
The 4D flow MRI scan had a field of view of 150 × 150 × 150 mm3 and a non-interpolated spatial resolution of 2.3 × 2.3 × 2.3 mm3. Echo time, repetition time, and flip angle were 2.2 ms, 5.2 ms, and 8°, respectively. To reduce the typically long acquisition times of 4D flow MRI, the scan was accelerated three times using k-t principal component analysis (Gyrotools, Zürich, Switzerland) resulting in a scan time of 12 min [22]. k-t principal component analysis acquisitions undersample k-space regularly over time, together with an interleaved training scan of the k-t space centre. The images are recovered during reconstruction by exploiting the relevant signal correlations available in the training data, which are represented as temporal basis functions and can be derived using a principal component analysis. The data was reconstructed using CRecon (Gyrotools, Zürich, Switzerland) with a k-t regularisation parameter of λ = 1. The velocity encoding was 100 cm/s.
Flow measurements
Visualisation of 4D flow MRI data and quantification of velocity and flow were done with GTFlow (Gyrotools, Zurich, Switzerland). For flow calculation, a region of interest (ROI) was chosen in the ascending aorta close to the sinotubular junction and downstream to the aortic valve. Net flow, forward flow, and backward flow were determined. The net flow was defined as the spatially averaged flow through the defined ROI:
$$ Q(t)={\int}_{\mathrm{ROI}}v\left(\mathbf{r},t\right)\ {d}^2\mathbf{r} $$
with v(r, t) being the velocity (pointing either in forward or backward direction) at position r and cardiac phase t.
Likewise, forward was defined as:
$$ Q{(t)}_{\mathrm{forward}}={\int}_{\mathrm{ROI}}v{\left(\mathbf{r},t\right)}_{v>0}\ {d}^2\mathbf{r} $$
and backward flow was defined as:
$$ Q{(t)}_{\mathrm{backward}}={\int}_{\mathrm{ROI}}v{\left(\mathbf{r},t\right)}_{v<0}\ {d}^2\mathbf{r}. $$
For the per cent quantification of the regurgitation fraction for one cardiac cycle T, forward flow volume was calculated as
$$ {V}_{\mathrm{forward}}={\int}_0^TQ{(t)}_{\mathrm{forward}}\ dt $$
and backward flow volume was calculated as:
$$ {V}_{\mathrm{backward}}={\int}_0^TQ{(t)}_{\mathrm{backward}}\ dt. $$
The per cent regurgitation fraction was based on the ratio RF = Vbackward/|Vforward|100. The stroke volume in millilitres was defined as SVflow = Vforward − |Vbackward|. The COflow in litres per minute was calculated by the product of the mean HR during the acquisition and SVflow.
In one heart, the 4D flow MRI scan was repeated three times within the same scan session at time points 0 min, 12 min, and 1 h 5 min.
Volume measurements
The LV volume was quantified for each heart using a 4D segmentation tool of velocity data from Medis (Medis medical imaging systems, Leiden, The Netherlands). Stroke volume was calculated as the difference between end-diastolic volume (EDV) and end-systolic volume (ESV). The per cent ejection fraction (EF) was defined as EF = SVvolume × 100/EDV. The size (long axis, short axis, and area) of the aortic annulus was measured during systole at the height of the aortic valve. Cardiac output was similarly calculated as the product of mean HR and SVvolume.
Statistical analysis
A statistical comparison between the flow and volumetric measurements was done using a Bland-Altman analysis and linear regression. Normal distribution of the data was tested using a Shapiro-Wilk test. The significance level was set to p < 0.05.