Emerging technologies

To avoid conflating related but distinct approaches, emerging technologies for SCI can be organized into four broad categories: invasive brain-spine interfaces, non-invasive brain-spine interfaces, spinal neuromodulation without cortical decoding, and other related neuromodulatory strategies that may influence locomotor recovery but do not themselves constitute BSIs. This distinction is important because true BSIs require both neural signal decoding and stimulation of spinal circuits, whereas many neuromodulation approaches influence spinal or supraspinal circuits without creating a real-time brain-to-spine bridge.

Conceptual framework of brain-spine interfaces

The development of BSIs and related neuromodulation devices represents an emerging area of investigation in SCI treatment. Whereas conventional BCIs or BMIs often decode neural signals to control external devices such as computers, cursors, or prostheses, BSIs are designed to restore functional communication between supraspinal motor centers and spinal sensorimotor circuits below the level of injury. In experimental and early clinical settings, BSIs are designed to translate neural activity related to intended movement into stimulation commands that modulate spinal circuits in real time [43]. The function of these machines is refined by artificial intelligence and machine learning, allowing for personalized rehabilitation regimens [44–46].

BSI specifically aims to establish a functional link between supraspinal motor centers and spinal sensorimotor circuits below the level of injury. Mechanistically, a BSI should be understood as a brain-to-spine control loop. First, cortical activity associated with intended movement is acquired from the motor cortex, either through implanted cortical electrodes in invasive systems or scalp-based recordings such as EEG in non-invasive systems. Second, decoding algorithms identify patterns of neural activity and translate them into predicted motor commands, gait-phase events, or limb trajectories, such as hip flexion, knee extension, foot strike, or foot-off. Third, the decoded output is converted into stimulation commands that drive targeted spinal stimulation, commonly through EES or tSCS [47, 48]. Fourth, spinal stimulation engages residual lumbosacral sensorimotor circuits, including dorsal-root afferents, interneuronal networks, and motor pools below the lesion. Fifth, activation of these circuits produces task-specific motor output when paired with attempted movement. Moreover, the resulting movement also produces proprioceptive, cutaneous, visual, and biomechanical feedback that may contribute to closed-loop adjustment and activity-dependent plasticity during rehabilitation (Fig. 1) [48–50].

Fig. 1: General closed-loop brain–spine interface framework for lower-limb restoration after SCI.The schematic represents a general BSI framework and demonstrates cortical motor-intention signal acquisition, real-time decoding into motor commands or gait events, targeted epidural or transcutaneous stimulation of the lumbosacral spinal segments, activation of residual lumbosacral sensorimotor circuits, and lower-limb motor output with ascending sensory feedback

Invasive brain-spine interfaces

Invasive BSIs represent the most direct form of brain-spine interfacing because they combine implanted cortical recording systems with targeted spinal stimulation to restore communication between motor intent and spinal sensorimotor circuits. The preclinical and early clinical literature should be separated carefully because different studies use different signal sources, decoding approaches, stimulation targets, and outcome measures.

Capogrosso et al. demonstrated a wireless BSI in a nonhuman primate model in which intracortical recordings from the motor cortex were decoded in real time to identify gait events and control epidural stimulation of lumbar spinal circuits. This work showed that cortical activity could be translated into stimulation commands that restored weight-bearing locomotor patterns in an experimental paralysis model. However, its limitations include its preclinical design, controlled experimental setting, and uncertain generalizability to chronic human SCI [50].

Bonizzato et al. further supported the mechanistic importance of closed-loop timing by showing that brain-controlled modulation of spinal circuits may enhance locomotor recovery compared with non-contingent or continuous stimulation in an animal model. This finding reinforces a key BSI principle: stimulation is most biologically meaningful when it is linked to neural intent or task phase rather than delivered as isolated background neuromodulation. However, these data remain preclinical and require confirmation in larger human studies [51].

Lorach et al. later translated this concept into a first-in-human/single-participant BSI system in a patient with chronic SCI. In that study, cortical activity related to intended lower-limb movement was recorded wirelessly and decoded into stimulation commands delivered to lumbosacral epidural electrodes. The system enabled more natural walking with crutch support and improved performance on tasks such as ramp walking and stair climbing. When the BSI was inactive, walking ability was lost despite cortical activity indicating gait initiation, supporting the importance of the decoded brain-to-spine bridge. The participant also demonstrated improvements in sensory, motor, and clinical assessments after BSI-mediated rehabilitation, suggesting possible additive neurological recovery [52].

Despite these encouraging findings, this evidence remains highly selected and should not be interpreted as broadly generalizable clinical efficacy. Key limitations include the single-participant design, need for multiple implanted devices, intensive calibration and rehabilitation, long-term durability concerns, and uncertain applicability across different injury levels, chronicity, residual descending pathways, and completeness of SCI. Larger prospective studies are needed before BSIs can be considered ready for routine clinical use.

Non-invasive brain-spine interfaces

Non-invasive BSI approaches seek to achieve a similar brain-to-spine connection without implanted cortical electrodes, most commonly by pairing electroencephalography-based decoding of motor intent with tSCS. In 2025, Atkinson et al. introduced and evaluated a novel non-invasive BSI that integrates electroencephalography decoding of motor intent with tSCS to facilitate movement. In this proof-of-concept study, six able-bodied participants performed voluntary knee extension tasks while the system detected µ- and β-band event-related desynchronization from the sensorimotor cortex to trigger tSCS in real time. The BSI achieved robust classification accuracy, with an average area under the curve of approximately 0.83 during cued tasks and 0.68 during uncued tasks, and was well tolerated without adverse effects. By demonstrating the feasibility of coupling cortical signals with spinal stimulation without surgical intervention, this work expands the potential for accessible neurorehabilitation strategies aimed at restoring voluntary motor function. However, this study primarily focuses on able-bodied participants rather than individuals with SCI, and thus these results demonstrate more proof-of-concept than direct clinical evidence [53].

Additionally, a recent translational proof-of-concept study used non-invasive human intention-related signals, including EOG/EEG-based control signals, to drive epidural spinal stimulation in an anesthetized macaque, producing left- and right-sided stepping-like lower-limb movements. The system used a portable non-invasive sensor and achieved an average decoding F1 score of 89.6% across four control commands. However, because the locomotor output was generated in an anesthetized macaque rather than a human SCI participant, this study should be interpreted as translational proof-of-concept [54].

Comparative evidence across brain-spine interface approaches

Overall, the current BSI literature can be organized by signal source, decoding strategy, stimulation modality, feedback loop, study population, functional outcome, and limitation. Invasive systems provide higher-resolution cortical signals and more spatially targeted epidural stimulation, but they require neurosurgical implantation and raise concerns regarding infection, hardware longevity, signal stability, and cost. Non-invasive systems reduce surgical burden by using EEG-based motor-intention decoding and transcutaneous stimulation, but they generally face lower signal resolution, reduced stimulation specificity, and greater susceptibility to artifact. Preclinical systems provide important mechanistic proof-of-concept, whereas human evidence remains limited to able-bodied feasibility studies, translational human-animal demonstrations, and highly selected early clinical cases. Therefore, the key translational question is not only whether BSI can generate movement under controlled conditions, but whether it can produce durable, reproducible, patient-centered functional gains across heterogeneous SCI populations.

A study-level comparison clarifies these distinctions. Capogrosso et al. used invasive cortical recordings in nonhuman primates to decode gait events and control lumbar epidural stimulation, demonstrating restoration of locomotor patterns in a preclinical model. Bonizzato et al. showed that brain-controlled modulation of spinal circuits could improve recovery compared with non-contingent stimulation in an animal model, supporting the importance of timing stimulation to neural intent. Lorach et al. provided first-in-human evidence that decoded cortical intent can drive lumbosacral epidural stimulation to support walking after SCI, but the report remains limited by its single-participant design. Atkinson et al. demonstrated the feasibility of a non-invasive EEG-tSCS interface in able-bodied participants, showing that motor-intention signals can trigger spinal stimulation without implanted cortical electrodes, although direct evidence in SCI patients remains limited. Mo et al. further supported proof-of-concept feasibility by using non-invasive human intention-related EOG/EEG signals decoded in real time to control epidural spinal stimulation in an anesthetized macaque, producing left- and right-sided stepping-like lower-limb movements. However, because the motor output occurred in an anesthetized nonhuman primate rather than in a human participant with SCI, the findings should be interpreted as translational proof-of-concept rather than direct clinical evidence (Table 1) [50–54].

Study type | Signal source / electrode | Cortical site | Stimulation target | Population/model | Main outcome | Limitation
Capogrosso et al. | Intracortical recordings | Motor cortex | Lumbar epidural stimulation | Nonhuman primates | Improved locomotor patterns | Preclinical model
Lorach et al. | Implanted cortical recording system | Motor cortex | Lumbosacral epidural stimulation | Single human participant with chronic SCI | Walking with crutch support; ramp/stair tasks | Single participant; highly selected
Atkinson et al. | EEG | Sensorimotor cortex | tSCS | Able-bodied participants | Feasibility of non-invasive BSI triggering | Not studied in SCI patients
Mo et al. | EOG/EEG | Non-invasive intention-related signals | Epidural stimulation | Human intention-related signals controlling stimulation in an anesthetized macaque | Stepping-like movements in anesthetized macaque | Translational proof-of-concept; not direct human SCI evidence

Spinal neuromodulation without cortical decoding

Spinal neuromodulation without cortical decoding represents a related but distinct strategy in which epidural or transcutaneous stimulation is used to modulate spinal circuits without real-time input from decoded brain activity [55]. These approaches helped establish the therapeutic potential of spinal circuit activation after SCI and provide an important foundation for BSI systems, but they do not themselves constitute BSIs unless stimulation is driven by decoded neural intent.

Related neuromodulation strategies

In 2024, Cho et al. studied lateral hypothalamic deep brain stimulation (DBSLH) in two individuals with incomplete spinal cord injury who had persistent gait deficits despite standard rehabilitation. Although not a BSI, lateral hypothalamic deep brain stimulation represents a related neuromodulatory strategy that may influence locomotor recovery after SCI. DBSLH produced immediate improvements in lower limb muscle activity, kinematics, endurance, and reduced perceived walking effort, and after three months of combined DBS and gait training, participants showed better walking performance and motor scores. Notably, these long-term gains persisted even when the DBS was turned off and were achieved without adverse effects on vital signs, weight, or hormones [56] (Table 2).

Approach | Role in SCI/BSI development | Evidence status | Key translational barriers
Epidural electrical stimulation without cortical decoding | Demonstrated that spinal sensorimotor circuits below the lesion can be activated to support standing, stepping, and voluntary movement in selected patients. | Small human studies, case reports, and mechanistic studies; not broadly generalizable. | Surgical invasiveness, patient selection, stimulation optimization, rehabilitation intensity, durability, complications, and cost.
Transcutaneous spinal cord stimulation | Established a noninvasive method for modulating spinal excitability through surface electrodes over targeted spinal segments. | Early clinical and rehabilitation studies with heterogeneous protocols and outcomes. | Variable targeting, durability of benefit, protocol standardization, patient selection, and need for intensive rehabilitation.
BCI/BMI cortical decoding | Enabled decoding of motor intent from cortical activity to control external devices such as computers, cursors, robotic limbs, or prostheses. | Established proof-of-concept and clinical research base, but often focused on external device control rather than spinal circuit restoration. | Signal stability, decoding accuracy, implant burden for invasive systems, training demands, and long-term usability.
Invasive brain-spine interfaces | Combine cortical signal acquisition, real-time decoding, and targeted spinal stimulation to reconnect motor intent with spinal sensorimotor circuits. | Strongest BSI-specific evidence, but still limited to preclinical studies and early first-in-human/single-participant reports. | Implant burden, decoding stability, long-term safety, infection risk, hardware durability, rehabilitation demands, cost, and regulation.
Non-invasive brain-spine interfaces | Pair EEG-based motor-intention decoding with noninvasive spinal stimulation, such as tSCS, to reduce surgical burden. | Proof-of-concept evidence, including able-bodied participant studies; limited direct evidence in SCI patients. | Lower signal resolution, decoding reliability, stimulation specificity, reproducibility, and need to demonstrate clinical benefit in SCI.
Spinal neuromodulation without cortical decoding | Shows that spinal circuits can be therapeutically activated without real-time brain-derived input, providing a foundation for BSI development. | Human and preclinical evidence for EES/tSCS; not a BSI unless stimulation is driven by decoded neural intent. | Lack of cortical intent integration, protocol heterogeneity, uncertain long-term outcomes, and individualized stimulation needs.
Related supraspinal neuromodulation, including lateral hypothalamic DBS | Suggests that non-BSI brain stimulation approaches may influence locomotor recovery and broader rehabilitation pathways after SCI. | Very early evidence from small cohorts; conceptually distinct from BSI. | Small sample sizes, unclear mechanisms, invasiveness, generalizability, patient selection, and need for larger controlled studies.
Clinical translation of BSI systems | Represents the movement from experimental closed-loop systems toward clinically deployable neurorehabilitation tools. | Not ready for routine clinical use; current evidence remains proof-of-concept or highly selected. | Patient selection, injury heterogeneity, chronicity, residual circuitry, long-term follow-up, ethics, privacy, cybersecurity, reimbursement, infrastructure, and regulation.