Coordinated control of miniature magnetic pumps for liquid sampling and delivery in a knee implant

SUMMARY

Total knee arthroplasty (TKA) is widely used to surgically treat knee arthritis; however, clinical results remain suboptimal, with issues such as infection, aseptic loosening, and implant failure often requiring invasive revision surgery. Existing implants also lack integrated sensing and drug-delivery capabilities due to the challenge of embedding wireless pumps within constrained spaces. Here, we report a next-generation TKA device that integrates miniature magnetic pumps and wireless sensors for closed-loop, on-demand liquid delivery. The pumps are independently actuated by external magnetic fields below 15 mT, enabling simultaneous biofluid sampling and therapeutic release. Joint biofluid properties are monitored continuously, and ~70 μL of liquid can be refilled and delivered via magnetically driven pumping. The implant withstands compressive stresses of ~26.7 MPa, exceeding physiological loads during daily activities. Closed-loop functionality is validated in an ex vivo porcine bone. This device enables continuous implant health monitoring and targeted anti-inflammatory therapy, advancing intelligent orthopedic implants.

In brief

This work introduces a fundamental mechanism for fully remote, wireless control of multiple millimeter-scale magnetic pumps for orthopedic implants. The pumps enable coordinated, controlled liquid sampling and release, addressing limitations of conventional knee implants that only sense and carry inflammation risk. The miniature pump design and magnetic actuation and charging system were validated in a benchtop phantom and ex vivo porcine model, demonstrating potential for real-time monitoring and targeted inflammation intervention through therapeutic release, thereby reducing surgical burden and enabling continuous implant surveillance.

Graphical Abstract

INTRODUCTION

A total knee arthroplasty (TKA) is a procedure that implants a medical device to replace an arthritic, painful knee joint.^1–3 This procedure is designed to restore knee function and alleviate pain in individuals suffering from severe osteoarthritis,⁴ rheumatoid arthritis,⁵ or post-traumatic cartilage damage. A TKA implant becomes necessary when severe pain, stiffness, or loss of function significantly impacts a person’s quality of life and cannot be effectively managed with non-surgical treatments. Following TKA implant placement, complications such as inflammation^6,7 pose significant risks, including infection, pain, swelling, implant failure, bone loss, reduced range of motion, allergic reactions, and thrombosis,⁸ which may ultimately require revision surgery.^9,10 Although preventive measures such as prophylactic antibiotics and anti-inflammatory treatments¹¹ exist, determining the optimal timing for these interventions remains uncertain. By the time symptoms arise and a computed tomography (CT) scan is needed to assess the surrounding areas,¹² inflammation may have already progressed to an advanced stage. Moreover, CT imaging requires a hospital visit, exposes patients to harmful radiation, and may not provide precise early detection of inflammation. Therefore, early recognition and treatment are crucial to ensuring implant success and patient recovery. However, detecting inflammation in knee implants remains a challenge due to the difficulty of directly sensing biomarkers in real-time.

Smart implants capable of early inflammation detection and management could provide an effective solution to this challenge. Integrating sensors into bone implants^13–19 may offer several advantages, including real-time monitoring of inflammatory biomarkers such as pH,²⁰ continuous assessment of implant stability, and early detection of complications such as infection or implant loosening.²¹ Advances in strain gauges,²² biosensors,²³ micro-electro-mechanical systems (MEMS) devices,¹³ and smart implant technologies²⁴ are driving the development of more functional and reliable sensor-integrated bone implants.^25,26 However, miniaturizing wireless sensors and enabling their untethered application within TKA implants remain critical yet unsolved challenges.

Currently, most sensors designed for TKA implants are limited to intraoperative use and rely on wired connections and bulky electronic systems. Recent studies on TKA implants have primarily focused on sensing pressure and motion,^{21,22,25–28} with limited capability for liquid delivery or sensing. Existing knee implant sensors primarily focus on pressure sensors,²² with proof-of-concept demonstrations showing limited functionality beyond mechanical load monitoring. So far, previous studies have not demonstrated closed-loop sensing and treatment of inflammation in knee implants, primarily due to the absence of miniature pumps and long-term sensing mechanisms for biofluid biomarkers. On the other hand, closed-loop drug release has been achieved only in opioid delivery systems,^29,30 which are generally larger in size and designed for applications in soft tissues rather than within knee implants. Therefore, overcoming these limitations is essential for developing next-generation smart knee implants that can autonomously detect and manage inflammation, ultimately improving patient outcomes and reducing the need for revision surgeries.

To address these challenges, we propose the integration of miniature magnetic pumps into TKA implants to enable liquid sampling and release for closed-loop sensing and drug-delivery functions. The device is fully powered using high-frequency magnetic fields via induction coupling, while the pumping mechanism within the microfluidic channel is controlled by low-frequency magnetic fields via magnetic torques. The microfluidic channel facilitates on-demand liquid sampling, with an integrated sensing unit that allows real-time pH or temperature monitoring for inflammation detection and controlled liquid release up to 100 μL. Additionally, the drug chamber can be refilled using a syringe needle, ensuring prolonged functionality. In addition, the sensing capability of the implant is tested in a knee phantom, where its accuracy is quantified and validated against high-precision sensors. Finally, we demonstrate the functionality of the sensing and pumping mechanisms in a porcine knee model ex vivo. This sensory knee implant may potentially allow for detecting inflammation and enabling local, on-demand therapeutic delivery, ultimately enhancing post-implant care and reducing complications.

RESULTS

Overall concept of the knee implant with wireless fluidic pumps and sensors

Our objective is to develop a smart polyethylene disk for total knee replacement that transforms the traditionally passive implant component into an active, multifunctional platform (Figure 1A). Unlike conventional TKA implants, where the plastic disk is a solid and inert element, our design integrates two wireless microfluidic pumps for liquid sampling and delivery, as well as electronic modules for wireless communication, remote charging, and biosensing. As illustrated in Figure 1B, the smart disk incorporates microfluidic channels and pumps that enable on-demand liquid handling, while a sensing module enables analysis of biofluids within the channel.

Figure 1. Concept of a smart knee implant for sensing biofluid properties and delivering a liquid drug.

(A) Schematic illustration of an arthritic knee and a knee implanted with a smart robotic device, enabling real-time detection of pathological conditions and on-demand therapeutic delivery. Created by Biorender.com.

(B) Illustration of the robotic knee implant disk, including the integrated wireless charging module, electronic sensing and communication units, liquid sampling unit, and drug-delivery unit.

(C) Optical image of the fully assembled smart implantable disk. Blue liquid: simulated synovial fluid.

(D) Optical image of the modular magnetic piston pump designed for controlled drug release.

(E) Optical image of the modular magnetic piston pump for biofluid sampling and the modular optical sensing unit for detecting biofluid properties.

(F) Optical image of the wearable magnetic actuation unit and wireless charging unit used to remotely control the pump operation and power the electronics in the implant, respectively.

(G) Flowchart illustrating the closed-loop operational workflow, including biofluid sampling, in situ sensing, and feedback-controlled therapeutic liquid release.

The device integrates microfluidic sampling, sensing, and drug delivery into a compact, multifunctional platform, as shown in the proof-of-concept device in Figure 1C. As shown in Figure 1D, a representative magnetic pump comprises a magnetic piston and a flexible magnetic valve. The piston is remotely actuated by an external magnetic field, enabling bidirectional fluid transport, either ejecting liquid for drug delivery or drawing in fluid for sampling. The drug-delivery chamber connects to the skin surface through an outlet port via the flexible magnetic valve, allowing the chamber to be refilled using a syringe and thereby supporting long-term therapeutic use. The sensing mechanism, shown in Figure 1E, operates by directing sampled liquid propelled by a magnetically actuated piston pump through a pair of LEDs and photodetectors. Variations in transmitted light are captured as voltage signals by the photodetectors. By incorporating functional sensing materials such as pH-responsive polymers, the system can detect inflammation-induced pH changes in real time. The pump dimensions are shown in Figure S1. This innovative design redefines the plastic disk from a passive load-bearing element into a smart diagnostic and therapeutic interface within the joint.

A wearable system is developed to provide both magnetic actuation and wireless power delivery. As shown in Figure 1F, the system includes two external magnets, each driven by a DC motor and motor driver, which rotate at programmable speeds. An onboard controller regulates motor activity and communicates wirelessly with a host computer via Bluetooth. This wearable unit enables wireless actuation of the two magnetic piston pumps and powers the implant’s sensing electronics via remote charging. Figure 1G outlines the coordination between the sensing and drug-delivery modules, which operate according to a flowchart-based logic. When the sensing signal from a specific sensor exceeds a predefined threshold, it triggers subsequent liquid release, enabling responsive and condition-specific therapy. By integrating real-time sensing with magnetic actuation, the system enables autonomous liquid sampling, pH detection, and closed-loop drug delivery based on detected pH, temperature, or other biofluid properties, offering a responsive and personalized therapeutic approach for inflammation management beyond existing smart knee implants, shown in Table 1.

Table 1.

Comparison of smart knee implants in the literature

Reference	Type of device	Delivery function	Sensing function	Wireless	Therapeutic Function
Arami et al. 2013 IEEE Biomedical Engineering [29]	Total knee implant	No	Motion sensing by magnetic tracker	Yes	No
Hasenkamp et al. 2013 Biomedical Microdevices [13]	Metal disk	No	Pressure sensing by strain gauge	No	No
Persona IQ® The Smart Knee® Implant (Zimmer AG) [28]	Total knee implant	No	Motion sensing by Inertia Measurement Unit	Yes	No
This work	Plastic spacer	Yes	Biofluid sensing by pumps and optical sensors	Yes	Yes

Mechanism of liquid sampling and liquid delivery by the fluidic pump

We further present the mechanism of fluidic sampling using a magnetic piston. Liquid sampling using a magnetic piston pump has not been previously demonstrated. Compared to our earlier work on drug release in dental implants,³¹ the added sampling capability and its coordination with liquid release represent a significant advancement toward closed-loop operation. To illustrate this dual functionality, we detail the magnetic control of the two pumps. As shown in Figure 2A, a rotating magnetic field actuates the magnetic piston, which consists of a miniature screw (size: M4, 3 mm length) coupled with a neodymium magnet (NdFeB, N45; 1 × 1 × 1 mm). When the external magnetic field rotates in the y-z plane, the piston moves vertically, driving the blue liquid through the microfluidic channel. For biocompatibility, the magnet is coated with parylene-C. Under this setup, up to about 100 μL of liquid can be transported within 12 s. The magnetic field strength required for piston actuation is around 13 mT, as shown in Figure 2B. Figure 2C demonstrates the volume of liquid ejected using a similar piston design, actuated by a reduced magnetic field of approximately 5 mT. See Figure S2 for the fabrication process and Table S1 for the dimensions of the pumps.

Figure 2. Mechanism of magnetically controlled pumps and valves for liquid sampling and release.

(A) Sequential images capturing the liquid release process driven by the magnetically actuated piston pump.

(B) Waveform of the applied actuation magnetic field used to drive the pump.

(C) Measured liquid ejection volume as a function of time, demonstrating controlled output.

(D) Sequential images illustrating the operation of a lockable flexible magnetic valve, actuated by an external magnetic field for opening and sealing.

(E) Schematic illustration of the drug reservoir refilling mechanism via syringe access.

(F) Video frames (Video S1) of the refilling setup showing the flexible valve and syringe interface.

The liquid chamber can be refilled on demand using a thin syringe needle. As shown in Figure 2D, a flexible magnetic valve is integrated into the liquid release pump to function as a refillable inlet. This valve is fabricated from a magnetized composite material, with magnetization aligned along its leaf structure. The opening angle α, which is defined as the angle between the magnetic leaf and the valve base, increases with field strength. This design allows the valve to open at an angle α > 30° under a horizontally applied magnetic field of approximately more than 13.9 mT. Figure 2D illustrates the actuation behavior of the valve. A gradually increased magnetic field is applied to determine when the valve starts to open. The valve opens at 6.6 mT in the direction shown in Figure 2D. To reseal the valve, a magnetic field of opposite polarity (y direction) is applied. The valve fully closes when the field magnitude reaches beyond 11.6 mT. Therefore, after finishing each refilling procedure, a magnetic field can be applied to reliably close the valve and ensure proper sealing performance.

To validate the functionality of the refilling module and its sealing performance, a drug refilling and delivery experiment is performed using a delivery pump integrated with the proposed refilling module. To support long-term drug delivery, a refillable module is developed and integrated into the delivery pump. As shown in Figure 2E, this module includes a magnetic valve and a dedicated refilling inlet. The module does not interfere with the normal drug release process. During delivery, the drug is ejected through the outlet as intended. When refilling is required, a syringe can be inserted through the skin into the inlet to replenish the internal drug reservoir. After refilling, the magnetic valve can be actuated to close and reseal the module by applying an external magnetic field, thereby preventing unintended leakage. A conical feature is incorporated to connect the 1.5 mm inlet to a 4 mm outer opening, improving needle alignment and guidance (Figure S3). Refilling experiments are performed using a syringe-based setup, with porcine tissue placed adjacent to the refill region to mimic tissue contact. The tapered geometry enables stable alignment and smooth needle insertion, allowing liquid injection into the internal reservoir. Repeated tests confirm consistent accessibility and reliability, with reduced sensitivity to insertion angle.

The entire process is illustrated in the video frames shown in Figure 2F and Video S1. At t = 2 s, the drug chamber is presumed empty following a prior release, and the piston is actuated downward to create space for refilling. At t = 15 s, an external syringe is inserted through the refilling inlet to inject the drug solution (blue-dyed water), completing the refill around t = 25 s. An upward magnetic field is then applied to reseal the inlet by closing the magnetic valve. At t = 37 s, a rotating magnetic field is used to initiate drug ejection, and by t = 39 s, the blue liquid is visibly released through the outlet. Importantly, no leakage is observed from the refilling inlet, confirming the excellent sealing performance of the magnetic valve. A small amount of fluid accumulates beneath the piston due to minor gaps between the piston and chamber wall. To address this, the piston is actuated downward and then back upward, allowing complete expulsion of the remaining liquid by t = 57 s. This also highlights the advantages of the magnetically actuated piston, which ensures reliable and repeatable motion for complete drug delivery.

Finally, we conducted 10 consecutive refilling and delivery cycles, as shown in Figure S4. In each cycle, liquid is first injected through the refill inlet, followed by magnetic actuation to close the valve and drive the piston pump for controlled delivery. This sequence is repeated for all cycles. The device maintains stable operation without leakage or functional degradation, demonstrating reliable sealing performance. Future work will evaluate long-term durability under extended cycling and physiologically relevant conditions.

Coordinated control of fluidic pumps for liquid sampling and delivery

Precise synchronization between the two magnetic pumps is critical for achieving closed-loop sensing and controlled drug delivery. Two decoupling strategies are developed to accommodate different spatial arrangements in the implantable device and to maximize magnetic actuation efficiency. Because the modular pumps are actuated by external magnets, it is important to maximize the positioning of the pumps within the implant and of the magnets externally to the knee to minimize the gap between the magnets and the pistons within the pump. In addition, the onboard magnetic sensor enables alignment between the external actuation system and the two magnetic pumps. We quantify the system’s tolerance to positional and angular misalignment and showed that actuation remains robust within a practical deviation range. By continuously measuring the local magnetic field, the sensor determines the orientation of the external actuation system. As shown in Figure S5, small angular deviations can be readily detected, enabling real-time re-alignment during operation. Minor misalignments (e.g., ~10° in-plane rotation) do not hinder pump actuation, as the magnetic components continued to effectively drive the piston magnet (Figure S6). To characterize this, magnetic field measurements are performed under both aligned and intentionally misaligned conditions using a three-axis sensor (B_x, B_y, and B_z) during continuous field rotation, with data acquired via LabVIEW (National Instruments). The measured misalignment angle (~8.97°) closely matches the angle obtained from magnetic field point-cloud analysis (~9.05°), validating the accuracy of the principal-component analysis (PCA)-based orientation estimation method.

The magnetic pumps are decoupled by the magnitude of the magnetic field, as shown in Figure 3 and Video S2. The first method shown in Figures 3A–3D illustrates two pumps placed in parallel and two actuation magnets fixed in the wearable magnetic actuation system to align with the pumps. In this configuration, each actuation magnet is able to control only one nearby pump. Figure 3A illustrates the coordinated control mechanism, where piston magnets in the two pumps are actuated by external magnetic fields of varying magnitudes. As shown in Figures 3B and 3C, an example waveform demonstrates the sequential activation: magnetic pump 2 is activated first under a magnetic field of approximately 8 mT, while magnetic pump 1 remains stationary due to a lower field variation of only 2.5 mT. Subsequently, pump 1 is activated under a 5 mT field, while pump 2 becomes inactive, again due to minimal field variation. This selective actuation enables staged operation of the pumps. To provide a more intuitive visualization, Figure 3D depicts the sequential actuation of the sampling pump (pump 1) and the delivery pump (pump 2). The red and blue dashed boxes highlight the magnetic piston regions in the sampling and delivery pumps, respectively, indicating their actuation states during sampling and delivery processes. The sampling process is presented from 0.5 to 19 s. The piston in the sampling pump moves upwards from the bottom to the top, while the piston in the delivery pump remains stationary. Subsequently, the delivery process is shown from 7.5 to 14.5 s, during which the piston in the delivery pump is actuated, while the sampling pump remains inactive. Figure 3D shows the resulting sequential motion where piston 1 operates from 0 to 5 s, while piston 2 activates from 7.5 to 14.5 s. The corresponding displacement profiles in Figure 3E clearly delineate pump 1’s sampling phase, followed by pump 2’s liquid release. These results validate the reliability of the first decoupled control strategy for independent pump actuation.

Figure 3. Mechanism of magnetically controlled drug delivery and liquid sampling.

(A) Illustration of the independent control of two miniature magnetic piston pumps for liquid sampling and liquid delivery. Two magnetic actuators are placed on opposite sides and can be independently controlled.

(B) Time-varying magnetic field as a function of time measured at the location of pump 1. The two magnetic actuators are turned on sequentially.

(C) Time-varying magnetic field as a function of time measured at the location of pump 2.

(D) Video frames (Video S2) showing the decoupled operation of the two magnetic piston pumps operated in a sequential manner in two cases.

(E) Traveled distance of the two pistons in the magnetic pumps as a function of time.

The magnetic pumps are decoupled by magnetic field rotating planes as shown in Figure 4 and Video S2. An alternative strategy involves decoupling pump activation through orthogonal rotation planes of the magnetic field. As depicted in Figure 4A, the two pumps are oriented with orthogonal rotational axes. When a rotating magnetic field is applied in the x-y plane (Figure 4B), the sampling pump’s piston rotates and translates to draw in fluid, while the piston in the drug release pump remains inactive, experiencing only minor vibration. Conversely, when the magnetic field rotates in the x-z plane, the drug release pump becomes active, while the sampling pump remains still. This selective control is visually confirmed by the video frames in Figure 4C: piston 1 operates from 0 to 35 s, first drawing in fluid, then ejecting it, followed by a resting phase until piston 2 activates between 35 and 60 s. To further demonstrate the effectiveness of the field-based decoupling strategy, Figure 4C presents the sequential actuation of the sampling pump (pump 1) and the delivery pump (pump 2). Blue-dyed water and yellow-dyed water were used to visually distinguish the sampling and drug release functions, respectively. At t = 6 s, pump 1 begins sampling as its piston moves downward, drawing blue fluid through the microchannel into the pump chamber. During this stage, pump 2 remains inactive, which is activated at t = 35 s. At t = 54 s, drug delivery is initiated by actuating pump 2 with a separate external magnet, causing its piston to eject the yellow fluid. Notably, the piston in pump 1 remains stationary throughout this process. Figure 4D presents the corresponding displacement profiles, clearly illustrating the sequential microfluidic operations of sampling, ejection, and drug delivery. These results confirm that magnetic actuation in orthogonal planes enables selective and interference-free control of each pump, validating the directional decoupling design. Distinct and independently addressable pumping cycles validate the precise, multi-channel control within a compact magnetic actuation framework.

Figure 4. Mechanism of coordinating two magnetic pumps by the applied rotating magnetic fields in different planes.

(A) Schematic illustrating the principle of independent actuation of two magnetic piston pumps integrated within a single device, enabling decoupled operation for sensing and drug delivery. Each pump contains a magnetization profile aligned with a distinct rotational plane of the applied magnetic field, allowing selective activation without mechanical or magnetic interference.

(B) Time-varying magnetic fields measured at the location of pump 1 when the magnetic field is applied in orthogonal planes to coordinate pump operation. By programming the rotation plane, frequency, and phase of the external magnetic field, each pump can be actuated independently or in a synchronized manner.

(C) Representative video frames (Video S2) demonstrating decoupled actuation of the two pumps, where activation of one pump produces negligible motion in the other, confirming effective magnetic isolation and selective control.

(D) Quantified piston traveled distances of the two pistons in the magnetic pumps as a function of time under coordinated magnetic actuation.

Wireless communication and sensing of biofluid properties

The overall electronic circuit design for wireless communication and charging is shown in Figure 5. A key function of the smart implant is to sense biofluid properties, such as pH and temperature, while enabling wireless sensing and communication. To achieve this, the implant integrates electronic units, which are detailed in Figure S7 and Table S2. Figure 5A shows the fully assembled smart implant device with the top cover removed for clearer visualization of the internal components. The electronic unit consists of several key components. First, a Bluetooth Low Energy System-on-a-Chip (BLE-SoC) board serves as the computational unit, enabling wireless communication. Second, a remote charging module powers the circuit via inductive coupling with an external charging coil, similar to the method reported in Ge et al.³² Third, a magnetic field sensor monitors the magnetic actuation signals to ensure precise control. Fourth, an LED and a phototransistor work in conjunction with smart materials within the fluidic channel to detect biofluid properties. Lastly, a user interface on a mobile device or computer is developed to control the electromagnetic actuation system and receive real-time sensor signals. This integrated system ensures seamless wireless operation, enabling efficient biofluid sensing and responsive actuation.

Figure 5. Mechanism of sensing biofluid properties by optical sensing and remote charging.

(A) Schematic of the wireless electronic board, which integrates a BLE-SoC module for wireless communication with a smartphone and a rectifier circuit that converts energy from the receiving coil into a regulated 3.3-V output.

(B) Data flow diagram of the sensory implant system, highlighting wireless power transfer and signal communication pathways.

(C) Optical image of the fully integrated electronic components.

(D) Charging voltage as a function of the vertical axial distance between the implant and the external wireless charging coil.

(E) Charging voltage profile over time during wireless charging.

(F) Illustration of the remote charging configuration of the wearable system.

(G) Optical image of the remote charging unit and magnetic actuation unit.

Figure 5B illustrates a flowchart of the system for explaining the process of the communication module. First, an external magnetic field generator provides energy via a high-frequency alternating magnetic field to a sensory implant, and a receiving (RX) coil and a remote charging unit harvest the energy and convert it into a constant operating voltage at 3.3 V to power the sensory implant. The sensory implant consists of an optical sensor pair, including a phototransistor board and an LED board to capture optical signals, as well as a magnetic sensor to capture magnetic signals. A BLE-SoC board manages communication with the mobile application and coordinates signal acquisition from the peripheral sensing units. Specifically, the phototransistor signal is received via the analog input, while the LED is powered by the analog output to provide constant light intensity. In addition, the magnetic sensor is interfaced with the BLE-SoC to acquire real-time magnetic field measurements along three axes. Finally, the magnetic field and phototransistor data acquired by the BLE-SoC are transmitted wirelessly to the mobile application and subsequently displayed on the user interface for real-time monitoring.

Remote charging is based on induction coupling between the RX and transmitting (TX) coils. As shown in Figure 5C, the remote charging circuit is designed with a compact footprint with voltage regulation and stabilization functions. When the TX coil is placed at the knee, the RX coil embedded in the implantable device can be wirelessly powered and can power the electronics with the harvested energy. Thus, it is critical to evaluate whether the wireless transferred power can provide sufficient energy to support device operation. We investigate the impact of the plane-to-plane separation distance among the TX and RX coils against the DC output voltage. Figure 5D presents the experimental setup used to investigate the wireless charging distance. Figure 5D also illustrates the RX output voltage variation with the increased charging distance. The output voltage is measured under no-load conditions to evaluate the maximum achievable supply level. The system supports a charging distance of up to 100 mm (Figure 5E), defined as the center-to-center separation between the TX and RX coils, which reflects the separation expected in practical applications. The result indicates that the separation distance is critical for reliable wireless power transfer. To validate remote charging functionality and system performance, we characterize the voltage output of the wireless power module. The device remains fully functional with remote charging applied continuously for 1 min, demonstrating its potential for sustained wireless energy transfer. The system maintains sufficient DC output for sensing even with intervening tissue between the RX and TX coils. As shown in Figure S8, the implantable device remains operational with porcine tissue between the TX and RX coils, successfully powering an onboard LED and activating the BLE-SoC. The device is placed within the TX coil and first tested in air, followed by tests with increasing tissue thickness. An ~11 mm tissue layer is initially placed above the device, with additional layers stacked to a total thickness of ~24 mm around the device. Successful power transfer is confirmed by LED illumination under all conditions. Lastly, Figures 5F and 5G show the 3D rendering of the remote charging unit and the optical image of the actual experimental setup for energy harvesting, respectively. See Table S3 for the components of the wearable magnetic actuation and remote charging system.

Sampling liquids and sensing biofluid properties

To enable biofluid sensing, we integrate optical electronics capable of detecting properties such as the pH of synovial fluid as a biomarker for inflammation. As shown in Figure 6A, the system employs a paired LED and phototransistor operating at a wavelength of 550 nm. Figure 6B shows the optical image of an optical sensor. When the smart material within the fluidic channel undergoes deformation, due to chemical or mechanical changes induced by the biofluid, this results in a measurable voltage change in the phototransistor. This mechanism enables real-time, wireless monitoring of biofluid conditions with high reliability and responsiveness. Figure 6C shows the basic mechanism of sensing liquid pH using a pH-responsive hydrogel³³ as well as optical sensor pairs with both the LED and phototransistor. An LED and a phototransistor are placed on opposite sides of the transparent polydimethylsiloxane (PDMS) microchannel. The hydrogel block is inserted into the microchannel and secured by a 1-mm-diameter pin that passes through the pinhole into the hydrogel block to anchor it in place. The hydrogel swells in response to varying pH levels and blocks the light transmission from the LED to the phototransistor. The corresponding change in light intensity is detected as a variation in the analog output of the phototransistor.

Figure 6. Characterization of the fluid sensing using a pH-responsive hydrogel.

(A) Schematic illustration of the sensing mechanism, where the hydrogel deforms in response to pH changes.

(B) Optical image of the sensing unit with integrated LED and photodetector while encapsulated by PDMS for sealing. Scale bar, 1 mm.

(C) Optical image of the fabricated pH-responsive hydrogel before and after applying a solution of pH = 3. Scale bar, 0.5 mm.

(D) Optical images of the hydrogel exposed to solutions with pH values of 4, 5, 6, and 7, respectively. Scale bar, 1 mm.

(E) Calibrated sensor response showing the percentage increase in signal as a function of pH.

(F) Comparison of pH values estimated by the hydrogel-based sensor versus a commercial pH meter, demonstrating strong correlation and accuracy.

The optical sensor and the microfluidic channel with the smart material are combined. The LED board and the phototransistor board are positioned face-to-face on opposite sides of the device, with a microchannel between them. To optimize the sensitivity of the optical sensor, the distance between the LED and the phototransistor is about 1 mm. The LED emits light onto the phototransistor, and the strength of the light after passing through the hydrogel varies the voltage of the phototransistor, thereby varying the output signal. As shown in Figure 6D, when liquid flows through the microchannel, the smart material inside, which is in a hollow square shape, deforms in response to biomarkers in the fluid, such as pH, which changes the transmitted light. This difference in optical path influences the amount of light reaching the phototransistor and, accordingly, changes the measured voltage. When the sampled liquid is acidic, the hydrogel swells to block more of the LED and the phototransistor. In contrast, when pH = 7, the size of the pH-responsive hydrogel tends to be smaller. Figure 6E compares the change in transistor output voltages at different pH values and illustrates the representative optical images of the hydrogel before and after immersion in hydrochloric acid solutions of different pH levels for 20 min. The figure clearly shows that hydrogel expansion increased as the pH level decreased. The hydrogel exhibits the greatest swelling ratio at pH = 4, with a 30.7% increase in area, while the minimal swelling ratio is observed at pH = 7, with only an 11.4% increase in area.

To validate the pH-sensing ability, we place the hydrogel blocks within the microfluidic channels with the LED and the phototransistor on each side. To evaluate the effect of hydrogel swelling on the sensor output, the microfluidic channel is placed between the LED and phototransistor with a pinned hydrogel block. We investigate the analog voltage change over time under different pH conditions, with each pH condition tested three times to assess repeatability. The voltage signals increase progressively and then reach a steady value in the range of 1,300–2,200 s, as the hydrogel block gradually expands, blocking more transmitted light until it reaches a saturated state. Lower pH levels induce greater swelling, resulting in larger voltage increases. Using the collected sensor signal and pH data, we build a model to map the liquid pH at steady state with the percentage increase of the phototransistor. Based on the average increase in the analog output of the phototransistor, a calibration curve is calculated. We further validate the calibrated model by testing the sensor signal with liquids of different pH values and comparing the estimated pH values with the ground truth measured with a pH meter. To validate the accuracy of the hydrogel-based sensing mechanism, four different pH levels of acidic solution are tested with a commercial pH sensing meter and the proposed pH sensor. The experiments are performed three times to prove the repeatability. The result shown in Figure 6F shows that the estimated mean pH value closely matches the ground-truth values.

This study focuses on the coordinated operation of two magnetic pumps for liquid sampling and release, with the potential integration of more durable sensors for reliable biofluid characterization under physiological conditions. We also evaluate the repeatability of the hydrogel’s pH-responsive swelling. Samples (4 × 2.6 mm) were immersed in HCl solutions of varying pH for 30 min, rinsed with deionized water, and dried at 70°C for 30 min. This swelling-drying cycle is repeated to assess reproducibility. As shown in Figure S9, the hydrogel exhibits consistent and repeatable pH-responsive behavior. To enhance robustness, future designs may incorporate microfluidic filters to remove proteins and particulates, as well as anti-biofouling surface coatings.³⁴ Future work will focus on long-term cycling stability and on integrating more durable materials with optical sensing.³⁵

Demonstration of closed-loop drug delivery triggered by biofluid sensing

The smart knee implant integrates a polymer disk with embedded microfluidic pumps and sensors to realize a fully closed-loop system for inflammation monitoring and treatment. The device periodically samples synovial fluid into an integrated microfluidic channel, where changes in acidity are detected using a pH-responsive hydrogel coupled with optical sensing. When the measured pH drops below 6.9, an indicator of potential inflammation or infection,³⁶ the system automatically triggers a delivery pump to release stored antibiotics. Therapeutic delivery is powered wirelessly by an external wearable magnetic actuation unit and can be performed multiple times per day as needed. Notably, the patient is required to wear the magnetic unit only during sensing and treatment cycles, minimizing daily burden and improving usability.

As shown in Figure 7A, the implantable disk is immersed in a synovial-fluid-mimicking environment (blue liquid). The closed-loop operation begins with fluid sampling by the microfluidic pump, followed by pH detection via deformation of the pH-responsive hydrogel within the sensing channel. An LED-phototransistor pair positioned on opposite sides of the hydrogel measures changes in light transmission caused by hydrogel deformation, producing a voltage signal that correlates with the local pH. When the detected pH falls below a predefined physiological threshold, the delivery module is activated, releasing therapeutic agents (yellow liquid) into the surrounding environment. This sequence demonstrates a fully integrated cycle of in situ sensing and feedback-controlled therapy.

Figure 7. Demonstration of closed-loop sensing and drug delivery controlled by a wearable actuation system.

(A) Illustration of the implantable device for liquid sampling, sensing, and delivery.

(B) Video frames (Video S3) of liquid release when the pH value is less than a threshold in a phantom.

(C) Liquid release volume of the device as a function of time.

(D) Optical image of testing the knee implant disk housing in a mechanical test setup with the load cell (2 kN). Scale bar, 1 cm.

(E) Mechanical stress of the knee implant before breaking when loaded using a tensile machine.

(F) Optical image of mounting the plastic disk on the porcine bone.

(G) X-ray images of the liquid release process before and after the liquid is released. Liquid: contrast agent. Scale bar, 4 mm.

(H) Biocompatibility test of the magnetic piston coated with parylene-C, including the control and experimental groups. Scale bars, 100 μm.

Figure 7B and Video S3 show the temporal sequence of the closed-loop operation, including liquid sampling, pH detection, and drug release. The sampling pump is activated at t = 14 s to draw synovial fluid (indicated by blue dashed lines) into the microfluidic channel for analysis by the pH sensing module. This step is initiated by an alternating magnetic field that actuates the sampling pump via magnet 1. Once the sensed signal exceeds a predefined threshold indicating a potential inflammatory or infectious event in the knee joint, the delivery pump is triggered. The microcontroller pauses the rotation of magnet 1 and subsequently actuates magnet 2 to drive the delivery pump. Drug release begins around t = 20 s and is completed by t = 27 s, resulting in a total released volume of 71 μL, as quantified in Figure 7C. Together, these results demonstrate the feasibility and effectiveness of closed-loop sensing and therapeutic delivery using pH as a clinically relevant biomarker.

Furthermore, the polymer disk serves as a critical structural component of the knee implant, making mechanical strength an essential design consideration. Because the sensing modules and pumps are embedded within the disk, internal cavities are required, which could potentially compromise compressive strength. To evaluate this effect, a compression test is conducted using a custom-built setup incorporating a load cell to measure applied stress on the disk. The experimental configuration is shown in Figure 7D. A testing probe with a cross-sectional area of 28.2 mm² is used to apply localized compressive stress to the disk surface. A linear screw mechanism incrementally advances the probe, progressively increasing the applied stress until failure.

As shown in Figure 7E, the disk cover fails at a peak compressive stress of 26.7 MPa, corresponding to the maximum compressive strength of the implant structure. Previous studies have reported peak contact pressures of up to 25 MPa in the human knee joint under extreme loading conditions.³⁷ The measured fracture strength of the polymer disk therefore exceeds these physiological stress levels, indicating that the integration of sensing and pumping modules does not compromise the mechanical integrity of the implant.

Lastly, to demonstrate the potential application of the proposed knee implant in biological tissue, the device was mounted on the femoral region of a porcine knee. Figures 7F and 7G present optical and X-ray images, respectively, capturing the liquid sampling and subsequent drug release. The implant is clearly visible on X-ray imaging, enabling direct visualization and monitoring of device status and operation using clinically available imaging modalities. In future implementations, implant positioning and functional status could be monitored noninvasively using CT imaging,³⁸ providing higher spatial resolution and 3D localization. To further assess biocompatibility, the magnetic piston component was coated with parylene-C prior to testing. Cell viability results are shown in Figure 7H, demonstrating high cell viability after exposure to the coated components and confirming the suitability of the parylene-C-encapsulated magnetic elements for in vivo applications.

DISCUSSION

In summary, we have presented a magnetically controlled strategy for independently actuating two miniature pumps, enabling a compact smart knee implant with integrated wireless sensing and closed-loop drug delivery. The implant is fully wirelessly powered using high-frequency magnetic fields, while low-frequency magnetic fields independently drive dual magnetic piston pumps for biofluid sampling and therapeutic release. An integrated microfluidic sensing module continuously monitors joint pH to detect inflammatory conditions and automatically triggers on-demand drug delivery. The system is refillable via a syringe-accessible valve and delivers up to ~70 μL per actuation under magnetic fields below 15 mT. Sensing accuracy has been validated in knee phantoms, and closed-loop sensing and pumping have been demonstrated in an ex vivo porcine knee model. The implant withstands compressive stresses of ~26.7 MPa, which exceeds physiological joint loads, thereby confirming mechanical robustness. Together, this work advances intelligent orthopedic implants by enabling real-time inflammation monitoring and targeted, refillable therapy via fully wireless magnetic control.

Potential improvements and extensions to the proposed system are outlined as follows. First, the current sensing modality focuses on pH, which is a well-established correlate of joint inflammation.^9,39,40 In healthy joints, synovial fluid typically exhibits a pH of 7.2–7.4, whereas in cases of joint infection, the pH has been reported to drop below 6.9.⁴¹ This shift provides a clinically relevant biomarker for joint health and has been used to aid the diagnosis of conditions such as prosthetic joint infection.^41,42 In this study, we primarily focused on demonstrating the coordinated operation of the two magnetic pumps for liquid release and sampling. Moving forward, the sampling pump can be integrated with more robust sensing elements for biofluid characterization. For instance, microfluidic filters could be incorporated to remove proteins and particulates, and anti-biofouling surface coatings may be applied to enhance long-term durability. In future iterations, additional inflammation-related biomarkers could be incorporated by integrating appropriate biochemical assays and sensing materials. In addition, the frequency of refilling could be reduced by increasing the volume of the liquid reservoir. The current syringe-based refilling mechanism could be further optimized with dedicated refilling tools, and the overall implant footprint could be reduced to allocate additional space for an enlarged drug chamber while maintaining mechanical integrity. These modifications would enhance therapeutic capacity without compromising implant robustness. Furthermore, in vivo validation will require the use of fully biocompatible materials and robust encapsulation strategies to ensure long-term safety and compatibility with surrounding tissues. The mechanical durability of the device must be evaluated under cyclic loading conditions that mimic physiological joint motion and load-bearing environments. Lastly, the implantation procedure also needs to be systematically validated using standard orthopedic tools and workflows, including assessments of insertion, positioning, fixation stability, and integration with existing implant systems.

Future work will focus on evaluating long-term biocompatibility, sensing stability, and therapeutic performance in porcine knee models under extended implantation periods. Beyond knee applications, the proposed sensory implant architecture could be adapted to other closed-loop orthopedic systems, representing a significant advance in real-time inflammation monitoring and on-demand therapy. More broadly, the platform can be expanded to detect additional biomarkers by incorporating smart materials responsive to temperature and other chemical indicators, enabling more comprehensive disease characterization. Lastly, the system may be translated to other anatomical sites, such as hip or glenoid implants.⁴³ By integrating closed-loop sensing with targeted drug delivery, this technology offers a promising pathway toward personalized, point-of-care therapies that may reduce or eliminate the need for secondary surgical interventions.

METHODS

Fabrication of the modular pump for liquid sampling and delivery

First, a cylinder (diameter: 6 mm, length: 15 mm) was printed with UV-curable resin (Clear Resin V4, Formlabs) in a stereo-photolithography (SLA) 3D printer (Form 3+, Formlabs) based on a 3D design prepared in Solidworks 2022 (Dassault Systemes). Then, the cylindrical tube was cleaned in isopropyl alcohol in an ultrasonic cleaner for 20 min. After cleaning, the cylinder was exposed to UV light at room temperature for 15 min to fully cure. Next, a 10-mm-deep hole was drilled in the cylinder by an M1.5 drill bit, and a through-hole was drilled at the base of the cylinder with an M0.5 drill bit to balance air pressure. Finally, an M2 tapping tool was used to tap a threaded hole of 9 mm deep.

Fabrication of the microchannel for liquid sampling and sensing

The microchannel was fabricated using a layer-by-layer assembly method. First, uncured liquid PDMS was poured onto a poly(methyl methacrylate) (PMMA) substrate with spacers (120 μm, six layers thick) and cured on a hot plate at 80°C for 30 min. The microchannel patterns were then cut using a UV laser machine (LPKF Protolaser U4). Next, the cut-out patterns were stacked and bonded with uncured liquid PDMS, followed by curing with a heat gun at 93°C for 30 s. A connector made of UV-curable resin (Clear Resin V4, Formlabs) was 3D printed using a SLA 3D printer (Form 3+, Formlabs) based on a 3D design created in SolidWorks 2022 (Dassault Systèmes). Finally, super glue (Loctite) was applied to bond the PDMS microchannels with the modular pump. To address potential leakage, uncured liquid PDMS was used to seal the connector and the microchannel, ensuring a secure and leak-proof assembly.

Preparation of a remote charging module

The remote charging module consists of an RX unit and a TX unit. The RX module comprises two components: the RX coil and a remote charging board, which has been integrated into the BLE-SoC board. To fabricate the RX coil, a resin core was 3D printed and subsequently wrapped with 0.1 mm copper wire, forming approximately 50 turns. The two ends of the copper wire were then connected to the RX charging unit on the electronic board, ensuring efficient power transfer. The TX unit consists of a driving circuit and a TX coil. The TX coil is a commercially available component (12V2A Long Distance, Tai-dacent). A 24-V DC power supply was used to drive the TX coil and its corresponding driver unit, enabling effective wireless energy transmission to the RX module.

Preparation of the electronic board and the sensing module with LED and photodetector

A BLE-SoC was used for wireless communication, and a remote charging unit was used to provide power to the system. Designed as a double-sided printed circuit board (PCB), the electronic board was compact and optimized for integration within the implantable device. The board was initially designed in EasyEDA (a software company) and featured two programming ports (SWDIO and SWDCLK), along with one analog input and a ground port for connecting to the sensor module. To fabricate the electronic board, an FR4 PCB was cut and etched using a laser machine (LPKF Protolaser U4). After patterning, all electronic components were soldered onto the PCB. A multimeter was then used to verify the circuit integrity and check for potential short circuits. Finally, the board was connected to a computer and programmed by uploading compiled firmware onto the BLE-SoC board. The BLE functionality was tested by connecting the board to a mobile phone, while a computer running MATLAB 2024a (Mathworks Inc.) was also used to establish a Bluetooth connection for further testing and data acquisition.

Preparation of a pH-responsive sensor setup using transmission-based optical sensing

To fabricate the sensor module, a 550 nm LED and a 550 nm phototransistor were selected for integration. The circuit of the sensing board was designed using EasyEDA. A single-sided FR4 board with a thickness of 1 mm was patterned and cut using the LPKF Protolaser U4. A 68-kΩ resistor was used as a reference resistor. A green LED (0603 size) was chosen and soldered onto the board. For the phototransistor, a 550 nm LED and an additional 68-kΩ resistor were also selected and soldered to ensure proper functionality and signal stability. Extra LED/phototransistor pairs of other wavelengths (630 nm) were also used during tests.

Preparation of pH-responsive hydrogel

A hydrogel precursor solution was prepared by mixing poly (ethylene glycol) diacrylate (PEGDA; Mn = 700, Sigma-Aldrich) and 2-(diisopropylamino) ethyl methacrylate (DPAEMA; Sigma-Aldrich) at a molar ratio of 1:2, along with 2 wt % 2,2-dimethoxy-2-phenylacetophenone (DMPA; Sigma-Aldrich) as a photoinitiator. Polyurethane particles (30 μm in diameter, Co-spheric) were mixed into the solution to block the light. The mixture was sonicated for 10 min to fully dissolve the DMPA and complete the precursor preparation. For UV curing, the precursor was infiltrated into a custom mold, constructed from two glass slides separated by a spacer via capillary action, and then exposed to 365 nm UV light for 45 min on each side.

The fabrication of the pH-responsive hydrogel followed a previously reported method,³³ which introduced a pH-independent swelling behavior. PEGDA (Mn = 700, Sigma-Aldrich) was mixed with DPAEMA (Sigma-Aldrich) at a molar ratio of 1:2. 2 wt % DMPA (Sigma-Aldrich) was added as a photoinitiator. The hydrogel precursor solution was sonicated for 10 min to fully mix. Next, the solution was mixed with fluorescent red polyurethane microparticles (45–53 μm in diameter, Cospheric), poured into a plastic culture dish, and exposed to 365 nm UV light for 6 min. Finally, an M0.5 drill bit and a razor blade were used to create a pinhole and a rectangular opening (2.5 × 1.5 mm) in the hydrogel block to block the light.

Procedure for testing the biocompatibility of the magnetic piston

Human fibroblast cells (CRL-2522) were obtained from the American Type Culture Collection (ATCC) and cultured in minimum essential medium (MEM; Gibco) supplemented with 10% (vol/vol) fetal bovine serum (FBS; Gibco). Fibroblasts were seeded into 6-well plates at a density of 5 × 10⁵ cells per well to facilitate initial attachment and spreading under standard incubation conditions (37°C, 5% CO₂, humidified atmosphere). After 24 h, the encapsulated magnetic piston was placed into the wells and co-cultured with the cells for an additional 72 h. Cell viability was assessed using the LIVE/DEAD Viability/Cytotoxicity Kit (Invitrogen, Thermo Fisher Scientific), and fluorescence signals were observed using a fluorescence microscope (Eclipse Ti-E, Nikon, Japan).

Preparation of the knee phantom

The knee phantom was digitally extracted from a human body model using SolidWorks 2022. The phantom model was then converted into a Standard Tessellation Language (STL) file and 3D printed using a fused deposition modeling (FDM) printer with thermoplastic polyurethane (TPU). To enhance stability and functionality of the wearable magnetic actuation system, a commercial belt was cut to the appropriate length and securely fixed to the knee phantom, ensuring proper fit and support during testing the magnetic actuation.

Preparation of the customized wearable actuation unit

To fabricate the wearable actuation unit, two DC motors and corresponding motor drivers were assembled. The DC motors and drivers were securely mounted onto two holders, which were then affixed to a flexible band designed to be worn around the knee. The DC motors were fastened to the holders using bolts and nuts to ensure stability during operation. Two permanent magnets were attached to the DC motors, with a bearing providing additional support for the magnets. To establish a connection between the magnet and the DC motor shaft, two couplers were bonded to the magnets with super glue, ensuring a strong and reliable attachment. For motor control, an Arduino Nano 33 BLE (Arduino) was used to regulate the DC motors. Two DC motor drivers (L298N mini, WWZMDiB) were used to drive the motors, with each motor connected to designated pins on the motor driver. Power was supplied to the motor drivers via an external power source, such as a battery or DC power supply. To enable wireless control, a smartphone running a BLE application (LightBlue) was used to establish a connection with the Arduino Nano 33 BLE, allowing for remote actuation of the wearable unit.

Statistics analysis

Data presentation (mean and standard deviations) and sample size (n) for each statistical analysis are mentioned in the figure captions. MATLAB 2024b (Mathworks) was used for the statistical analysis.

RESOURCE AVAILABILITY

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Xiaoguang Dong (xiaoguang.dong@vanderbilt.edu).

Materials availability

This study did not generate new unique reagents.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.device.2026.101203.

Highlights.

• A smart knee implant for liquid sampling, sensing, and delivery

• Coordinated magnetic pumps for independent control of liquid sampling and release

• Wireless actuation, charging, and communication for on-demand liquid pumping and sensing

THE BIGGER PICTURE.

A knee implant is essential for patients who suffer from joint damage caused by genetic disorders or injuries. However, current implants face critical limitations, particularly the risk of long-term inflammation, which often requires invasive open surgery for intervention. These implants typically lack integrated sensing and drug-delivery capabilities due to the technical challenges of embedding wireless components within the compact design of an implant. Here, we report a novel knee implant integrating wireless sensors and miniature pumps for closed-loop temperature sensing and on-demand drug delivery. The system enables remote access and control of sensing and actuation functions, offering a compact and fully implantable solution. We demonstrate the implant’s ability to monitor biofluid properties such as temperature and deliver liquid up to about 70 μL through magnetically actuated pumping at a relatively small magnetic field of less than 15 mT. A refillable drug reservoir, accessed via a thin syringe needle through an integrated valve, supports long-term treatment without the need for replacement surgery. The implant disk can withstand a compression of about 26.7 MPa, exceeding the 20 MPa experienced during everyday life, demonstrating mechanical strength. Encapsulation of the electronics and magnetic pistons in biocompatible materials allows biocompatibility to be validated via an in vitro cell viability test. Finally, we validate the functionality of the closed-loop system in a porcine bone model ex vivo. In the long term, this integrated approach provides real-time monitoring and targeted intervention for inflammation by releasing therapeutic agents such as antibiotics, thereby reducing surgical burden and enabling continuous surveillance of implant health.

Data and code availability

All data generated or analyzed during this study are included in this published article and its supplemental information files.