A Wireless Self-Fine-Tuning Artificial Neural Pathway Implant for Near-Real-Time Close-Loop Motor Modulation

A Wireless Self-Fine-Tuning Artificial Neural Pathway Implant for Near-Real-Time Close-Loop Motor Modulation | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article A Wireless Self-Fine-Tuning Artificial Neural Pathway Implant for Near-Real-Time Close-Loop Motor Modulation Milin Zhang, Mengchun Sun, Ziyao Zhao, Chaochao Li, Yuwei Zhang, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6244028/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Artificial neural pathway (ANP) implants hold transformative potential for empowering individuals with disabilities, which has recently revolutionized the clinical treatment of spinal cord injury through epidural electrical stimulation (EES). Classical EES operates with a pre-set program and needs manual modification, consuming both manpower and time. Although existed studies proposed kinematic image processing to configurate EES or a brain-spine interface to control motions automatically, prompt tuning on motor modulation in an open environment is still inaccessible. Here, we show a wireless self-fine-tuning ANP implant to establish a rapid self-feedback mechanism appropriate for free activities to achieve precise motor modulation by EES. The fine-tuning pathway was attached to the writing pathway using a wireless implanted bidirectional neural interface featuring compact size, low input noise, and high energy efficiency, with different functionalities including reading, writing, and bidirectional control. The sciatic neural signals are fed into the pathway as rapid feedback for self tuning. Without manual interventions, we restored gaits of hindlegs in paralyzed animals with EES by a self-fine-tuning, near-real-time, and close-loop motor modulation mechanism. Physical sciences/Engineering/Electrical and electronic engineering Physical sciences/Engineering/Biomedical engineering Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Artificial neural pathway (ANP) implants hold transformative potential for empowering individuals with disabilities by reestablishing communication between the broken neural pathway, realising the restoration of the lost functions, which enhances the mobility and quality of life. For those with sensory impairments, such as hearing loss, implants like cochlear devices can restore the ability to perceive sound 1,2 , significantly improving communication and social interaction. Furthermore, these technologies can aid in managing neurological disorders by modulating brain activity, offering relief from symptoms of conditions like Parkinson’s disease and epilepsy 3,4 . By translating neural signals into digital commands, implants facilitate communication for individuals with severe disabilities, such as amyotrophic lateral sclerosis, allowing them to express themselves and interact with the world 5,6 . There are two key pathways in a typical ANP implant, i.e. the reading pathway and the writing pathway (Fig. 1a). The reading pathway interfaces with the neural networks located above the injury site, while the writing pathway integrates with the neural circuits situated below the lesion. In recent years, ANP, working synergistically with a control central plant, has been applied in clinical practice to restore complex motor functions, e.g. upper limp mobility restoration using neuroprosthesis (Fig. 1b) and the recovery of walking ability in patients with permanent paralysis caused by spinal cord injury (SCI) (Fig. 1c). SCI is a serious and intractable central nervous system injury that leads to significant impairment of physical functions, particularly sensory and motor capabilities. It was reported that about 250,000 to 500,000 people suffer from SCI every year globally 7 . Recent advancement in ANP has revolutionized the clinical treatment of SCI with epidural electrical stimulation (EES) to become the gold standard in association with neurorehabilitation 8 . The extensive clinical experience of treating intractable pain with EES also promotes its application in motor recovery. Classical EES integrated only the writing pathway, operating with a pre-set programme and manual modification of stimulus parameters rather than automatic correction according to patient’s kinestate (Fig. 1d), which consumes much manpower and weakens modulating effects. Capogrosso et al . proposed configuration of EES through real-time processing of gait kinematics and confirmed its feasibility 9 . However, it is only suitable for the motions in a closed experimental environment, since real-time gait recording by cameras is hard to manage for free motions in an open environment like real life scenes. Lorach et al . introduced an extra reading pathway implanted in the skull to acquire motion attention through the acquired neural activities, enabling a whole close loop of brain, spinal cord, and computer, allowing patients to control motions autonomously 10 (Fig. 1e). However, this close loop still lacks a “supervisor” that can rapidly check if the modulated motions are correct and accurate, since as a “commander”, the brain has to spend hundreds of milliseconds on visual feedback after wrong motions occurring 11 , limiting the promptness and precision of EES modulation to some extent. Therefore, we introduced a self-fine-tuning pathway into the ANP system (Fig. 1f) to establish a rapid self-feedback mechanism appropriate for free activities to achieve precise motor modulation by EES with self tuning. The fine-tuning pathway was attached to the writing pathway using a wireless implanted bidirectional neural interface featuring compact size, low input noise, and high energy efficiency, with different functionalities including reading, writing, and bidirectional control. The writing implant allows a wider stimulus voltage range and a higher upper limit of stimulus current amplitude for modulation; the reading implant enables wireless acquisition to collect neural activities from sciatic nerve or surface electromyograph (sEMG) sensors; and the bidirectional implant enables self-fine-tuning feedback for the modulator according to acquired neighbouring activities. Furthermore, compared with the time-consuming image processing and manual modification, sciatic neural signals spend only about 14 ms detecting errors and thus are fed into the self-fine-tuning pathway as rapid feedback for precise modulation tuning. Without manual interventions, we restored gaits of hindlegs in paralyzed animals with EES by a self-fine-tuning, near-real-time, and close-loop motor modulation mechanism. The scalability and adaptability of the self-fine-tuning ANP make it accommodate varying degrees of paralysis and individual patient needs. Different combinations of implant locations enable the generation of pathways that coordinate with the natural neural circuit for different scenarios, such as walking, activities of upper limbs, and urination. Ultimately, this innovative design holds the potential to significantly improve the quality of life for paralysis patients by offering more robust and flexible treatment options. Results Self-fine-tuning ANP implant for bidirectional interactions The self-fine-tuning ANP implant comprises two functional modules, that is, motion function encoding through modulation (Fig. 2 a) and neural activity recording for fine tuning (Fig. 2 b). A fine-tuning unit (Fig. 2 c) associated with these two functional modules was implanted subcutaneously for efficient data transfer with other distributed modules (Fig. 2 d). The motion function encoding module (Fig. 2 a) supports a voltage output up to ± 12 V by integrating a boost converter. The high voltage output capability ensures effective stimulation in case of a possibly increased local impedance after chronic electrode implantation. It features 16 current source channels with a 14-bit resolution and allows a delivered current from 1 µA to 16.384 mA. The 16 channels can be configured independently for simultaneous stimulation in various combinations of single-ended mode or differential mode. Single-ended mode allows a 12-V voltage upper limit, which is increased to 24 V in differential modes by pairing two current channels. The time resolution is precise down to 1 µs, and the system can generate arbitrary stimulation patterns lasting from microseconds to seconds. For the sciatic neural signal recording module (Fig. 2 c), a boost converter configures a ± 2.5 V voltage at the front end. An analog-to-digital converter converts the amplified signals into digital signals that will be transmitted to the fine-tuning unit via Bluetooth by the controller. The input noise of acquisition end is only 1.7 µV (0.2–4 kHz) with a 40-dB amplifier gain and a 100-dB common-mode rejection ratio, permitting satisfactory measurement accuracy even with an increased local impedance. sEMG sensors (Fig. 2 d) were applied for acquiring sEMG signals on beagle’s hindlegs (Extended Data Fig. 1 ). The system employs narrowband continuous phase modulation on each chip's corresponding channel. Different signals are assigned to different frequency bands and transmitted with frequency-division multiplexing. Each chip collects raw signals through surface electrodes and then modulates them into different frequency bands. A frequency shaping filter concentrates the energy within each band, reducing crosstalk among frequency bands effectively. In this way, the acquired signals can be sent simultaneously, greatly reducing the transmission delay compared to other transmission methods like time-division multiplexing. In a multi-chip transmission scenario, the system's effective wireless transmission range with accurate data delivery is about 8 m. The fine-tuning unit (Fig. 2 c) implemented logic for both the pulse generation for motion modulation and the self fine tuning according to a preset motor law map. Both the acquired sciatic neural signals and the sEMG signals were sent to the implant device as acquired responses. The acquired responses were fed into the self fine tuning module. A tuning strategy was generated according to a comparison between the preset motor law map and the acquired responses, and a set of optimized parameters were re-sent to the controller of the implant to perform stimulus inputs. Decoding motor laws of hindlegs in healthy gait We investigated the motor patterns of hindlegs in the healthy rat and beagle subjects, respectively, as standard references for subsequent motor modulation as well as the self-fine-tuning logic. For beagle (Fig. 3 a; Supplementary Video 1), we divided a single gait into two phases based on the paw-ground positional relationship: swing (no contact) and stance (with contact) (Fig. 3 b). Based on synchronously recorded kinematics, sEMG signals, and sciatic neural signals during consistent gait cycles, we found that the swing-to-stance phase ratio was approximately 1: 3 in beagle’s hindlegs (Fig. 3 b). All three signals exhibited significant periodical changes corresponding to gait cycles (Fig. 3 c–e). Furthermore, sEMG signals were phase-dependent, highlighting essential muscles for each posture: the contraction of both quadriceps femoris (QF) and biceps femoris (BF) determines the swing phase; the stance phase mainly relies on the QF contraction; and the transition from stance to swing is driven by the contraction of both BF and gastrocnemius muscle (GM) (Fig. 3 d). The range of hindleg-joint activation was also demonstrated: hip joint 95–130°, knee joint 95–125°, ankle joint 95–140°, and foot angle (paw to ground) 40–110° (Fig. 3 e). Similarly, the changes in EMG signals, sciatic neural signals, and joint activation also corresponded to gait cycles in rats (Extended Data Fig. 2 ; Supplementary Video 2). Configuration of stimulus patterns A single gait cycle of a beagle’s hindleg was segmented into three main motions, including full leg flexion (swing phase), weight acceptance (stance phase), and propulsion (transition from stance to swing) 8 . For each segment, the root mean square (RMS) ratio of the sEMG signals for the QF, tibialis anterior (TA), BF, and GM was 2.5: 1: 2.1: 1.1 (full leg flexion), 1.8: 1.2: 1.1: 1 (weight acceptance), and 1.3: 1: 2.7: 2 (propulsion). The optimal configuration of cathode and anode, along with stimulus parameters for each segment was confirmed on both sides. This ensured that the stimulus pattern could activate ipsilateral muscles according to the RMS ratio observed under healthy conditions while minimizing feedback from contralateral muscles, achieving high specificity (Fig. 4 ; Supplementary Video 3, 4). Immediate motor recovery after EES When we applied EES using an optimal stimulus pattern, beagle’s paralyzed hindlegs experienced immediate motor recovery. The absence of voluntary movement transformed into regular gait cycle, clearly divided into swing and stance phases (Supplementary Video 5), with a phase ratio of approximately 1: 3 (Fig. 5 a), consistent with a healthy gait. No significant sciatic neural signals, sEMG signals, and joint motions were observed when EES was off. However, with EES on, all these kinematic parameters exhibited periodical changes corresponding to gait cycles (Fig. 5 b). Each kinematic parameter showed a significant difference before and after EES activation, with no difference between a paralyzed beagle with EES on and a health beagle in the same experimental conditions (Fig. 5 c), confirming that EES effectively restored motion in the beagle's paralyzed hindlegs. Motor functions of SCI rats were also successfully recovered by EES (Extended Data Fig. 3 a; Supplementary Video 6). Sensitive and reliable sciatic neural signal feedback for self fine tuning Various factors can affect motor modulation by EES, such as scar, effusion, electrode displacement, and the overall physical state of the animals. These factors may alter local resistance and distort modulated motions, especially with chronic electrode implantation. To maintain effective modulation, the stimulus parameters should be adjusted in real time according to kinestate, using a self-feedback and close-loop modulation mechanism. While real-time motion capture is feasible in controlled environments, it is challenging during free movements in real-world settings. In contrast, signals delivered by wireless Bluetooth are more convenient and universal. EMG signals and peripheral neural signals are widely applied as feedback in neuroprosthetic, with EMG providing motor feedback and neural signals offering sensory feedback. Our bidirectional interface enables the collection and delivery of sEMG signals and sciatic neural signals without spatiotemporal limitations. The sciatic nerve, being mixed, controls motor functions; however, its use as a feedback mechanism in motor modulation by EES has not been reported yet. Since the peripheral nerves transmit EES signals from the spinal cord to target muscles, we observed that sciatic neural signals exhibited periodical changes with gait cycles (Fig. 3 c). This led us to speculate that these sciatic neural signals could potentially reflect kinestate. We tested the qualitative relationships between sciatic neural signals and EES signals as well as between sEMG signals and sciatic neural signals in three segments. We found that the sciatic neural signal increased in parallel at an exponential rate with increasing stimulus. sEMG signals similarly showed a positive correlation with sciatic neural signals before reaching saturation at tetanic contraction (Fig. 6 a). Therefore, we confirmed that the sciatic nerve, acting as a “fast road” between the spinal cord and target muscles, can reflect both outputted EES intensity and activated kinestate. This makes it valuable as a feedback mechanism during close-loop modulation by EES. Furthermore, since the target muscle contraction and activity generation are the termination of entire modulation process, using either sEMG signals or motion image as feedback means a long-distance feedback path and a long-time feedback period, which will inevitably have negative influence on modulation effects. Therefore, we anticipated a self-fine-tuning modulation mechanism that could immediately identify and correct errors, reducing the impact of incorrect stimulus parameters on motor effects, essentially, a near-real-time modulation system. To imitate potential stimulus bias during actual motor modulation, we designed a specific scenario: while modulating “full leg flexion”, we randomly outputted a deviated stimulus parameter and used the sciatic neural signal and sEMG signal as feedback condition, respectively. Once the signal was sent back to the laptop, the algorithm identified errors and automatically adjusted the stimulus amplitude until the feedback signal reached its standard RMS (as in healthy gait), indicating completed error correction (Fig. 6 b). We compared the modulating results based on two feedback conditions and found that after an incorrect EES parameter was outputted, sciatic neural signals presented a significantly shorter delay in detecting the error than sEMG signals (Fig. 6 b; Extended Data Fig. 4 ). A more natural and smoother motion trajectory was achieved through the feedback modulation with sciatic neural signals (Fig. 6 b; Supplementary Video 7). This confirms that the sciatic neural signal can serve as sensitive and reliable self-fine-tuning feedback for achieving near-real-time close-loop motor modulation. Near-real-time close-loop motor modulation with self fine tuning Using sciatic neural signal as feedback for modulation tuning, we established a near-real-time self-fine-tuning motor modulation mechanism (Fig. 7 a). To demonstrate how this mechanism operates and confirm its feasibility, we designed an experimental scenario. First, we calculated the RMS of sciatic neural signals during full leg flexion and propulsion activated by optimal stimulus patterns, respectively, and used them as the standard RMS for self-fine-tuning modulation. We then delivered a weak initial stimulus. After the sciatic neural signals were fed back to the laptop, an algorithm evaluated the amplitude and automatically adjusted the stimulus parameter. This process was repeated until the RMS of the sciatic neural signals matched the standard RMS (Fig. 7 b). Finally, the mechanism stopped modulation and maintained a stable stimulus output. Through this procedure, we observed that when the RMS of sciatic neural signals reached the pre-recorded standard pattern, the sEMG signals also approached an ideal state (Fig. 7 c). This demonstrates the effectiveness of feedback modulation using sciatic neural signals. The increase in sEMG signals and optimized kinematics of beagle’s hindlegs indicate that a near-real-time, self-fine-tuning modulation mechanism based on sciatic neural signals can effectively regulate muscle contraction and enhance locomotor performance (Fig. 7 d; Supplementary Video 8). Similar modulated effects were also achieved on rats (Extended Data Fig. 3 b; Supplementary Video 9). Since the beagle was supported by a vertical suspension device, no body weight was exerted during the weight acceptance phase, and we used a stable stimulus to activate motion without feedback modulation. Discussion Functional restoration has always been a key focus of rehabilitation research following nerve injury. To circumvent the challenging nerve regeneration, ANP that can bypass the injured site and reestablish the neural communication has been increasingly used as a clinical therapeutic strategy for some nervous system diseases 1 – 6 . As a breakthrough, ANP using EES managed motor recovery after SCI by temporospatially activating specific combination of spinal motor neuron pools 8 , 12 . Recently, the brain-spine interface has enabled the cortex to directly control the neural circuits below injured spinal level, achieving complete close-loop motor modulation 10 . However, motion bias always occurs inevitably even though the brain issues correct instructions. Therefore, to guarantee satisfactory modulation effects, checking and adjusting modulation promptly in response to kinestate is crucial but currently lacks solutions. Aiming to address this, we developed an ANP implant integrating a self-fine-tuning strategy. This enabled us to establish a near-real-time sciatic-neural-signal-feedback modulation mechanism with self-calibration capability, particularly suited for free activities in open environment. The neural interface applied to motor function recovery after SCI needs to accommodate the scenarios where the local impedance of electrode pads increases gradually during chronic implantation, resulting in complex and variable stimulus patterns. This requires the stimulation unit in the implant to possess a high stimulation current limit and more stimulation channels. The implant proposed in this paper enables a stimulation current limit of 16.32 mA and 16 stimulation channels in any spatiotemporal combination, surpassing existed works 13 – 19 (Supplementary Table 1). Additionally, we integrated the stimulation and acquisition units into a single chip, further reducing the size of this implant. Capogrosso et al . achieved close-loop motor modulation with EES by configuring stimulation based on real-time motor kinematics of rats and non-human primates 9 . This approach is feasible and effective in an experimental setting. However, this image-processing-based close-loop tuning requires real-time motion recording using high-resolution cameras, which is challenging for community-based activities. In contrast, wireless electrical signals delivery could be more practical. To enable electrical-signal-based feedback modulation, a bidirectional neural interface with both stimulating and recording capabilities is dispensable. Currently, most neural interfaces approved for EES are unidirectional. A recent commercial bidirectional interface modulates stimulus intensity in a close-loop manner by evaluating local impedance 20 . However, since all approved spinal neural interfaces are designed for pain treatment, their pre-set stimulation patterns do not accommodate the complex and flexible configurations needed for motor modulation. This limitation may explain why close-loop motor modulation with EES for free movements has not yet been realized in clinical practice. We suggest that updated devices will advance the modulation technologies. Self-fine-tuning modulation based on electrical signal feedback is widely used in neuroprosthetic, utilizing both EMG signals (motor feedback) and neural signals (sensory feedback), and has been confirmed to be effective 21 . The sciatic nerve, a main nerve innervating the lower limbs, regulates both motor and sensory functions. Song et al . controlled rabbit ankle motions through decoding the activity recorded in the sciatic nerve, demonstrating functional electrical stimulation 22 . Hwang et al . achieved close-loop motor function restoration in paralyzed rats by recording and stimulating sciatic nerve using an implanted nerve cuff electrode 23 . Since the sciatic nerve, delivering the signals downstream from the spinal cord, is located above the muscles and joints in the modulation pathway, it enables a much shorter feedback path than muscle and motion. Therefore, our study highlights the potential of using sciatic neural signals as feedback for near-real-time self-fine tuning, which is difficult to achieve through sEMG signal or motion image feedback. Given their wireless delivery, high sensitivity, and rapid response, sciatic neural signals hold promising application prospect for enhancing EES-based close-loop motor modulation. We expect to reconstruct gait using concise configurations and parameters. Thus, we only configured paired cathode and anode and altered stimulation amplitude to restore expected motions, based on which we selected the optimal pattern. As a result, some configurations that could achieve similar effects but be more complex might be ignored. Then, we proposed and confirmed a novel concept, that is, introducing a self-fine-tuning pathway into the motor modulation of EES through sciatic neural signal feedback. Its feasibility reveals that the signals of nerve branches innervating target muscles directly could have considerable potentials to become feedback conditions especially for modulating fine movements. Certainly, it must be considered carefully due to inevitably increased number of implants and operation difficulty 24 . Future work must also address higher integration of whole system, that is, integrate the algorithms of data processing and parameter regulating to the implant module, so that the needless delay resulting from the data transmission and processing outside the system could be avoided, further improving the time sensitivity of the feedback mechanism. In conclusion, we successfully restored hindleg gaits in paralyzed rats and beagles after SCI using a self-fine-tunning ANP implant, achieving near-real-time close-loop motor modulation with EES. This designed ANP implant characterized by its small size, low noise, and high efficacy offers a feasible and effective self-tuning method for further optimizing the clinical application of ANP in the future. Methods Animal Experiments were conducted on male Sprague Dawley rats (10 weeks, 300 g, n = 10) and beagles, including males (12–14 months, 9–11 kg, n = 5) and females (12–14 months, 8–10 kg, n = 2). The rat experiments were performed at Chinese Institute for Brain Research (CIBR) in Beijing and approved by the Animal Care and Use Committee at CIBR (AP# CIBR-IACUC-074). The beagle experiments were conducted at Huafuyuan Biological Technology Co., Ltd, Beijing, and approved by local Ethical Committee for Laboratory Animal Welfare (HFYIACUC20230216001). Locomotor training Before kinematic testing, rats and beagles underwent locomotor training on treadmill (#C300, Reao, China for beagles). The training lasted for 30 min daily for one week with a speed of 0.144 km/h for rats and 1.5 km/h for beagles. After SCI, partial vertical support was provided to assist with walking. Implantation of chronic sciatic nerve electrodes An incision was made along the sciatic nerve on the right hindleg. The sciatic nerve was dissected through the intramuscular space and gently isolated using a nerve dissector. For beagles, the isolated sciatic nerve was wrapped with a 4-mm-diameter cuff electrode featuring 1 × 4 contacts and a 500-mm-long wire (Extended Data Fig. 5 a, b; 6 a, b). For rats, a 2-mm-diameter cuff electrode with 1 × 1 contact and a 250-mm-long wire was used (Extended Data Fig. 7 a). The cuff electrodes were provided by Kedou Brain Computer Technology Co. Ltd., Suzhou, China. Implantation of chronic EMG leads in rats The procedure followed the method introduced in a previous study 9 . Briefly, we removed about 1 mm of the isolation layer from one end of the silicone-coated stainless-steel wires (#KD–998, Kedou Brain Computer Technology Co. Ltd., Suzhou, China). The exposed end was implanted into the target muscle and sutured in place using 7–0 unabsorbable sutures (Extended Data Fig. 7 b). Implantation of bidirectional neural interface For beagles, the bidirectional neural interface was encapsulated into a titanic box with a wireless charge coil and implanted subcutaneously at interscapular region (Extended Data Fig. 6 c, d). For rats, we prepared a protective shell secured with cranial nails on the skull and enclosed the bidirectional neural interface within the shell (Extended Data Fig. 7 c, d). All wires of electrodes were connected to the neural interface through subcutaneous tunnels. Signal recordings and analysis Colored markers were attached at the greater trochanter, lateral condyle, lateral malleolus, and toe 25 . Two digital video cameras (#C930c Business WebCam, Logitech) were positioned on both sides of treadmill to capture the locomotion of hindlegs. For beagles, four sEMG sensors were placed on the QF, BF, TA, and GM, respectively. The external ends of cuff electrodes and implanted EMG leads were connected to a bilateral neural interface in acquisition mode. The cameras, sEMG sensors, and neural interfaces recorded locomotion synchronously, with signals transmitted wirelessly via Bluetooth. We isolated a single gait cycle and extracted relevant kinematic parameters to analyze locomotor patterns by DeepLabCut and Electrophysiology Studio (#NeX Std, Beijing Ningju Technology Co. Ltd, Beijing, China). Autopsy To explore the anatomy of the spine and spinal cord, a beagle sacrificed for other experiment was used for autopsy. We exposed the L3–L7 spinous processes and vertebral laminae to observe the anatomic characteristics. Using appropriate surgical tools, we performed a laminectomy and dissected the dura, revealing the lumbosacral spinal cord and bilateral spinal nerves. The conus medullaris was identified at the L7 segment, with no discrepancy between the spinal cord and vertebra segments (Extended Data Fig. 8a, b). Imaging data The beagle was anesthetized with an intramuscular injection of tiletamine hydrochloride and zolazepam hydrochloride (Virbac, France, 7–25 mg/kg) before imagological examinations and was positioned prone during the procedure. The beagle underwent spinal structural magnetic resonance (MR) imaging using a 3.0T MR scanner (MAGNETOM Spectra, Siemens Healthineers) with 16-channel body and 32-channel spine array coils (Extended Data Fig. 8c–f). The standard MR imaging protocol included three pulse sequences, all performed without gadolinium-based contrast: a) two-dimensional (2D) sagittal T1-weighted turbo spin-echo (TSE) with a repetition time (TR) of 400.0 ms, echo time (TE) of 9.5 ms, and voxel size of 0.9 × 0.9 × 4.0 mm; b) 2D sagittal and axial T2-weighted TSE with a TR of 3000.0 ms, TE of 96.0 ms, and voxel size of 0.9 × 0.9 × 4.0 mm; c) three-dimensional (3D) coronal T2-weighted sampling perfection with application optimized contrast using different flip angle evolution (SPACE) with a TR of 3000.0 ms, TE of 197.0 ms, and voxel size of 1.0 × 1.0 × 1.2 mm. The 2D sagittal T1-weighted TSE and 2D sagittal and axial T2-weighted TSE were used for imaging the spine and spinal cord, while the 3D coronal T2-weighted SPACE was used for imaging the lumbosacral plexus nerves. The total scan time was approximately 30 minutes. Detailed acquisition parameters for the above three MR pulse sequences are showed in Supplementary Table 2. The beagle underwent spinal 3D CT scanning without gadolinium-based contrast on a SOMATOM Perspective spiral CT scanner (Siemens Healthineers) with 64 rows and 128 slices (Extended Data Fig. 8g–k). The detailed acquisition parameters were quality reference, 280 ms; CARE Dose4D, 130 kV; slice thickness, 0.6 mm; acquisition, 64 × 0.6 mm; pitch, 1.2; scanning time, 4.85 s; rotating time, 1.0 s; delay, 3.0 s; phase-encoding direction, head to feet. The total scan time was approximately 10 minutes. 3D Slicer was used to process the data for 3D reconstruction (Extended Data Fig. 8l, m) and measure the length of L3–L7 segment, the width of the lumbar vertebral canal, and the height of the lumbar epidural space 26 . Surgical procedures The hair on the thoracic and lumbar back, as well as the bilateral hindlegs, was removed. For beagles, atropine sulfate (Ruicheng, China) was used by intramuscular injection (0.04–0.1 ml/kg) to inhibit gland secretion 15 min before general anesthesia. Anesthesia was induced with an intramuscular injection of tiletamine hydrochloride and zolazepam hydrochloride (Virbac, France) at a dose of 7–25 mg/kg. The L3 spinous process was identified under X-ray. 0.5 ml methylene blue solution was injected percutaneously into the supraspinous ligament for marking. After endotracheal intubation, beagle was maintained anesthetized with 1.5–1.8% isoflurane in balanced oxygen during the surgical procedures. Sodium lactate ringer’s injection was administered intravenously for fluid therapy during the operation. For rats, anesthesia was induced with 5% isoflurane, and maintenance with a dose of 1.5–3% isoflurane in balanced oxygen. The surgical areas were disinfected by iodophor. Partial laminectomies were performed between each pair of adjacent vertebral laminae at the L3–L7 segment in beagles and at the T13–L2 segment in rats. A customized electrode (Beijing Ningju Technology Co. Ltd, Beijing, China, Extended Data Fig. 5 c, d) was implanted into epidural space to cover the L3–L7 segment in beagles (Extended Data Fig. 6 e) and the T13–L2 segment in rats (Extended Data Fig. 7 e). The orientation of electrode was adjusted through the gaps created by partial laminectomies and secured by suturing it to the paravertebral muscles. In beagles, a subfascial negative pressure drainage tube was placed in the incision to drain the effusion. The fascia was closed with 4–0 absorbable sutures, and the skin was closed with 4–0 unabsorbable sutures in rats and 2–0 unabsorbable sutures in beagles. Laminectomies and crush injury were conducted at T9 segment. The spinal cord was fully crushed for 2 s with forceps with a width of 0.1 mm in the last 5 mm of tips 27 . The fascia was closed with 4–0 absorbable sutures, and the skin was closed with 4–0 unabsorbable sutures in rats and 2–0 unabsorbable sutures in beagles. The animal was housed individually in a warm, dry environment with at libitum access to water and food. The incisions were disinfected and dressed daily. For beagles, ceftiofur (Sivea, China) was administered intramuscularly at 0.1 ml/kg every three days, twice, for infection prevention. Tolfenamic acid (Sivea, China) was given intramuscularly at 0.1 ml/kg every 48 hours for anti-inflammation and pain relief. The drainage device was removed based on daily drainage volume. For rats, ceftiofur sodium was injected intramuscularly at 4 mg/kg once daily for the first three postoperative days, then every other day for eight days. Urethral catheterization for beagle and manual urination for rats were performed twice daily until voluntary urination resumed. Stitches were removed two weeks postoperatively. Selection of stimulus patterns The entire procedure was performed under general anaesthesia. Based on previous studies, we positioned the cathode on the lateral sides of the electrode and the anode around it to control the stimulating current flow and target specific areas of activated motor neurons. We initiated a 0.5 mA stimulus at a frequency of 40 Hz through a bidirectional neural interface, increasing the stimulus in 0.05 mA increments 25 , 28 , 29 . A chronic cuff electrode recorded sciatic neural signals, while chronic EMG leads captured the EMG signals in rats. Wired needle electrodes penetrating muscles recorded the EMG signals in beagle. The sciatic nerve and rat EMG signals were transmitted to a laptop via the bidirectional neural interface. Wired needle electrodes were connected to a recorder that transmitted the beagle's EMG signals. Activated motions were recorded by two digital video cameras (#C930c Business WebCam, Logitech). We screened the cathode and anode configurations to identify those activating the correct motions for full leg flexion, weight acceptance, and propulsion. For each configuration, we calculated the normalized RMS of sEMG signals of each muscle against stimulus amplitude as well as the selectivity index, which was defined as the RMS ratio of sEMG signals between primary and secondary muscles 30 ; and for each segment, we chose the configuration achieving highest selectivity index as the optimal one (Extended Data Fig. 9a). When the selectivity index ≥ 90%, the corresponding stimulus amplitude range was considered appropriate for activating the motion. Within this range, the stimulus amplitude that minimally activated contralateral muscles was selected as the optimal intensity (Extended Data Fig. 9b). Algorithm for close-loop modulation The modulation process was modeled as a single-input, single-output system, where the stimulus amplitude served as the input control signal and the recorded RMS of the sciatic neural signals acted as the feedback parameter. This enabled the targeted modulation of both propulsion and full leg flexion. In the self-fine-tunning ANP implant, the stimulation and acquisition units transmitted stimulus parameters and sciatic neural signals to the server, respectively. The server performed gait phase segmentation based on the timestamp synchronization between the stimulation and acquisition units, computed the corresponding RMS of sciatic neural signals, and updated the stimulus parameters through the modulation algorithm (Fig. 7 a). Assuming the stimulus amplitude was \(\:{s}_{i,j}\) ​, where 𝑖 ≥0 represented the number of iterations, and 𝑗 = 0, 1 denoted different modulated actions (propulsion and full leg flexion). The stimulus amplitude increment after each update was \(\:ste{p}_{i,j}={s}_{i,j}-{s}_{i-1,j}\) , and the RMS of sciatic neural signals obtained from each measurement was \(\:{N}_{i,j}\) , with the target RMS being \(\:{N}_{opt,j}\) . The iterative process of the entire modulation algorithm was described by the following relationship: $$\:{\gamma\:}_{i,j}=\text{ln}(1+\left(e-1\right)\raisebox{1ex}{$\text{ln}{N}_{i,j}$}\!\left/\:\!\raisebox{-1ex}{$\text{ln}{N}_{opt,j}$}\right.)$$ $$\:\raisebox{1ex}{$ste{p}_{i+1,j}$}\!\left/\:\!\raisebox{-1ex}{$ste{p}_{i,j}$}\right.={\gamma\:}_{i,j}\bullet\:\raisebox{1ex}{${N}_{opt,j}{N}_{i-1,j}$}\!\left/\:\!\raisebox{-1ex}{${N}_{i,j}^{2}$}\right.\:$$ $$\:{s}_{i+1,j}={s}_{i,j}+ste{p}_{i+1,j}$$ \(\:{\gamma\:}_{i,j}\) represented the correction factor in the iterative algorithm, primarily used to adjust the magnitude of \(\:ste{p}_{i+1,j}\) calculated based on the exponential relationship between the RMS of sciatic nerve signals and the stimulus amplitude. As \(\:{N}_{i,j}\) ​ approached \(\:{N}_{opt,j}\) ​, \(\:\gamma\:\:\) tended to 1. \(\:ste{p}_{1,j}={s}_{1,j}-{s}_{0,j}\) ​ was a predefined value used as the initial stimulus increment for estimating the exponential relationship in the iteration process. It was primarily determined based on the magnitude of the optimal stimulus parameters, with a typical range of 0.05–0.1 mA in experiments. After iteration, when $$\:\raisebox{1ex}{$\left|{N}_{opt,j}-{N}_{i,j}\:\right|$}\!\left/\:\!\raisebox{-1ex}{${N}_{opt,j}$}\right.<5\%$$ the modulation algorithm stopped updating and outputted stable stimulus amplitude \(\:{s}_{i,j}\) . Statistical analysis All data are reported as the mean ± standard deviation. We used Prism (version 9.0, GraphPad Software, Inc., San Diego, CA) for quantifications and plotting, and DataGraph (Version 5.4, Visual Data Tools, Inc., Chapel Hill, NC) to draw stick figures and diagrams of sciatic nerve and EMG signals and joint motions. For two-group comparisons, a two-tailed t-test was performed when normality was confirmed. Statistical significance was set at P < 0.05. Declarations Data Availability The data supporting the results in this study are available in the Article and its Supplementary Information. The raw and analysed datasets generated during the study are available for research purposes from the corresponding authors upon reasonable request. Data Availability The data supporting the results in this study are available in the Article and its Supplementary Information. The raw and analysed datasets generated during the study are available for research purposes from the corresponding authors upon reasonable request. Acknowledgements We thank Z. Zhu and C. Zhang for support in the postoperative care of beagles and X. Liu, C. Sun, and T. Zhou for support in designing the self-fine-tuning ANP implant. Author Contributions M.S., Z.Z., and C.L. designed the experimental concept. M.S. and C.L. carried out all the animal surgeries and kinematic testing with the assistance from Z.C., Y.Z., T.S., and Y.C.. M.S. and Z.Z. analyzed the data. Z.Z. designed the algorithm and operated the modulation with the assistance from M.S. and X.H.. Y.Z. and Y.W designed the self-fine-tuning ANP implant. M.S. and Z.Z. prepared the manuscript with input from C.L., Y.Z., Y.W., Z.C., and Y.Z.. A.W. and S.B. provided the schematic diagrams of electrodes. J.P., X.Y., and M.Z co-supervised the project. Competing Interests The authors declare no competing interests. References Dieter, A. et al. Towards the optical cochlear implant: optogenetic approaches for hearing restoration. EMBO Mol. Med. 12, e11618 (2020). Schvartz-Leyzac, K.C. et al. Cochlear Health and Cochlear-implant Function. J. Assoc. Res. Otolaryngol. 24, 5–29 (2023). Karikari, E. & Koshechkin, K. A. Review on brain-computer interface technologies in healthcare. Biophys. Rev. 15, 1351–1358 (2023). Nojima, I. et al. Brain-Computer Interface Training Based on Brain Activity Can Induce Motor Recovery in Patients with Stroke: A Meta-Analysis. Neurorehabil. Neural. Repair. 36, 83–96 (2022). Luo, S. et al. Brain-Computer Interface: Applications to Speech Decoding and Synthesis to Augment Communication. Neurotherapeutics. 19, 263–273 (2022). Vansteensel, M. J. et al. Towards clinical application of implantable brain-computer interfaces for people with late-stage ALS: medical and ethical considerations . J. Neurol. 270, 1323–1336 (2023). Collaborators GBDSCI. Global, regional, and national burden of spinal cord injury, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet Neurol. 22, 1026–1047 (2023). Rowald, A. et al. Activity-dependent spinal cord neuromodulation rapidly restores trunk and leg motor functions after complete paralysis. Nat. Med . 28, 260–271 (2022). Capogrosso, M. et al. Configuration of electrical spinal cord stimulation through real-time processing of gait kinematics. Nat. Protoc. 13, 2031–2061 (2018). Lorach, H. et al. Walking naturally after spinal cord injury using a brain-spine interface. Nature. 618, 126–133 (2023). Zheng, J. & Meister, M. The unbearable slowness of being: Why do we live at 10 bits/s?. Neuron. 113, 192–204 (2025). Anderson, M. A. et al. Natural and targeted circuit reorganization after spinal cord injury. Nat. Neurosci. 25, 1584–1596 (2022). Zhou, A. et al. A wireless and artefact-free 128-channel neuromodulation device for closed-loop stimulation and recording in non-human primate. Nat. Biomed. Eng. 3, 15–26 (2019). Wang, Y. et al. 26.3 A Closed-Loop Neuromodulation Chipset with 2-Level Classification Achieving 1.5Vpp CM Interference Tolerance, 35dB Stimulation Artifact Rejection in 0.5ms and 97.8% Sensitivity Seizure Detection. IEEEISSCC. pp. 406–408 (2020). ElAnsary, M. et al. Bidirectional peripheral nerve interface with 64 second-order opamp-less ΔΣ ADCs and fully integrated wireless power/data transmission. IEEE Journal of Solid-State Circuits . 56, 3247-3262 (2021). Lee, J. et al. Neural recording and stimulation using wireless networks of microimplants. Nat Electron. 4, 604–614 (2021). Liu, F. et al. A fully implantable opto-electro closed-loop neural interface for motor neuron disease studies. IEEE Transactions on Biomedical Circuits and Systems . 16, 752-765 (2022). Chen, J. C. et al. A wireless millimetric magnetoelectric implant for the endovascular stimulation of peripheral nerves. Nat. Biomed. Eng. 6, 706–716 (2022). Burton, A. et al. Fully implanted battery-free high power platform for chronic spinal and muscular functional electrical stimulation. Nat. Commun . 1, 7887 (2023). Vallejo, R. et al. A New Direction for Closed-Loop Spinal Cord Stimulation: Combining Contemporary Therapy Paradigms with Evoked Compound Action Potential Sensing. J. Pain Res. 14, 3909–3918 (2021). Farina, D. et al. Toward higher-performance bionic limbs for wider clinical use. Nat. Biomed. Eng. 7, 473–485 (2023). Song, K. I. et al. Compact Neural Interface Using a Single Multichannel Cuff Electrode for a Functional Neuromuscular Stimulation System. Ann. Biomed. Eng. 47, 754–766 (2019). Hwang, Y. E. et al. Closed-Loop Control of Functional Electrical Stimulation Using a Selectively Recording and Bidirectional Nerve Cuff Interface. IEEE Trans. Neural. Syst. Rehabil. Eng. 32, 504–513 (2024). Habibollahi, M. et al. Active Neural Interface Circuits and Systems for Selective Control of Peripheral Nerves: A Review. IEEE Trans. Biomed. Circuits Syst. 18, 954–975 (2024). Bonizzato, M. et al. Multi-pronged neuromodulation intervention engages the residual motor circuitry to facilitate walking in a rat model of spinal cord injury. Nat. Commun. 12, 1925 (2021). Fedorov, A. et al. 3D Slicer as an Image Computing Platform for the Quantitative Imaging Network. Magnetic Resonance Imaging. 30, 1323–1341 (2012). Li, Y. et al. Microglia-organized scar-free spinal cord repair in neonatal mice. Nature. 587, 613–618 (2020). Slawinska, U. et al. Comment on "Restoring voluntary control of locomotion after paralyzing spinal cord injury". Science. 338, 328; author reply (2012). Barra, B. et al. Epidural electrical stimulation of the cervical dorsal roots restores voluntary upper limb control in paralyzed monkeys. Nat. Neurosci. 25, 924–934 (2022). Wagner, F. B. et al. Targeted neurotechnology restores walking in humans with spinal cord injury. Nature. 563, 65–71 (2018). Additional Declarations There is NO Competing Interest. Supplementary Files InventoryofSupportingInformation.docx Article File SupplementaryData.pdf Article File SupplementaryVideo1.mp4 Decoding motor laws of beagle’s hindlegs in healthy gait. SupplementaryVideo2.mp4 Decoding motor laws of rat’s hindlegs in healthy gait. SupplementaryVideo3.mp4 Configuration of stimulus patterns on beagle’s left hindleg. SupplementaryVideo4.mp4 Configuration of stimulus patterns on beagle’s right hindleg. SupplementaryVideo5.mp4 Beagle’s immediate motor recovery after EES. SupplementaryVideo6.mp4 Rat’s immediate motor recovery after EES. SupplementaryVideo7.mp4 Sciatic neural signal feedback versus sEMG signal feedback on beagle. SupplementaryVideo8.mp4 Near-real-time close-loop self-fine-tuning motor modulation on beagle. SupplementaryVideo9.mp4 Near-real-time close-loop self-fine-tuning motor modulation on rat. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6244028","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":443075937,"identity":"cc97dcec-88f5-4db9-a14a-89f0a96769a1","order_by":0,"name":"Milin Zhang","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0001-7544-1837","institution":"Tsinghua University","correspondingAuthor":true,"prefix":"","firstName":"Milin","middleName":"","lastName":"Zhang","suffix":""},{"id":443075938,"identity":"a8ae530b-e90e-495e-ad03-7bc07498e389","order_by":1,"name":"Mengchun Sun","email":"","orcid":"","institution":"Nankai University","correspondingAuthor":false,"prefix":"","firstName":"Mengchun","middleName":"","lastName":"Sun","suffix":""},{"id":443075939,"identity":"16668b3a-68a8-481e-bd08-0298e0f3657f","order_by":2,"name":"Ziyao Zhao","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Ziyao","middleName":"","lastName":"Zhao","suffix":""},{"id":443075940,"identity":"f8e758ed-280c-481f-805b-ad618c638cb2","order_by":3,"name":"Chaochao Li","email":"","orcid":"","institution":"Chinese PLA General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Chaochao","middleName":"","lastName":"Li","suffix":""},{"id":443075941,"identity":"f272cd8c-fba1-4d8f-b84f-3dc5aacaae61","order_by":4,"name":"Yuwei Zhang","email":"","orcid":"","institution":"Beijing Ningju Technology","correspondingAuthor":false,"prefix":"","firstName":"Yuwei","middleName":"","lastName":"Zhang","suffix":""},{"id":443075942,"identity":"d9f3600e-5b9d-4c74-8ee4-accaa2638219","order_by":5,"name":"Xinyi Hong","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Xinyi","middleName":"","lastName":"Hong","suffix":""},{"id":443075943,"identity":"280c4d8a-9721-4812-8ae9-59b12ca4c36d","order_by":6,"name":"Yusong Wu","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Yusong","middleName":"","lastName":"Wu","suffix":""},{"id":443075944,"identity":"3653d665-c98a-4a67-908f-e7e4205674c3","order_by":7,"name":"Zhe Chen","email":"","orcid":"","institution":"Beijing Ningju Technology","correspondingAuthor":false,"prefix":"","firstName":"Zhe","middleName":"","lastName":"Chen","suffix":""},{"id":443075945,"identity":"4ec61608-335a-49be-8895-51e9a3f3c6f7","order_by":8,"name":"Yue Zhong","email":"","orcid":"","institution":"Beijing Ningju Technology","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Zhong","suffix":""},{"id":443075946,"identity":"3d9a17e4-4a70-456a-8c0e-d111d66c0962","order_by":9,"name":"Tianqi Su","email":"","orcid":"","institution":"Harbin Medical University Cancer Hospital","correspondingAuthor":false,"prefix":"","firstName":"Tianqi","middleName":"","lastName":"Su","suffix":""},{"id":443075947,"identity":"75ddc93d-e611-4efe-967c-9fea1c8bacf2","order_by":10,"name":"Yuhui Cui","email":"","orcid":"","institution":"The First Medical Center of Chinese PLA General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuhui","middleName":"","lastName":"Cui","suffix":""},{"id":443075948,"identity":"eb1ec9ae-8d69-485a-8315-be375675949d","order_by":11,"name":"Anhe Wang","email":"","orcid":"","institution":"State Key Laboratory of Biochemical Engineering, Institute of Process Engineering, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Anhe","middleName":"","lastName":"Wang","suffix":""},{"id":443075949,"identity":"103b4b3b-fe2d-4bdd-a787-999df4751a7f","order_by":12,"name":"Shuo Bai","email":"","orcid":"https://orcid.org/0000-0001-5360-3399","institution":"Institute of Process Engineering, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Shuo","middleName":"","lastName":"Bai","suffix":""},{"id":443075950,"identity":"21b01180-2ba6-42be-99db-22dddc2d2617","order_by":13,"name":"Jiang Peng","email":"","orcid":"","institution":"Chinese PLA General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiang","middleName":"","lastName":"Peng","suffix":""},{"id":443075951,"identity":"3d8213d5-8754-49f4-a8bd-b45f7e957183","order_by":14,"name":"Xinguang Yu","email":"","orcid":"","institution":"The First Medical Center of PLA General Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xinguang","middleName":"","lastName":"Yu","suffix":""}],"badges":[],"createdAt":"2025-03-17 11:20:26","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6244028/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6244028/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":80780900,"identity":"164b319a-85f1-4e53-88ce-b8e387435462","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":199530,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eArtificial neural pathway (ANP) functionality demonstration.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Paradigm of ANP consisting of reading and writing pathways. ANP bypasses damaged neural connections and reestablishes neural signal communication through the body. \u003cstrong\u003eb,\u003c/strong\u003e ANP enabling the amputee to control the neuroprosthesis by autonomous thoughts. \u003cstrong\u003ec,\u003c/strong\u003e ANP assisting in delivering the motion attention from the motor cerebral cortex to the intact spinal cord below the lesion. \u003cstrong\u003ed,\u003c/strong\u003e Typical epidural electrical stimulation (EES) operating with a pre-set program. \u003cstrong\u003ee,\u003c/strong\u003e Typical ANP using EES allowing the patients with spinal cord injury to control motions autonomously. \u003cstrong\u003ef,\u003c/strong\u003e Self-fine-tunning ANP implant realising a rapid and precise self-feedback mechanism appropriate for free activities and further optimizing the modulation effects of EES. Abbreviation: ANP, artificial neural pathway; PN, peripheral nerve; sEMG, surface electromyograph. Illustrations created using BioRender (https://biorender.com).\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/d717fb88180f7839daec466a.png"},{"id":80780889,"identity":"0c4b18b4-ba71-4aa8-9e9c-93221147daf8","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":401650,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSchematic diagram of motor modulation by a self-fine-tuning artificial neural pathway (ANP) implant.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Paradigm of self-fine-tuning ANP. \u003cstrong\u003eb,\u003c/strong\u003e Motion encoder generating corresponding stimulus output against obtained parameters. \u003cstrong\u003ec,\u003c/strong\u003e Sciatic neural recorder sending the collected signals to the controller of the implant after analog-to-digital conversion. \u003cstrong\u003ed,\u003c/strong\u003e Fine-tuning unit (vitro) receiving real-time responses from the recorder and performing fine-tuning algorithm to optimize parameters. \u003cstrong\u003ee,\u003c/strong\u003e Surface electromyograph sensors acquiring the activities of correlated muscles to construct motor law map and verify the modulation along with motion recording. Abbreviations: sEMG, surface electromyograph; Amp., amplitude; Freq., frequency; PMU, power management unit; LNA, low noise amplifier; ADC, analog-to-digital converter. Partial illustrations created using BioRender (https://biorender.com).\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/d6c60daea3b74bfd040e68dc.png"},{"id":80780887,"identity":"6f7fca9c-a546-44d5-962f-476358435cd5","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":388885,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDecoding motor laws of beagle’s hindleg in healthy gait cycles.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Schematic diagram of the experimental platform, consisting of treadmill running at 1.5 km/h; camera capturing the beagle’s gait; surface electromyograph (sEMG) sensors collecting the sEMG signals from the hindlegs; and chronic cuff electrode recording sciatic neural signals. \u003cstrong\u003eb,\u003c/strong\u003e Motion capture and stick diagram. Red sticks indicate the swing phase; blue sticks indicate the stance phase. Swing phase lasts 0.25\u0026nbsp;±\u0026nbsp;0.02 s (\u003cem\u003en\u003c/em\u003e = 7 beagles); stance phase lasts 0.72\u0026nbsp;±\u0026nbsp;0.05 s (\u003cem\u003en\u003c/em\u003e = 7 beagles). \u003cstrong\u003ec,\u003c/strong\u003e Sciatic neural signals showing periodical changes consistent with gait cycles. \u003cstrong\u003ed,\u003c/strong\u003e Periodical and phase-dependent sEMG signals of hindlegs. \u003cstrong\u003ee,\u003c/strong\u003e Periodical changes in joint angles during walking. Abbreviations: sEMG, surface electromyography; SN, sciatic nerve; QF, quadriceps femoris; BF, biceps femoris; TA, tibialis anterior muscle; GM, gastrocnemius muscle. Illustrations in \u003cstrong\u003ea\u003c/strong\u003e, \u003cstrong\u003ec\u003c/strong\u003e, \u003cstrong\u003ed\u003c/strong\u003e, and \u003cstrong\u003ee\u003c/strong\u003e created using BioRender (https://biorender.com).\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/24c47de4f96bc19e5bc44ea9.png"},{"id":80781400,"identity":"c133d3c6-73ef-4c84-b25c-ee92e1dede9d","added_by":"auto","created_at":"2025-04-17 04:41:37","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":298577,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfiguration of stimulus patterns. \u003c/strong\u003eThree segments depend on the contraction of highlighted muscles, respectively. The root mean square ratio of the activated surface electromyograph signals matches that of a healthy condition (full leg flexion, 2.5: 1: 2.1: 1.1; weight acceptance, 1.8: 1.2: 1.1: 1; propulsion, 1.3: 1: 2.7: 2; \u003cem\u003en\u003c/em\u003e = 10 steps). Stimulus patterns with optimal configuration of cathode and anode, along with stimulus parameters, ideally activate ipsilateral muscles and minimally affect contralateral ones. Abbreviations: QF, quadriceps femoris; BF, biceps femoris; TA, tibialis anterior muscle; GM, gastrocnemius muscle; RMS, root mean square; sEMG, surface electromyography; EES, epidural electrical stimulation; L, left; R, right. Illustrations created using BioRender (https://biorender.com).\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/88d5fdd0ad141a98781fae0a.png"},{"id":80780881,"identity":"963aa8d1-1213-41ec-bbec-124fd10a6ebc","added_by":"auto","created_at":"2025-04-17 04:33:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":453545,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eImmediate motor recovery after epidural electrical stimulation (EES). a,\u003c/strong\u003e Beagle with spinal cord injury shows no significant gait with EES off. Meaningful and stable gait cycles of the hindlegs occur immediately after EES is turned on. Red sticks indicate swing phase; blue sticks indicate stance phase. Swing phase lasts 0.28 ± 0.02 s (\u003cem\u003en\u003c/em\u003e = 7 beagles); stance phase lasts 0.84 ± 0.05 s (\u003cem\u003en\u003c/em\u003e = 7 beagles). \u003cstrong\u003eb,\u003c/strong\u003eComparisons of sciatic neural signals, surface electromyograph (sEMG) signals, and joint motions between EES off and on. When EES is off, the paralyzed hindlegs show no significant sciatic neural signals, sEMG signals, or joint motions. After EES is turned on, these three kinematic parameters exhibit periodical changes like the conditions in the healthy gait cycles. \u003cstrong\u003ec,\u003c/strong\u003eBar plots comparing the root mean square of sEMG signals and the joint motions between paralysis with EES on and off, and between the healthy gait and the paralysis with EES on, respectively (\u003cem\u003en\u003c/em\u003e = 7 beagles). Abbreviations: EES, epidural electrical stimulation; QF, quadriceps femoris; BF, biceps femoris; TA, tibialis anterior muscle; GM, gastrocnemius muscle. Two-tailed t-test; ns, not significant; ****, \u003cem\u003eP \u003c/em\u003e\u0026lt;0.001.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/5ea6757205bd70557279d554.png"},{"id":80781629,"identity":"1c8732a2-2e23-46fd-8832-0b238cafa074","added_by":"auto","created_at":"2025-04-17 04:49:37","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":479400,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cbr\u003e\n\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSciatic neural signal serving as feedback condition in self-fine-tuning motor modulation. a,\u003c/strong\u003e Qualitative relationship among the stimulus amplitude, the sciatic neural signals, and the surface electromyograph (sEMG) signals. The sciatic neural signals show positive exponential correlation with stimulus amplitude in each segment (full leg flexion, R\u003csup\u003e2 \u003c/sup\u003e= 0.9715; weight acceptance, R\u003csup\u003e2 \u003c/sup\u003e= 0.9912; propulsion, R\u003csup\u003e2 \u003c/sup\u003e= 0.9925); the sEMG signals also demonstrate a positive correlation with stimulus amplitude in each segment before tetanic contraction (full leg flexion R\u003csup\u003e2 \u003c/sup\u003e= 0.9912; weight acceptance R\u003csup\u003e2 \u003c/sup\u003e= 0.9881; propulsion R\u003csup\u003e2 \u003c/sup\u003e= 0.9672) and then become saturated. \u003cstrong\u003eb,\u003c/strong\u003e Comparison of feedback effects using the sciatic neural signals and the sEMG signals. Incorrect stimulus is detected at 14.00 ± 3.10 ms after output by sciatic neural signals but at 70.61 ± 3.72 ms by sEMG signals, indicating a significant delay of sEMG signal feedback (\u003cem\u003en\u003c/em\u003e = 21 cycles: 3 cycles per beagle, 7 beagles totally). Stick diagrams illustrate that delayed feedback via sEMG signals affects the motion trajectory more than the rapid feedback by sciatic neural signals. Bar plots compare the joint motions between the normal steps and the steps modulated by the sciatic neural signals and the sEMG signals, respectively (\u003cem\u003en\u003c/em\u003e = 7 beagles). Abbreviations: SN, sciatic nerve; QF, quadriceps femoris; BF, biceps femoris; TA, tibialis anterior muscle; GM, gastrocnemius muscle; RMS, root mean square; sEMG, surface electromyography. Two-tailed t-test; ns, not significant; ****, \u003cem\u003eP \u003c/em\u003e\u0026lt;0.001. Illustrations in \u003cstrong\u003eb\u003c/strong\u003e created using BioRender (https://biorender.com).\u0026nbsp;\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/3c52274f72ea022a2eb066d4.png"},{"id":80780894,"identity":"5e0e72a8-e063-419c-bbe3-86cf42168378","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":444928,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSelf-fine-tuning motor modulation with sciatic neural signal feedback. a,\u003c/strong\u003e Procedure of self-fine-tuning modulation. Red star indicates the targeting pre-recorded standard root mean square (RMS). \u003cstrong\u003eb,\u003c/strong\u003e RMS of sciatic neural signals achieves its standard pattern after 5-cycle stimulus output. Surface electromyograph signals increase synchronously with sciatic neural signals and approaches their ideal state as the RMS of sciatic neural signals achieves a standard pattern. \u003cstrong\u003ec,\u003c/strong\u003e A near-real-time, close-loop, and self-fine-tuning modulation mechanism based on sciatic neural signal feedback effectively regulates muscle contraction and enhances locomotor performance. Data are presented as mean ± standard deviation (\u003cem\u003en\u003c/em\u003e = 7 beagles). Abbreviations: SN, sciatic nerve; QF, quadriceps femoris; BF, biceps femoris; TA, tibialis anterior muscle; GM, gastrocnemius muscle; RMS, root mean square; sEMG, surface electromyography. Illustration in \u003cstrong\u003ea\u003c/strong\u003e created using BioRender (https://biorender.com).\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/dd8efff31ef03d8c6c71166e.png"},{"id":102749147,"identity":"66cfcf78-37de-4355-acfe-f8eeb5c40582","added_by":"auto","created_at":"2026-02-16 09:12:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3755644,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/3fe57673-798e-4a80-9b46-632ce8aefcfb.pdf"},{"id":80780877,"identity":"cb4ea06e-d9db-44b8-8d42-61707a2f2766","added_by":"auto","created_at":"2025-04-17 04:33:36","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2620296,"visible":true,"origin":"","legend":"Article File","description":"","filename":"InventoryofSupportingInformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/417f4fd76f467daf238090a4.docx"},{"id":80780891,"identity":"d8b33e5a-7b50-4ff3-9030-de9b03eee97f","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":102887,"visible":true,"origin":"","legend":"Article File","description":"","filename":"SupplementaryData.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/bcbc14bb5aca6664f2c9f194.pdf"},{"id":80780892,"identity":"30013fb0-495b-479d-a887-45f2fb04bf99","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"mp4","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":14801823,"visible":true,"origin":"","legend":"Decoding motor laws of beagle\u0026#x2019;s hindlegs in healthy gait.","description":"","filename":"SupplementaryVideo1.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/4e30432abb8c7023cf728cf0.mp4"},{"id":80781402,"identity":"9f9d14dc-7110-4c25-8baf-5f1f12466ec4","added_by":"auto","created_at":"2025-04-17 04:41:37","extension":"mp4","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":3489711,"visible":true,"origin":"","legend":"Decoding motor laws of rat\u0026#x2019;s hindlegs in healthy gait.","description":"","filename":"SupplementaryVideo2.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/d81aab0335319d7be0decbf2.mp4"},{"id":80780915,"identity":"f207f550-37fe-44c0-9f74-a58aa44bc3d8","added_by":"auto","created_at":"2025-04-17 04:33:38","extension":"mp4","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":53563883,"visible":true,"origin":"","legend":"Configuration of stimulus patterns on beagle\u0026#x2019;s left hindleg.","description":"","filename":"SupplementaryVideo3.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/4e332534cb277be2668d9d84.mp4"},{"id":80780914,"identity":"6c9f99fe-b961-4fc9-b607-c2305a2967d8","added_by":"auto","created_at":"2025-04-17 04:33:38","extension":"mp4","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":56746964,"visible":true,"origin":"","legend":"Configuration of stimulus patterns on beagle\u0026#x2019;s right hindleg.","description":"","filename":"SupplementaryVideo4.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/a4d79feb3cccd48f4c15e029.mp4"},{"id":80780896,"identity":"1b56ee89-6139-4cd5-b656-ecb76e7dac91","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"mp4","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":28180210,"visible":true,"origin":"","legend":"Beagle\u0026#x2019;s immediate motor recovery after EES.","description":"","filename":"SupplementaryVideo5.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/f87ed93ab400249f642184e5.mp4"},{"id":80782204,"identity":"361c3f4c-1821-4dd1-a19b-c8feadf5095b","added_by":"auto","created_at":"2025-04-17 04:57:37","extension":"mp4","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":28622766,"visible":true,"origin":"","legend":"Rat\u0026#x2019;s immediate motor recovery after EES.","description":"","filename":"SupplementaryVideo6.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/de11b9d15d41a6d5f998e501.mp4"},{"id":80780920,"identity":"f4ed3783-70a9-4b5c-ab30-3ef2ed0f14c9","added_by":"auto","created_at":"2025-04-17 04:33:39","extension":"mp4","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":53260994,"visible":true,"origin":"","legend":"Sciatic neural signal feedback versus sEMG signal feedback on beagle.","description":"","filename":"SupplementaryVideo7.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/ad314a7ad028a07289e7cac2.mp4"},{"id":80781406,"identity":"e9c25ee3-66bc-4697-862b-7abaf5206671","added_by":"auto","created_at":"2025-04-17 04:41:37","extension":"mp4","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":23308937,"visible":true,"origin":"","legend":"Near-real-time close-loop self-fine-tuning motor modulation on beagle.","description":"","filename":"SupplementaryVideo8.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/a1d6fccdd3228bcb84d48f94.mp4"},{"id":80780902,"identity":"c3d09918-d1c9-4bef-9fe5-52090acc0a85","added_by":"auto","created_at":"2025-04-17 04:33:37","extension":"mp4","order_by":11,"title":"","display":"","copyAsset":false,"role":"supplement","size":17708914,"visible":true,"origin":"","legend":"Near-real-time close-loop self-fine-tuning motor modulation on rat.","description":"","filename":"SupplementaryVideo9.mp4","url":"https://assets-eu.researchsquare.com/files/rs-6244028/v1/3247d0938e279560f2d909bb.mp4"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"A Wireless Self-Fine-Tuning Artificial Neural Pathway Implant for Near-Real-Time Close-Loop Motor Modulation","fulltext":[{"header":"Introduction","content":"\u003cp\u003eArtificial neural pathway (ANP) implants hold transformative potential for empowering individuals with disabilities by reestablishing communication between the broken neural pathway, realising the restoration of the lost functions, which enhances the mobility and quality of life. For those with sensory impairments, such as hearing loss, implants like cochlear devices can restore the ability to perceive sound\u003csup\u003e1,2\u003c/sup\u003e, significantly improving communication and social interaction. Furthermore, these technologies can aid in managing neurological disorders by modulating brain activity, offering relief from symptoms of conditions like Parkinson\u0026rsquo;s disease and epilepsy\u003csup\u003e3,4\u003c/sup\u003e. By translating neural signals into digital commands, implants facilitate communication for individuals with severe disabilities, such as amyotrophic lateral sclerosis, allowing them to express themselves and interact with the world\u003csup\u003e5,6\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThere are two key pathways in a typical ANP implant, i.e. the reading pathway and the writing pathway (Fig. 1a). The reading pathway interfaces with the neural networks located above the injury site, while the writing pathway integrates with the neural circuits situated below the lesion. In recent years, ANP, working synergistically with a control central plant, has been applied in clinical practice to restore complex motor functions, e.g. upper limp mobility restoration using neuroprosthesis (Fig. 1b) and the recovery of walking ability in patients with permanent paralysis caused by spinal cord injury (SCI) (Fig. 1c).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSCI is a serious and intractable central nervous system injury that leads to significant impairment of physical functions, particularly sensory and motor capabilities. It was reported that about 250,000 to 500,000 people suffer from SCI every year globally\u003csup\u003e7\u003c/sup\u003e. Recent advancement in ANP has revolutionized the clinical treatment of SCI with epidural electrical stimulation (EES) to become the gold standard in association with neurorehabilitation\u003csup\u003e8\u003c/sup\u003e. The extensive clinical experience of treating intractable pain with EES also promotes its application in motor recovery. Classical EES integrated only the writing pathway, operating with a pre-set programme and manual modification of stimulus parameters rather than automatic correction according to patient\u0026rsquo;s kinestate (Fig. 1d), which consumes much manpower and weakens modulating effects. Capogrosso \u003cem\u003eet al\u003c/em\u003e. proposed configuration of EES through real-time processing of gait kinematics and confirmed its feasibility\u003csup\u003e9\u003c/sup\u003e. However, it is only suitable for the motions in a closed experimental environment, since real-time gait recording by cameras is hard to manage for free motions in an open environment like real life scenes. Lorach \u003cem\u003eet al\u003c/em\u003e. introduced an extra reading pathway implanted in the skull to acquire motion attention through the acquired neural activities, enabling a whole close loop of brain, spinal cord, and computer, allowing patients to control motions autonomously\u003csup\u003e10\u003c/sup\u003e (Fig. 1e). However, this close loop still lacks a \u0026ldquo;supervisor\u0026rdquo; that can rapidly check if the modulated motions are correct and accurate, since as a \u0026ldquo;commander\u0026rdquo;, the brain has to spend hundreds of milliseconds on visual feedback after wrong motions occurring\u003csup\u003e11\u003c/sup\u003e, limiting the\u0026nbsp;promptness and precision of EES modulation to some extent.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTherefore, we introduced a self-fine-tuning pathway into the ANP system (Fig. 1f) to establish a rapid self-feedback mechanism appropriate for free activities to achieve precise motor modulation by EES with self tuning. The fine-tuning pathway was attached to the writing pathway using a wireless implanted bidirectional neural interface featuring compact size, low input noise, and high energy efficiency, with different functionalities including reading, writing, and bidirectional control. The writing implant allows a wider stimulus voltage range and a higher upper limit of stimulus current amplitude for modulation; the reading implant enables wireless acquisition to collect neural activities from sciatic nerve or surface electromyograph (sEMG) sensors; and the bidirectional implant enables self-fine-tuning feedback for the modulator according to acquired neighbouring activities. Furthermore, compared with the time-consuming image processing and manual modification, sciatic neural signals spend only about 14 ms detecting errors and thus are fed into the self-fine-tuning pathway as rapid feedback for precise modulation tuning. Without manual interventions, we restored gaits of hindlegs in paralyzed animals with EES by a self-fine-tuning, near-real-time, and close-loop motor modulation mechanism.\u003c/p\u003e\n\u003cp\u003eThe scalability and adaptability of the self-fine-tuning ANP make it accommodate varying degrees of paralysis and individual patient needs. Different combinations of implant locations enable the generation of pathways that coordinate with the natural neural circuit for different scenarios, such as walking, activities of upper limbs, and urination. Ultimately, this innovative design holds the potential to significantly improve the quality of life for paralysis patients by offering more robust and flexible treatment options.\u0026nbsp;\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003eSelf-fine-tuning ANP implant for bidirectional interactions\u003c/h2\u003e \u003cp\u003eThe self-fine-tuning ANP implant comprises two functional modules, that is, motion function encoding through modulation (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ea) and neural activity recording for fine tuning (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). A fine-tuning unit (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ec) associated with these two functional modules was implanted subcutaneously for efficient data transfer with other distributed modules (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ed).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe motion function encoding module (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ea) supports a voltage output up to ± 12 V by integrating a boost converter. The high voltage output capability ensures effective stimulation in case of a possibly increased local impedance after chronic electrode implantation. It features 16 current source channels with a 14-bit resolution and allows a delivered current from 1 µA to 16.384 mA. The 16 channels can be configured independently for simultaneous stimulation in various combinations of single-ended mode or differential mode. Single-ended mode allows a 12-V voltage upper limit, which is increased to 24 V in differential modes by pairing two current channels. The time resolution is precise down to 1 µs, and the system can generate arbitrary stimulation patterns lasting from microseconds to seconds.\u003c/p\u003e \u003cp\u003eFor the sciatic neural signal recording module (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ec), a boost converter configures a ± 2.5 V voltage at the front end. An analog-to-digital converter converts the amplified signals into digital signals that will be transmitted to the fine-tuning unit via Bluetooth by the controller. The input noise of acquisition end is only 1.7 µV (0.2–4 kHz) with a 40-dB amplifier gain and a 100-dB common-mode rejection ratio, permitting satisfactory measurement accuracy even with an increased local impedance.\u003c/p\u003e \u003cp\u003esEMG sensors (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ed) were applied for acquiring sEMG signals on beagle’s hindlegs (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The system employs narrowband continuous phase modulation on each chip's corresponding channel. Different signals are assigned to different frequency bands and transmitted with frequency-division multiplexing. Each chip collects raw signals through surface electrodes and then modulates them into different frequency bands. A frequency shaping filter concentrates the energy within each band, reducing crosstalk among frequency bands effectively. In this way, the acquired signals can be sent simultaneously, greatly reducing the transmission delay compared to other transmission methods like time-division multiplexing. In a multi-chip transmission scenario, the system's effective wireless transmission range with accurate data delivery is about 8 m.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe fine-tuning unit (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ec) implemented logic for both the pulse generation for motion modulation and the self fine tuning according to a preset motor law map. Both the acquired sciatic neural signals and the sEMG signals were sent to the implant device as acquired responses. The acquired responses were fed into the self fine tuning module. A tuning strategy was generated according to a comparison between the preset motor law map and the acquired responses, and a set of optimized parameters were re-sent to the controller of the implant to perform stimulus inputs.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDecoding motor laws of hindlegs in healthy gait\u003c/h2\u003e \u003cp\u003eWe investigated the motor patterns of hindlegs in the healthy rat and beagle subjects, respectively, as standard references for subsequent motor modulation as well as the self-fine-tuning logic. For beagle (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea; Supplementary Video 1), we divided a single gait into two phases based on the paw-ground positional relationship: swing (no contact) and stance (with contact) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb). Based on synchronously recorded kinematics, sEMG signals, and sciatic neural signals during consistent gait cycles, we found that the swing-to-stance phase ratio was approximately 1: 3 in beagle’s hindlegs (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb). All three signals exhibited significant periodical changes corresponding to gait cycles (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec–e). Furthermore, sEMG signals were phase-dependent, highlighting essential muscles for each posture: the contraction of both quadriceps femoris (QF) and biceps femoris (BF) determines the swing phase; the stance phase mainly relies on the QF contraction; and the transition from stance to swing is driven by the contraction of both BF and gastrocnemius muscle (GM) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed). The range of hindleg-joint activation was also demonstrated: hip joint 95–130°, knee joint 95–125°, ankle joint 95–140°, and foot angle (paw to ground) 40–110° (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee). Similarly, the changes in EMG signals, sciatic neural signals, and joint activation also corresponded to gait cycles in rats (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003e; Supplementary Video 2).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eConfiguration of stimulus patterns\u003c/h3\u003e\n\u003cp\u003eA single gait cycle of a beagle’s hindleg was segmented into three main motions, including full leg flexion (swing phase), weight acceptance (stance phase), and propulsion (transition from stance to swing)\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. For each segment, the root mean square (RMS) ratio of the sEMG signals for the QF, tibialis anterior (TA), BF, and GM was 2.5: 1: 2.1: 1.1 (full leg flexion), 1.8: 1.2: 1.1: 1 (weight acceptance), and 1.3: 1: 2.7: 2 (propulsion). The optimal configuration of cathode and anode, along with stimulus parameters for each segment was confirmed on both sides. This ensured that the stimulus pattern could activate ipsilateral muscles according to the RMS ratio observed under healthy conditions while minimizing feedback from contralateral muscles, achieving high specificity (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e; Supplementary Video 3, 4).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eImmediate motor recovery after EES\u003c/h3\u003e\n\u003cp\u003eWhen we applied EES using an optimal stimulus pattern, beagle’s paralyzed hindlegs experienced immediate motor recovery. The absence of voluntary movement transformed into regular gait cycle, clearly divided into swing and stance phases (Supplementary Video 5), with a phase ratio of approximately 1: 3 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea), consistent with a healthy gait. No significant sciatic neural signals, sEMG signals, and joint motions were observed when EES was off. However, with EES on, all these kinematic parameters exhibited periodical changes corresponding to gait cycles (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb). Each kinematic parameter showed a significant difference before and after EES activation, with no difference between a paralyzed beagle with EES on and a health beagle in the same experimental conditions (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec), confirming that EES effectively restored motion in the beagle's paralyzed hindlegs. Motor functions of SCI rats were also successfully recovered by EES (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea; Supplementary Video 6).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eSensitive and reliable sciatic neural signal feedback for self fine tuning\u003c/h3\u003e\n\u003cp\u003eVarious factors can affect motor modulation by EES, such as scar, effusion, electrode displacement, and the overall physical state of the animals. These factors may alter local resistance and distort modulated motions, especially with chronic electrode implantation. To maintain effective modulation, the stimulus parameters should be adjusted in real time according to kinestate, using a self-feedback and close-loop modulation mechanism. While real-time motion capture is feasible in controlled environments, it is challenging during free movements in real-world settings. In contrast, signals delivered by wireless Bluetooth are more convenient and universal. EMG signals and peripheral neural signals are widely applied as feedback in neuroprosthetic, with EMG providing motor feedback and neural signals offering sensory feedback. Our bidirectional interface enables the collection and delivery of sEMG signals and sciatic neural signals without spatiotemporal limitations. The sciatic nerve, being mixed, controls motor functions; however, its use as a feedback mechanism in motor modulation by EES has not been reported yet. Since the peripheral nerves transmit EES signals from the spinal cord to target muscles, we observed that sciatic neural signals exhibited periodical changes with gait cycles (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec). This led us to speculate that these sciatic neural signals could potentially reflect kinestate. We tested the qualitative relationships between sciatic neural signals and EES signals as well as between sEMG signals and sciatic neural signals in three segments. We found that the sciatic neural signal increased in parallel at an exponential rate with increasing stimulus. sEMG signals similarly showed a positive correlation with sciatic neural signals before reaching saturation at tetanic contraction (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea). Therefore, we confirmed that the sciatic nerve, acting as a “fast road” between the spinal cord and target muscles, can reflect both outputted EES intensity and activated kinestate. This makes it valuable as a feedback mechanism during close-loop modulation by EES.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFurthermore, since the target muscle contraction and activity generation are the termination of entire modulation process, using either sEMG signals or motion image as feedback means a long-distance feedback path and a long-time feedback period, which will inevitably have negative influence on modulation effects. Therefore, we anticipated a self-fine-tuning modulation mechanism that could immediately identify and correct errors, reducing the impact of incorrect stimulus parameters on motor effects, essentially, a near-real-time modulation system. To imitate potential stimulus bias during actual motor modulation, we designed a specific scenario: while modulating “full leg flexion”, we randomly outputted a deviated stimulus parameter and used the sciatic neural signal and sEMG signal as feedback condition, respectively. Once the signal was sent back to the laptop, the algorithm identified errors and automatically adjusted the stimulus amplitude until the feedback signal reached its standard RMS (as in healthy gait), indicating completed error correction (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb). We compared the modulating results based on two feedback conditions and found that after an incorrect EES parameter was outputted, sciatic neural signals presented a significantly shorter delay in detecting the error than sEMG signals (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb; Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). A more natural and smoother motion trajectory was achieved through the feedback modulation with sciatic neural signals (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb; Supplementary Video 7). This confirms that the sciatic neural signal can serve as sensitive and reliable self-fine-tuning feedback for achieving near-real-time close-loop motor modulation.\u003c/p\u003e\n\u003ch3\u003eNear-real-time close-loop motor modulation with self fine tuning\u003c/h3\u003e\n\u003cp\u003eUsing sciatic neural signal as feedback for modulation tuning, we established a near-real-time self-fine-tuning motor modulation mechanism (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea). To demonstrate how this mechanism operates and confirm its feasibility, we designed an experimental scenario. First, we calculated the RMS of sciatic neural signals during full leg flexion and propulsion activated by optimal stimulus patterns, respectively, and used them as the standard RMS for self-fine-tuning modulation. We then delivered a weak initial stimulus. After the sciatic neural signals were fed back to the laptop, an algorithm evaluated the amplitude and automatically adjusted the stimulus parameter. This process was repeated until the RMS of the sciatic neural signals matched the standard RMS (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb). Finally, the mechanism stopped modulation and maintained a stable stimulus output. Through this procedure, we observed that when the RMS of sciatic neural signals reached the pre-recorded standard pattern, the sEMG signals also approached an ideal state (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec). This demonstrates the effectiveness of feedback modulation using sciatic neural signals. The increase in sEMG signals and optimized kinematics of beagle’s hindlegs indicate that a near-real-time, self-fine-tuning modulation mechanism based on sciatic neural signals can effectively regulate muscle contraction and enhance locomotor performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ed; Supplementary Video 8). Similar modulated effects were also achieved on rats (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb; Supplementary Video 9). Since the beagle was supported by a vertical suspension device, no body weight was exerted during the weight acceptance phase, and we used a stable stimulus to activate motion without feedback modulation.\u003c/p\u003e \n\n "},{"header":"Discussion","content":"\u003cp\u003eFunctional restoration has always been a key focus of rehabilitation research following nerve injury. To circumvent the challenging nerve regeneration, ANP that can bypass the injured site and reestablish the neural communication has been increasingly used as a clinical therapeutic strategy for some nervous system diseases\u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e–\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. As a breakthrough, ANP using EES managed motor recovery after SCI by temporospatially activating specific combination of spinal motor neuron pools\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Recently, the brain-spine interface has enabled the cortex to directly control the neural circuits below injured spinal level, achieving complete close-loop motor modulation\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. However, motion bias always occurs inevitably even though the brain issues correct instructions. Therefore, to guarantee satisfactory modulation effects, checking and adjusting modulation promptly in response to kinestate is crucial but currently lacks solutions. Aiming to address this, we developed an ANP implant integrating a self-fine-tuning strategy. This enabled us to establish a near-real-time sciatic-neural-signal-feedback modulation mechanism with self-calibration capability, particularly suited for free activities in open environment.\u003c/p\u003e\u003cp\u003eThe neural interface applied to motor function recovery after SCI needs to accommodate the scenarios where the local impedance of electrode pads increases gradually during chronic implantation, resulting in complex and variable stimulus patterns. This requires the stimulation unit in the implant to possess a high stimulation current limit and more stimulation channels. The implant proposed in this paper enables a stimulation current limit of 16.32 mA and 16 stimulation channels in any spatiotemporal combination, surpassing existed works\u003csup\u003e\u003cspan additionalcitationids=\"CR14 CR15 CR16 CR17 CR18\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e–\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e (Supplementary Table\u0026nbsp;1). Additionally, we integrated the stimulation and acquisition units into a single chip, further reducing the size of this implant.\u003c/p\u003e\u003cp\u003eCapogrosso \u003cem\u003eet al\u003c/em\u003e. achieved close-loop motor modulation with EES by configuring stimulation based on real-time motor kinematics of rats and non-human primates\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. This approach is feasible and effective in an experimental setting. However, this image-processing-based close-loop tuning requires real-time motion recording using high-resolution cameras, which is challenging for community-based activities. In contrast, wireless electrical signals delivery could be more practical. To enable electrical-signal-based feedback modulation, a bidirectional neural interface with both stimulating and recording capabilities is dispensable. Currently, most neural interfaces approved for EES are unidirectional. A recent commercial bidirectional interface modulates stimulus intensity in a close-loop manner by evaluating local impedance\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. However, since all approved spinal neural interfaces are designed for pain treatment, their pre-set stimulation patterns do not accommodate the complex and flexible configurations needed for motor modulation. This limitation may explain why close-loop motor modulation with EES for free movements has not yet been realized in clinical practice. We suggest that updated devices will advance the modulation technologies.\u003c/p\u003e\u003cp\u003eSelf-fine-tuning modulation based on electrical signal feedback is widely used in neuroprosthetic, utilizing both EMG signals (motor feedback) and neural signals (sensory feedback), and has been confirmed to be effective\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. The sciatic nerve, a main nerve innervating the lower limbs, regulates both motor and sensory functions. Song \u003cem\u003eet al\u003c/em\u003e. controlled rabbit ankle motions through decoding the activity recorded in the sciatic nerve, demonstrating functional electrical stimulation\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Hwang \u003cem\u003eet al\u003c/em\u003e. achieved close-loop motor function restoration in paralyzed rats by recording and stimulating sciatic nerve using an implanted nerve cuff electrode\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. Since the sciatic nerve, delivering the signals downstream from the spinal cord, is located above the muscles and joints in the modulation pathway, it enables a much shorter feedback path than muscle and motion. Therefore, our study highlights the potential of using sciatic neural signals as feedback for near-real-time self-fine tuning, which is difficult to achieve through sEMG signal or motion image feedback. Given their wireless delivery, high sensitivity, and rapid response, sciatic neural signals hold promising application prospect for enhancing EES-based close-loop motor modulation.\u003c/p\u003e\u003cp\u003eWe expect to reconstruct gait using concise configurations and parameters. Thus, we only configured paired cathode and anode and altered stimulation amplitude to restore expected motions, based on which we selected the optimal pattern. As a result, some configurations that could achieve similar effects but be more complex might be ignored. Then, we proposed and confirmed a novel concept, that is, introducing a self-fine-tuning pathway into the motor modulation of EES through sciatic neural signal feedback. Its feasibility reveals that the signals of nerve branches innervating target muscles directly could have considerable potentials to become feedback conditions especially for modulating fine movements. Certainly, it must be considered carefully due to inevitably increased number of implants and operation difficulty\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. Future work must also address higher integration of whole system, that is, integrate the algorithms of data processing and parameter regulating to the implant module, so that the needless delay resulting from the data transmission and processing outside the system could be avoided, further improving the time sensitivity of the feedback mechanism.\u003c/p\u003e\u003cp\u003eIn conclusion, we successfully restored hindleg gaits in paralyzed rats and beagles after SCI using a self-fine-tunning ANP implant, achieving near-real-time close-loop motor modulation with EES. This designed ANP implant characterized by its small size, low noise, and high efficacy offers a feasible and effective self-tuning method for further optimizing the clinical application of ANP in the future.\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eAnimal\u003c/h2\u003e\u003cp\u003eExperiments were conducted on male Sprague Dawley rats (10 weeks, 300 g, \u003cem\u003en\u003c/em\u003e = 10) and beagles, including males (12–14 months, 9–11 kg, \u003cem\u003en\u003c/em\u003e = 5) and females (12–14 months, 8–10 kg, \u003cem\u003en\u003c/em\u003e = 2). The rat experiments were performed at Chinese Institute for Brain Research (CIBR) in Beijing and approved by the Animal Care and Use Committee at CIBR (AP# CIBR-IACUC-074). The beagle experiments were conducted at Huafuyuan Biological Technology Co., Ltd, Beijing, and approved by local Ethical Committee for Laboratory Animal Welfare (HFYIACUC20230216001).\u003c/p\u003e\u003ch2\u003eLocomotor training\u003c/h2\u003e\u003cp\u003eBefore kinematic testing, rats and beagles underwent locomotor training on treadmill (#C300, Reao, China for beagles). The training lasted for 30 min daily for one week with a speed of 0.144 km/h for rats and 1.5 km/h for beagles. After SCI, partial vertical support was provided to assist with walking.\u003c/p\u003e\u003ch2\u003eImplantation of chronic sciatic nerve electrodes\u003c/h2\u003e\u003cp\u003eAn incision was made along the sciatic nerve on the right hindleg. The sciatic nerve was dissected through the intramuscular space and gently isolated using a nerve dissector. For beagles, the isolated sciatic nerve was wrapped with a 4-mm-diameter cuff electrode featuring 1 × 4 contacts and a 500-mm-long wire (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea, b; \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea, b). For rats, a 2-mm-diameter cuff electrode with 1 × 1 contact and a 250-mm-long wire was used (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea). The cuff electrodes were provided by Kedou Brain Computer Technology Co. Ltd., Suzhou, China.\u003c/p\u003e\u003ch2\u003eImplantation of chronic EMG leads in rats\u003c/h2\u003e\u003cp\u003eThe procedure followed the method introduced in a previous study\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Briefly, we removed about 1 mm of the isolation layer from one end of the silicone-coated stainless-steel wires (#KD–998, Kedou Brain Computer Technology Co. Ltd., Suzhou, China). The exposed end was implanted into the target muscle and sutured in place using 7–0 unabsorbable sutures (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eb).\u003c/p\u003e\u003ch2\u003eImplantation of bidirectional neural interface\u003c/h2\u003e\u003cp\u003eFor beagles, the bidirectional neural interface was encapsulated into a titanic box with a wireless charge coil and implanted subcutaneously at interscapular region (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ec, d). For rats, we prepared a protective shell secured with cranial nails on the skull and enclosed the bidirectional neural interface within the shell (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec, d). All wires of electrodes were connected to the neural interface through subcutaneous tunnels.\u003c/p\u003e\u003ch2\u003eSignal recordings and analysis\u003c/h2\u003e\u003cp\u003eColored markers were attached at the greater trochanter, lateral condyle, lateral malleolus, and toe\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Two digital video cameras (#C930c Business WebCam, Logitech) were positioned on both sides of treadmill to capture the locomotion of hindlegs. For beagles, four sEMG sensors were placed on the QF, BF, TA, and GM, respectively. The external ends of cuff electrodes and implanted EMG leads were connected to a bilateral neural interface in acquisition mode. The cameras, sEMG sensors, and neural interfaces recorded locomotion synchronously, with signals transmitted wirelessly via Bluetooth. We isolated a single gait cycle and extracted relevant kinematic parameters to analyze locomotor patterns by DeepLabCut and Electrophysiology Studio (#NeX Std, Beijing Ningju Technology Co. Ltd, Beijing, China).\u003c/p\u003e\u003ch2\u003eAutopsy\u003c/h2\u003e\u003cp\u003eTo explore the anatomy of the spine and spinal cord, a beagle sacrificed for other experiment was used for autopsy. We exposed the L3–L7 spinous processes and vertebral laminae to observe the anatomic characteristics. Using appropriate surgical tools, we performed a laminectomy and dissected the dura, revealing the lumbosacral spinal cord and bilateral spinal nerves. The conus medullaris was identified at the L7 segment, with no discrepancy between the spinal cord and vertebra segments (Extended Data Fig.\u0026nbsp;8a, b).\u003c/p\u003e\u003ch2\u003eImaging data\u003c/h2\u003e\u003cp\u003eThe beagle was anesthetized with an intramuscular injection of tiletamine hydrochloride and zolazepam hydrochloride (Virbac, France, 7–25 mg/kg) before imagological examinations and was positioned prone during the procedure.\u003c/p\u003e\u003cp\u003eThe beagle underwent spinal structural magnetic resonance (MR) imaging using a 3.0T MR scanner (MAGNETOM Spectra, Siemens Healthineers) with 16-channel body and 32-channel spine array coils (Extended Data Fig.\u0026nbsp;8c–f). The standard MR imaging protocol included three pulse sequences, all performed without gadolinium-based contrast: a) two-dimensional (2D) sagittal T1-weighted turbo spin-echo (TSE) with a repetition time (TR) of 400.0 ms, echo time (TE) of 9.5 ms, and voxel size of 0.9 × 0.9 × 4.0 mm; b) 2D sagittal and axial T2-weighted TSE with a TR of 3000.0 ms, TE of 96.0 ms, and voxel size of 0.9 × 0.9 × 4.0 mm; c) three-dimensional (3D) coronal T2-weighted sampling perfection with application optimized contrast using different flip angle evolution (SPACE) with a TR of 3000.0 ms, TE of 197.0 ms, and voxel size of 1.0 × 1.0 × 1.2 mm. The 2D sagittal T1-weighted TSE and 2D sagittal and axial T2-weighted TSE were used for imaging the spine and spinal cord, while the 3D coronal T2-weighted SPACE was used for imaging the lumbosacral plexus nerves. The total scan time was approximately 30 minutes. Detailed acquisition parameters for the above three MR pulse sequences are showed in Supplementary Table\u0026nbsp;2.\u003c/p\u003e\u003cp\u003eThe beagle underwent spinal 3D CT scanning without gadolinium-based contrast on a SOMATOM Perspective spiral CT scanner (Siemens Healthineers) with 64 rows and 128 slices (Extended Data Fig.\u0026nbsp;8g–k). The detailed acquisition parameters were quality reference, 280 ms; CARE Dose4D, 130 kV; slice thickness, 0.6 mm; acquisition, 64 × 0.6 mm; pitch, 1.2; scanning time, 4.85 s; rotating time, 1.0 s; delay, 3.0 s; phase-encoding direction, head to feet. The total scan time was approximately 10 minutes.\u003c/p\u003e\u003cp\u003e3D Slicer was used to process the data for 3D reconstruction (Extended Data Fig.\u0026nbsp;8l, m) and measure the length of L3–L7 segment, the width of the lumbar vertebral canal, and the height of the lumbar epidural space\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eSurgical procedures\u003c/h2\u003e\u003cp\u003eThe hair on the thoracic and lumbar back, as well as the bilateral hindlegs, was removed. For beagles, atropine sulfate (Ruicheng, China) was used by intramuscular injection (0.04–0.1 ml/kg) to inhibit gland secretion 15 min before general anesthesia. Anesthesia was induced with an intramuscular injection of tiletamine hydrochloride and zolazepam hydrochloride (Virbac, France) at a dose of 7–25 mg/kg. The L3 spinous process was identified under X-ray. 0.5 ml methylene blue solution was injected percutaneously into the supraspinous ligament for marking. After endotracheal intubation, beagle was maintained anesthetized with 1.5–1.8% isoflurane in balanced oxygen during the surgical procedures. Sodium lactate ringer’s injection was administered intravenously for fluid therapy during the operation. For rats, anesthesia was induced with 5% isoflurane, and maintenance with a dose of 1.5–3% isoflurane in balanced oxygen. The surgical areas were disinfected by iodophor.\u003c/p\u003e\u003cp\u003ePartial laminectomies were performed between each pair of adjacent vertebral laminae at the L3–L7 segment in beagles and at the T13–L2 segment in rats. A customized electrode (Beijing Ningju Technology Co. Ltd, Beijing, China, Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec, d) was implanted into epidural space to cover the L3–L7 segment in beagles (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ee) and the T13–L2 segment in rats (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ee). The orientation of electrode was adjusted through the gaps created by partial laminectomies and secured by suturing it to the paravertebral muscles. In beagles, a subfascial negative pressure drainage tube was placed in the incision to drain the effusion. The fascia was closed with 4–0 absorbable sutures, and the skin was closed with 4–0 unabsorbable sutures in rats and 2–0 unabsorbable sutures in beagles.\u003c/p\u003e\u003cp\u003eLaminectomies and crush injury were conducted at T9 segment. The spinal cord was fully crushed for 2 s with forceps with a width of 0.1 mm in the last 5 mm of tips\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. The fascia was closed with 4–0 absorbable sutures, and the skin was closed with 4–0 unabsorbable sutures in rats and 2–0 unabsorbable sutures in beagles.\u003c/p\u003e\u003cp\u003eThe animal was housed individually in a warm, dry environment with at libitum access to water and food. The incisions were disinfected and dressed daily. For beagles, ceftiofur (Sivea, China) was administered intramuscularly at 0.1 ml/kg every three days, twice, for infection prevention. Tolfenamic acid (Sivea, China) was given intramuscularly at 0.1 ml/kg every 48 hours for anti-inflammation and pain relief. The drainage device was removed based on daily drainage volume. For rats, ceftiofur sodium was injected intramuscularly at 4 mg/kg once daily for the first three postoperative days, then every other day for eight days. Urethral catheterization for beagle and manual urination for rats were performed twice daily until voluntary urination resumed. Stitches were removed two weeks postoperatively.\u003c/p\u003e\u003ch2\u003eSelection of stimulus patterns\u003c/h2\u003e\u003cp\u003eThe entire procedure was performed under general anaesthesia. Based on previous studies, we positioned the cathode on the lateral sides of the electrode and the anode around it to control the stimulating current flow and target specific areas of activated motor neurons. We initiated a 0.5 mA stimulus at a frequency of 40 Hz through a bidirectional neural interface, increasing the stimulus in 0.05 mA increments\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. A chronic cuff electrode recorded sciatic neural signals, while chronic EMG leads captured the EMG signals in rats. Wired needle electrodes penetrating muscles recorded the EMG signals in beagle. The sciatic nerve and rat EMG signals were transmitted to a laptop via the bidirectional neural interface. Wired needle electrodes were connected to a recorder that transmitted the beagle's EMG signals. Activated motions were recorded by two digital video cameras (#C930c Business WebCam, Logitech).\u003c/p\u003e\u003cp\u003eWe screened the cathode and anode configurations to identify those activating the correct motions for full leg flexion, weight acceptance, and propulsion. For each configuration, we calculated the normalized RMS of sEMG signals of each muscle against stimulus amplitude as well as the selectivity index, which was defined as the RMS ratio of sEMG signals between primary and secondary muscles\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e; and for each segment, we chose the configuration achieving highest selectivity index as the optimal one (Extended Data Fig.\u0026nbsp;9a). When the selectivity index ≥ 90%, the corresponding stimulus amplitude range was considered appropriate for activating the motion. Within this range, the stimulus amplitude that minimally activated contralateral muscles was selected as the optimal intensity (Extended Data Fig.\u0026nbsp;9b).\u003c/p\u003e\u003ch2\u003eAlgorithm for close-loop modulation\u003c/h2\u003e\u003cp\u003eThe modulation process was modeled as a single-input, single-output system, where the stimulus amplitude served as the input control signal and the recorded RMS of the sciatic neural signals acted as the feedback parameter. This enabled the targeted modulation of both propulsion and full leg flexion. In the self-fine-tunning ANP implant, the stimulation and acquisition units transmitted stimulus parameters and sciatic neural signals to the server, respectively. The server performed gait phase segmentation based on the timestamp synchronization between the stimulation and acquisition units, computed the corresponding RMS of sciatic neural signals, and updated the stimulus parameters through the modulation algorithm (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea).\u003c/p\u003e\u003cp\u003eAssuming the stimulus amplitude was \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{s}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e​, where 𝑖 ≥0 represented the number of iterations, and 𝑗 = 0, 1 denoted different modulated actions (propulsion and full leg flexion). The stimulus amplitude increment after each update was \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ste{p}_{i,j}={s}_{i,j}-{s}_{i-1,j}\\)\u003c/span\u003e\u003c/span\u003e, and the RMS of sciatic neural signals obtained from each measurement was \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e, with the target RMS being \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{opt,j}\\)\u003c/span\u003e\u003c/span\u003e. The iterative process of the entire modulation algorithm was described by the following relationship:\u003c/p\u003e\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{\\gamma\\:}_{i,j}=\\text{ln}(1+\\left(e-1\\right)\\raisebox{1ex}{$\\text{ln}{N}_{i,j}$}\\!\\left/\\:\\!\\raisebox{-1ex}{$\\text{ln}{N}_{opt,j}$}\\right.)$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:\\raisebox{1ex}{$ste{p}_{i+1,j}$}\\!\\left/\\:\\!\\raisebox{-1ex}{$ste{p}_{i,j}$}\\right.={\\gamma\\:}_{i,j}\\bullet\\:\\raisebox{1ex}{${N}_{opt,j}{N}_{i-1,j}$}\\!\\left/\\:\\!\\raisebox{-1ex}{${N}_{i,j}^{2}$}\\right.\\:$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:{s}_{i+1,j}={s}_{i,j}+ste{p}_{i+1,j}$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{\\gamma\\:}_{i,j}\\)\u003c/span\u003e \u003c/span\u003e represented the correction factor in the iterative algorithm, primarily used to adjust the magnitude of \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ste{p}_{i+1,j}\\)\u003c/span\u003e\u003c/span\u003e calculated based on the exponential relationship between the RMS of sciatic nerve signals and the stimulus amplitude. As \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e​ approached \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{N}_{opt,j}\\)\u003c/span\u003e\u003c/span\u003e​, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\gamma\\:\\:\\)\u003c/span\u003e\u003c/span\u003etended to 1. \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ste{p}_{1,j}={s}_{1,j}-{s}_{0,j}\\)\u003c/span\u003e\u003c/span\u003e​ was a predefined value used as the initial stimulus increment for estimating the exponential relationship in the iteration process. It was primarily determined based on the magnitude of the optimal stimulus parameters, with a typical range of 0.05–0.1 mA in experiments. After iteration, when\u003c/p\u003e\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$$\\:\\raisebox{1ex}{$\\left|{N}_{opt,j}-{N}_{i,j}\\:\\right|$}\\!\\left/\\:\\!\\raisebox{-1ex}{${N}_{opt,j}$}\\right.\u0026lt;5\\%$$\u003c/div\u003e\u003c/div\u003e\u003cp\u003ethe modulation algorithm stopped updating and outputted stable stimulus amplitude \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{s}_{i,j}\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eAll data are reported as the mean ± standard deviation. We used Prism (version 9.0, GraphPad Software, Inc., San Diego, CA) for quantifications and plotting, and DataGraph (Version 5.4, Visual Data Tools, Inc., Chapel Hill, NC) to draw stick figures and diagrams of sciatic nerve and EMG signals and joint motions. For two-group comparisons, a two-tailed t-test was performed when normality was confirmed. Statistical significance was set at \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data supporting the results in this study are available in the Article and its Supplementary Information. The raw and analysed datasets generated during the study are available for research purposes from the corresponding authors upon reasonable request.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data supporting the results in this study are available in the Article and its Supplementary Information. The raw and analysed datasets generated during the study are available for research purposes from the corresponding authors upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Z. Zhu and C. Zhang for support in the postoperative care of beagles and X. Liu, C. Sun, and T. Zhou for support in designing the self-fine-tuning ANP implant.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eM.S., Z.Z., and C.L. designed the experimental concept. M.S. and C.L. carried out all the animal surgeries and kinematic testing with the assistance from Z.C., Y.Z., T.S., and Y.C.. M.S. and Z.Z. analyzed the data. Z.Z. designed the algorithm and operated the modulation with the assistance from M.S. and X.H.. Y.Z. and Y.W designed the self-fine-tuning ANP implant. M.S. and Z.Z. prepared the manuscript with input from C.L., Y.Z., Y.W., Z.C., and Y.Z.. A.W. and S.B. provided the schematic diagrams of electrodes. J.P., X.Y., and M.Z co-supervised the project.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDieter, A. et al. Towards the optical cochlear implant: optogenetic approaches for hearing restoration. \u003cem\u003eEMBO Mol. Med.\u003c/em\u003e\u003cstrong\u003e12,\u003c/strong\u003e e11618 (2020).\u003c/li\u003e\n\u003cli\u003eSchvartz-Leyzac, K.C. et al. Cochlear Health and Cochlear-implant Function. \u003cem\u003eJ. Assoc. Res. Otolaryngol.\u003c/em\u003e\u003cstrong\u003e24,\u003c/strong\u003e 5\u0026ndash;29 (2023).\u003c/li\u003e\n\u003cli\u003eKarikari, E. \u0026amp; Koshechkin, K. A. Review on brain-computer interface technologies in healthcare. \u003cem\u003eBiophys. Rev.\u003c/em\u003e\u003cstrong\u003e15,\u003c/strong\u003e 1351\u0026ndash;1358 (2023).\u003c/li\u003e\n\u003cli\u003eNojima, I. et al. Brain-Computer Interface Training Based on Brain Activity Can Induce Motor Recovery in Patients with Stroke: A Meta-Analysis. \u003cem\u003eNeurorehabil. Neural. Repair.\u003c/em\u003e\u003cstrong\u003e36,\u003c/strong\u003e 83\u0026ndash;96 (2022).\u003c/li\u003e\n\u003cli\u003eLuo, S. et al. Brain-Computer Interface: Applications to Speech Decoding and Synthesis to Augment Communication. \u003cem\u003eNeurotherapeutics.\u003c/em\u003e\u003cstrong\u003e19,\u003c/strong\u003e 263\u0026ndash;273 (2022).\u003c/li\u003e\n\u003cli\u003eVansteensel, M. J. et al. Towards clinical application of implantable brain-computer interfaces for people with late-stage ALS: medical and ethical considerations\u003cem\u003e. J. Neurol.\u003c/em\u003e\u003cstrong\u003e270,\u003c/strong\u003e1323\u0026ndash;1336 (2023).\u003c/li\u003e\n\u003cli\u003eCollaborators GBDSCI. Global, regional, and national burden of spinal cord injury, 1990\u0026ndash;2019: a systematic analysis for the Global Burden of Disease Study 2019. \u003cem\u003eLancet Neurol.\u003c/em\u003e\u003cstrong\u003e22,\u003c/strong\u003e 1026\u0026ndash;1047 (2023).\u003c/li\u003e\n\u003cli\u003eRowald, A. et al. Activity-dependent spinal cord neuromodulation rapidly restores trunk and leg motor functions after complete paralysis. \u003cem\u003eNat. Med\u003c/em\u003e. \u003cstrong\u003e28,\u003c/strong\u003e 260\u0026ndash;271 (2022).\u003c/li\u003e\n\u003cli\u003eCapogrosso, M. et al. Configuration of electrical spinal cord stimulation through real-time processing of gait kinematics. \u003cem\u003eNat. Protoc.\u003c/em\u003e\u003cstrong\u003e13,\u003c/strong\u003e 2031\u0026ndash;2061 (2018).\u003c/li\u003e\n\u003cli\u003eLorach, H. et al. Walking naturally after spinal cord injury using a brain-spine interface. \u003cem\u003eNature.\u003c/em\u003e\u003cstrong\u003e618,\u003c/strong\u003e 126\u0026ndash;133 (2023).\u003c/li\u003e\n\u003cli\u003eZheng, J. \u0026amp; Meister, M. The unbearable slowness of being: Why do we live at 10 bits/s?. \u003cem\u003eNeuron. \u003c/em\u003e\u003cstrong\u003e113, \u003c/strong\u003e192\u0026ndash;204 (2025). \u003c/li\u003e\n\u003cli\u003eAnderson, M. A. et al. Natural and targeted circuit reorganization after spinal cord injury. \u003cem\u003eNat. Neurosci.\u003c/em\u003e\u003cstrong\u003e25,\u003c/strong\u003e 1584\u0026ndash;1596 (2022).\u003c/li\u003e\n\u003cli\u003eZhou, A. et al. A wireless and artefact-free 128-channel neuromodulation device for closed-loop stimulation and recording in non-human primate. \u003cem\u003eNat. Biomed. Eng.\u003c/em\u003e\u003cstrong\u003e3, \u003c/strong\u003e15\u0026ndash;26 (2019).\u003c/li\u003e\n\u003cli\u003eWang, Y. et al. 26.3 A Closed-Loop Neuromodulation Chipset with 2-Level Classification Achieving 1.5Vpp CM Interference Tolerance, 35dB Stimulation Artifact Rejection in 0.5ms and 97.8% Sensitivity Seizure Detection. \u003cem\u003eIEEEISSCC.\u003c/em\u003e pp. 406\u0026ndash;408 (2020).\u003c/li\u003e\n\u003cli\u003eElAnsary, M. et al. Bidirectional peripheral nerve interface with 64 second-order opamp-less \u0026Delta;\u0026Sigma; ADCs and fully integrated wireless power/data transmission. \u003cem\u003eIEEE Journal of Solid-State Circuits\u003c/em\u003e. \u003cstrong\u003e56,\u003c/strong\u003e 3247-3262 (2021).\u003c/li\u003e\n\u003cli\u003eLee, J. et al. Neural recording and stimulation using wireless networks of microimplants. \u003cem\u003eNat Electron.\u003c/em\u003e\u003cstrong\u003e4,\u003c/strong\u003e 604\u0026ndash;614 (2021). \u003c/li\u003e\n\u003cli\u003eLiu, F. et al. A fully implantable opto-electro closed-loop neural interface for motor neuron disease studies. \u003cem\u003eIEEE Transactions on Biomedical Circuits and Systems\u003c/em\u003e. \u003cstrong\u003e16,\u003c/strong\u003e 752-765 (2022).\u003c/li\u003e\n\u003cli\u003eChen, J. C. et al. A wireless millimetric magnetoelectric implant for the endovascular stimulation of peripheral nerves. \u003cem\u003eNat. Biomed. Eng.\u003c/em\u003e\u003cstrong\u003e6,\u003c/strong\u003e 706\u0026ndash;716 (2022).\u003c/li\u003e\n\u003cli\u003eBurton, A. et al. Fully implanted battery-free high power platform for chronic spinal and muscular functional electrical stimulation. \u003cem\u003eNat. Commun\u003c/em\u003e. \u003cstrong\u003e1,\u003c/strong\u003e 7887 (2023).\u003c/li\u003e\n\u003cli\u003eVallejo, R. et al. A New Direction for Closed-Loop Spinal Cord Stimulation: Combining Contemporary Therapy Paradigms with Evoked Compound Action Potential Sensing. \u003cem\u003eJ. Pain Res.\u003c/em\u003e\u003cstrong\u003e14,\u003c/strong\u003e 3909\u0026ndash;3918 (2021).\u003c/li\u003e\n\u003cli\u003eFarina, D. et al. Toward higher-performance bionic limbs for wider clinical use. \u003cem\u003eNat. Biomed. Eng. \u003c/em\u003e\u003cstrong\u003e7,\u003c/strong\u003e 473\u0026ndash;485 (2023).\u003c/li\u003e\n\u003cli\u003eSong, K. I. et al. Compact Neural Interface Using a Single Multichannel Cuff Electrode for a Functional Neuromuscular Stimulation System. \u003cem\u003eAnn. Biomed. Eng.\u003c/em\u003e\u003cstrong\u003e47,\u003c/strong\u003e 754\u0026ndash;766 (2019).\u003c/li\u003e\n\u003cli\u003eHwang, Y. E. et al. Closed-Loop Control of Functional Electrical Stimulation Using a Selectively Recording and Bidirectional Nerve Cuff Interface. \u003cem\u003eIEEE Trans. Neural. Syst. Rehabil. Eng.\u003c/em\u003e\u003cstrong\u003e32,\u003c/strong\u003e 504\u0026ndash;513 (2024).\u003c/li\u003e\n\u003cli\u003eHabibollahi, M. et al. Active Neural Interface Circuits and Systems for Selective Control of Peripheral Nerves: A Review. \u003cem\u003eIEEE Trans. Biomed. Circuits Syst.\u003c/em\u003e\u003cstrong\u003e18,\u003c/strong\u003e 954\u0026ndash;975 (2024).\u003c/li\u003e\n\u003cli\u003eBonizzato, M. et al. Multi-pronged neuromodulation intervention engages the residual motor circuitry to facilitate walking in a rat model of spinal cord injury. \u003cem\u003eNat. Commun.\u003c/em\u003e\u003cstrong\u003e12, \u003c/strong\u003e1925 (2021).\u003c/li\u003e\n\u003cli\u003eFedorov, A. et al. 3D Slicer as an Image Computing Platform for the Quantitative Imaging Network. \u003cem\u003eMagnetic Resonance Imaging. \u003c/em\u003e\u003cstrong\u003e30,\u003c/strong\u003e 1323\u0026ndash;1341 (2012). \u003c/li\u003e\n\u003cli\u003eLi, Y. et al. Microglia-organized scar-free spinal cord repair in neonatal mice. \u003cem\u003eNature.\u003c/em\u003e\u003cstrong\u003e587, \u003c/strong\u003e613\u0026ndash;618 (2020).\u003c/li\u003e\n\u003cli\u003eSlawinska, U. et al. Comment on \u0026quot;Restoring voluntary control of locomotion after paralyzing spinal cord injury\u0026quot;. \u003cem\u003eScience.\u003c/em\u003e\u003cstrong\u003e338, \u003c/strong\u003e328; author reply (2012). \u003c/li\u003e\n\u003cli\u003eBarra, B. et al. Epidural electrical stimulation of the cervical dorsal roots restores voluntary upper limb control in paralyzed monkeys. \u003cem\u003eNat. Neurosci.\u003c/em\u003e\u003cstrong\u003e25,\u003c/strong\u003e 924\u0026ndash;934 (2022).\u003c/li\u003e\n\u003cli\u003eWagner, F. B. et al. Targeted neurotechnology restores walking in humans with spinal cord injury. \u003cem\u003eNature.\u003c/em\u003e\u003cstrong\u003e563, \u003c/strong\u003e65\u0026ndash;71 (2018).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6244028/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6244028/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial neural pathway (ANP) implants hold transformative potential for empowering individuals with disabilities, which has recently revolutionized the clinical treatment of spinal cord injury through epidural electrical stimulation (EES). Classical EES operates with a pre-set program and needs manual modification, consuming both manpower and time. Although existed studies proposed kinematic image processing to configurate EES or a brain-spine interface to control motions automatically, prompt tuning on motor modulation in an open environment is still inaccessible. Here, we show a wireless self-fine-tuning ANP implant to establish a rapid self-feedback mechanism appropriate for free activities to achieve precise motor modulation by EES. The fine-tuning pathway was attached to the writing pathway using a wireless implanted bidirectional neural interface featuring compact size, low input noise, and high energy efficiency, with different functionalities including reading, writing, and bidirectional control. The sciatic neural signals are fed into the pathway as rapid feedback for self tuning. Without manual interventions, we restored gaits of hindlegs in paralyzed animals with EES by a self-fine-tuning, near-real-time, and close-loop motor modulation mechanism.\u003c/p\u003e","manuscriptTitle":"A Wireless Self-Fine-Tuning Artificial Neural Pathway Implant for Near-Real-Time Close-Loop Motor Modulation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-17 04:33:31","doi":"10.21203/rs.3.rs-6244028/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"88f999b4-b66f-4615-aecc-0940a834faa6","owner":[],"postedDate":"April 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":47170894,"name":"Physical sciences/Engineering/Electrical and electronic engineering"},{"id":47170895,"name":"Physical sciences/Engineering/Biomedical engineering"}],"tags":[],"updatedAt":"2026-02-16T08:06:21+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-17 04:33:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6244028","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6244028","identity":"rs-6244028","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}