Benchtop Models of Patient-Specific Intraventricular Flow During Heart Failure and LVAD Support

Abstract

The characterization of intraventricular flow is critical to evaluate the efficiency of fluid transport and potential thromboembolic risk but challenging to measure directly in advanced heart failure (HF) patients with left ventricular assist device (LVAD) support. The study aims to validate an in-house mock loop (ML) by simulating specific conditions of HF patients with normal and prosthetic mitral valves (MV) and LVAD patients with small and dilated left ventricle volumes, then comparing the flow-related indices result of vortex parameters, residence time (RT), and shear-activation potential (SAP). Patient-specific inputs for the ML studies included heart rate, end-diastolic and end-systolic volumes, ejection fraction, aortic pressure, E/A ratio, and LVAD speed. The ML effectively replicated vortex development and circulation patterns, as well as RT, particularly for HF patient cases. The LVAD velocity fields reflected altered flow paths, in which all or most incoming blood formed a dominant stream directing flow straight from the mitral valve to the apex. RT estimation of patient and ML compared well for all conditions, but SAP was substantially higher in the LVAD cases of the ML. The benchtop system generated comparable and reproducible hemodynamics and fluid dynamics for patient-specific conditions, validating its reliability and clinical relevance. This study demonstrated that ML is a suitable platform to investigate the fluid dynamics of HF and LVAD patients and can be utilized to investigate heart–implant interactions.

1 Introduction

Heart failure (HF) affects approximately 5.8 million Americans and is projected to cost the U.S. healthcare system at least $70 billion annually by 2030 [1,2]. HF patients with mitral (MV) or aortic valve (AV) dysfunction, who are unsuitable for surgery, often opt to receive minimally invasive transcatheter valve replacement (TMVR or TAVR). Previous studies reported a 30-day mortality rate of 6%–18% and a stroke rate of 5%–14%, for TMVR, a mortality rate of 2.1%, a stroke rate of 1.4%, and a moderate paravalvular leakage rate of 15.3% for TAVR, [3,4]. Bioprosthetic valve replacements have a higher incidence of valvular deterioration and leakage, while mechanical valves pose greater thromboembolic (TE) risk [5,6].

Left-ventricular assist devices (LVADs) are mechanical pumps surgically connected to the heart and the aorta and are a popular alternative to a heart transplant for treating advanced HF. LVADs alleviate HF symptoms by unloading the left ventricle (LV) and reducing myocardial wall stress. Some patients experience a partial restoration of ventricular structure and function called reverse remodeling that can lead to myocardial recovery and LVAD explant [7]. However, the level of ventricular remodeling sufficient to allow LVAD removal is still controversial [8], and pump weaning and explantation strategies are heterogeneous [9–11]. Approximately 2% of LVAD patients achieve sufficient myocardial recovery to allow pump removal [12]; however, wean-off will be the future therapeutic goal with rapid advancement in LVAD technology and management.

Intraventricular flow examination is crucial to evaluate heart–implant interactions for developing strategies to improve patient outcomes. Advanced HF patients, especially those with LVAD support, are challenging to study due to many comorbidities [13]. Previous studies consist primarily of in vitro mock loop (ML) experiments [14–16] and in-silico simulations [17–20], except for one clinical [21] and two preclinical in vivo studies [22,23]. Rossini et al. Reconstructed the flow fields in LVAD-implanted patients using color-Doppler transthoracic echocardiography and compared it to those found in HF patients and normal subjects [21]. However, this comparison was complicated by the need to shift the imaging plane from the standard echocardiographic view in some LVAD implanted patients, due to ultrasound artifacts from the inflow cannula at the LV apex [21]. Aigner et al. used echo particle image velocimetry (echo-PIV) to visualize flow in the isolated porcine hearts during Medtronic HeartWare ventricular assist device support [22]. The study was well designed and executed but had several limitations, including the anatomical and functional difference between healthy porcine hearts and HF human hearts, and the underestimation of incoming velocity and shear [22]. Finally, Schinkel et al. visualized intraventricular flow fields using echo-PIV and ultrasound-enhancing agents, obtaining promising results [23]. However, the study displayed several limitations, including hindered off-axis apical view, underestimation of incoming velocity, and potential risk of ejecting ultrasound-enhancing agents [23]. Overall, the in vivo flow measurements are sensitive to imaging artifacts, and can be time-consuming and operator-dependent. On the other hand, in vitro and in-silico models allow for inexpensive, reproducible, controllable experimentation while providing significant support evidence.

Due to the lack of clinical gold standards, many researchers have utilized in vitro and in-silico studies to investigate fluid dynamics of heart–implant interactions. These models presented various limitations, such as oversimplified ventricular structures and motions in ML [14,24], and numerical simulations failed to adequately capture the flow complexity [17–20]. Under a specific context of use and appropriate credibility analysis, a simple model can be developed and serve as the foundation for a more complex and realistic model. Hence, the ML can be designed to capture cardiac fluid interaction of heart valve or LVAD patients and serves as a tool for testing multiple assumptions and justification regarding translation to the clinical setting. Moreover, computational models can be developed, calibrated, and validated from results obtained using controlled benchtop experiments.

In this study, an in-house ML is designed to resemble specific conditions of HF patients with regular and prosthetic MV and LVAD patients with small and dilated LV volumes. TE-related flow indices, including residence time (RT), shear-activation potential (SAP), and vortex properties, were calculated and compared with reported clinical studies [21,25–28]. By validating the capability and accuracy of the benchtop system in simulating various hemodynamics and flow dynamics conditions, the study demonstrated that ML is a suitable platform to study intraventricular flow and various potential impacts of treatment strategies for HF and LVAD patients and is used to validate computational models.

2 Methods

2.1 Clinical Image Acquisition.

Patients with written informed consent (Institutional Review Boards at San Diego State University and Sharp Memorial Hospital) underwent color-Doppler and B-mode echocardiographic examination using a Vivid 7 scanner and a phased-array 2–4 MHz transducer (General Electric Healthcare, Chicago, IL). Reported data, including ejection fraction (EF), heart rate (HR), pump speed, LV end-diastolic volume (EDV), end-systolic volume (ESV), E/A ratio, AV opening, and cardiac output, were from five HeartMate II (HM2) candidates: HF conditions of patient (PC) 1–4, and LVAD conditions (after the pump implantation) of PC1 and PC5. PC3 and PC4 had prosthesis MV (unknown valve types); after LVAD implantation, PC1 experienced significant LV volume reduction, while PC5 did not. Images from the long-axis 3-chamber view were processed to generate two-dimensional (2D) velocity fields using the echocardiographic flow vector mapping method, as previously validated in vitro [29] and in vivo [28] (see Appendix A for more details). In HF patients, the long-axis three-chamber view could be easily obtained, facilitating the replication of a similar view in the ML study (Fig. 1(a), Appendix A). However, following LVAD implantation, the presence of shadows and artifacts caused by the metal inflow cannula at the LVAD apex posed challenges in obtaining a comparable view. Consequently, patient echocardiographic images often exhibited shifted imaging windows. In some instances, the MV inflow jet appeared at the center of the image, while the boundaries of the LV, pump inlet, and LVOT were not visible, as shown in Fig. 1(b).

Fig. 1.

B-mode echocardiographic image and reconstruction velocity vector fields in the apical long-axis three-chamber view of (a) heart failure and (b) LVAD patients. (MV: mitral valve, AV: aortic valve, LV: left ventricle).

2.2 Mock Loop Studies.

The ML, constructed based on a three-element Windkessel model, replicated HF hemodynamics and the interaction of the native ventricle and LVAD [14]. To match clinical cases, the experimental design used three transparent silicone LV models with severely, moderately, and slightly dilated volumes (with ESV of 180 ml, 150 ml, and 100 ml, respectively, see Supplemental Fig. S1 available in the Supplemental Materials on the ASME Digital Collection). A transparent inflow cannula and HM2 were connected to the apex, while the outflow graft was replaced with 16 mm-diameter Tygon tubing and connected to the ascending aorta. All studies used porcine bioprosthetic heart valves (Medtronic) in both mitral and aortic positions, except the HF ML6 and ML7 which used a mitral tilting-disk mechanical valve (Medtronic) to produce reverse flow patterns as seen in PC3 and PC4.

A programable, linear displacement piston pump attached to the tank imposed a volume displacement waveform derived from the native heart to make the silicone LV “beat” [14]. Pressure (medical grade, disposable Transpak IV sensor (ICU Medical)) and flow (Transonic ME-PXL Series sensor (Transonic Systemic, Inc.)) transducers recorded LV pressure (LVP), aortic root pressure (AoP), total aortic flow (Q_Ao), and LVAD flow (Q_LVAD) at 200 Hz (LabChart, AD Instruments). The average of Q_Ao in one minute was equivalent to cardiac output. A mixture of 40% glycerol and de-ionized water (3.72 cP at 20 °C) filled the circulating fluid to match blood density and viscosity. Particle image velocimetry (davis software, lavision) captured the midplane velocity fields from a thin laser light sheet illuminated by neutrally buoyant fluorescent particles (20- $\mu$m, PMMA-RhB) (Fig. 2(a), Appendix A). Trigger image pairs were acquired at 40 Hz; fifty image sets were collected and phase-averaged for each time point in the cycle [14]. The HF studies imitated dilated cardiomyopathy conditions with severely dilated LV volume, 15–24% EF, and systemic pressure matching a class IV New York Heart Association [30]. For LVAD studies, a range of EF (24–39%), HM2 speed (8.6–9.8k rpm), and LV volume (slightly to severely dilated) were available to match specific patients.

Fig. 2.

Schematic of the mock loop setup from the top-down view: (a) the standard imaging plane captures the long-axis three-chamber view, providing a visual representation of the flow patterns typically observed in heart failure patients and (b) the imaging plane is modified to visually correspond to the flow patterns observed in R1 (MV: mitral valve, AV: aortic valve): (a) standard imaging plane and (b) modified imaging plane

2.3 Data Analysis.

The average transvalvular pressure (TVP), flow through the AV (Q_AV), energy equivalent pressure (EEP), and surplus hemodynamic energy (SHE) were calculated from flow and pressure signals (Appendix B). The temporal waveform and average of vortex properties and velocity pulsatility index (VPI) were calculated from the measured velocity fields [28] (Appendix B) (Supplemental Fig. S2 available in the Supplemental Materials on the ASME Digital Collection). After integrating a forced advection equation for ten cardiac cycles, the instantaneous RT and SAP maps for the midplane displayed convoluted regions of residual flow (RT > 2s) and elevated SAP (>200/s) [21,31]. Refer to Appendix C for the estimation of RT and SAP. Vortex properties, areas of residual flow and high SAP (as percentages of the midplane area), spatial average, and maxima of RT and SAP were compared with patient cases.

2.4 Comparison With Published Clinical Studies.

A comparison of hemodynamics, vortex properties, RT, and SAP was performed with five published clinical studies [21,25–28] to supplement the small number of patient cases in this study. Figures 7 and 8 combined the average and standard deviation results of clinical studies [21,25–28], specific patient cases (HF = 2 and LVAD = 4), and ML cases (HF = 7 and LVAD = 7).

Fig. 7.

Hemodynamics comparison (averages ± standard deviations) of mock-loop versus clinical studies: (a)aortic pressure, (b) ejection fraction, (c) surplus hemodynamic energy, (d) energy-equivalent pressure, (e)velocity pulsatility index, and (f) left ventricle volume (MV: mitral valve, LVV: left ventricle volume, EDV: end-diastolic volume, ESV: end-systolic volume)

Fig. 8.

Mock-loop versus clinical studies comparison (averages ± standard deviations). (a–d) Vortex properties: (a)circulation, (b) kinetic energy, (c) radius, and (d) symmetry CCW/CW (MV: mitral valve, LVV: left ventricle volume). (e–g) RT and SAP at the beginning of systole after 5 s of integration: (e) space-averaged and maximum RT, (f) space-averaged and maximum SAP, (g) size of regions with RT > 2 s and SAP > 200/s.

To investigate the variability in LVAD results, the velocity, RT, and SAP of two LVAD patients in Rossini et al. [21] (case 9596 and 5362 denoted as R1 and R2 for short) were directly compared to LVAD ML3 and ML7 (see Supplemental Table S2 available in the Supplemental Materials and Fig. 6). To resemble the flow patterns observed in R1, the imaging plane of ML3 was modified by switching the positions of the laser and camera, as illustrated in Fig. 2(b). This adjustment ensured that the mitral jet is positioned at the center of the LV chamber, while simultaneously obscuring the view of the LVOT.

Fig. 6.

Clinical data [21] (R) and specific ML matching conditions at three cardiac instants (MV, AV, LVAD (pump inlet)). (a) Velocity maps, RT, and shear activation potential maps for the midplane area after 10 cycles of integration. (b) Time-varying vortex circulation (diastole starts at t = 0, second filling phase occurs at 55%, and systole starts at 70% of the cardiac cycle).

3 Results

3.1 Patient-Specific Studies.

Table 1 summarized patient hemodynamics and the matched ML conditions of HF and LVAD studies. The EF, E/A ratio, EDV, and ESV compared well with those of each patient within the design limitations of the system. For LVAD studies, systemic resistances were adjusted to reproduce similar cardiac output and AV opening as recorded in patients under the same HM2 support levels.

Table 1.

Hemodynamics of end-stage heart failure and LVAD PC and matching ML conditions

Heart failure studies		EDV (mL)	ESV (mL)	EF (%)	CO (L/min)	HR (bpm)	E/A ratio
Study 1 normal/bioprosthesis MV	PC1	197	154	22	2.9	69	1
	ML3	221	180	20	2.8	68	1
	PC2	246	200	19	3.7	82	1.5
	ML4	225	180	20	3.5	78	1.5
Study 2 prosthesis/tilting-disk MV	PC3	209	171	18	3.5	92	0.9
	ML6	233	180	22	3.6	68	1
	PC4	203	172	16	2.5	82	2.6
	ML7	209	180	15	2.2	78	2.6

LVAD studies		EDV (mL)	ESV (mL)	CO (L/min)	HR (bpm)	E/A ratio	HM2 speed (rpm)	AV opening
Study 3 slightly dilated LVV	PC1	120	90	4.5–5	82	1	9800	No
	ML1	141	100	4.5	79	1	9800	No
Study 4 severely dilated LVV	PC5	210	186	4.5–5	70	2	9600	Minimal
	ML6	238	180	4.5	70	2	9600	Minimal

MV: mitral valve, EDV: end-diastolic volume, ESV: end-systolic volume, EF: ejection fraction, HR: heart rate, LVV: left ventricle volume, CO: cardiac output, AV: aortic valve.

3.1.1 Heart Failure Studies.

Figure 3 displayed the velocity fields at three cardiac events for four patients and the corresponding ML studies. In the presence of a bioprosthetic MV, vortex evolution in PC1 and ML3 followed the characteristic pattern observed in dilated cardiomyopathy patients [28]: an asymmetric vortex ring formed by the incoming mitral flow with a strong dominant clockwise (CW) vortex circulating within the LV during diastole (Fig. 3(a)). Normal vortical flow has been suggested to play a beneficial role in facilitating early LV filling by preserving fluid momentum [32,33] and minimizing energy dissipation rate [34]. It also helped to prevent intraventricular blood stagnation [35] by effectively washing out residual blood mass within a few heartbeats [36–38].

Fig. 3.

Intraventricular velocity fields of (a and b) heart failure and (c and d) LVAD patients (PC) and ML studies (MV: mitral valve, AV: aortic valve, LVAD: pump inlet, LVV: left ventricle volume)

Figure 4 illustrated the time-varying vortex circulation for all PC-ML pairs beginning with MV opening, followed by early and late diastolic filling and stopping at the end-systole event. As shown in Fig. 4(a), the CW circulation of ML3 peaked during E-wave filling, reduced during diastasis, and peaked again at the A wave; the counterclockwise (CCW) vortex appeared at the early E- and A-wave phases but decayed quickly. For PC1, the CCW vortices formed momentarily in early diastole, while the CW vortices persisted during diastole and dissipated gradually. Overall, the average CW circulation and kinetic energy (KE) for the ML were 46–87% higher than its counterpart (Table 2(a)). PC3 and ML6 velocity fields shown in Fig. 3(b) reflected a reverse vortex pattern with a dominant and persistent CCW vortex. The CCW vortices for this pair displayed 40–50% larger radii with 44–71% higher circulation strengths and KE than their counterparts as shown in Figs. 4(a) and 4(b).

Fig. 4.

Time-varying clockwise and counterclockwise vortex circulations of heart failure and LVAD patients (PC) and ML studies (diastole starts at t = 0, second filling phase occurs at 55%, and systole starts at 70% of the cardiac cycle) (LVV: left ventricle volume)

Table 2.

Heart failure and LVAD studies (a) vortex parameters, and (b) residence time and shear activation potential properties (average $\pm$ standard deviation) for patient cases (PC) and ML conditions

		Heart failure studies				LVAD studies
		Normal/bioprosthesis MV		Prosthesis/tilting-disk MV		Slightly/moderately dilated LVV		Severely dilated LVV
Vortex parameters		PC1 and PC2	ML #1 − 5	PC3 and PC4	ML #6 − 7	PC1	ML #1 − 4	PC5	ML #5 − 7
(a) Intraventricular CW and CCW vortex properties
Circulation (×10–3 m2/s)	CW	19.70 $\pm$2.29	21.07 $\pm$14.06	8.16 $\pm$3.64	5.92 $\pm$0.46	10.35	37.70 $\pm$15.60	7.83	29.80 $\pm$14.84
	CCW	3.09 $\pm$1.33	7.54 $\pm$3.27	19.51 $\pm$4.29	14.55 $\pm$5.56	4.94	13.32 $\pm$6.07	3.22	7.16 $\pm$2.16
Kinetic energy (mJ/m)	CW	20.04 $\pm$4.82	27.55 $\pm$28.71	5.71 $\pm$4.82	5.74 $\pm$2.20	9.15	101.4 $\pm$40.13	5.25	39.05 $\pm$28.34
	CCW	2.66 $\pm$2.31	14.85 $\pm$15.03	19.66 $\pm$11.83	10.27 $\pm$5.88	5.13	47.59 $\pm$23.21	2.00	16.39 $\pm$11.37
Radius (cm)	CW	0.89 $\pm$0.04	0.79 $\pm$0.22	0.61 $\pm$0.14	0.50 $\pm$0.09	0.67	1.00 $\pm$0.28	0.67	0.94 $\pm$0.18
	CCW	0.37 $\pm$0.04	0.51 $\pm$0.17	0.99 $\pm$0.06	1.02 $\pm$0.13	0.51	0.67 $\pm$0.14	0.51	0.51 $\pm$0.12
Symmetry CCW/CW		0.43 $\pm$0.04	0.80 $\pm$0.22	1.64 $\pm$0.45	2.48 $\pm$0.29	0.74	0.74 $\pm$0.38	0.82	0.57 $\pm$0.07
		Heart failure studies				LVAD studies
		Normal/bioprosthesis MV		Prosthesis/tilting-disk MV		Slightly/moderately dilated LVV		Severely dilated LVV
Parameters		PC1 and PC2	ML #1 − 5	PC3 and PC4	ML #6 − 7	PC1	ML #1 − 4	PC5	ML #5 − 7
(b) RT and SAP properties
RT (s)	Ave	2.19 $\pm$0.25	1.93 $\pm$0.39	3.54 $\pm$0.34	3.16 $\pm$0.55	1.98	1.09 $\pm$0.34	3.26	1.36 $\pm$0.73
	Max	5.17 $\pm$0.03	4.12 $\pm$0.79	5.42 $\pm$0.19	5.12 $\pm$0.19	4.56	3.10 $\pm$0.96	7.13	4.24 $\pm$1.85
SAP (s–1)	Ave	136.4 $\pm$17.91	291.3 $\pm$135.15	216.5 $\pm$180.0	363.0 $\pm$165.5	130.7	303.6 $\pm$203.4	70.79	171.5 $\pm$101.45
	Max	1168 $\pm$475.4	1129 $\pm$373.8	848.2 $\pm$756.9	1137 $\pm$506.2	573.1	1325 $\pm$581.2	311.4	1014 $\pm$292.6
Area of regions with RT > 2 s (%)		45.22 $\pm$7.51	45.88 $\pm$21.97	82.99 $\pm$14.41	73.02 $\pm$8.67	46.45	10.04 $\pm$8.29	68.87	15.79 $\pm$14.58
Area of regions with SAP > 200/s (%)		66.08 $\pm$15.27	68.17 $\pm$22.63	56.51 $\pm$46.86	80.03 $\pm$4.99	59.86	55.02 $\pm$32.01	11.51	39.75 $\pm$21.04

MV: mitral valve, LVV: left ventricle volume.

The RT and SAP maps and corresponding mean values displayed marked differences among the cases studied (Fig. 5). For PC1 and ML3 (Fig. 5(a)), low EF and dilated ventricle volume impaired wash-out, resulting in large regions of residual flow (i.e., RT > 2 cycles [36] localized in the LV center and along the LVOT. Shear entering the chamber with the filling jet was trapped and exposed to high RT, thus, causing shear-exposed fluid accumulation and regions of elevated SAP over 200/s. The reverse vortex pattern further worsened LV washout as reflected in the PC3 and ML6, leading to large regions of high RT (>4 s) covering more than 80% midplane area. As a result, retained blood within the LV experienced continuous exposure to shear, leading to the formation of regions characterized by a very high shear rate (>1000/s). Higher RT and residual flow areas of PC3 and ML6 were also reflected in Table 2(b).

Fig. 5.

Left ventricle maps of RT and SAP for the midplane area after ten cycles of integration are shown for heart failure and LVAD patients (PC) and ML studies. Space-averaged values were reported (RT in red and SAP in yellow) (MV: mitral valve, LVV: left ventricle volume).

3.1.2 Left Ventricular Assist Device Studies.

The two LVAD cases demonstrated significant variations in LV volumes (Table 1). Specifically, PC1 experienced a substantial 70% reduction in LV volume after 7 days of LVAD support, while the LV of PC5 remained considerably dilated even after two years with LVAD assistance. Similar to the HF condition, the incoming mitral flow developed into a vortex ring structure but was drawn toward the apex, producing stasis regions along the LVOT. For the slightly dilated LV comparison of PC1 and ML1, vortex circulation displayed a flattened-out pattern with attenuated E- and A-wave peaks (Figs. 3(c) and 4(c)). PC1 had stronger vortex circulation and KE than PC5 (Table 2(a)). Corresponding ML conditions produced a similar flow pattern but exhibited higher velocity values. The average of vortex circulation and KE for ML was notably higher than the corresponding PC for LVAD cases.

The RT map of PC1 showed a fluid stream directed from the MV toward the LVAD. Residual flow regions were observed at the LV apex and along the septal wall, resulting in elevated RT and SAP (Fig. 5). PC5 displayed worse fluid transport and wash-out capacity with larger RT and residual area than PC1 despite having a similar LVAD support level, but SAP and elevated SAP areas were small. ML1, with a smaller LV volume, exhibited lower RT and residual area than ML6, consistent with patient cases. The high residual flow regions, located along the LVOT and LV base due to none/minimal AV opening, were also associated with high SAP. Overall, the patterns and average values of RT for ML were quite comparable to PC, especially in the smaller LV volume condition, but the SAP was 1.8–5 times higher.

3.1.3 Published Clinical Left Ventricular Assist Device Cases.

To illustrate the capability of the ML in producing high-risk TE regions of LVAD patients and to explore the variability in the previous RT and SAP results, ML3 and ML7 were designed to match two specific published patient cases in Rossini et al. [21]. In order to visually align the observed flow patterns in R1, where the mitral jet appeared at the center of the LV chamber and obstructed view of the LVOT, the imaging plane of ML3 was adjusted accordingly (Fig. 2(b)). The velocity maps showed that incoming mitral flow was continuously drawn toward the LVAD inlet at the LV apex. Residual flow regions localized along the free walls, resulting in shear-exposed fluid accumulation and elevated SAP regions (Fig. 6(a)). While the residual flow region was similar in R1 and ML3, the ML condition had elevated shear appearing in front of the filling jet in early diastole and effectively washing out by the end of the cycle. The vortex trajectory of ML3 showed two distinct E-wave and A-wave peaks with higher circulation strength (Fig. 6(b)).

The R2 condition with dilated LV volume had images resembling the long-axis three-chamber view, showing incoming mitral flow close to the free wall and a partial view of the AV. R2 and ML7 presented a similar vortex trajectory and strength pattern: the CW vortex persisted and resided on the anterior half of the ventricle, and the CCW vortex decayed rapidly along the free wall. ML7 exhibited higher shear along the edges of the mitral jet and accumulation of residual flow and SAP regions along the LVOT.

3.2 Population Comparison

3.2.1 Heart Failure.

Dilated cardiomyopathy patients with HF symptoms generally exhibited depressed cardiac function and low cardiac output. Clinical studies reported an average AoP of 60 mmHg, EF of 20% or less, cardiac output of 3.0–3.5 L/min, and high SHE, matching the ranges of ML results (Fig. 7). In many patients, the heart progressively dilated to compensate for an increased demand for blood from other organs and tissues. The changes in LV geometry and function led to the formation of a larger, more persistent prograde vortex [28]. The inertia associated with this strong swirl was suspected to prevent efficient inflow–outflow redirection, leading to increased energy dissipation and SHE [25,35] and contributing to longer RT of blood [28,39,40]. These observations aligned with the results obtained from the ML studies (Fig. 8). Flow patterns with a dominant CW vortex and smaller than 1 CCW/CW vortex symmetry were common in patients with regular/bioprosthetic MV, while a reversed pattern of larger CCW vortex was noted in cases of mechanical MV (Figs. 8(a)–8(d)). The spatial average and maximum values of RT and SAP and areas of high RT and SAP demonstrated good agreement between clinical, PC, and ML studies (Figs. 8(f)–8(g)).

3.2.2 Left Ventricular Assist Device.

Left ventricular assist device implantation provided an alternative pathway for blood to move from the heart into the aorta: the inlet located at the LV apex and the outlet attached to the ascending aorta, bypassing the AV. Chronic unloading of the heart allowed reverse remodeling, in which the average LV volume decreased by 35%, LV mass by 20%, and EF increased within six months of LVAD support. Clinical studies reported that AoP increased by 20 mmHg, EF increased by 7–12%, VPI decreased slightly, and SHE decreased significantly by approximately 90% (Fig. 7). Larger vortex circulation and KE were noted in ML conditions, especially those with smaller LV volumes (Figs. 8(a) and 8(b)).

During LVAD support, RT, residual, and elevated SAP regions decreased markedly, while the average and maximum SAP remained unchanged (Figs. 8(f) and 8(g)). Lower RT and residual areas were noted in smaller LV volume cases. RT was 32–58% higher, and SAP was 57–70% lower in reported PC.

4 Discussion

4.1 Measurement of Velocity Fields in the Left Ventricle Midplane.

The velocity fields in both patients and the ML was acquired in the LV midplane. A pilot stereoscopic PIV study was conducted using two cameras, employing the same LV model, which resulted in the recording of three-dimensional velocity fields (Vx and Vy representing in-plane components, while Vz representing the out-of-plane velocity component) (Supplemental Fig. S3A available in the Supplemental Materials on the ASME Digital Collection). The average in-plane velocity (V), calculated as the root mean squared of Vx and Vy, was plotted against the average Vz for one cardiac cycle (Supplemental Fig. S3B available in the Supplemental Materials). During the peak E- and A-waves, Vz accounted for 12–30% of V, indicating that the in-plane velocity components were dominant and sufficient in capturing the flow patterns. These findings align with previous research, which demonstrated that midplane 2D imaging adequately captured the asymmetric mitral vortex pattern, as identified in studies utilizing MRI and echocardiography [29]. Additionally, a recent clinical study comparing 2D vector flow mapping and 4D flow MRI in patients suggested that in-plane velocity data provides adequate information to capture the primary features of intraventricular blood transport [41].

4.2 Summary of the Findings.

This study revealed that the intraventricular flow patterns were highly influenced by factors such as the type and orientation of valvular prostheses, ventricular geometry, and the level of LVAD support. In HF, specific types or orientations of prosthetic MV led to the formation of irregular vortex structures and impaired fluid washout, while other configurations had minimal impact on vortex development and fluid transport, consistent with previous findings [31,42]. In LVAD patients, regions of stasis were observed at the LV apex and along the LVOT due to the presence of the LVAD inlet and minimal aortic valve opening, which posed an increased risk of thromboembolism [19,35]. The combination of altered flow patterns and the presence of mechanical devices such as LVADs created unique flow dynamics and potential hemodynamic challenges that need to be carefully considered in the management of these patients.

Before LVAD implantation, obtaining the image of the LV midplane in patients was easily achieved using the standard long-axis three-chamber view. This facilitated the replication of a similar view in the ML setup. However, following LVAD implantation, the echo plane had to be shifted to minimize artifacts from the pump inlet at the apex. Consequently, the three-chamber view was compromised, making it challenging to replicate in the ML setup. Multiple attempts were made by moving the setup camera and laser to visually obtain similar flow patterns observed in each LVAD patient. However, this approach would be difficult to replicate for future studies or by other research groups. Therefore, the report primarily focused on the standard imaging plane shown in Fig. 2(a), and the discrepancies in the results were explained accordingly. Furthermore, to demonstrate the significant impact of a shift in the imaging plane on the observed velocity patterns, an example of a modified imaging plane (Fig. 2(b)) in ML3 was provided.

The accurate characterization of in vivo LVAD flow presented challenges due to the requirement for a precise three-chamber midplane view, which was often constrained by the angle of the transducer and the presence of pump artifacts [29]. The hindered view of the LVAD inflow cannula resulted in the overestimation of RT in some LVAD patients, such as PC1 and PC5, who exhibited sizable residual flow regions around the LV apex. Conversely, in cases where a clear three-chamber view was obtained, such as R2 and ML7, the LVAD was shown to significantly improve fluid transport, resulting in minimal RT and residual flow. Furthermore, the offset view of the MV in some LVAD patients only captured a portion of the structure and trajectory of the incoming jet, leading to an underestimation of the strength and KE of the jet. This, in turn, could result in miscalculations of shear and SAP since the highest shear typically appeared at the front of the filling jet, associated with the vortex ring development [31]. In-vivo studies commonly faced limitations in measuring low velocities and shear, which could explain the discrepancy observed in the results of PC5, which exhibited high RT regions with very low SAP. This also clarifies why clinical cases often reported lower vortex circulation, KE, and SAP values [21].

4.3 Significance.

Nonphysiological blood flow patterns within the LV, chronically activating platelets and causing regional stasis and high shear rate, are linked to TE and stroke [43–46]. Regional stasis accelerates platelet aggregation by allowing the accumulation of activated platelets and thrombin [45,46], while a high shear rate activates platelets accumulatively with repeated cycles [47,48]. The prediction of TE and stroke are notoriously difficult due to the sensitivity to patient-specific conditions [49–51]. The quantification of fluid dynamics and indices can provide additional insight into transport efficiency and highlight TE-risk regions, for example, SAP and blood RT evaluated the intensity of the shear stress and the cumulative time of exposure contributing to platelet activation [31,52], or the combination of RT and vortex KE can accurately distinguish patients with LV thrombus [53,54]. While the small patient sample and experimental nature of this approach may limit the ability to provide optimal values of these indices, the metrics obtained through the ML could still be valuable in providing insights into the complex intraventricular flow of HF and LVAD patients. These metrics can aid in the development of optimal management strategies for individual patients, allowing for personalized monitoring of pump support and the balance between thromboembolic and hemolytic adverse outcomes.

The HeartMate3 LVAD, which incorporates magnetic impellers and an artificial pulse, still reports relatively high incidence rates of bleeding (10%) and stroke (20%) within the first year of implantation [55]. These adverse events can be attributed to pump-related platelet dysfunction, which is associated with exposure to high shear stress and altered intraventricular flow. The combination of these factors promotes bleeding and creates a surface conducive to thrombin generation [43,56]. Furthermore, the risk of apical thrombosis, particularly in patients with reduced LV size [19,20], indicates that TE remains a prevalent concern even in patients with favorable ventricular remodeling. The flow dynamics in LVAD-supported hearts are complex and multifactorial, dependent on various factors such as the insertion angle [19,57] and depth [58] of the inflow cannula, outflow graft configuration, pump speed variation, the patient's etiology and response [59,60], and more. Given that the TE risk evolves over the course of LVAD support, ML can serve as a crucial tool for assessing variations in flow indices associated with different inflow cannula configurations, support levels, ventricular sizes, or the synchronizing effects of the artificial pulse and heart rate.

The rate of cardiac recovery of LVAD patients is often challenging to predict and optimize [61,62]. Different centers use parameters obtained from echocardiography and right heart catheterization to assess cardiac improvement, but the lack of standardized acceptance criteria leads to highly varied protocols and management strategies [9–11]. The ML could be utilized to investigate the relationships between flow indices and patient outcomes by reproducing fluid dynamics of a specific patient case under a wide range of LVAD operating conditions and exploring any indications of LV unloading and fluid transport improvement. The results informed fluid transport efficiency and high TE-risk regions aiding in the development of management strategies to optimize heart–implant interaction.

This study demonstrates the ability of ML to simulate the complex intraventricular flow in patients with HF and left ventricular assist devices LVADs. The ML system enables the acquisition of high-resolution images at small time steps to directly measure the in-plane 2D velocity field, in contrast with in vivo echocardiographic acquisitions, which can be limited by resolution, contrast agent behavior, and one-dimensional velocity data in the case of color-Doppler [63]. The ML approach offers a way to obtain detailed flow measurements without the need for invasive procedures or safety concerns, making it a simpler and more cost-effective alternative to traditional patient studies. It serves as a suitable platform for studying the potential impact of treatment strategy or can be utilized as a diagnostic tool to identify problematic regions responsible for adverse events in HF and LVAD patients. By characterizing flow indices related to TE, such as RT and SAP, ML can provide evidence of the hemocompatibility of cardiac implants, which is crucial for informing clinical decision-making and regulatory evaluations of cardiac implants. Lastly, the ML results obtained in this study can be applied in the development of computational models or used as a validation tool for computational simulations. By comparing the ML results with computational simulations, the accuracy and reliability of the models can be assessed and improved.

5 Limitations

All efforts were made to design the benchtop studies to match specific patient conditions within the limitations of the ML system. The 2D intraventricular velocity fields were reconstructed from echo color Doppler using a previously described and validated method [29], providing a good approximation of flow patterns in the apical long-axis view. Despite the inherent limitations of characterizing the complex three-dimensional features of LV flow transport using a planar approximation, recent clinical data suggested that the main LV transport boundaries obtained from in-plane color-Doppler velocities compare well with those obtained from 4D flow MRI [41]. The method has yet to be quantitatively validated in LVAD patients due to the challenges of performing MRI [21]. Replicating imaging views of LVAD patients in the ML setup proved to be challenging due to ultrasound artifacts and patient motion. As a result, this study focused primarily on reporting the results obtained from the standard three-chamber views. The discrepancies in the results were thoroughly explained to provide a comprehensive understanding of the findings.

A small sample size was appropriate since this study aimed to demonstrate MLs capability in modeling the intraventricular flow of specific HF and LVAD patients. All LVAD studies were conducted with inflow cannulas aligned parallel to the septal wall with no protrusion since no information regarding cannula positions and angles was noted in patient cases, and patient-specific chamber shapes were not matched in the silicone models. This and deviations in imaging plane could explain the differences encountered between some clinical and ML cases, e.g., PC5 versus ML6. Future studies with volumetric three-dimensional PIV on patient-specific geometries will address these limitations.

6 Conclusion

This study demonstrated that the ML produced comparable hemodynamics and fluid dynamics to specific patient cases as well as published studies, thus, validating the reliability and clinical relevance of the system. The customizable benchtop model allowed physical testing of cardiovascular implants in a controlled environment that simulated a range of usage conditions. Additionally, the ML system enabled detailed measurement of intraventricular flow at lower costs and with no associated risks compared to patient studies. This had the potential to advance the understanding of intraventricular flow in HF and LVAD patients, inform treatment strategies and clinical evaluation, and aid in the development of computational models.

Acknowledgment

The authors want to acknowledge the following individual and groups who helped with the patient data-collecting process:

• Robert Adamson, Suzanne Chillcott, Brian Jaski, Jennifer Key, Suzan Lerum, Rebecca Price, and Marcia Stahovich from Sharp Memorial Hospital, San Diego.

Funding Data

This project was funded in part by the FDA Critical Path Initiative and by appointment to the Research Participation Program at the Division of Biomedical Physics, Office of Science and Engineering Laboratories, Center for Devices and Radiological Health, U.S. Food and Drug Administration, administered by the Oak Ridge Institute for Science, and Education through an interagency agreement between the U.S. Department of Energy and FDA. Funding from the U.S. NHLBI (NCAI-UCCAI-2017-06-6, 1 R01 HL160024-01, 1R01HL158667-01A1) is also acknowledged.

Conflict of Interest

The authors have no conflicts to disclose.

Data Availability Statement

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

Disclaimer

The mention of commercial products, their sources, or their use in connection with material reported herein is not to be construed as either an actual or implied endorsement of such products by the Department of Health and Human Services.

Human Studies/Informed Consent

All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation at San Diego State University and Sharp and Sharp Memorial Hospital (United States) and with the Helsinki Declaration of 1975, as revised in 2000 (5). Informed consent was obtained from all patients included in the study.

Supplementary Material

Supplementary Material

Supplementary Figures

Appendix A

Velocity Fields Measurement in the Left Ventricle Midplane

Under the assumption of planar flow, this study focused on the large flow structure in the apical long-axis plane of the LV. The selection of this view was informed by prior demonstrations highlighting its ability to provide clinically relevant parameters. Notably, Thompson et al. [64] verified that accurate pressure differences can be calculated using 2D acquisition, mainly because the influence of momentum fluxes perpendicular to this plane is negligible. Furthermore, this particular view enables the observation of LV vortex development and trajectory, with exhibited favorable agreement with flow patterns documented in previous clinical studies utilizing phase contrast MRI [32,36] and various echocardiography modalities [28,29,65].

Patient Studies.

The 2D + t vector velocity maps were reconstructed from Color-Doppler and B-mode in the apical long-axis three-chamber view (passing through the LV apex, the center of the mitral, and the aortic valves) [21,29]. This view can be readily obtained in cases of heart failure, enabling the replication of a similar view in the mock loop study as shown in Fig. 9.

Fig. 9.

Apical long-axis three-chamber view (left) patient B-mode image (right) camera view of the ventricle bag

Mock Loop Studies.

In the mock loop study, a laser light sheet illuminated particles at the LV center, cutting through the mitral and aortic valves, to achieve an apical long-axis view that closely resembled the view obtained in patient studies [14].

Appendix B

Transvalvular pressure (mmHg)

$$\text{TVP}\left(t\right)=\text{AoP}\left(t\right)\hspace{0.17em}–\hspace{0.17em}\text{LVP}\left(t\right)\hspace{0.17em}$$ (B1)

where AoP is the aortic root pressure and LVP is the left ventricle pressure.

Flow through the aortic valve (Q_AV) (L/min)

$$Q_{\text{AV}}\left(t\right)=Q_{\text{Ao}}\left(t\right)\hspace{0.17em}–\hspace{0.17em}{Q}_{\text{LVAD}}(t)$$ (B2)

where Q_Ao is the total aortic flow and Q_LVAD is the LVAD flow.

Energy equivalent pressure (ergs/cm³)

$$\mathrm{EEP}\hspace{0.17em}=\hspace{0.17em}\frac{\int (\mathrm{AoP}\hspace{0.17em}\times \hspace{0.17em}{Q}_{\mathrm{Ao}})}{\int Q_{\mathrm{Ao}}}$$ (B3)

Surplus hemodynamic energy (mmHg)

$$\text{SHE}=1332\times (\text{EEP}\hspace{0.17em}–\hspace{0.17em}{\text{AoP}}_{\text{ave}})$$ (B4)

Velocity pulsatility index

$$\mathrm{VPI}\hspace{0.17em}=\hspace{0.17em}\frac{V_{\max }-V_{\min }}{V_{\mathrm{ave}}}$$ (B5)

where V is the resultant velocity.

Appendix C

Blood Residence Time and Shear Rate

Blood RT was calculated using the modified advection equation

$${\partial }_t{T}_R+{\nabla }_{2D}·\left({\mathbf{v}}_{\mathrm{PIV}}{T}_R\right)=1+\left({\nabla }_{2D}·{\mathbf{v}}_{\mathrm{PIV}}\right)T_R$$ (C1)

where v_PIV is velocity field obtained from the PIV measurements, ${\nabla }_{2D}·\left(\right)$ stands for the in-plane divergence operator, e.g., ${\nabla }_{2D}·(u,v)={\partial }_{x}u+{\partial }_{y}v$. The correction on the right-hand side of Eq. (C1) ensures that the $T_R$ value transported by the flow does not accumulate or spread in a nonphysical manner in local flow “compression” or “expansion” regions, where the in-plane divergence of the PIV velocity field is not negligible.

The potential of SAP platelet $={\mathrm{\Sigma }}^{\frac{1}{\alpha -1}}$, where Σ is governed by an advection equation

$${\partial }_t\mathrm{\Sigma }+\nabla ·\left(\mathit{v}\mathrm{\Sigma }\right)={\dot{\gamma }(x,y,t)}^{\alpha }$$ (C2)

forced by the shear rate $\dot{\gamma }=\sqrt{\mathbf{t}\mathbf{r}\left[{S(x,y,t)}^2\right]}$, with $S$ being the symmetric part of the velocity gradient tensor.

Both equations were completed by homogeneous Dirichlet initial and boundary conditions at the mitral inlet region. The equations were solved numerically using a third-order WENO scheme to discretize spatial fluxes and a third-order Runge–Kutta scheme to integrate in time, as described in our previous publication [54]. The power exponent $\alpha \approx 2$, which reflects the relative importance of shear intensity versus shear exposure time, was obtained by fitting the model to Hellums' compendium of platelet activation experiments (see Ref. [31] for details). This fit also suggests that platelet activation may occur for $\mathrm{SAP}\hspace{0.17em}\sim 3\times {10}^4\hspace{0.17em}{\mathrm{s}}^{-1}$.

Nomenclature

AoP =

aortic root pressure

AV =

aortic valve

CCW =

counter-clockwise

CW =

clockwise

EDV =

end-diastolic volume

EEP =

energy equivalent pressure

EF =

ejection fraction

ESV =

end-systolic volume

HF =

heart failure

HM2 =

HeartMate II

HR =

heart rate

KE =

kinetic energy

LV =

left ventricle

LVAD =

left ventricular assist device

LVOT =

left ventricle outflow tract

LVP =

left ventricle pressure

ML =

mock loop

MV =

mitral valve

PC =

patient case

PIV =

particle image velocimetry

Q_Ao =

total aortic flow

Q_AV =

flow through the aortic valve

Q_LVAD =

LVAD flow

RT =

residence time

SAP =

shear-activated potential

SHE =

surplus hemodynamic energy

TAVR =

transcatheter aortic valve replacement

TE =

thromboembolic

TMVR =

transcatheter mitral valve replacement

TVP =

transvalvular pressure

VPI =

velocity pulsatility index