Experimental Validation of a Cardiac Simulator for in vitro Evaluation of Prosthetic Heart Valves

Objective This work describes the experimental validation of a cardiac simulator for three heart rates (60, 80 and 100 beats per minute), under physiological conditions, as a suitable environment for prosthetic heart valves testing in the mitral or aortic position. Methods In the experiment, an aortic bileaflet mechanical valve and a mitral bioprosthesis were employed in the left ventricular model. A test fluid of 47.6% by volume of glycerin solution in water at 36.5ºC was used as blood analogue fluid. A supervisory control and data acquisition system implemented previously in LabVIEW was applied to induce the ventricular operation and to acquire the ventricular signals. The parameters of the left ventricular model operation were based on in vivo and in vitro data. The waves of ventricular and systemic pressures, aortic flow, stroke volume, among others, were acquired while manual adjustments in the arterial impedance model were also established. Results The acquired waves showed good results concerning some in vivo data and requirements from the ISO 5840 standard. Conclusion The experimental validation was performed, allowing, in future studies, characterizing the hydrodynamic performance of prosthetic heart valves.

Although the main goal of most cardiac simulators is to mimic left ventricular and systemic circulation, pulse duplicators conception and control loop design can be different according to the experimental purpose. Once the evaluation of left ventricular assist devices is performed, for instance, based on adaptive estimation of the aortic pressure and suitable response regarding the left ventricular contractility variation [10,[17][18][19][20] , the operation of cardiac simulators in this case is designed to an automatic variation of cardiovascular parameters (according to the Frank-Starling mechanism), requiring a full closed-loop control. However, in terms of prosthetic heart valves testing, the cardiac simulators are used only to conduct cyclic operation with good repeatability, in some predicted ventricular conditions [12] . In this case, the cardiac simulator control system can be simplified, but the ventricular model is expected to mimic some anatomical  (2), optical platform made of acrylic apt to laser velocimetry applications (3), mitral valve (4), aortic valve (5), flow probe (6), characteristic resistance (7), adjustable compliance (8), adjustable peripheral resistance (9), pre-atrial reservoir (10), digital thermostat (11), temperature sensor (12), heater (13), work fluid reservoir (14), microcomputer to run the supervisory control and data acquisition system (15), servomotor drive (16), servomotor (17), linear slide table (18), hydraulic cylinder (19), DAQ module (20), signal conditioners (21), invasive blood pressure transducers (22), and flowmeter (23). system using LabVIEW 2011 (National Instruments Corp., Austin, TX, USA), described previously [43] , was applied for ventricular operation and to acquire the ventricular signals for each HR. Figure 1 shows the schematic model of the left heart and systemic circulation.
According to Figure 1, n. 20, a multifunction data acquisition (DAQ) module (NI USB-6212 BNC: 16-Bit, 400 kS/s, National Instruments Corp., Austin, TX, USA) was used connected to the instrumentation and the microcomputer running the LabVIEW. The left ventricular model (Figure 1, n. 1) was designed so that the geometry, size and valves positioning were similar to the natural left heart anatomy. The volumetric capacity of the ventricular chamber (up to 220 mL) was combined with the viability of laser velocimetry applications. The optical platform (Figure 1, n. 3) is completely exposed to the atmosphere [42] .
Through the DAQ module and LabVIEW, a pulse counter input was established to quantify the rotation of the servomotor shaft via an encoder. Every shaft revolution was discretized in 1,000 pulses. Thereby, the left ventricular model operation during cardiac cycles was referred to the servo motor encoder signals, allowing knowing the residual left ventricular volume (LVV) as a function of time, which is important to establish the LVP versus LVV diagram of the left ventricular model.
Supervisory control system allows modulating the servomotor drive parameters of the shaft rotation, such as velocity, acceleration, number of revolutions and waiting times. It allows the simulator to induce the SV, the HR and proper duration of systole and diastole, for instance. Indirectly, it also enables to establish the EDV and the ESV.
Arterial impedance (a Windkessel model based on three elements, Figure 1, n. 7-9) is not a function of a full closed-loop control system. Therefore, manual adjustments of compliance and resistances were required. Moreover, the test fluid level in the pre-atrial reservoir (Figure 1, n. 10) determines the atrial pressure [3] . Also, air volumes can be injected into the work fluid reservoir (Figure 1, n. 14 confined air) in order to adjust the sensitivity of the flexible membrane actuation.

Experimental Procedure
An aortic bileaflet mechanical valve (CarboMedics Inc., Austin, TX, USA, 27 mm diameter) and a stented tricuspid mitral bioprosthesis of bovine pericardium (confidential information, 31 mm diameter) were used in the left ventricular model. The aortic and mitral valves were positioned as shown in Figure 2.
A glycerol-water mixture with 47.6% by volume of glycerin solution in water (with normal saline solution to allow the electromagnetic flowmeter operation) at 36.5±0.5ºC was used as a blood analogue fluid. It implies a dynamic viscosity of approximately 4 mPas [44] . The supervisory control system was configured to send parameters to the servomotor driver fixing SV, ESV, HR, and DiaD, according to Table 1. The same SysD of 360 ms was fixed for the three predicted HR. Thus, changes in HR were established by different DiaD (Table 1), which is significantly more influenced by HR [27,28] .
Regarding the data acquisition system, the flow and pressure analog inputs were established, respectively, with first order lowpass filters (Butterworth) of 10 Hz and 20 Hz [12] . The flow signals were also associated with a median filter of 100 elements, in order to reduce the signal noise from the electromagnetic flowmeter. Sample rate was 1 kHz, based on 1,000 samples.
The arterial compliance was adjusted to near 2.2 mL/ mmHg [9] and the peripheral resistance was slightly higher as HR was increased.

RESULTS
Through LabVIEW, waves of LVP, AoP, AoF, and SV were acquired. In order to verify the cardiac simulator ability to replicate some ventricular and systemic circulation characteristics, Figures 3, 4 and 5 were obtained, respectively for 60, 80 and 100 bpm.
All responses from cardiac simulator (according Figures 3 to 5) were inserted in Table 1 for suitable comparison. Afterload was slightly higher as HR was increased, at a constant preload [40] .
LVP versus LVV diagram (PxV diagram) were plotted as shown in Figure 6, with each loop averaged over five consecutive cycles.
Rapid variations at AoP wave were observed at the instant of the mechanical aortic valve closure, in accordance with the strong impact and the possibility of momentary partial reopening of the leaflets [5,47] . At the same time, corresponding wave oscillations were noticed in the LVP shown in Figures 3 and  4, indicating a rapid interruption of these descending pressures. However, it did not appear at 100 bpm (LVP of Figure 5), when the duration of diastole was shorter than at 60 and 80 bpm (see DiaD in Table 1). This means that, although the duration of systole was the same for the three HR, the flow dynamics regarding the isovolumetric relaxation phase (beginning of diastole) at 100 bpm affected differently the AoP wave. As a consequence, the LVP was also different.

Pressure Versus Volume Diagram
According to Figure 6, the PxV diagrams (each loop averaged over five consecutive cycles) demonstrate EDV of 120 mL [26] . The values of SV and ESV (Table 1) were similar to some studies including different HR [19,36] .
There were no phase in which the pressure increased without changing the intraventricular volume, as would be expected from the isovolumetric contraction (beginning of systole).   In this cardiac simulator, the ventricular filling is not strictly passive, as in projects with a distinct design [10,11,24] , but runs according to the hydraulic piston return and flexible membrane deflection -besides the air volumes injected into the work fluid reservoir, like a viscoelastic impedance adapter [5] . Anyway, some other cardiac simulators have achieved good results in this sense, regardless of the type of project [19,20,48,49] , though not always using a blood analog fluid [20,48] .

Ejection Phase and Cardiac Output
The EPL for all HR was 250 ms. Although the reference value was 210 ms [31] , it is possible find in vitro studies with EPL up to 300 ms [22] . At 60 bpm, the peak flow rate in the aorta was close to 25 L/min, consistent with Dasi et al. [15] , in which values of 24 L/min were found. The peak flow rate was lower as HR was increased, according lower values of SV for 80 and 100 bpm. The CO and SV waves were reciprocally consistent as a function of time (and also concerning the LVP data).
The CO was 3.8, 4.2 and 4.9 L/min, respectively for 60, 80 and 100 bpm. These values were lower than expected by theoretical assessment (CO = SV * HR), when aortic regurgitation is neglected. However, the regurgitant volume expected in the aortic bileaflet Carbomedics valve of 27 mm at 70 bpm is about 7.5 mL/beat [23] . Thus, the CO results seemed also consistent.

Other Features
Although the duration of systole was 360 ms, the period from the beginning of the isovolumetric contraction until the end of the isovolumetric relaxation varied depending on HR (Figures 3  to 5). All the values were higher than those mentioned in the literature concerning natural valves, i.e., 320 ms at 70 bpm [31] . However, this duration is influenced by the strong impact of the metallic leaflets on the aortic valve closure [5,47] .
In the flow through large arteries and heart chambers, blood behaves as a Newtonian fluid, with shear rates higher than 100 sec -1 [13,37] . Although, in the vicinity of the hinges -prone to flow disturbances and recirculation -, and where the size of the flow domain is similar to the magnitude of the blood cell size, the non-Newtonian effects must be considered [14,50] .

CONCLUSION
All acquired waves (i.e., LVP, AoP, and AoF responses obtained for 60, 80 and 100 bpm) showed good repeatability for the cardiovascular parameters and prosthetic valves used. Despite some limitations, the cardiac simulator is suitable for in vitro evaluation of prosthetic heart valves. Hence, the cardiac simulator was validated to these conditions, in accordance with the human physiological parameters.
Hydrodynamic testing of prosthetic heart valves can be started, once the cardiac simulator operating parameters allow valid experimental comparisons of flow through mitral or aortic prostheses.

Future Studies
Since the optical accesses were provided in the ventricular model, it is possible to apply, in future works, the PIV and LDA systems [13,15,23] . These results can be used to obtain a computational model of the flow [14,15] . Further studies may also consider a development of a full closed-loop control for the simulator, where its responses should be evaluated by the PxV diagram for several physiological conditions dynamically [11,19,20,48,49,51] .

Limitations
Some limitations of the cardiac simulator are inherent to the in vitro condition. The left ventricular model is not completely flexible and cannot simulate the twisting motion that occurs in the human heart. Furthermore, the ventricular filling is not passive. These conditions may affect the LVP, AoP and CO waves, for instance, which were also attenuated by the low-pass filters used (cut-off frequencies of 20 Hz for LVP and AoP, and 10 Hz for CO).
Some well-known variables from the human physiological literature (regarding the natural heart valves) served as parameters for this validation experiment. However, they were not strictly met, since cardiac prostheses were used.
In this work, we did not satisfy the requirements from the ISO 5840 concerning the report of the prosthetic valves behavior or the analysis of their hydrodynamic performance, as