Aortic stenosis (AS) is the narrowing of the aortic valve orifice due to reduced mobility of the valve leaflets. It arises as a result of inflammatory processes akin to those driving atherosclerosis, whereby endothelial damage due to mechanical stress and other biological processes induces fibrosis, thickening, and calcification of the valve leaflets (1,2). Although AS affects the elderly population disproportionately, its onset and progression can be dramatically accelerated by existing underlying congenital defects -such as bicuspid aortic valve (BAV) disease -which occurs when two aortic valve leaflets are fused together (3). Hemodynamically, the narrowing of the aortic valve orifice gradually leads to elevated transaortic pressure gradients (4,5). The increased afterload (or pressure overload) results in higher left ventricular (LV) systolic pressures and a reduction in the volume ejected at each heartbeat (stroke volume, SV) leading to drops in cardiac output and the onset of symptoms such as angina and exertional syncope (6,7). In two thirds of AS patients, pressure overload drives LV remodeling, resulting in loss of LV compliance and eventually in diastolic and/or systolic dysfunction (8)(9)(10). This complication of AS causes higher mortality and rehospitalization rates after aortic valve replacement, and may eventually lead to heart failure (11)(12)(13).
AS currently affects approximately 1.5 million people in the US and is associated with a five-year survival rate of 20% from the onset of symptoms, if untreated (14,15). To date, there are no effective pharmacological treatments for AS, and it is estimated that 80,000-85,000 aortic valve replacement procedures are performed every year in the USA (16,17). The prosthetic aortic valve market (valued at $6.9 billion in 2021) is rapidly expanding and is projected to reach $19.7 billion by 2031 (18). Next-generation prosthetic aortic valves are currently under development, aiming to enhance hemodynamic performance, durability, and long-term safety (19). Unfortunately, the majority of hydrodynamic models currently used for functional evaluation of prosthetic valves rely on rigid, idealized components and fail to recreate patient-specific anatomies and hemodynamics (20). The limitations of current models emphasize the need for high-fidelity patient-specific platforms that meet the increasingly rigorous International Standard guidelines for the evaluation of cardiovascular implants (21,22).
Recently, hydrodynamic platforms that integrate patient-specific aortic replicas have been developed for studies of congenital heart disease (23), aortic dissection (24,25), and AS (26,27). Kovarovic et al. proposed a patient-specific model that integrates molded replicas of patientspecific aortic root and calcific valve geometries obtained from computed tomography (CT) data with a rigid pulse duplicator system (27). Similarly, Haghiashtiani et al. combined image-guided aortic root models with a rigid pulsatile pumping system (28). In their work, they leveraged the multi-material 3D printing (MM3DP) approach first demonstrated by Hosny et al. (29) to manufacture anatomical models of calcified valves, with the advantage of enhanced prototype versatility compared to molding techniques. Nevertheless, the performance of these hydrodynamic models is largely dependent upon the availability of biomechanically relevant 3D printable polymers and the fidelity of the leaflet reconstructions. As such, although these MM3DP heart valve models can be valuable in understanding the effects of calcifications on leaflet flexibility for TAVR valve sizing applications (29), the differences between the intrinsic mechanical properties of the flexible materials that are employed in commercial multi-material 3D-printers, and those of the native heart valves, can greatly compromise their ability to reliably recapitulate patient-specific hemodynamics.
Due to these potentially compounding errors, and the inherent lack of real-time tunability of the as-fabricated 3D-printed valve geometries to compensate for these differences, several design-manufacturing-testing iterations would be required to obtain a high-fidelity hydrodynamic system, hindering the translatability of these models to clinical or industrial settings, which can critically rely on rapid turnaround times. Furthermore, by relying on traditional pumps or pulse duplicators, these systems are unable to model diastolic dysfunction (DD) caused by LV remodeling processes secondary to AS, which is observed in most of these patients, severely limiting the clinical relevance of these models.
Leveraging our previous work, in which we demonstrated the ability of a non-patientspecific aortic sleeve to recreate the hemodynamics of AS in a porcine model (30), we propose a soft robotics-enabled 3D-printed anatomical hydrodynamic system that is capable of recreating AS and congenital defects in a patient-specific fashion. In addition, using an analogous design workflow, we develop a patient-specific soft robotic LV sleeve that allows us to mimic changes in cardiac function observed in these patients, simulating longitudinal disease progression. We demonstrate that our soft robotic aortic sleeve can be controlled to recreate patient-specific hemodynamics of AS more accurately than current methods. Moreover, we showcase the ability of our model to mimic DD resulting from loss of LV compliance, and to predict hemodynamic changes associated with treatment. This soft robotics-enabled model of both aortic and LV hemodynamics of relevance in AS demonstrates the advantage of increased tunability over more traditional approaches, paving the way towards high-fidelity testing platforms for the evaluation of cardiac devices, personalized device selection, and outcome prediction.
We retrospectively selected 15 patients with a diagnosis of AS who had undergone transthoracic or transesophageal echocardiography, or a combination thereof, as well as CT imaging for hemodynamic and anatomic evaluation. Our patient cohort had a broad spectrum of functional and structural characteristics relevant to AS, as summarized in Table 1. In this study population of 15 patients (6 female; age: 78 13 years; BSA range: 1.67-2.23 m 2 ), the aortic annular diameter ranged from 22 to 32 mm. Four of the selected patients had a bicuspid aortic valve anatomy, 6 were diagnosed with severe AS, 14 showed some evidence of aortic regurgitation, 8 displayed thickening of the LV wall, and 4 had a left ventricular ejection fraction (LVEF) lower than 40% (31). Further details can be found in table S1.
Figure 1 summarizes the workflow and overall architecture of our model developed based on patient hemodynamics and imaging. Clinical data, including CT, echography, and catheterization data, were obtained from AS patients (Fig. 1A). We first segmented the CT images to create 3D anatomic models of patients' LV and aortas (Fig. 1B,2A), which we 3D-printed with a soft elastomeric photopolymer resin (Fig. 1C, fig. S1). We then used CT images to design patientspecific soft robotic LV and aortic sleeves (Fig. 1C). When combined with the patient specific 3D models and our in vitro hydrodynamic model (Fig. 1D, fig. S2), the soft robotic LV sleeve provided the contractile force necessary to generate patient-specific systolic pressure and flows as well as modulation of LV compliance seen in the spectrum of pressure overload, whereas the soft robotic aortic sleeve provided morphologic mimicry and recapitulation of patient specific hemodynamics (Movie S1). Ultimately, this personalized model allowed for testing of hemodynamic changes induced by transcatheter aortic valve replacement (TAVR) under different conditions (Fig. 1D). An overview of the workflow, main findings, and applications of our model is described in Movie 1.
The platform designed and developed in this work is ultimately intended for high-fidelity testing and evaluation of medical devices for AS, procedural planning and outcome prediction, as well as product selection personalized to each patient's anatomy, hemodynamics, and disease state. As such, we demonstrated its potential utility by implanting a transcatheter aortic valve replacement (TAVR) and conducted retrospective clinical validation in a subset of our study cohort.
The soft robotic aortic sleeve enabled us to recreate the valve lesion morphology of each individual patient with high fidelity. Figure 2A shows a comparison, for each patient, of the aortic valve cine CT images and of the aortic cross-sections of our model under actuation of the aortic sleeve, captured using an endoscopic camera in the system. These images demonstrate that our sleeve can qualitatively recreate a range of patient-specific anatomies, including those of degenerative AS and BAV, with high accuracy. We then superimposed the CT and aortic sleeve images of the orifice area for each patient and calculated the Sørensen-Dice similarity coefficient (DSC) for a quantitative comparison (33). An example of superimposed CT and aortic sleeve images is illustrated in Fig. 2B, where the white area denotes the region of perfect overlap between the two images, whereas the cyan or red areas denote regions of image mismatch. All images used for calculation of DSC score can be found in fig. S3. The average DSC across the 15 patients modeled was 0.88 0.05 (Fig. 2C), demonstrating the overlap between native valve morphology seen on CT and our personalized aortic sleeve.
By recreating patient-specific aortic valve anatomies, our soft robotic aortic sleeve was able to induce aortic valve hemodynamics as measured via echography. We validated our model by comparing the clinical parameters measured in these patients for AS evaluation with those obtained by our system. We also compared the performance of our model to one containing valves fabricated using the MM3DP approach by Hosny et al. (29), whereby CT images were used to generate 3D-printed patient-specific aortic sinus and leaflet anatomies in tandem with rigid calcium-like patterns corresponding to the patients' mineralized nodules. To document the performance of the MM3DP valves in a similar context, we merged them with the patient-specific model of the LV and ascending aorta to perform functional hydrodynamic studies (fig. S4). Fig. 3A shows images of the LV and aortic sleeves of the soft robotic model, highlighting their main components, including the inflatable pockets and inelastic fabric. It also shows the MM3DP valves with calcium-like nodules and their corresponding CT data for the subset of patients used for hydrodynamic validation (Patients 7-11).
Our model accurately recapitulated the critical hemodynamic parameters of AS with high accuracy for each patient. For our analysis, we considered the mean (Pmean; Fig. 3B) and maximum transaortic pressure gradients (Pmax; Fig. 3C), the peak aortic flow velocity (vmax; Fig. 3D) and the stroke volume (SV; Fig. 3E). The soft robotic model more closely matched the clinical targets, with a compounded average absolute deviation of 7.7 % for all metrics, compared to the MM3DP approach (13.9 %). Specifically, we found variations of Pmean = 6. S3). By tuning the actuation pressures of the LV sleeve during diastole, LV compliance and diastolic function can be modulated (Fig. 4C). This approach allows us to recreate elevations in the LV enddiastolic pressure (LVEDP; Fig. 4D) associated with concentric remodeling secondary to AS.
We simulated the hemodynamics of four patients with reported LV catheterization data (Patients 10-13, see table S1) to demonstrate the ability of the soft robotic system to recreate cardiac hemodynamics of patients with different degrees of LV remodeling. Measurements of the LV and aortic hemodynamics, including LVEDP (Fig. 4G), systolic LVP (LVPS; Fig. 4H), LV ejection fraction (LVEF; Fig. 4I), and systolic (AoPS; Fig. 4J), diastolic (AoPD; Fig. 4K), and mean (AoPm; Fig. 4L) aortic pressures, and comparison with clinical data further corroborate the ability of our soft robotic system to create a high-fidelity model of each patients' LV and aortic hemodynamics.
For each of these metrics, we computed the absolute deviation from the corresponding clinical values, obtaining: 1.2 0.9 % (LVEDP), 0.6
Overall, this approach showcases the ability of our platform to recreate LV hemodynamics and changes in LV compliance associated with remodeling processes and DD secondary to AS with high fidelity.
We demonstrated use of our model for the evaluation of post-TAVR hemodynamics in a subset of patients via implantation of the self-expanding Evolut R (Medtronic, Minneapolis, MN) and the balloon-expandable SAPIEN 3 (Edwards Lifesciences, Irvine, CA) prostheses. S2.
Finally, we investigated the degree of paravalvular leak and regurgitation in a group of patients receiving an undersized valve versus recipients of an appropriately sized prosthesis. Fig. 5I illustrates representative color flow mapping Doppler images for the two groups, highlighting more substantial paravalvular leak associated with undersized implants. Analogously, calculation of the aortic regurgitation index (ARI) through catheterization shows less optimal TAVR performance for the undersized group than for the appropriately sized valve (Fig. 5J; ARI = 22.0 vs. 45.4 ; p = 9.810 -5 ). These findings are consistent with the literature associating lower values of ARI with higher mortality in patients with aortic valve disease (35,36).
In this work, we present the development of a patient-specific hydrodynamic model driven by tunable soft robotic tools, of relevance in AS and LV remodeling. We demonstrated the ability of this model to recreate patient-specific anatomies of degenerative stenotic aortic valves and of congenital BAV disease (Fig. 2A). Biomechanical mimicry of the AS lesion is paramount to accurately recreate local flow hemodynamics in a high-fidelity platform. In this work, calculation of the DSC between the model and CT images demonstrated that the patient-specific aortic sleeve can achieve enhanced mimicry (DSC = 0.88 ) compared to commercial aortic banding techniques (DSC = 0.47) and non-specific aortic sleeve (DSC = 0.72) (30). Furthermore, our platform recapitulates the hemodynamics of AS (Fig. 3 B-E) with greater accuracy compared to other systems, with a mean absolute deviation equal to 7.7 % (n = 15), which is lower than that achieved via the use of MM3DP valves produced on commercial multi-material 3D-printers (13.9 %, n = 5; Fig. 3 B-E).
In the context of developing a clinically relevant hydrodynamic testing platform, the incorporation of molded or MM3DP valves depends more heavily on the resolution of the patients' CT images and on the quality of image segmentation when compared to our more adaptable soft robotic platform. In these systems (molded and MM3DP), any mismatch between the patients' anatomies and the aortic root replicas can only be improved by manual editing of the digital valvular geometries. This time-consuming iterative process compromises the utility of these hydrodynamic models in a clinical setting, where TAVR procedures are often performed within a day from initial CT imaging. Conversely, in our soft robotic model, actuation can be tuned in real-time to obtain high-fidelity mimicry of the patients' hemodynamics, by modulating the aortic root diameter (Fig. 3F). The controllability of molding or MM3DP-based hydrodynamic models is further limited by differences between the mechanical properties of the manufactured valve material and those of the native leaflet tissue. Although the investigation of valve kinetics was beyond the scope of this work, the dynamics of the aortic sleeve in this model could be controlled to mimic the motion of stenotic valve leaflets, as previously demonstrated (30). Unlike other models, our platform was also able to recreate the anatomies of congenital valvular defects, such as BAV (Patients 12-15 in Fig. 2A and Fig. 3B-E), which is a primary driver of AS in the younger population (3). The MM3DP approach developed by Hosny et al. (29) relies on an algorithm that computes an idealized geometry of a tricuspid aortic valve from CT image landmarks, and is yet to be broadened to recreate BAV anatomies or other congenital aortic valve defects.
Integration of a controllable soft robotic LV sleeve (Fig. 3A, Fig. 4A-B) is a critical step towards the development of clinically relevant hydrodynamic models of AS and other cardiovascular conditions. Traditional hydrodynamic models leverage displacement-control pumps, which eject a prescribed amount of volume into the circulation. Firstly, this is not representative of cardiac physiology. Secondly, it makes it challenging to recreate the hemodynamics of conditions where the afterload is altered. In the context of AS, these systems would not be able to capture drops in SV associated with a higher afterload (Fig. 3E), leading to overestimated values of flow velocities predicted by the continuity equation (Fig. 3D, G) (37,38). Conversely, by mimicking the biomechanics of the native heart in both physiology and disease, we overcome these limitations and recreate both pressure and flow in a more clinically relevant manner (Fig. 4) (39)(40)(41)(42).
The personalized LV sleeve design enables us to modulate LV compliance to simulate the hemodynamic effects of cardiac remodeling secondary to AS in a patient-specific manner (Fig. 4C). Particularly, we were able to simulate alterations in LV filling pressures and DD in patients with various degrees of remodeling due to pressure overload (Fig. 4D-M). This model of modulation of ventricular compliance can be used to represent different states of disease progression, which has not been shown previously. As thickening and subsequent decrease of LV compliance are estimated to occur in more than two-thirds of patients with AS (8), it is paramount that preclinical models of AS can correctly recapitulate changes in LV diastolic biomechanics and hemodynamics associated with pressure overload.
Our model was shown to predict hemodynamic changes associated with treatment with established TAVR prostheses. We used clinical transaortic pressure gradient data in a subset of patients to retrospectively validate our system as a platform for hemodynamic outcome prediction (Fig. 5 D-E). Further, we measured changes in LVP and AoP (Fig. 5B), LV PV loops (Fig. 5C), EOA (Fig. 5F), vmax (Fig. 5G), and SV (Fig. 5H), resulting from simulating intervention in our model. We found that changes measured in this study are consistent with the literature of large-population studies of TAVR outcome (43)(44)(45).
This research thus has the potential to enable medical device companies to test and optimize their devices reliably across a spectrum of clinical cases, broadening the usability of devices to those patients for whom current TAVR designs are not suitable, beneficial, or safe to use. In the clinic, it would provide physicians with a platform for device selection, and procedural planning and outcome prediction. Furthermore, it may provide clinicians with a tool to improve techniques for TAVR delivery to minimize risk of coronary obstruction or valve migration and optimize device selection for patients with complex anatomies or sizes that fall between recommended use ranges for a given device. Finally, it may help identify subgroups for which TAVR could be the beneficial and performed safely within patient populations -such as BAV patients -that are currently ineligible for TAVR and have been traditionally excluded from major trials comparing surgical versus transcatheter interventions (46,47).
The continuous development of new techniques for the treatment of AS accentuates the need for high-fidelity systems that can be utilized as training platforms (48,49). In vitro models developed to date lack the anatomical and functional accuracy required to make them a suitable alternative to in vivo models, which remain difficult to realize in large numbers, due to ethical concerns and elevated costs (50,51). The model proposed in this work has the potential to contribute to advances in TAVR interventional training. Firstly, our patient-specific approach allows for enhanced anatomical accuracy and enables the recapitulation of morphologies of a variety of AS lesions and congenital defects of the aortic valve that other systems are unable to mimic. Furthermore, our model provides the advantage of recreating aortic hemodynamics and secondary LV dysfunction with elevated controllability, potentially allowing to simulate changes in cardiac function as they may occur during intervention (52). The usability of this model as a training platform can be enhanced further by augmenting anatomical fidelity through integration of 3D-printed elements modeling vascular access points for TAVR procedures.
Despite the many advantages offered by our hydrodynamic platform, there are three main limitations that should be considered. Firstly, the position of the aortic sleeve on the ascending aorta may affect the compliance of the 3D-printed model. Particularly, the aortic sleeve causes a drop in distension of the aortic segment corresponding to the position of the sleeve. In addition, any aortic segments that are proximal to the aortic sleeve will experience LV (rather than aortic) pressures, which are elevated in AS. Together, these factors may lead to local differences in aortic distension compared to those measured physiologically. Secondly, although our model was shown to capture LV pressures and flows (Fig. 4C) with high accuracy compared to other hydrodynamic models (53)(54)(55), the isovolumic regions of LV PV loops display a non-zero net flow towards or outside of the LV. This observation is a result of the position of the aortic flow probe, which could only be placed distal to the aortic valve plane due to the irregular geometry of the 3D-printed aortic anatomies. Finally, the dataset used in this study did not provide indications of the patients' systolic-diastolic ratio, which limited our ability to modulate the exact dynamics of ventricular contraction and may have influenced hemodynamic measurements.
Improvements to this study would involve automation of the sleeve design process and 3D printing techniques to further reduce the manufacturing time of the patients' replicas to maximize clinical utility. Furthermore, obtaining access to a broader spectrum of transcatheter valve prostheses and a larger clinical database would permit a prospective validation of our TAVR prediction study. Finally, the utilization of alternative 3D printing materials with enhanced optical properties would enable patient-specific 4D flow visualization and turbulence formation through particle image velocimetry (PIV) studies.
Eventually, use of this soft robotics-driven model can be broadened to simulate the hemodynamics of other valvular heart diseases and conditions that affect LV function, including restrictive cardiomyopathies and heart failure -both with reduced and preserved ejection fraction. We are hopeful that this model can pave the way towards high-fidelity patient-specific tunable models with a translational potential poised to improve clinical care of the millions of people worldwide affected by AS and other cardiovascular conditions.
Anonymized CT and echocardiographic clinical data from fifteen patients with AS were obtained retrospectively via Institutional Review Boards (IRB) approval at the Massachusetts General Hospital. Using echocardiographic measurements of the left ventricular (LV) diameter during systole and diastole, we screened for chest CT images acquired during diastole. This approach allowed us to design the geometry of our patient-specific soft robotic LV sleeve in its pre-actuation state. Conversely, images of the aortic valve were taken during peak systole from the patients' aortic valve cine images, thus enabling the development of patient-specific soft robotic aortic sleeves that could recreate the morphology of the stenotic leaflets during systole. Each patient's 3D-printed anatomical model was integrated with the LV and aortic sleeves into a hydrodynamic flow loop, with added pressure and flow sensors, and an endoscopic camera for hemodynamic evaluation. Hemodynamic parameters relevant in AS were measured using pressure-volume catheters, flow probes, an endoscopic camera, and continuous wave and color flow mapping Doppler, as described below. Results were compared with the patients' clinical data, as well as established methods based on MM3DP. Finally, hemodynamic changes due to implantation of Evolut R (Medtronic, Minneapolis, MN) and SAPIEN 3 (Edwards Lifesciences, Irvine, CA) valves were evaluated.
CT and aortic valve cine images (slice thickness t = 1.0-1.5 mm, x-ray depth D = 10 mm) were segmented on Mimics Research software (v.21.0.0.406, Materialise, NV) by thresholding, multiple-slice editing, and auto-interpolation. The geometrical axes of the cine images were reoriented to ensure visualization of the valve leaflets orthogonal to the direction of flow.
For each patient, the 3D anatomy of the LV and aorta (ascending through descending segment) was exported from the CT images as a shell stereolithography (STL) file with wall thickness equal to 1.3 mm. A thickness value lower than that of the human aorta was chosen to compensate for any mismatch in mechanical properties between the 3D printing photopolymer resin (Elastic 50A; Formlabs, Somerville, MA) and those of the native aorta (fig. S1). Since the mechanics of the LV are defined by actuation pressures of the LV sleeve, a thickness value of 1.3mm for the 3D-printed LV wall was chosen for ease of manufacturing. Each STL file was then imported to Preform software (v3.21, Formlabs) and the architecture of the support material was manually adjusted to avoid any overhang and to minimize the presence of internal support material. Each anatomy was then printed on a Form 3B Stereolithography (STL) 3D-printer (Formlabs Inc.) with a layer thickness of 0.1 mm.
Each anatomical (STL) LV model was used for the design of a patient-specific soft robotic LV sleeve in SolidWorks (2019, Dassault Systèmes, Vélizy-Villacoublay, France). The outer surface of the LV was offset by 10 mm to generate guide tracings for the sleeve geometry and divided the tracings into four circumferential quadrants (each approximately 90 degrees apart) and. These quadrants were flattened to a plane to create the contours of the molds for manufacturing. The flat tracings were then extruded by the same offset (10 mm) and each mold was 3D-printed from a rigid photopolymer (Veroblue, Stratasys, Eden Prairie, MN) with an inkjet-based Objet30 3Dprinter (Stratasys, Eden Prairie, MN). Similarly, the contours of the aortic valve leaflets were exported from CT images and converted into flat geometries, which were then extruded, and 3Dprinted for manufacturing of the aortic sleeve.
Analogously to the manufacturing technique previously described by our group (30,56), for each of the four LV molds and three aortic molds (or two for bicuspid valve anatomies) per patient, two sheets of Thermoplastic Polyurethane (TPU, HTM 8001-M 80A shore polyether film, 0.012" thick, American Polyfilm, Inc., Branford, CT) were vacuum-formed (Dental Vacuum Former, Yescom, City of Industry, CA) into the shape of the molds. Each pair of TPU sheets was then heatsealed at 320F for 8 seconds on a heat press transfer machine using negative acrylic molds to create enclosed and inflatable geometries. For each sleeve, these inflatable pockets were then heat-sealed using a similar process as that described above to a 200-Denier TPU-coated fabric (Oxford fabric, Seattle fabrics Inc., Seattle, WA), which was designed to fit around their respective LV or aortic anatomy. Further, holes were opened through the fabric on one side of each of the pockets to connect soft tubes (latex rubber 1/16" ID 1/8" OD tubing, McMaster-Carr, Elmhurst, IL) as actuation lines through PVC connectors (polycarbonate plastic double-barbed tube fitting for 1/16" tube ID, McMaster-Carr, Elmhurst, IL).
A 3D-printed (Objet30, Stratasys) rigid skeleton with four arms (one for each pair of adjacent pockets of the LV sleeve) was designed to secure each LV sleeve to the corresponding 3D-printed geometry. A belt-like securing mechanism with Velcro-adhesive was integrated in the design of the fabric of each aortic sleeve to guarantee secure attachment around the 3D-printed aorta (Fig. 3A).
In this work, the MM3DP approach developed by Hosny et al (29) was optimized for hydrodynamic testing and used for comparison with our model. We used the algorithm developed in their work to generate the aortic valve leaflet geometries from landmarks of abdominal CT images with superimposed calcium-like nodules. However, instead of limiting the vascular anatomy to the aortic sinus, we integrated the valve leaflets and nodules into the entire LV and aortic (through the aortic arch segment) anatomies. Firstly, this approach allowed us to conduct functional tests of their model of AS. Secondly, it enabled us to integrate their approach with our strategy of LV actuation, allowing for a fairer comparison between the two models of AS. The LV, aorta, aortic valve leaflets, and calcium nodules were printed simultaneously using an Objet 500-Connex3 3D-printer (Stratasys) using the same printing techniques as what are described in the original publication (29). To do this, we created a small (2-5 mm in diameter) hole in proximity to the LV apex to remove any support material laid down during the manufacturing process. The hole was then sealed using the same resin material and UV light was applied manually.
For each patient, a closed loop was set up for hydrodynamic studies (viscosity of medium, =1,0 cP) (22). The loop was assembled by connecting the 3D-printed anatomy to a series of soft PVC plastic tubing (3/8"-5/8" ID, 5/8-1" OD, McMasterr-Carr), two variable-resistance ball valves to mimic arterial and venous resistance, and two custom-made acrylic compliance chambers to recreate peripheral compliance. A unidirectional mechanical valve (Regent bileaflet mechanical prosthesis, 19AGN-751 standard cuff, Abbott Laboratories, Chicago, IL) was connected in proximity to the venous system. Two clamp-on perivascular flow probes (PS series, Transonic, Ithaca, NY) were used to measure flow immediately distal to the descending aorta (LV outflow) and distal to the surgical valve in the LV (LV inflow). The flow probes were connected to a twochannel flowmeter console (400-series, Transonic), which was in turn connected to an 8-channel Powerlab system (ADInstruments, Sydney, Australia) for data acquisition and recording. Two straight-tip 5F pressure-volume (PV) catheters were inserted through two adjustable catheter connectors to measure pressures at the LV and at the ascending aorta, distal to the aortic sleeve. The catheters were connected to a Transonic ADV500 PV System and to the Powerlab (ADInstruments). Given the mismatch in electrical impedance between the 3D printing material and that of the native cardiac tissue, the PV catheters could not be used to reliably measure volumes inside the LV. An endoscopic camera was inserted in the system to visualize the cross-sectional profiles of the aorta during actuation for subsequent calculations of the valve EOA.
The system was actuated pneumatically through the soft robotic LV sleeve, which was connected to a control box and associated GUI, where input pressure tracings could be defined (Fig. 4C). The aortic sleeve was actuated hydraulically using a syringe pump (70-3007 PHD ULTRA™ Syringe Pump Infuse/Withdraw, Harvard Apparatus, Cambridge, MA). The actuation pressures and volumes of the soft robotic sleeves were modulated to achieve the values of SV and Pmax for each individual patient, as well as LV and aortic pressure values when known. Systolic and diastolic actuation pressures of the LV sleeves ranged between 8-13 psi and 0-6 psi for systole and diastole, respectively, whereas actuation volumes equal to 20 -40 mL were used for actuation of the aortic sleeves.
The Epiq CVx cardiovascular ultrasound system (Philips, Amsterdam, Netherlands) was used in tandem with the X5-1 transducer (Philips) for echocardiographic evaluation of each patientspecific model to assess the degree of AS and, in some instances, LV function and paravalvular leak. As the transaortic pressure gradients and the EOA could be more accurately measured using the PV catheters and the endoluminal camera respectively, echocardiography was primarily used to compute the peak flow velocity (vmax) through the aortic valve (Fig. 3D, fig. S6, Fig. 5G). The probe was positioned directly on the 3D-printed geometry, leveraging the anatomical curvature between the ascending aorta and proximal arch to align the ultrasonic beam with the direction of flow for continuous wave Doppler imaging. Tracings of the aortic flow velocity were obtained, and the peak value of each tracing (vmax) was calculated. In a subset of patient models, color flow mapping Doppler images were obtained for visualization of the flow through the soft robotic aortic sleeve and stenotic MM3DP valves for comparison. In addition, 2D movies of the LV in long-axis view, of the MM3DP valve, and of the SE valve prosthesis (Evolut R, Medtronic) were recorded to visualize actuation of the soft robotic LV sleeve (Fig. 4A-B) and mobility of the MM3DP and TAVR leaflets. Finally, color flow mapping Doppler images were taken to provide a qualitative comparison of the degree of paravalvular leak between the patient models with an appropriately sized and an undersized valve (Fig. 5I).
A study of the differences in paravalvular leak and ARI between the undersized TAVR and appropriately sized TAVR groups was conducted simulating implantation of valves of various sizes in patients with a spectrum of annular dimensions. We used undersized valves in patients 2, 3, 6, 7, 10, 14; and appropriately sized valves in patients 3, 4, 6, 7, 11, and 14. The Evolut R valve (Medtronic) was delivered manually from a distal opening in the anatomy to the point of constriction or slightly supra-annularly, whereas the SAPIEN 3 valve (Edwards Lifesciences) was delivered using the Edwards transfemoral balloon catheter (Edwards Lifesciences).
In this study, we computed the ARI as per Equation ( 1):
where AoPD and AoPS are the diastolic and systolic aortic pressures, respectively, and the LVEDP is the end-diastolic LV pressure. All parameters in this equation were measured from PV catheters. Other metrics including Pmean, Pmax, EOA, vmax, SV, LVP, LVV, and AoP were calculated.
Data were primarily visualized and acquired by LabChart (Pro v8.1.16, ADInstruments). All the input signals were filtered using the default 50Hz band-stop filter. Signals include the LV and aortic pressures for calculation of the transaortic pressure gradient, the flow rates out of and into the LV for calculation of the SV and LVEF. From these data, the LVEDP, LVPs, AoP, LVV could be extracted. Analogously, the actuation pressure of the LV sleeve was displayed and recorded on LabChart. Data analysis and visualization was performed through an automated algorithm developed on MATLAB R2020a (MathWorks, Natick, MA). Values for peak flow velocity measured using continuous wave Doppler were included in the analysis. For catheterization data, average values and standard deviations were calculated for ten consecutive heart cycles after the soft robotic sleeve actuation degrees were successfully tuned to recreate the patients' cardiac and aortic hemodynamics. Echocardiography data was averaged across three heart cycles. A two-tailed t-test was performed (MATLAB R2020a) to determine significance between the undersized and the correctly sized TAVR groups (Fig. 5J) using a 95% confidence interval (p < 0.05). For each patient in the analysis, we considered the average ARI value calculated across five consecutive heart cycles. A Kolmogorov-Smirnov test (MATLAB R2020a) confirmed a standard normal distribution of the ARI averages within each of the two groups (Fig. 5J).
Images of the aortic cross-sections were processed using the Image Processing and Computer Vision MATLAB toolbox (MATLAB R2020a) for calculation of the EOA. These images and those obtained from the patients' CT were binarized and each pair of images (one pair per patient) was cross-registered. The rigid distortion option, enabling only translation and rotation of the moving image, was utilized for image registration, and the Sorensen-Dice similarity coefficient (DSC) was calculated for each patient model.