Self-powered in-stent restenosis diagnosis via magnetoelastic stents

Abstract

Widely used in millions of atherosclerosis treatments, conventional metal stents, although pervasive, only provide mechanical support to narrowed arteries. However, many patients experience in-stent restenosis after implantation. Here we developed smart magnetoelastic stents that preserve mechanical functionality while enabling self-powered hemodynamic monitoring for continuous and timely diagnosis of in-stent restenosis. Using a clinical catheter, the smart stent is deployed in the swine carotid artery for in vivo hemodynamic sensing, enabling effective detection of induced stenosis through artificial intelligence-assisted signal interpretation. In vivo and in vitro studies demonstrate the biosafety of the smart stent through immune profiling, human cytokine analysis and single-cell RNA sequencing. These results underscore the smart stent’s potential for seamless integration into biological systems as a reliable diagnostic tool. This platform technology could potentially revolutionize current stent technology and contribute to improved strategies for managing atherosclerosis.

Diseases caused by arterial atherosclerosis^1–3, such as myocardial infarction, stroke and peripheral vascular disease, are the leading causes of death and disability in the United States, affecting millions of people annually⁴. Atherosclerosis is customarily treated through angioplasty with stenting, during which an implantable metal stent mechanically opens the arterial segment^5,6, restoring blood flow through the affected arteries. Despite advances in drug-eluting stents⁷, in-stent restenosis (ISR)^8–10 still occurs after surgery, due to the excess growth of plaque, the proliferation of smooth muscle cells and/or inflammatory scarring within the treated arteries (Supplementary Note 1 and Supplementary Fig. 1)⁷. With millions of stents implanted worldwide annually, ISR still poses a substantial global public health challenge. Early detection of ISR is critical to averting life-threatening complications such as heart attacks and acute coronary syndrome^11–13. Although there are many standard methods for detecting ISR (Supplementary Fig. 2), such as computed tomography angiography¹⁴, magnetic resonance imaging¹⁵, digital subtraction angiography¹⁶ and Doppler ultrasound¹⁷, these are largely ineffectual on a widespread scale due to a combination of intermittent measurement, high cost of equipment, operator dependence, employment of contrast agents, use of invasive techniques, patient exposure to ionizing radiation and inaccessibility in rural, remote or resource-limited areas^18–20. These obstacles make timely detection of ISR exceedingly unlikely, with ISR often eluding detection until it culminates in severe stenosis or complete occlusion of the artery, at which point repeat surgery becomes a necessity²¹.

These challenges prompted our development of smart stents as a transformative platform technology for continuous, timely and cost-effective diagnosis of ISR. Conventional metal stents primarily provide mechanical support to open the blocked arteries, whereas the smart stent goes beyond that by integrating an intrinsically waterproof and biocompatible magnetoelastic mesh layer for continuous hemodynamic sensing. Using the magnetoelastic mesh layer, the smart stent works in a self-powered manner and sensitively converts the hemodynamics into high-fidelity electrical signals, which are continuously received by a soft coil. To demonstrate its efficacy for ISR diagnosis, the smart stents were delivered and deployed in vivo into the carotid arteries of Yorkshire swine using clinical catheters, under real-time fluoroscopy in a minimally invasive manner. Subsequent analysis of the smart stent’s in vivo hemodynamic sensing signals successfully identified the induced stenosis. Artificial intelligence (AI)-assisted signal interpretation also demonstrated the capability to classify abnormal blood flow patterns, supporting scalable and efficient surveillance of stenosis in future clinical applications involving massive real-time hemodynamic data.

Biosafety studies were conducted, including culturing the smart stents with human peripheral blood mononuclear cells (PBMCs). Subsequent cytokine profile analysis showed no detectable abnormalities in key human inflammatory cytokines compared with controls. Single-cell RNA sequencing (scRNA-seq) revealed no notable changes in the expression of major inflammatory markers across immune cell clusters relative to controls, supporting immunocompatibility. In addition, animal studies using rat models support biocompatibility, as comprehensive immune profiling of T cell phenotypes and macrophage polarization does not reveal evidence of increased immune activation or altered polarization states compared with controls. Together, these results suggest the overall biosafety of the smart stent for applications in living systems and support its further translational development. We anticipate that this technology has the potential to extend current stent-based approaches and contribute to improved strategies for ISR diagnosis in millions of patients affected by atherosclerosis.

Working mechanism

Traditional metal stents have largely focused on mechanically supporting the blocked arteries⁵, whereas the developed smart stent consolidates a clinically approved metal stent with an inner pressure-sensitive magnetoelastic mesh layer (Supplementary Fig. 3). Thus, the device concurrently achieves a distinctive characteristic of self-powered and continuous hemodynamic sensing while maintaining an open passageway for blood flow (Fig. 1a). At the baseline state, the magnetoelastic mesh within the smart stent maintains a stable magnetic field (Fig. 1b). Pulsatile blood flow applies time-varying pressure to the mesh, leading to a temporary variation in the magnetic field (Fig. 1c). During the subsequent phase of the pressure cycle (Fig. 1d), the magnetic field of the magnetoelastic mesh recovers to its baseline level. Together, these cyclic variations in magnetic flux induce high-fidelity electrical signals in a soft coil positioned outside the artery for hemodynamic sensing, as the magnetic field of the smart stent can penetrate the tissues²². Analyzing these signals enables the detection of ISR and monitoring of postoperative conditions. The incorporation of the magnetoelastic mesh introduces the smart stent with the ISR diagnosis capability without compromising the original compatibility to the angioplasty procedure (Fig. 1e) nor its mechanical expanding functionality (Fig. 1f and Supplementary Fig. 4). Furthermore, this self-powered working mechanism decreases dependence on external power supplies, potentially facilitating the future clinical translation of the smart stent (Fig. 1g).

Fig. 1 |. A smart magnetoelastic stent for self-powered ISR diagnosis.

a, Schematics of atherosclerosis and the smart stent designed to mechanically support the artery while enabling self-powered, continuous hemodynamic sensing for ISR diagnosis. b–d, Working principle of the smart stent. b, At baseline, the magnetoelastic mesh within the smart stent maintains a stable magnetic field. c, Pulsatile blood flow applies time-varying pressure to the smart stent, causing temporary changes in the magnetic field that induce time-series electrical signals in the coil. d, During the subsequent phase of the pressure cycle, the magnetic field recovers to its baseline level. Inset axes: y axis, signal amplitude; x axis, time (t). e–g, Key features of the smart stent: compatibility with catheter delivery procedures (e); preservation of the original mechanical expansion function (f); and integration of a magnetoelastic mesh enabling self-powered, continuous hemodynamic sensing for ISR diagnosis (g). h, Photo of the various smart stents designed for different blood vessels. Scale bar, 3 mm. i, Smart stent being pushed out from a 6-Fr clinical catheter. Scale bar, 2.5 mm. Inset shows an enlarged view. Panels a–g have been partially created in BioRender. Liu, Z. (2025) https://biorender.com/tbie25f.

Manufacturing compatibility and adaptability

The smart stent comprises two components: a clinically approved metal stent substrate and an inner magnetoelastic mesh layer (Supplementary Fig. 5), securely bonded to maintain adhesion even when bent or twisted (Supplementary Note 2). This robust interface ensures that, during both device delivery and under high blood flow conditions, the magnetoelastic mesh remains firmly in place without delamination. The device’s straightforward design supports scalable manufacturing, and its use of readily available raw materials combined with compatibility with industrial-scale fabrication technologies further enhances its potential for mass production and practical clinical deployment. Furthermore, the smart stent’s customizable fabrication enables adaptation to diverse anatomies and lesion sites. Through adjusting the dimensions of the clinical metal stent and the structures of the magnetoelastic mesh layer, specialized devices can be fabricated, such as the tapered carotid stent, which preserves the natural tapering of the carotid artery after expansion, and the covered stent for aortic aneurysm repair (Fig. 1h). This adaptability supports the potential applicability of the smart stent’s wide applicability in treating a variety of cardiovascular and cerebrovascular diseases. In addition, the smart stent is compatible with a clinical catheter delivery system (Fig. 1i). Such features establish the smart stent as a compatible, versatile and adaptable solution for diverse clinical settings, suggesting its potential as a transformative tool for ISR diagnosis.

Structural optimization

We characterize the magnetomechanical coupling of the smart stent. The smart stent consists of a magnetoelastic mesh layer (Fig. 2a) bonded to a clinical metal stent. We measured the surface magnetic flux map of the mesh (Fig. 2b), revealing the porous structure embedded within the layer. This magnetic flux arises from the magnetoelastic composite (Fig. 2c), which is composed of SiO₂-coated magnetic particles embedded within a polymer matrix (Fig. 2d). The SiO₂ coating (Fig. 2e) contributes to material stability and supports safety^22,23. Upon pulsed magnetization, these magnetic particles maintain a stable and permanent magnetic field (Fig. 2f) within the magnetoelastic mesh layer, which could be deformed by the blood flow-induced dynamic pressure, causing detectable shifts in magnetic flux.

Fig. 2 |. Characterization of the smart magnetoelastic stent.

a, Photo of the magnetoelastic mesh layer. Scale bar, 2 mm. b, Normalized magnetic field mapping of the magnetoelastic mesh layer. Scale bar, 300 μm. c, Schematic illustration of magnetic particles embedded within a polymer matrix to form the magnetoelastic mesh. d, Transmission electron microscopy image of magnetic particles. Scale bar, 25 nm. Representative of three experiments. e, Transmission electron microscopy image of a magnetic particle showing a SiO₂ coating layer on the surface. Scale bar, 3 nm. Representative of three experiments. f, Magnetic hysteresis loop of the magnetic particles. g, Force required to stretch the magnetoelastic mesh layer and a solid membrane of identical dimensions without a mesh structure to 100% strain. Data are presented as mean ± s.d.; n = 3 experiments. h, Magnetic field variations under subtle pulsatile pressures. Data show the original readouts (light lines) and the corresponding processed signals (dark lines). i, Dependence of the magnetic flux density on thickness. Data are presented as mean ± s.d.; n = 5 experiments. j, Variation of magnetic field components in the smart stent across different magnetization directions. The z axis and x axis are defined as the vertical and axial direction of the smart stent, respectively. Data are presented as mean ± s.d.; n = 3 experiments. k, Consistent magnetic field of the magnetoelastic mesh maintained throughout the cycling process. n = 15 measurements from three experiments. Significance was determined by a two-tailed t-test. NS, not significant. Box plots show individual values, central line (mean), box limits (25th and 75th percentiles) and whiskers (minimum and maximum). l, Resistance variation of the soft coil in response to applied frequency. m, Photo of the fabricated soft coils with different dimensions. Scale bar, 3 mm. μT, microtesla; emu, electromagnetic unit of magnetic moment; mT, millitesla; Oe, Oersted; B, magnetic flux density; ΔB, change in magnetic flux density.

The mesh structure of the magnetoelastic layer was selected for three key advantages (Supplementary Note 3). First, the mesh design mimics the lattice architecture of the metal stent substrate used in this study, enabling integration into a unified construct and ensuring compatibility with catheter delivery procedures (Supplementary Note 3). Second, the mesh configuration prevents occlusion of arterial side branches covered by the smart stent. When deployed in regions containing arterial bifurcations, often termed ‘jailed’²⁴, the mesh design permits preserved blood flow into these vessels. By contrast, a solid membrane could partially or completely obstruct such branches, potentially causing ischemic complications (Supplementary Fig. 6 and Supplementary Note 4). Third, the mesh structure enhances mechanical responsiveness. By reducing the overall stiffness of the magnetoelastic layer, the mesh design requires less force to be stretched compared to a solid membrane of identical dimensions without a mesh structure (Fig. 2g). This increased mechanical responsiveness leads to detectable magnetic field variation under the subtle forces generated by pulsatile flow (Fig. 2h), thereby enhancing the fidelity and quality of the hemodynamic sensing signals.

We further examined the thickness of the magnetoelastic mesh. Compared with thicker layers, thinner layers, for example approximately 200 μm, allow easier loading of the smart stent onto the catheter tip and reduce disturbance to local hemodynamics after deployment (Supplementary Note 5). Excessively thick meshes may obstruct the catheter lumen and impede guidewire passage, thereby reducing compatibility with standard clinical catheter systems. However, thinner meshes substantially reduce the overall magnetic flux density (Fig. 2i), as the reduced thickness disrupts the alignment of internal magnetic particles²⁵. Based on these design principles, once the metal stent substrate is selected, mesh architectures with optimized designs can be developed by balancing anatomical constraints, biosafety considerations and sensing performance (Supplementary Note 2).

We also examined the magnetization orientations of the magnetoelastic mesh (Supplementary Fig. 7) by fabricating smart stents with three distinct configurations: axial, vertical and radial magnetization. We initially assessed the surface magnetic field distribution at the central cross-section of the smart stents with varying magnetization directions (Supplementary Fig. 8). The reconstructed data plots (Supplementary Fig. 9) show that the smart stent with vertical magnetization produces a pronounced peak value, accompanied by a spatially varying distribution. In addition, the radially magnetized configuration demonstrates a magnetic field with sufficient intensity and uniformity. We further quantified magnetic flux variations under the applied deformation, where the z axis is defined as the vertical direction and the x axis is defined as the axial direction of the smart stent. The absolute changes in magnetic field components along the x, y and z axes were recorded (Fig. 2j), with both the vertically and radially magnetized configurations exhibiting sufficient variation in the z axis component. Given the direction of blood flow-induced pressure and the position of the soft coil, either radially or vertically magnetized configurations were selected for the subsequent studies.

In addition, we evaluated the stability of the magnetoelastic mesh. We used a peristaltic pump to circulate fluid through the magnetoelastic mesh of the smart stent deployed in soft tubing, simulating physiological arterial pulsations. The magnetic field remained stable after prolonged peristaltic cycles (Fig. 2k). These findings confirm the long-term stability of the magnetic properties. Finally, the resistance and impedance of the soft coil were thoroughly investigated, with results shown in Fig. 2l. Notably, the resistance remained stable across a wide frequency range up to 10,000 Hz, enabling reliable detection of magnetic field variations induced by blood flow. In addition, the fabrication of the soft coils is customizable and compatible with industry-scale manufacturing processes (Fig. 2m). The optimally designed soft coils, matched to the corresponding smart stents, could be mass produced as ready-to-use components (Supplementary Table 1), supporting practicality and scalability for future clinical applications (Supplementary Note 2).

Characterization of the magnetoelastic stent in vitro

We characterized the hemodynamic sensing, stenosis detection and catheter delivery capabilities of the smart stent in vitro using a pump-driven pulsatile flow loop (Fig. 3a and Supplementary Note 6). Typical current signal waveforms generated by the smart stent are displayed in Fig. 3b. These waveforms originated from flow-induced pressure variation, causing the magnetic field fluctuations, thereby generating peak current 1 (PC1) and peak current 2 (PC2), respectively. An increase in pulse pressure²⁶ exerted on the smart stent enhances magnetic flux variation. This dynamic response amplifies the induced current signal, quantified as the absolute difference between PC1 and PC2, in direct proportion to the increase in pulse pressure (Fig. 3c), demonstrating the smart stent’s capability for hemodynamic sensing.

Fig. 3 |. In vitro investigation of smart magnetoelastic stents.

a, Schematics of deployed smart stent. b, Generated current signal from the smart stent highlighting key features. The yellow band indicates the time period of peak induction. c, The linear relationship between induced current signals and varying pulse pressure. The red line represents the linear regression fit, with the 95% confidence band (CB) and the 95% prediction band (PB) shown. d, Bland–Altman plots showing mean and ± 1.96 s.d. for calculated heart rate compared to reference values. e, Correlation of the smart stent’s current outputs (red) and the PC1 to PC2 time interval (gray) with different heart rates. Data are presented as mean ± s.d.; n = 9 pulsations. f, Comparison of current outputs (red) and peak-to-peak time (gray) of the smart stent with and without porcine tissue interposed between the vessel model containing the deployed smart stent and the external coil receiver. Data are presented as mean ± s.d.; n = 9 pulsations. g, Current signals indicating experimentally induced vessel stenosis. Inset is the enlarged waveforms of the generated current signals. h, Generated current values of the smart stent under normal flow and experimentally induced stenosis conditions. Data are presented as mean ± s.d.; n = 20 pulsations. i, Photo of guide catheters (6–8 Fr) configured for left and right coronary arteries. Scale bar, 6 mm. j, Photo of a guide catheter with a smart stent loaded at the tip. Scale bar, 0.9 mm. Inset is the enlarged view of the tip. k, Photo of a sheath catheter loaded with a smart stent and a catheter introducer. Scale bar, 1 mm. l, Schematic illustration of the catheter-based delivery procedure for deploying the smart stent into the target artery. AL, Amplatz Left; AR, Amplatz Right. Panels a and l have been created in BioRender. Liu, Z. (2025) https://biorender.com/tbie25f.

Heart rate (beats per minute (BPM)), simulated by the pulsatile pump, can be calculated from the generated current signals by analyzing the time interval between PC1 in two successive pulse waveforms (Supplementary Fig. 10). The heart rate measurements obtained from the smart stent-generated current signals align well with reference measurements (Fig. 3d). In addition, the results indicate that both the PC1–PC2 current value interval and the PC1–PC2 time interval showed relatively consistent behavior with limited fluctuation across different heart rates (Fig. 3e). To further evaluate transdermal sensing capabilities, a porcine tissue was positioned between the silicone vessels (with deployed smart stent) and the external coil receiver. This translated into a reduction in the current value generated, whereas the peak-to-peak time remained comparable across conditions (Fig. 3f). Finally, we evaluated the smart stent’s response to hemodynamic changes. Inducing stenosis in the silicone vessel reduced flow and correspondingly decreased the current signal (Fig. 3g), as reflected by a lower PC1–PC2 current value (Fig. 3h). Upon releasing the induced stenosis, the current signals promptly recovered, demonstrating the smart stent’s rapid responsiveness to hemodynamic variations. These results support the smart stent’s capability to detect stenosis.

We evaluated the compatibility of our smart stent with clinical catheter systems (Fig. 3i and Supplementary Note 7). Smart stents of various diameters were successfully loaded into the tip of Launcher guide catheters (Medtronic) for both left and right coronary artery configurations (LA6AL75, LA6AR10, LA7AL75, LA7AR10, LA8AL75 and LA8AR10) (Fig. 3j) as well as into AXS Infinity LS neurovascular catheters (Stryker Neurovascular) (Fig. 3k). The loaded smart stent could then be pushed out using a smaller catheter or introducer for deployment (Fig. 3l). These results indicate that our smart stent is compatible with common 6–8-Fr catheter systems (Supplementary Table 2), supporting its potential for clinical translation.

Characterization of the magnetoelastic stent in vivo

We conducted in vivo studies in a swine model to explore the smart stent’s clinical potential (Fig. 4a and Supplementary Note 8). The studies involved implanting the smart stent within swine arteries (Fig. 4b and Supplementary Fig. 11) and performing self-powered, continuous hemodynamic sensing (Fig. 4c).

Fig. 4 |. In vivo demonstration of smart magnetoelastic stents for stenosis diagnosis on a swine model.

a, Schematic illustration of in vivo studies using the swine model. b,c, Photograph (b) and schematic illustration (c) of the smart stent implanted in the swine carotid artery. Scale bar, 3 mm. d–f, Images of smart stent implantation using a catheter-based approach. A clinical catheter loaded with the smart stent was navigated over a glidewire under real-time fluoroscopy to the targeted position (d), and then the smart stent was deployed from the catheter using the catheter introducer to stabilize the proximal smart stent in position (e) while the catheter was withdrawn to release the smart stent (f). Scale bars, 7.5 mm. g, Diagnostic angiogram proximal to the smart stent deployment site. Scale bar, 7.5 mm. h, Pre-contrast image from a position proximal to the deployed smart stent in the left circumflex coronary artery. i, Post-contrast image from the same proximal position in the left circumflex coronary artery. The contrast dye flows through the artery, passing through the smart stent and extending to the coronary artery bifurcation. This demonstrates a clear vessel lumen and indicates the absence of blood clot formation at the site of the smart stent deployment. Scale bars, 9 mm. j, The smart stent-generated blood flow signals in vivo. Data are presented as the mean (red) with the error bar (gray). k, Fluoroscopy image showing the microwire alongside the deflated balloon catheter and the deployed smart stent. Scale bar, 5.5 mm. l, Fluoroscopy image showing the precise placement of the inflated balloon catheter within the smart stent to induce in-stent stenosis. Scale bar, 5.5 mm. m, Current signals indicating experimentally induced stenosis. Inset is the enlarged waveforms of the generated current signals. n, Current signals recorded during experimentally induced distal and proximal stenosis. Inset is the fluoroscopy images showing stenosis induced at distinct anatomical locations. o, Generated current values (red) and peak-to-peak times (gray) of the smart stent under the patency and instent stenosis situation. Data are presented as mean ± s.d.; n = 20 measurements. p, Acquired peak current values of pulsation signals from the smart stent under distal stenosis (red) and proximal stenosis (blue) conditions. The dashed line denotes the baseline current level, serving as a reference for assessing signal amplitude changes across different conditions. The yellow band highlights the period of induced stenosis, corresponding to balloon inflation. The red and blue lines aid visualization of trends in the data points. q, Alignment of the soft coil with the implanted smart stent in the swine carotid artery visualized under real-time fluoroscopy. Scale bar, 5 mm. Panels a and c have been partially created in BioRender. Liu, Z. (2025) https://biorender.com/tbie25f.

First, we delivered and deployed the smart stent by using a catheter approach. The clinical catheter’s access is gained through the established femoral arterial sheath (Supplementary Fig. 12 and Extended Data Fig. 1a,b). Specifically, the clinical catheter loaded with the smart stent was navigated over a glidewire (Extended Data Fig. 1c), under real-time fluoroscopy (Supplementary Fig. 13), from the femoral arterial sheath to the targeted position of the swine artery (Fig. 4d). The smart stent was then deployed by implementing the catheter introducer to stabilize the proximal smart stent in position (Fig. 4e) while withdrawing the catheter to release the smart stent into the artery (Fig. 4f and Supplementary Video 1). Angiographic imaging with iodinated contrast dye (Supplementary Fig. 14) from a position proximal to the smart stent deployment site was obtained (Fig. 4g), indicating no apparent flow limitation or damage to the vessel due to implantation. In addition, the mesh structure of the smart stent ensures side branch access, preventing obstruction of blood flow to small side vessels (Extended Data Fig. 1d,e). These results support the smart stent’s compatibility with angioplasty procedures, allowing its deployment in diverse vascular locations (Extended Data Fig. 1f,g). We also demonstrated our smart stent system’s applicability to the coronary artery interventions (Supplementary Note 9). Using a 6-Fr Launcher catheter system (Supplementary Fig. 15), the smart stent was successfully deployed into the left circumflex coronary artery via a percutaneous approach (Fig. 4h). Angiographic imaging with iodinated contrast dye showed intact blood flow through the stented segment (Fig. 4i), with no evidence of slow-flow phenomena or filling defects compared to baseline images prior to deployment or to the non-stented portions of the artery (Supplementary Fig. 16). These results indicate that the smart stent did not induce appreciable obstruction or turbulence at the macroscopic level at the coronary artery, including in a jailed coronary artery bifurcation. No acute thrombosis was observed, and the arteries remained patent, with no evidence of clot formation. The established delivery and deployment method, leveraging a clinical catheter system, offers a widely applicable approach with potential for integration into clinical practice.

Next, the implanted smart stents not only mechanically supported the swine carotid artery opening (Supplementary Video 2) but also enabled self-powered and continuous hemodynamic sensing. The smart stents effectively and sensitively captured hemodynamic changes, as demonstrated by the blood flow signals generated (Fig. 4j). Specifically, the heart rate in BPM calculated from the smart stent-generated blood flow signals aligns closely with the reference values (Supplementary Fig. 17). We investigated the smart stent for ISR diagnosis by inducing stenosis in the carotid artery that housed the implanted device. An endovascular balloon catheter was positioned within the target artery containing the smart stent via the femoral arterial sheath (Fig. 4k). The balloon was strategically placed at specific locations over a microwire under fluoroscopic guidance (Extended Data Fig. 2a–d) and inflated to induce controlled stenosis at distinct sites (Fig. 4l). The smart stent sensing signals were recorded in real time, including responses to the induced in-stent stenosis (Fig. 4m), as well as distal (far from the heart) and proximal (closer to the heart) stenosis (Fig. 4n), respectively. The results show that stenosis led to hemodynamic changes, which could be detected by the smart stent. Specifically, the induced in-stent stenosis resulted in reduced blood flow, as reflected by a decrease in the smart stent’s peak-to-peak current amplitude (Fig. 4o). In addition, slight fluctuations in the peak-to-peak time interval were observed, which may be attributed to the presence of the endovascular balloon catheter within the smart stent, partially impeding its mechanical responsiveness under pulsatile flow. These findings suggest that the smart stent can detect stenosis-induced hemodynamic changes through analysis of sensing signals. Additionally, we induced varying degrees of stenosis by inflating the balloon to different extents under live fluoroscopy (Extended Data Fig. 2e–h). The hemodynamic signals recorded by the smart stent decreased correspondingly with high resolution, effectively capturing the progression of stenosis (Extended Data Fig. 2i). These results suggest that smart stent’s high-fidelity sensing signals have the potential to correlate with established clinical metrics (Supplementary Note 10), such as the narrowest diameter²⁷, minimal lumen area (MLA)²⁸ and fractional flow reserve (FFR)²⁹, for stenosis evaluation.

We also investigated the smart stent’s sensing signals in relation to the location of stenosis (Supplementary Fig. 18). Distal stenosis resulted in increased current signal values (Fig. 4p, red). This effect may be related to downstream obstruction: in distal stenosis, blood flows through the smart stent before encountering the blockage, which may lead to localized pressure buildup that enhances smart stent deformation and amplifies the sensing signal. Conversely, proximal stenosis (Fig. 4p, blue), by obstructing blood flow before it reaches the smart stent, decreases the blood flow, leading to reduced current signal values. These results support the smart stent’s ability to detect stenosis and differentiate between proximal and distal lesions relative to its location (Supplementary Note 10).

We further explored AI-assisted interpretation of smart stent sensing signals^30–32, particularly considering that future clinical translation of the smart stent will continuously generate large volumes of real-time hemodynamic data that would be infeasible for clinicians to manually interpret due to time and workload constraints (Supplementary Note 11). To address this, in vivo smart stent blood flow sensing data were first combined with simulated signals to construct a dataset (Extended Data Fig. 3a). Second, an AI-assisted hemodynamic signal interpretation pipeline was developed. The blood flow sensing signals were initially subjected to standardized time-series preprocessing (Extended Data Fig. 3b), including resampling, windowing, class balancing, normalization and optional continuous wavelet transform (CWT) processing. Multiple neural network architectures were subsequently trained, including one-dimensional convolutional neural network (CNN), one-dimensional CNN + long short-term memory (LSTM) and two-dimensional CNN (Extended Data Fig. 3c), to classify the blood flow sensing signals (Extended Data Fig. 3d). Testing results and performance metrics, including Precision, Recall, F1 Score and Specificity (Supplementary Note 12), demonstrated that these models distinguished sensing signals corresponding to different conditions. These findings highlight the potential of integrating AI algorithms with the smart stent to enable automated, continuous surveillance of abnormal flow dynamics and timely detection of stenosis. It is important to note, however, that the real-world prevalence of stenosis is expected to be much lower³³ than in our current dataset. This discrepancy may reduce the model’s precision in future clinical deployment. Therefore, future work will focus on validating and recalibrating the AI-assisted hemodynamic signal interpretation pipeline using prevalence-weighted evaluation to ensure robust, real-world applicability.

In addition, we investigated the robustness and practical applicability of the smart stent. To ensure consistent signal morphology, we used live fluoroscopy to precisely align the implanted smart stent with the soft coil (Fig. 4q). This alignment is particularly critical for practical applications, as any misalignment may introduce artifacts into the sensing signals. Also, increasing the distance between the smart stent and the soft coil attenuated the magnetic field variation, resulting in reduced sensing signal amplitude (Supplementary Note 13). To address variability arising from individual anatomical differences, such as arterial wall thickness or tissue depth, a personalized calibration procedure can be performed immediately after stent deployment (Supplementary Note 14). This calibration can also serve as a baseline for future postoperative adjustments. Notably, the smart stent detects stenosis by identifying abnormal flow patterns relative to its own baseline over time. Our approach does not rely on comparisons across different devices or anatomical sites with varying locations. Finally, we conducted independent swine studies using the same platform across multiple timepoints and anatomical variations, yielding consistent results that support the smart stent’s scalability and reliability and suggest its potential for clinical translation.

Biosafety of the smart magnetoelastic stent

Long-term biosafety within the host environment is crucial for clinical translation. To assess the feasibility and potential of prolonged implantation, we evaluated the smart stent’s biosafety through both in vitro and in vivo studies (Supplementary Note 15).

We first cultured human PBMCs with either the smart stents or the clinically approved stent (control group). After 3 days of incubation, the culture supernatants were collected, and the human cytokine profiles were analyzed. The results showed that the concentrations of key pro-inflammatory and anti-inflammatory cytokines³⁴, including IFNγ (Fig. 5a), IL-2 (Fig. 5b), IL-6 (Fig. 5c), IL-10 (Fig. 5d), TNF (Extended Data Fig. 4a) and VEGF-A (Extended Data Fig. 4b), exhibited no statistically significant differences between the two groups. The similar levels of these cytokines indicate that the smart stent did not induce additional adverse immunological activation in vitro relative to the medical devices used as controls, supporting its biocompatibility.

Fig. 5 |. In vitro biosafety studies.

a, Concentration of human IFNγ. Data are presented as mean ± s.d.; n = 4 experiments. Significance was determined by a two-tailed t-test. b, Concentration of human IL-2. Data are presented as mean ± s.d.; n = 4 experiments. Significance was determined by a two-tailed t-test. c, Concentration of human IL-6. Data are presented as mean ± s.d.; n = 4 experiments. Significance was determined by a two-tailed t-test. d, Concentration of human IL-10. Data are presented as mean ± s.d.; n = 4 experiments. Significance was determined by a two-tailed t-test. e, UMAP analysis of immune cell populations, including T cells, monocytes, B cells and natural killer (NK) cells. f, Expression levels of LTA across immune cell clusters. g, Expression levels of CD40 across immune cell clusters. h, Expression levels of IL1B across immune cell clusters. i, Expression levels of CCL4 across immune cell clusters.

We further performed scRNA-seq on PBMCs cultured with either the smart stent or the clinically approved stent (control group) to assess potential immunogenic differences at the cellular level. The uniform manifold approximation and projection (UMAP) analysis of immune cell populations, including T cells, monocytes, B cells and natural killer cells, revealed similar distributions between the two groups (Fig. 5e and Extended Data Fig. 4c). Furthermore, similar expression levels of key inflammatory markers were observed within each immune cell cluster, including LTA (Fig. 5f), CD40 (Fig. 5g), IL1B (Fig. 5h) and CCL4 (Fig. 5i). Additional analysis of T cell cluster indicated similar distributions of naive CD4⁺ T cells, naive CD8⁺ T cells and effector memory CD4⁺ T cells in the smart stent and control groups (Extended Data Fig. 5). These findings further support the immunological safety and biocompatibility of the smart stent.

We conducted animal studies to evaluate the in vivo biocompatibility. In the smart stent group, a magnetoelastic layer was implanted in the femoral vessel region of rats, whereas the control group underwent the same surgical procedures with the implantation of a medical membrane (Supplementary Note 15). Positron emission tomography/computed tomography (PET/CT) scans with [¹⁸F] fluorodeoxyglucose (FDG) at 28 days after implantation revealed similar uptake values between the smart stent group and the control group in these living rats (Supplementary Fig. 19). These similar uptake values suggest the absence of additional inflammation localized to the magnetoelastic layer implantation site. In addition, we performed comprehensive immune profiling to evaluate T cell phenotypes and macrophage polarization. At the experimental endpoint, we collected both the local muscle tissue at the implantation site as well as blood samples from both experimental and control groups. Fluorescence-activated cell sorting (FACS) analysis was conducted to statistically assess immune responses in the smart stent group and the control group (Fig. 6a). Immune profiling included quantification of CD4⁺ and CD8⁺ T cell populations in the blood (Fig. 6b) and muscle (Fig. 6c) samples, naive and activated CD4⁺ T cells in the blood (Fig. 6d) and muscle (Fig. 6e) samples, naive and activated CD8⁺ T cells in the blood (Fig. 6f) and muscle (Fig. 6g) samples, regulatory and helper T cells in the blood (Fig. 6h) and muscle (Fig. 6i) samples as well as the mean fluorescence intensity (MFI) of pro-inflammatory (M1) and anti-inflammatory (M2) macrophage markers in the blood (Fig. 6j) and muscle (Fig. 6k) samples. No statistically significant differences were observed between the smart stent and control groups across the measured immune parameters, suggesting that implantation of the magnetoelastic layer was not associated with appreciable immune activation compared to the medical devices used as controls.

Fig. 6 |. In vivo biosafety studies.

a, Gating strategies used for FACS analysis to evaluate T cell phenotypes. b,c, Percentages of CD4⁺ (left) and CD8⁺ (right) T cells in blood (b) and muscle (c) samples. d,e, Percentages of naive (left) and activated (right) CD4⁺ T cells in blood (d) and muscle (e) samples. f,g, Percentages of naive (left) and activated (right) CD8⁺T cells in blood (f) and muscle (g) samples. h,i, Percentages of regulatory (left) and helper (right) T cells in blood (h) and muscle (i) samples. j,k, MFI of pro-inflammatory (CD80⁺, left) and anti-inflammatory (CD163⁺, right) macrophage markers expressed on the CD11b/c⁺ rat macrophages in blood (j) and muscle (k) samples. Box plots show individual values, central line (mean), box limits (25th and 75th percentiles) and whiskers (minimum and maximum); n = 8 experiments (b–k). Significance was determined by a two-tailed t-test. T_reg, regulatory T.

Finally, we performed elemental analysis in the animal study. The measured Fe concentrations in collected kidney and heart tissues from the smart stent group and the control group showed no statistically significant differences (Supplementary Fig. 20). Given that the magnetoelastic mesh layer is the only implanted component that contains additional Fe, the similar iron levels between the two groups suggest the absence of detectable systemic accumulation in major organs, supporting the biosafety of the magnetoelastic mesh.

In brief, these studies indicate that the magnetoelastic mesh does not induce detectable additional immune or inflammatory responses compared to the medical devices used as controls. Given that the smart stent is also compatible with catheter delivery procedures, these findings support its biosafety, suggesting no evidence of additional risks or exacerbation of inherent risk factors associated with stenting (angioplasty) procedures (Supplementary Note 16).

Discussion

We developed a smart magnetoelastic stent as a self-powered diagnostic platform for continuous ISR diagnosis, extending the functionality of conventional metal stents beyond mechanical support. The device was delivered and deployed in vivo using a clinical catheter and identified induced stenosis through AI-assisted signal analysis. Comprehensive in vivo and in vitro studies, including immune profiling, cytokine analysis and scRNA-seq, supported the biosafety of the smart stent. To further contextualize its performance, we conducted an evaluation highlighting its attributes across platform design, clinical translation and biosafety (Extended Data Fig. 6). Together, these attributes suggest the potential of the smart stent as a practical diagnostic technology for atherosclerosis management (Supplementary Note 17).

This work successfully demonstrated the smart stent’s hemo-dynamic sensing and stenosis detection capabilities as well as its compatibility with catheter-based delivery. Future work will focus on translating the smart stent to more complex anatomies, such as coronary arteries with calcified lesions, distal targets or tortuous vessels (Supplementary Note 18). Achieving this will require loading the smart stent into smaller-diameter clinical catheters to facilitate access to challenging or distal coronary segments. To preserve signal fidelity in the dynamic coronary setting, the development of adhesive coils will be critical to minimize motion artifacts by reducing relative movement (Supplementary Note 19). Furthermore, signal transmission through the sternum and ribs could be enhanced by integrating an implantable intermediate transmission module, enabling improved signal readout by positioning the receiving coil closer to the smart stent.

It is important to note that the current results of our developed AI-assisted hemodynamic signal interpretation pipeline are based on a well-defined dataset collected under controlled laboratory conditions during the preclinical stage. In future clinical translation³⁵, as larger and more diverse datasets with realistic prevalence of stenosis become available, we plan to further investigate the performance of different model architectures and explore both transfer learning and training-from-scratch strategies within our pipeline to rigorously evaluate model generalizability and robustness. In addition, investigating AI-assisted interpretation of smart stent sensing signals to distinguish subtypes of stenosis, such as different locations and varying severity levels, could enhance the platform’s diagnostic potential across diverse clinical scenarios. Furthermore, integrating more advanced architectures³⁶, such as transformer-based models, may further enhance the performance, scalability and reliability of AI-assisted signal interpretation, particularly when applied to large-scale, heterogeneous and personalized clinical data.

In addition, our in vivo studies in swine models suggest the immediate blood compatibility of the smart stent, with no observable evidence of acute thrombosis. Future work could focus on further reducing thrombogenicity during prolonged in vivo application, for example, by loading antithrombotic agents onto the stent surface (Supplementary Note 20). Finally, our in vivo rat studies and in vitro studies support the biosafety of the smart stent. Future work could include long-term studies in larger animal models or advanced organoid systems to further assess smart stent–vascular interactions, which will accelerate the clinical translation.

We envision that, with continued advancements in research and progress through regulatory pathways (Supplementary Note 21), the smart stent technology could be translated into clinical applications, potentially offering long-term benefits to patients and helping to reduce the financial burden (Supplementary Note 22). Beyond its potential clinical utility, the smart stent may also serve as a platform technology for cardiovascular research, enabling studies of complex coronary hemodynamics, including variations in microvascular resistance under changing metabolic demand (Supplementary Note 23). Moving beyond cardiovascular and cerebrovascular care, the smart stents may be extended to various clinical scenarios, such as ureteral stents to ensure the flow of urine in cases of obstructed ureters³⁷ and gastrointestinal stents to track food passing through the esophagus³⁸. Furthermore, integrating smart stents with drug-eluting technologies holds the potential to decrease the risk of ISR³⁹. The use of smart stents in conjunction with external magnetic fields could enable a new field of endovascular robotics, inducing mechanical vibrations to remove the re-stenotic plaque or wirelessly deliver drugs. We anticipate that our smart stents could help catalyze the development of transdisciplinary research strategies with expansive applications across cardiovascular diseases.

Methods

Fabrication

(1) Smart magnetoelastic stents. Customized printing or spin-coating techniques were used to fabricate the magnetoelastic tubes or membranes, consisting of SiO₂-coated magnetic particles embedded in silicone (Supplementary Note 2). A laser cutter (Universal Laser Systems, PLS6.150D) was used to remove excess material, creating the desired geometry and shape of the magnetoelastic mesh layer. Medical instant adhesive (Henkel, LOCTITE medical device instant adhesive) could be used as needed to bond the magnetoelastic mesh layer to the clinically approved self-expanding metal stent (Stryker Neurovascular, Wingspan Stent System). The smart stent was magnetized at impulse fields (2.65 T) by an impulse magnetizer (ASC Scientific, IM-10–30). (2) Soft coils. The soft coil was fabricated by encapsulating a custom-fabricated copper coil, manufactured to the specified dimensions, wire diameter and number of turns, within a soft silicone matrix.

Structural and mechanical characterization

The structural features of the smart magnetoelastic stent were examined using transmission electron microscopy (Titan Krios High-Res Cry-oEM). Magnetic flux density was quantified using a digital Gauss meter (Kanetec Co., Ltd., TM-801) and a custom-designed Hall sensor array (Melexis, MLX 90393). Magnetic hysteresis characteristics were evaluated using a superconducting quantum interference device magnetometer (Quantum Design, MPMS3). Mechanical properties were assessed using a universal testing machine (Chatillon Force Measurement).

Electrical performance measurement

Electrical characterization was performed using a system comprising a pulsatile pump (FlowTek 125) and a low-noise current preamplifier (Stanford Research Systems, SR570) (Supplementary Note 6). Templates for silicone vessels and stenosis models were produced using a high-speed 3D printer (Creality K1 Max).

Loading the smart magnetoelastic stent

The smart stent was loaded into clinical catheters (Stryker Neurovascular, AXS Infinity LS; Medtronic, Launcher guide catheter) using customized crimping tools (Supplementary Note 7). During this process, the diameter of the smart stent tip segment was reduced to less than the inner diameter of the clinical catheter. The reduced-diameter segment was aligned and inserted into the catheter tip, followed by sequential compression of the remaining segments until the stent was fully loaded into the catheter.

Machine learning

Smart stent-generated sensing signals were labeled according to their experimental origin and corresponding condition and subsequently divided into training and testing subsets at the start of the analysis pipeline to prevent data contamination (Supplementary Note 12). A standardized preprocessing pipeline was established to ensure data consistency and reproducibility, comprising the following sequential steps: frequency detection and resampling, windowing, class balancing, normalization, train/validation split and optional CWT processing. Multiple neural network architectures were then trained, including one-dimensional CNN, one-dimensional CNN + LSTM and two-dimensional CNN, to classify the blood flow sensing signals across distinct physiological states. Model performance was evaluated on the testing set using standard classification metrics, including Precision, Recall, F1 Score and Specificity. All training and evaluation procedures were executed on a single L40s GPU.

Animal studies on swine

Yorkshire swine (n = 5), 40–60 kg and 3–4 months old, sourced from Premier BioSource, were used in these studies. Each swine received daily antiplatelet therapy consisting of aspirin (325 mg) and clopidogrel (75 mg) before smart stent implantation, following standard practice. During the procedure, the animals were maintained under general anesthesia with continuous monitoring of vital parameters, including heart rate, blood pressure, oxygen saturation and body temperature. Before stent deployment, intravenous heparin was administered to achieve a target activated clotting time of 200–300 seconds. Smart stents were implanted into the target arteries (Supplementary Note 24). After deployment, hemodynamic sensing signals were recorded using the implanted smart stent in conjunction with the soft coil and a current preamplifier (SR570). All procedures were approved by the Institutional Animal Care and Use Committee of the University of California, Los Angeles (UCLA).

Endovascular approach for implantation

Access to the target artery was gained through the established femoral arterial sheath (Supplementary Note 25). A standard catheter delivery system (Stryker Neurovascular, AXS Infinity LS) with a loaded smart stent was navigated over a glidewire under real-time fluoroscopic guidance. Once positioned in the artery of interest, the catheter introducer was used to stabilize the proximal end of the smart stent and to propel the device as needed. With the proximal end of the smart stent secured by the introducer, the catheter was then carefully withdrawn. As the catheter was retracted, it gradually released the smart stent, allowing it to expand and to conform to the inner walls of the artery.

Deploying the smart magnetoelastic stent into the coronary artery

The smart magnetoelastic stent was loaded into a 6-Fr Launcher catheter system (Medtronic), which was advanced through an 8-Fr femoral access sheath. To prevent ventricular fibrillation/arrhythmias that commonly occur during coronary procedures in swine, amiodarone (1.5 mg kg⁻¹ intramuscular) was administered at the start of the procedure, with an additional approximately 3 mg kg⁻¹ intravenous if arrhythmias occurred. Lidocaine (2 mg kg⁻¹ intravenous) was given every 20 minutes throughout the endovascular intervention. The 6-Fr catheter with a loaded smart stent was navigated over a guidewire under fluoroscopic guidance. Once positioned in the coronary artery, a smaller catheter was used to stabilize the proximal end of the smart stent and to propel the device forward as needed into the vessel lumen. With the proximal end of the smart stent secured, the 6-Fr catheter was then carefully withdrawn. As the catheter was retracted, it gradually released the smart stent, allowing it to expand and to conform to the inner walls of the left circumflex coronary artery. To facilitate sensing signal collection, a median sternotomy was performed to expose the heart, allowing the soft coil to be positioned in close proximity to the deployed smart stent.

Diagnostic angiogram

Arterial access was obtained via the femoral artery in swine, after which a 5-Fr diagnostic glide catheter and glidewire were advanced to selectively access various neck arteries as needed (Supplementary Note 26). Once the catheter was appropriately positioned, the glidewire was withdrawn, and angiography was performed by injecting an iodinated contrast agent (Omnipaque (iohexol) 300). Fluoroscopic images were acquired to assess contrast distribution within the vasculature, and digital subtraction angiography was used to enhance visualization of the contrast-filled vessels. Throughout the procedure, catheters were continuously flushed with heparinized saline (5,000 U of heparin in 1 liter of 0.9% NaCl) to minimize the risk of thrombosis.

Swine model with vessel stenosis

An endovascular balloon microcatheter (Stryker Neurovascular, TransForm Occlusion Balloon Catheter) was introduced over a microwire and was positioned in the lumen of the vessel at the location of interest (proximal, distal and in-stent stenosis) using fluoroscopic guidance (Supplementary Note 27). The balloon was inflated in a small, incremental fashion to induce internal vessel stenosis under live fluoroscopy.

PBMC biosafety study

Human PBMCs from healthy donors were sourced through the UCLA CFAR Virology Core Laboratory under federal and state compliance, with all samples fully deidentified. The in vitro biosafety study was assessed by co-culturing the devices with PBMCs in cell culture plates, using lymphocyte culture medium which comprised RPMI 1640 supplemented with 10% (v/v) FBS, and 1% (v/v) penicillin–streptomycin–glutamine. PBMCs were maintained in culture for 3 days, after which cell viability was quantified via flow cytometry, and live cells were identified as e506 live-cell viability-negative cells. Culture supernatants were collected to assess secreted cytokine profiles using a 38-plex Luminex assay (Millipore), performed in accordance with the manufacturer’s protocol, and fluorescence was analyzed on a Luminex 200 platform.

Single-cell RNA-seq

Human PBMCs from healthy donors were cultured with the devices for 3 days following the same protocol described above. Freshly collected samples were immediately transported to the UCLA Technology Center for Genomics & Bioinformatics Core for library preparation and scRNA-seq. Cell counts were determined using a Countess II automated cell counter (Invitrogen/Thermo Fisher Scientific). A total of 10,000 cells per experimental group were loaded onto the Chromium platform (10x Genomics), and libraries were generated using the Chromium Next GEM Single Cell 3′ Kit v3.1 and Chromium Next GEM Chip G Single Cell Kit (10x Genomics) according to the manufacturer’s instructions. Library quality was assessed using D1000 ScreenTape on a 4200 TapeStation (Agilent Technologies), and sequencing was performed on an Illumina NovaSeq platform with the NovaSeq S4 Reagent Kit (100 cycles; Illumina). For cell clustering and annotation, the merged digital expression matrix was generated by Cell Ranger on the 10x cloud analysis platform, which was then analyzed using the Seurat R package (version 4.3.0) following recommended workflows. Low-quality cells were filtered, and the expression matrix was normalized using the NormalizeData function. Variable features were identified across datasets using FindVariableFeatures and SelectIntegrationFeatures. Batch effects were corrected using FindIntegrationAnchors and IntegrateData based on the selected feature genes. The integrated dataset was then processed through the standard Seurat pipeline for dimensionality reduction, clustering and gene expression analysis.

Animal studies on rats

Sprague–Dawley rats (12 weeks old, Charles River Laboratories) were used in these studies. During surgical procedures, rats were anesthetized with isoflurane delivered via inhalation (Supplementary Note 28). The femoral vessel region was surgically exposed, and the surrounding tissues were carefully separated to allow access. The device was then implanted in the femoral vessel region of rats. The surgical wound was closed with a 4–0 nylon suture. After surgery, animals were monitored daily for the first 3 days and every other day thereafter. Postoperative analgesia was maintained through continued administration of Carprofen gel cups. At the study endpoint, rats were euthanized, and relevant tissues and blood samples were collected for subsequent analyses. All procedures were approved by the UCLA Institutional Animal Care and Use Committee.

PET/CT on rat models

Rats were maintained under anesthesia during administration of FDG, which was injected through the caudal vein using a 29-gauge needle (Supplementary Note 29). Detailed three-dimensional PET imaging was performed using a PET scanner (GNECT) to visualize FDG distribution. Images were analyzed in AMIDE and Dragonfly 2022 software.

FACS analysis

(1) Sample preparation. Rat tissue samples were minced and mechanically homogenized and then filtered and treated with red blood cell lysis to obtain a single-cell suspension. For rat blood samples, red blood cells were lysed directly to generate a single-cell suspension. (2) Antibodies. For immune profiling, a customized panel of fluorophore-conjugated antibodies was used. These antibodies included Pacific Blue-conjugated rat CD45 (1:200; clone OX-1; cat. no. 202226), PE-Cy7-conjugated rat CD3 (1:200; clone 1F4; cat. no. 201421), FITC-conjugated rat CD4 (1:200; clone W3/25; cat. no. 201505), APC-conjugated rat CD8α (1:200; clone G28; cat. no. 200610), PE-conjugated rat CD62L (1:200; clone OX-85; cat. no. 202912), APC-conjugated rat CD25 (1:200; clone OX-39; cat. no. 202113), FITC-conjugated rat CD11b/c (1:200; clone OX-42; cat. no. 201805) and PE-conjugated rat CD80 (1:200; clone 3H5; cat. no. 200205), all obtained from BioLegend. A FITC-conjugated donkey anti-rabbit IgG secondary antibody (1:500; minimal cross-reactivity; BioLegend, 406403) was used to detect a rabbit recombinant monoclonal CD163 antibody (1:200; clone EPR19518; Abcam, ab182422). Cell viability was assessed using Fixable Viability Dye eFluor506 (1:500; eBioscience). (3) FACS analysis. Cells were stained according to established flow cytometry protocols and in accordance with the manufacturers’ recommendations for each antibody. Isotype-matched controls were included to assess staining specificity. Experiments were performed on a MACSQuant Analyzer 10 (Miltenyi Biotec) using standard operating settings, and data were analyzed with FlowJo version 9 (BD Biosciences) software.

Element analysis

The elemental analysis was performed by using Inductively Coupled Plasma Mass Spectrometry analysis (PerkinElmer, NexION 2000). Each sample was measured in triplicate, with background signals subtracted during data processing.

Statistics and reproducibility

Statistical analyses were performed using OriginLab. Statistical significance was assessed using a two-tailed t-test. Data are presented as mean ± s.d. unless otherwise indicated. Biosafety studies were analyzed in a blinded manner. Conclusions were drawn based on results from multiple experiments.

Extended Data

Extended Data Fig. 1 |. Catheter delivery of the smart stent.

a-b, Illustration (a) and photo (b) showing clinical catheter access via the established femoral arterial sheath for smart stent delivery. Scale bar, 12 mm. c, Delivery catheter navigation over glide wire. Real-time fluoroscopy is used to visualize the progression and position of the glide wire and catheter. Once the glide wire is successfully navigated to the desired location, it serves as a track along which the clinical catheter loaded with the smart stent can be advanced. Scale bar, 4 mm. d-e, Diagnostic angiograms evaluating a side branch jailed by the smart stent. (d) Post-implantation and (e) pre-implantation angiograms demonstrate that the smart stent does not obstruct flow to the side branch. Preserving anterograde blood flow in jailed branches is important for the utilization of the smart stent in a variety of vascular locations. The smart stent’s mesh structure is specially designed to prevent the complete obstruction of flow across the smart stent to the side branches. Scale bars, 3 mm. f-g, Versatile deployment of the smart stent in multiple locations. Notably, the appearance of the smart stent in the image (g) is lighter (fainter) in color compared to image (f). This variation in visualization is attributed to differences in tissue thickness and the distinct anatomical areas being imaged. Scale bars, 6 mm. Panel a has been partially created in BioRender. Liu, Z. (2025) https://biorender.com/tbie25f.

Extended Data Fig. 2 |. Induced stenosis in the pig carotid artery in vivo.

a, Fluoroscopic image showing the normal patent state with the deflated balloon catheter positioned in the pig carotid artery in vivo. b, Precision placement of the balloon catheter at the upstream position to induce the proximal stenosis. c, Precision placement of the balloon catheter inside the smart stent to induce the in-stent stenosis. d, Precision placement of the balloon catheter at the downstream position to induce the distal stenosis. Scale bars, 4 mm (a-d). e-h, Images showing incremental inflation of the balloon to induce controlled stenosis within the smart stent, starting with mild narrowing (e), gradually occluding more of the lumen (f–g), and finally reaching a severe level of stenosis (h). Scale bars, 5.6 mm (e-h). i, Hemodynamic sensing signals recorded during gradual balloon inflation to simulate progressively increasing stenosis severity.

Extended Data Fig. 3 |. AI-assisted interpretation of smart stent sensing signals.

a, Representative smart stent sensing signals included in the dataset. b, Standardized time-series preprocessing pipeline, including resampling, windowing, class balancing, normalization, train–validation split, and optional continuous wavelet transform (CWT) processing. c-d, Multiple neural network architectures were subsequently trained, including 1D Convolutional Neural Network (CNN), 1D CNN + Long Short-Term Memory (LSTM), and 2D CNN (c), to classify the blood flow sensing signals (d). Abbreviations: Conv., convolutional layer; FC layer, fully connected layer.

Extended Data Fig. 4 |. Quantification of human inflammatory cytokines and cell type markers.

a, Concentration of human TNF. b, Concentration of human VEGF-A. Data are presented as mean ± s.d. in a and b; n = 4 experiments. Significance is determined by a two-tailed t-test. N.S., not significant. c, Cell-type gene markers for each cluster, with the percentage of expressing cells and the average expression level used to identify cell types.

Extended Data Fig. 5 |. Clustering analysis of T cell subpopulations.

a, Uniform Manifold Approximation and Projection (UMAP) analysis of T cells, including naive CD4⁺ T cells, naive CD8⁺ T cells, and effector memory CD4⁺ T cells. Naive CD4⁺ T cells were identified as CD3E⁺, CD4⁺, CCR7^high, TCF7^high, SELL^high, CD69^low, FAS⁻, IL7R⁺ T cells. Naive CD8⁺ T cells were identified as CD3E⁺, CD8A⁺, CCR7^high, TCF7^high, SELL^high, CD69^low, FAS⁻, IL7R⁺ T cells. Effector memory CD4⁺ T cells were identified as CD3E⁺, CD4⁺, CCR7^low, TCF7^low, SELL^low, CD69^high, FAS⁺, IL7R⁺ T cells. b, Cell-type gene markers for each cluster, with the percentage of expressing cells and the average expression level used to identify cell types.

Extended Data Fig. 6 |. Evaluation of the attributes of the smart stent.

The smart stent represents a self-powered diagnostic platform for continuous ISR diagnosis, extending the functionality of conventional metal stents beyond mechanical support. An evaluation is presented to outline its attributes across (1) platform technology, (2) clinical translation, and (3) biosafety. Together, these attributes suggest the potential of the smart stent as a practical diagnostic technology that could complement existing stent technologies and support improved management of atherosclerosis. Partially created in BioRender. Liu, Z. (2025) https://biorender.com/tbie25f.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s44161-025-00773-4.

Data availability

Data supporting the results in this study are available within the article and its Supplementary Information. Single-cell RNA-seq datasets generated for this paper are located in the Gene Expression Omnibus repository under accession number GSE312793. Source data are provided with this paper.

Code availability

The machine learning code in this study is available via GitHub at https://github.com/JCLABShare/STENT.