Towards the development of a shape memory polymer for individualized endovascular therapy of intracranial aneurysms using a 3D-printing/leaching method

Abstract

Endovascular treatment of intracranial aneurysms (ICA) aims to occlude the aneurysm space for preventing ICA growth/rupture. Modern endovascular techniques are still limited by lower complete occlusion rates, frequently leading to aneurysm growth, rupture and re-operation. In this work, we propose shape memory polymer (SMP)-based embolic devices that could advance the effectiveness of ICA therapy by facilitated individualized ICA occlusion. Specifically, we develop an 3D-printing/leaching method for the fabrication of 3D-SMP devices that can be tailored to patient-specific aneurysm geometries that are obtained from computed tomography angiography. We demonstrate that this method allows the fabrication of highly porous, compressible foams with unique shape memory properties and customizable microstructure. In addition, the SMP foams exhibit great shape recovery, anisotropic mechanical properties, and the capability to occlude in-vitro models with individualized geometries. Collectively, this study indicates that the proposed method will have the potential to advance the translation of coil- and stent-free embolic devices for individualized treatment of saccular ICAs, targeting complete and long-term durable aneurysm occlusion.

1. Introduction

Intracranial aneurysms (ICAs) are focal dilations of the vascular wall that are estimated to affect 3–5% of the general population.^[1,2] Most ICAs remain asymptomatic and unruptured;^[2] however, 7.9 (95% CI: 6.9 – 9.0) per 100,000 persons will suffer from aneurysm-related subarachnoid hemorrhage (SAH) every year,^[3] which can lead to death or long-term disability. The risk of ICA rupture is related to geometrical, anatomical, demographical, among others.^[4,5] When unruptured aneurysms are diagnosed, this risk can be mitigated via different therapeutic methods in the clinic.

Modern ICA therapeutic methods include microsurgical clipping and endovascular embolization. The latter has been demonstrated to have superb effectiveness and improved safety outcomes, especially for the older patients. In addition, its lower invasiveness makes it the preferred surgical option for treating an ICA.^[6–8] Procedure-wise, endovascular embolization uses biocompatible materials to prevent flow into the aneurysm space (sac). This method was first introduced by Guglielmi et al. through the use of platinum coils^[9,10] and has been further developed with assisting devices, including stents and balloons. In addition to Guglielmi detachable coils (GDCs), other endovascular techniques have emerged recently, expanding the applicability of endovascular therapy to aneurysm geometries that cannot be treated with bare coils or coils with assisted devices.^[11]

Despite the great advance in the ICA endovascular technologies, the occlusion effectiveness of state-of-the-art endovascular devices is far below optimal, especially when treating large ICAs with complex 3D geometries. For example, the GDCs have exhibited 50 – 66 % complete occlusion rates,^[12–15] while hydrogel-coated coils showed complete occlusion rates of ~ 69 %,^[14] not significantly superior to the GDCs, according to meta-analysis.^[16] In addition, other endovascular devices, such as the the Woven EndoBridge and flow diverters, only reach complete occlusion rates of ~ 80 %.^[17–23] Thus, improving these limited complete occlusion rates is of paramount importance, as incomplete aneurysm occlusion can potentially lead to progressive aneurysm growth and the subsequent rupture.^[24]

To this end, individualized endovascular devices that mimic the patient-specific aneurysm geometry emerge as potential candidates to improve the effectiveness of endovascular therapy. For example, we have proposed the use of shape memory polymers as biocompatible alternatives for the occlusion of ICAs. By means of fabricating patient-specific porous SMPs, we will be able to promote complete occlusion immediately after treatment and thus, prevent aneurysm recurrence.

In our previous work, we have characterized the potential of an SMP for the fabrication of porous endovascular devices.^[25,26] Here, we present the fabrication of a patient-specific porous SMP using a novel 3D-printing/leaching method. Moreover, we characterize the thermo-mechanical properties of this 3D-printed SMP, shorted as 3DSMP, so that we can fine tune the properties of the SMP materials to simulate the occlusion of in-vitro patient-specific aneurysms.

2. Results

2.1. Fabricated 3DSMP Foams and Microstructure

The 3DSMP foams were fabricated with a method that combines 3D-printing and leaching. Briefly, polyvinyl alcohol (PVA) cubes were first 3D printed using a grid infill pattern (Figure 1a) and were then used as templates for the SMP synthesis. Here, the freshly mixed SMP solution was poured on the PVA templates, and the specimens were cured using different thermal treatments in a nitrogen-rich environment until they became solid (see Section 5). Next, the PVA filament templates were leached in a water bath using sonication, leaving polyurethane foams that mimicked the infill microstructure of the original print (Figure 2a.1 and Figure 2b.1). Scanning electron microscopy (SEM) micrographs demonstrated that the SMP solution cured around the PVA infill filaments (Figure 2a.2 and Figure 2a.3), allowing the replication of the pattern and creating the microchannels after sonication (Figure 2b.2 and Figure 2b.3).

Figure 1:

(a) Schematics showing the steps for the fabrication of SMP foams using templates 3D printed with polyvinyl alcohol (PVA) (see Section 1). (b) Pipeline for the production of individualized PVA templates and phantoms obtained from patient’s CTA imaging. The white arrow indicates the parent vessel, which has a diameter of 2.45 mm. (The 3D model was enlarged twofold.)

Figure 2:

Photographs and SEM micrographs of the 3DSMP foams from the (a) top view and (b) side view.

To evaluate the overall microstructure (i.e., porosity) of the foams relating to the 3D printing condition, we further fabricated 3DSMP foams using PVA templates with different infill densities. The PVA templates had an average wall thickness of 0.402±0.007 mm, 0.382±0.005 mm, and 0.449 ± 0.009 mm for the 40 %, 50 %, and 60 % infill densities, respectively (Figure 3). These template structures yielded 3DSMP foams with consistently increased wall thickness: 0.465 ± 0.019 mm for the 40 %, and 0.434 ± 0.012 mm for the 50 % infill densities (Figure 3b, $p<0.05$), but with no significant change in the 60 % templates (0.438±0.008 mm, Figure 3b, $p>0.05$). The different infill densities allowed to obtain 3DSMPs with different porosities. Overall, the 3DSMP foams precisely mimicked the PVA template with minor defects, such as increased surface roughness and randomly clogged pores (not shown).

Figure 3:

(a) SEM micrographs of PVA templates (top) and 3DSMP foams (bottom) at different infill densities (columns). (b) Measurements of the wall thickness between the PVA templates and 3DSMP foams.

2.2. Differential Scanning Calorimetry

To determine the effect of the PVA-leaching process on the glass transition temperature ($T_g$) of the SMP, differential scanning calorimetry (DSC) was performed. 3DSMP foams were fabricated using different monomer molar ratios (Table 1), as described previously.^[25] First, we observed that controlling the monomer molar ratios allowed the fine tuning of the resultant $T_g$ of the material, where an increased concentration of TEA leads to a decrease in $T_g$ (Figure 4a). Further, we observed that the $T_g$ of the 3DSMP foams was significantly lower than their solid counterparts (Figure 4a), indicating that our PVA-leaching method affected the thermal behavior of the resultant polyurethane matrices. Using the measured $T_g$ from the solid and 3D foams, we performed a 2-term exponential regression for each material type using the TEA molar ratio as a predictor for $T_g$. Using the predicted values from these models, we fabricated a new molar ratio (named SMP-X in Table 1) associated with a predicted $T_g\approx {41}^{\circ }\mathrm{C}$ (Figure 4b). This temperature was selected as a target temperature desirable for in-vivo delivery of the foam that will not cause thermal damage to the artery wall and aneurysm tissues. Experimental testing of the SMP-X ratio (TEA molar ratio of ~ 0.22) demonstrated that our model accurately predicted the $T_g$ of the 3DSMP foams. This monomer ratio was used for the subsequent testing of the shape recovery and mechanical properties of the material, and in the in-vitro aneurysm embolization demonstration experiment (see Section 2.6).

Table 1:

Monomer ratios used for SMP synthesis and thermal analysis.

SMP Ratio	Molar Ratio
	HDI	HPED	TEA
SMP5	1.0	0.35	0.19
SMP6	1.0	0.3	0.26
SMP7	1.0	0.25	0.33
SMP8	1.0	0.2	0.4
SMP-X	1.0	0.32	0.22

Figure 4:

(a) Experimental $T_g$ measurements of the solid SMP and 3DSMP. (b) $T_g$ exponential models (i.e., as a function of TEA molar ratio) used to predict the TEA content of 3DSMP foams with $T_g\approx {41}^{\circ }\mathrm{C}$.

2.3. Attenuated Total Reflectance — Fourier Transform Infrared Spectroscopy (ATR-FTIR) Characterizations

We further performed ATR-FTIR to verify the chemical integrity of the material after the 3D-printing/leaching process. We previously characterized the ATR-FTIR spectrum of our pristine SMP material,^[26] with peaks at: (i) 1683 cm⁻¹: urethane carbonyl C=O stretching, (ii) 1542 cm⁻¹: urethane C-N-H deformation vibrations, (iii) 1246 cm⁻¹: C-N stretching vibrations, (iv) 1142 cm⁻¹: C-O-C stretching vibrations, and (v) 3306 cm⁻¹ : N-H stretching vibrations (Figure 5a).^[27,28] This spectrum is characteristic of highly cross-linked polyurethanes, as described by Stern.^[28] However, the 3DSMP foams exhibited some noticeable differences from highly cross-linked polyurethanes in their FTIR spectrum. Namely, we observed the appearance of a peak at 1616 cm⁻¹ (Figure 5b), adjacent to the 1683 cm⁻¹ peak, which can be observed in linear polyurethanes.^[28] In addition, we observed more prominent peaks at 3327 cm⁻¹, related to O-H vibrations, and peaks at 1046 and 1142 cm⁻¹ associated with C-O hydroxyl stretching. These changes in the FTIR spectrum might be associated with the urethane links formed by PVA and HDI during the SMP curing and/or the interactions with the added PVA filament, attributing to the portions of the material with linear polyurethane segments.

Figure 5:

(a) FTIR spectra of solid SMP and 3DSMP of the SMP-X ratio. (b) Zoomed spectra showing the altered urethane chemistry of the 3DSMP foams due to the 3D-printing & PVA-leaching process.

2.4. Compressive Mechanical Properties of the 3DSMP Foams

The mechanical properties of the 3DSMP foams were characterized using a compressive mechanical tester. We fabricated specimens with the SMP-X monomer ratio (Table 1) using PVA templates with a grid infill pattern (Figure 2) at different infill densities (40 %, 50 % and 60 %, Figure 3). We tested the compressive behavior of the material along the $X$ - and $Z$-axis to study the effect of infill pattern density on the overall mechanical behavior of the foams. First, we observed that the stress-strain curves exhibited a non-linear behavior for both axes (Figure 6a). Specifically, we observed that both axes exhibited a semi-linear behavior up to ~ 60 % strain; then, the stress increased exponentially. We also observed an anisotropic behavior, where the $Z$-axis was stiffer than the $X$-axis (Figure 6a), due to the higher material density in the $Z$-axis that is associated with the short separation between printed lattices (Figure 2b.2). In addition, we observed that the mechanical properties of the material also depended on the infill density and compressive direction (Figure 6). Namely, the peak stress at ~ 90 % compressive strain along the $X$-axis was homogenous across the infill densities with an average of 0.92 ± 0.11 MPa ($p>0.05$, Figure 6b.1); however, the peak stress along the $Z$-axis was lower in the 40 % infill density group than in the other groups, albeit with no significance (1.44 ± 0.18 MPa vs. 3.48 ± 0.61 MPa, $p=0.15$, Figure 6b.2). The peak stress measurements decreased with cyclic compression, indicating the plastic deformation of the foams. This cumulative stress reduction (CSR) was not dependent on the infill density nor the compressive direction; the CSR for both directions was −13.10 ± 3.23 % ($p>0.05$, Figure 6b.3–4). The average elastic modulus for the $X$-axis was 0.08 ± 0.01 MPa with no significant differences between infill density groups (Figure 6b.5). For the $Z$-axis, the average elastic modulus for the 60 % infill density group was significantly higher than for the other groups (0.22 ±0 .02 vs 0.14 ± 0.01, $p<0.01$, Figure 6b.6). Overall, the $Z$-axis exhibited a higher elastic modulus than the $X$-axis $(p<0.01)$.

Figure 6:

(a) Representative compressive stress-strain curves of the 3DSMP foams at different infill densities, and (b) derived mechanical parameters: peak stress, CSR, and elastic modulus.

2.5. Shape Recovery Properties of 3DSMP-X Foams

As demonstrated in the DSC results (see Section 2.2), the material exhibited glass transitions that allow the shape reprogramming and recovery. To characterize this behavior, we performed shape recovery experiments that provided insight into the capacity of the material to store its deformed geometry and the degree of recovery from that reprogrammed, deformed geometry. We fabricated 3DSMP cubes, which were first heated above $T_g$ (i.e., $T=T_g+{10}^{°}\mathrm{C}$.), compressed them uniaxially, and induced shape storage by cooling the material below $T_g$. Then, we measured the shape recovery under different thermal conditions: (i) at room temperature $\left(T={23}^{°}\mathrm{C}\right)$, we observed that the material was able to maintain the shape, i.e., < 5% shape recovery in ~ 53 s, and that the average shape recovery was 55 % after 10 min (Figure 7a); (ii) in a −20 °C freezer, we observed that the 3DSMP cubes maintained their deformed configuration, exhibiting only 0.68% shape recovery after 4 days of storage in the freezing condition (Figure 7b); and (iii) lastly, we triggered shape recovery with a heating ramp up to $T=T_g+{10}^{°}\mathrm{C}$, where we observed that > 95 % shape recovery was accomplished in ~ 70 s (Figure 7c and Figure S1).

Figure 7:

Shape recovery properties of the 3DSMP foams: Shape storage at (a) room temperature $\left(T={23}^{°}\mathrm{C}\right)$ and (b) $T=-{20}^{°}\mathrm{C}$. (c) Shape recovery triggering using a heating ramp at $T=T_g+{10}^{°}\mathrm{C}$.

2.6. Patient-Specific 3DSMP Foams: Proof-of-Concept Demonstration of Image-Based Fabrication and Phantom Occlusion

To demonstrate the use of the proposed 3D-printing/leaching method in patient-specific modeling, we considered aneurysm phantom models with geometries derived from patient computed-tomography angiography (CTA) data. We fabricated aneurysm phantoms and 3DSMP foams using two different patient geometries (Figure 8). The phantoms were used to simulate the aneurysm space, which was successfully occluded by the patient-specific foams. We observed that complete occlusion of the aneurysm models was achieved for both 3DSMP geometries. Specifically, aneurysm geometry #1 is a bifurcation aneurysm with an elongated geometry, which allowed a facile delivery and complete occlusion of the aneurysm space and neck. On the other hand, aneurysm geometry #2 was a side-walled aneurysm with a “cherry-like” geometry, with a more spherical and irregular shape that made delivery more difficult due to the overhanging portions of the foam, which led to additional repositioning of the foam during delivery. Overall, both aneurysm geometries were successfully occluded using our patient-specific 3DSMP foam.

Figure 8:

Proof-of-concept demonstration of the occlusion of aneurysm phantoms with patient-specific geometries using 3DSMP foams. Aneurysm geometries were enlarged threefold from original CTA imaging to facilitate delivery of the foam.

3. Discussion

3.1. Overall Findings

In this study, we fabricated porous polyurethane 3D structures using a 3D printing/leaching method, inspired by the method proposed by Hernández-Córdova et al.^[29] This method overcomes previous limitations related to the direct extrusion of our polyurethane formulation in a liquid state. Specifically, because of the use of TEA (a triol) in our polyurethane formulation, the SMP presents a high degree of crosslinking, making traditional fused deposition modeling (FDM) 3D-printing method not applicable for the fabrication of patient-specific foams. Therefore, this study provides a significant advance in our goal to fabricate novel endovascular devices considering patient-specific geometries that can potentially provide complete and long-lasting occlusion of ICAs. We found that combining PVA FDM 3D-printing with leaching provided a low-cost, facile technique for the fabrication of CAD modeled SMPs with a highly porous structure. Overall, the comprehensive thermomechanical characterization of the 3DSMP foams led to the following findings:

1. The SMP solution is able to adhere to the PVA 3D-printed grid structure and, after curing, it creates a sheath that surrounds the infill filament. After leaching, the 3DSMP foam precisely mimics the filament grid pattern, as well as the overall 3D geometry of the PVA template (Figures 2 and 3).

2. The glass transition temperature $\left(T_g\right)$, of the resultant 3DSMP foams can be fine-tuned by judiciously adjusting the monomer molar ratios, as described previously (Figure 4).^[25] Also, 3DSMP foams exhibit a significantly lower $T_g$ than the pristine SMP.

3. The FTIR spectra of the 3DSMP and solid SMP materials demonstrated the synthesis of highly crosslinked polyurethane. In addition, the 3DSMP FTIR spectrum exhibits features that are more typical of linear polyurethane, which may arise from the unwanted PVA-HDI urethane linkage. The formation of these linear urethane links might explain the lower $T_g$ observed in the DSC experiments for the 3DSMP foams, due to higher polymer chain mobility.

4. The 3DSMP foams exhibited an anisotropic mechanical behavior where the $X$-axis of the grid-pattern prints was significantly more compliant than the $Z$-axis. This is due to the lower material density in this direction, as observed in Figure 2. Specifically, the grid design on the $X-Y$ plane facilitates compression along these axes, while the short distance between the lattices along the $Z$-axis made the material stiffer in this direction. In addition, this anisotropic behavior provides an opportunity for the design of compression patterns for the guided compression of the patient-specific endovascular devices. Further, we also demonstrated the control of infill density (i.e. porosity) of the 3DSMP material, which could lead us to achieve fine-tuning of the mechanical parameters of the material. Overall, the 3DSMP foams exhibited great compressibility on the $X-Y$ plane with a fine-tunable porosity. 5) Finally, we demonstrated the use of the 3D-printing/leaching method for the fabrication of patient-specific ICA embolization foams. The patient-specific 3DSMP foams were used to occlude aneurysm models with two patient-specific geometries from patient medical image data.

3.2. Study Limitations

This study, while advancing the development of individualized embolic materials for ICA endovascular treatment, faces some limitations. First, the SMP solution infiltration was not in a well-controlled environment that guaranteed homogenous infiltration, which might lead to the formation of closed pores in the final 3DSMP foam. We recommend that infill densities higher than 60 % are avoided to prevent this issue. In addition, the formation of the hollow channels, created by the PVA filament leaching, can also potentially lead to regions prone to rupture after cyclic compression. Our experiments indicate that 10 cycles of compression did not induce noticeable damages to the polymeric structure, as the specimens were able to completely recover their original shape after the compression test (data not shown); however, for its future applications, we will limit cyclic compression prior to delivery to the aneurysm to prevent the breaking of the inner lattice structure.

Also, the TEA molar ratio of the SMP-X that was determined in our DSC experiments utilized a higher HPED ratio. During synthesis, HPED was the most viscous monomer of our formulation, which in high concentrations can make monomer dissolution difficult and time consuming. Previously, we have reported the monomer mixing times as the period from initial contact between monomers to the change in solution transparency from a “hazy” consistency to a completely clear solution. We later found that monomer mixing should be extended for monomer ratios containing high HPED concentrations, as this can lead to local variations in $T_g$, affecting the overall thermomechanical behavior of the final product.

Further, our proof-of-concept occlusion of aneurysm models provided insight into the potential aneurysm geometries that can be treated with the 3DSMP devices. Aneurysm #1 had an elongated geometry with a wide neck and a narrow tip, which allowed for printing a template without overhanging portions. These features yielded an easy-to-manipulate and smooth 3DSMP foam. On the other hand, aneurysm geometry #2 did exhibit some overhanging portions, due to a very narrow neck, making foam delivery difficult and requiring adjustments to its position during delivery. In fact, we aimed at performing 3DSMP foam delivery to a giant aneurysm geometry (see the 3D model in Figure 1b) with a relatively narrow parent artery. However, occlusion of this geometry was not possible due to the currently unfeasible compression degree that would be required to transport the foam through the relatively narrow artery to the aneurysm sac. This limitation in compressibility also played a key role in the decision to enlarge the aneurysm geometries to facilitate delivery. In addition, this geometry exhibited protrusions that, in a lattice-based structure, can be structurally unstable and lead to foam rupture.

Overall, the developed 3DSMP device exhibits remarkable potential for the treatment of saccular bifurcation aneurysms, such as in aneurysm #1; other geometries will require further improvement and optimization of the infill structure.

3.3. Future Directions

The application of the method presented in this study opens doors to opportunities to test the potential of individualized embolic materials for ICAs. Using this technique, we plan to develop a closed flow loop for the in vitro demonstration of occlusion, and experimental flow dynamics characterization of treated and untreated aneurysms, similarly to the work performed by Baidya et al.^[30] In addition, we will develop a triggering mechanism for shape recovery. In our previous work, we have used carbon nanotubes for enhancing the material’s electric conductivity for inducing Joule-heating of porous SMPs;^[26] we plan to incorporate similar actuation methods in future studies.

Moreover, we plan to improve the infill structure presented in this study, based on the geometrical limitations discussed in Section 3.2. Specifically, we will develop printing algorithms for more structurally stable complex geometries, as well as infill patterns for guided compression. These developments will be based on in silico simulations of foam delivery.

Furthermore, we will translate this device from its current in vitro conditions, to an in vivo setting that provides information on key biological processes involved in aneurysm embolization. First, we will explore the capacity of this material to promote platelet aggregation, a fundamental process in the thrombogenesis that is characteristic of GDCs.^[9,10] We will also explore the interaction of the material with cell lines that will neighbor the material in vivo, such as endothelial and smooth muscle cells. Lastly, we will test the effectiveness of this individualized approach in animal models, such as the elastase-induced aneruysm model in rabbits that we have presented previously.^[31] These efforts will provide sustained major progress on the translation of individualized endovascular ICA therapy.

4. Conclusion

In this study, we have developed highly porous crosslinked polyurethane SMPs using a 3D-printing/leaching-based fabrication method. This approach allows for the fabrication of 3D structures using a grid infill pattern with controllable anisotropic compressibility, tunable $T_g$ and the capacity to occlude aneurysm models. The 3DSMP foams fabricated in the present method have been shown to provide a significant improvement to our previously characterized porous SMPs. Our future investigations will include combining the 3DSMP foams with conductive materials to obtain a Joule-heating-triggered SMP with patient-specific geometries, thus advancing the translational potential of this endovascular therapy to obtain complete and durable occlusion of ICAs.

5. Experimental Section

3D Printing of PVA Templates:

Sacrificial templates were 3D-printed using a PVA filament (MatterHackers, $d=1.75\mathit{\text{mm}}$) in a commercial 3D-printer (Creality CR-10S, China). Templates were fabricated in two different geometries: (i) cubic specimens $(10\times 10\times 10\mathit{\text{mm}})$ for material characterization, and (ii) two patient-specific geometries (Figure 8). Cubic specimens were designed using CAD software (Solidworks, Dessault Systems, USA) and exported to a 3D-printing slicing software for infill design (Cura, Ultimaker, Netherlands). Here, templates were fabricated with a grid infill pattern at different densities (40 %, 50 % and 60 %). Top and bottom layers of the print were removed to expose the print infill for SMP solution infiltration. Additionally, two outer walls were printed around the infill structure to aid in the removal of excess cured SMP. Secondly, patient-specific geometries were designed using computed tomography angiography (CTA) images of human ICAs obtained at the University of Oklahoma’s Health Sciences Center (IRB #7932).The CTA images were segmented using Amira (Thermofisher & Zuse Institute Berlin), surface smoothing was performed using Meshmixer (Autodesk, San Rafael, CA), and subsequently 3D-modeled for printing in SolidWorks.^[32] Furthermore, a rigorous calibration protocol was performed to determine the optimal printing parameters that provided great surface quality and dimensional accuracy for each type of geometry (Table S1).

Preparation of 3DSMP Foams:

Polyurethane SMP solution was synthesized using hexamethylene diisocyanate (HDI, ≥98.0 %, Sigma-Aldrich) as the hard segment, N,N,N0,N0-tetrakis (hydroxypropyl) ethylenediamine (HPED, ≥98.0 %, Alfa Aesar) as the catalyst, and triethanolamine (TEA, ≥99.0 %, Sigma-Aldrich) as the soft segment. HDI, HPED and TEA were mixed at different molar ratios (Table 1), as described previously,^[25,33,34] using a magnetic stirrer in a nitrogen-protected environment until the monomer solution turned clear. Then, the solution was evenly poured onto 10 × 10 × 10 mm PVA 3D-printed templates. The uncured SMP solutions were maintained at −20 °C in a vacuum container (FoodSaver, Newell Brands, USA) for 1 h to slow polymerization and to allow complete infiltration of the solution into the template. Then, the specimens were cured at 70 °C for 1 h and 100 °C for an additional hour each under a nitrogen-rich environment using a vacuum oven (BOV-20, Being Instrument, USA). The cured blocks were next immersed in a water bath sonicator (Branson 1800, Branson Ultrasonics Corporation, USA) and sonicated at 50 °C for 24 h to dissolve the PVA templates and residuals. Finally, the 3D foams (3DSMP) were dried in a vacuum oven at 60 °C until mass measurements stabilized (indicating no further water evaporation). The fabricated specimens were stored in a desiccator in vacuum to preserve thermal properties of the material.

Attenuated Total Reflectance - Fourier Transform Infrared Spectroscopy (ATR-FTIR):

3DSMP foams were characterized with ATR-FTIR to verify that polyurethane synthesis was not altered during the 3D-printing process. The ATR-FTIR spectrum was obtained with a spectrometer (Nicolet iS50R, Thermo Scientific, USA) with a $L$-alanine doped triglycine sulfate (DTGS) and mercury cadmium telluride (MCT) detector. The 3DSMP foams were compared to solid specimens to analyze the differences in the morphology of characteristic polyurethane peaks. The spectrum analysis was performed using an in-house MATLAB program (Mathworks, USA).

Scanning Electron Microscopy (SEM):

Micrographs of the 3DSMP specimens were obtained to characterize the microstructural and morphological features of the foams. The specimens were imaged in a field-emission environmental scanning electron microscope (Thermo Quattro S, Thermofisher, USA) at an accelerating voltage of $20\mathit{\text{kV}}$. Architectural features, such as the respective wall thickness and fiber thickness, were measured using ImageJ Fiji (National Institutes of Health, USA).^[35]

Differential Scanning Calorimetry (DSC):

DSC was performed to determine the effect of the PVA-leaching process in the $T_{\mathrm{g}}$ of the synthesized 3DSMP foams. The specimens $(n=3)$ at the monomer ratios described in Table 1 were examined under DSC testing. The experimental groups included the 3DSMP specimens and solid SMP specimens, as control. Specifically, the specimens were tested in a DSC device that was calibrated with tin, indium and biphenyl(DSC2500, TA Instruments, USA). The experiment consisted of three heating/cooling cycles at a range of −20 °C to 120 °C at a 5 °C/min heating rate and 10 °C/min cooling rate with isothermal stabilization at the end of each cycle for 5 min. The glass transition temperature $\left(T_{\mathrm{g}}\right)$ was measured from the second heating cycle using the TA Instruments Trios software v5.1.1.45672. The obtained $T_{\mathrm{g}}$ measurements were then used to obtain two-parameter exponential models (MATLAB R2022a) of the leached and solid specimens. Using the predicted values of the models, a monomer ratio with $T_{\mathrm{g}}\approx {41}^{°}\mathrm{C}$ was formulated. The resultant SMP ratio (SMP-X in Table 1) was tested in DSC to verify its $T_{\mathrm{g}}$.

Mechanical Characterization - Uniaxial Compressive Testing:

Using the SMP-X monomer ratio $\left(T_{\mathrm{g}}\approx {41}^{°}\mathrm{C}\right)$, 3DSMP foams ($n=6$) were subjected to cyclic compressive mechanical testing. Compressive deformation was applied to the specimens using a plate mechanical tester (Instron 5969, USA) at a compressive rate of 4 mm/min up to a compressive strain of 80 %. Specimens were kept in a heating chamber at $T=T_{\mathrm{g}}+{10}^{°}\mathrm{C}$ during mechanical testing. Mechanical parameters were calculated and processed in MATLAB R2022a (Mathworks, USA): (i) peak stress was defined as the measured stress at 80 % strain; (ii) cumulative stress relaxation (CSR) was quantified as the percent reduction of peak stress at each compression cycle; and, lastly, (iii) the elastic modulus was calculated as the slope of the linear fit obtained from the data points up 60 % strain (i.e. 800 data points at an acquisition rate of 10 Hz) of the stress-stretch curve.

Shape Recovery Assessments:

SMP-X specimens (n = 3) were heated in a vacuum oven (BOV-20, Being Instrument, USA) at $T=T_{\mathrm{g}}+{10}^{°}\mathrm{C}$ to induce the rubbery state of the SMP. The specimens were then compressed to 80 % strain with fixed plates and cooled down to −20 °C to fix the programmed geometry. Then, the compression plates were released and shape recovery was tracked using a digital camera (Nikon d5600 – 50 mm lens - 23.5 × 15.6 mm CMOS sensor, Japan) under three different thermal conditions (i) room temperature ($T={23}^{°}\mathrm{C}$), (ii) freezing conditions $\left(T=-{20}^{°}\mathrm{C}\right)$, and (iii) a thermal ramp from room temperature to $T=T_{\mathrm{g}}+{10}^{°}\mathrm{C}$. Shape recovery was measured using Eq. (1),

$$\mathit{\text{SR}}\%=\frac{h_{\left(t\right)}-h_c}{h_o-h_c}\times 100$$ (1)

where $\mathit{\text{SR}}\%$ is the percentage of shape recovery, $h_{\left(t\right)}$ is the height of the foam at a given time, $h_c$ is the height of the compressed foam, and $h_o$ is the original (pre-compression) height of the foam.

Fabrication of Patient-Specific Aneurysm Models and In Vitro Proof-of-Concept:

Occlusion Demonstration Aneurysm geometries were obtained from human CTA images and 3D-printed using PVA water-soluble filament, as described above. Then, polydimethylsiloxane (PDMS, Sylgard 184, DOW, USA) solution was poured on the mold, degassed using a vacuum pump, and cured at 70 °C for 1 h in a vacuum oven. Next, PVA was washed out in a water bath and sonicator at 50 °C. Finally, the aneurysm phantoms were dried at 60 °C for 3h. The 3DSMP foams with patient-specific geometries were compressed and transported into the aneurysm phantom. Shape recovery was triggered using an external heat lamp and final occlusion was captured using a digital camera (Nikon d5600 – 50 mm lens - 23.5 × 15.6 mm CMOS sensor, Japan).

Statistical Analysis:

Quantitative measurements were expressed as mean ± standard error of the mean. Normality was first tested using the Shapiro-Wilk test and Q-Q plots. For one-factor designs, one-way ANOVA or Kruskal-Wallis were used to test differences between groups. For two-factor designs, two-way ANOVA or aligned-rank transform tests were used, as applicable.^[36] Pair-wise comparisons were performed using Tukey-HSD test. Differences between groups were deemed significant when $p<0.05$. Statistical tests were performed in Prism v9.5.1. (GraphPad Software, USA).