The electrical stimulation of the nervous system has shown great clinical potential in injury and pathology, yet experimentally driven practice makes it challenging to identify effective design choices and personalized stimulation protocols. This review outlines emerging model-based optimization frameworks that address these challenges by leveraging biophysical digital twins of neural interfaces. Enabling acceleration strategies and complementary data-driven approaches are also highlighted, along with key factors that currently limit clinical translation.

Restoring lower-limb function in patients with severe spinal cord injury (SCI) remains challenging. Spinal cord stimulation may enhance and reinstate lower-limb movements, but it is either used in open-loop control or its control depends upon residual motor functions, limiting its applicability in severely paralyzed individuals. The decoding of motor intentions from cortical signals may provide an interesting alternative in such cases. Electroencephalography (EEG) is an ideal solution since it is noninvasive and has been employed diffusely in the past to decode upper-limb movement intentions. Nonetheless, its application in lower-limb control remains underexplored. In this study, we investigated whether EEG can be used to decode lower-limb movement correlates in four SCI patients with varying injury severity during attempted left/right hip flexion or knee extension across four experimental sessions. We performed statistical analysis of event-related desynchronization/synchronization and machine learning classification to evaluate single and multi-window decoding performance. Our results suggest that EEG signals can often differentiate lower-limb movement attempts from rest, whereas decoding of left vs right and hip vs knee movements was more elusive. Left vs right decoding accuracy was improved through multi-window decoding, showing multiple sessions with above-chance results. In one patient, it was possible to attain above-chance three-class decoding (left/right/rest). Discriminating hip and knee movements proved more challenging. These findings establish a baseline for EEG decoding of lower-limb motor attempts in severely paralyzed individuals and pave the way for the development of brain-controlled neuroprosthetic systems.

Naturalistic sensory feedback is crucial for improving the functionality and acceptance of upper-limb prostheses. Electrical stimulation of peripheral nerves via implanted electrodes has shown a viable approach to providing real-time sensory information to hand prostheses. Although the loss of digits accounts for most upper-limb amputations, no existing digital prosthesis incorporates sensory feedback. This would be particularly important in the case of thumb amputation, as it leads to major functional impairments. Here, we present a computational model to assess the feasibility of using a cuff electrode to provide sensory feedback to thumb prostheses. We propose the distal index-thumb branch of the median nerve as an ideal implantation site and derive a plausible topography for its fascicular organization. We then employ surrogate models of fiber activation to enable rapid and accurate evaluation of the implant performance. Our results demonstrate that while the system can achieve selective sensory feedback to the thumb at coarse levels, delivering fine-grained sensory information remains challenging. Future developments in cuff electrode design and stimulation protocols will be necessary to enhance feedback resolution while preventing unintended activation of the first lumbrical muscle.Clinical Relevance— Our in-silico analysis supports the development of distal implants based on cuff electrodes for providing sensory feedback to thumb prostheses.

The use of self-paced omnidirectional treadmills equipped with body weight support and virtual reality environments in the clinical practice may allow the training of walking in safe and repeatable conditions across different pathologies. One premise for this to be true is that the conditions of overground and treadmill walking may be similar, so that the latter can be used as an early training to develop the former. Here, we study the equivalence in the spatio-temporal parameters of gait between walking overground and on the Moonwalker, an omnidirectional treadmill equipped with body weight support and virtual reality. Additionally, we show that such gait parameters could be used to distinguish between healthy and parkinsonian patients with similar accuracies for the overground and treadmill condition. Overall, this work contributes to justify the introduction of self-paced omnidirectional treadmill in the clinical and rehabilitative practice.

Electrical stimulation of peripheral nerves offers a way to restore sensory-motor functions and treat drug-resistant conditions affecting internal organs. Understanding the fascicular organization of the implanted nerves is essential for enhancing the selective neuromodulation of the targeted bodily functions. In fact, this knowledge can inform the development of computational models that can be used to optimize electrode design and stimulation protocols. Traditionally, peripheral nerve topographies are segmented manually to highlight fascicle contours, resulting in a labor-intensive and error-prone process. In this study, we present a UNet-based deep neural network for automatic segmentation of fascicles from nerve histological sections, trained on original data from different nerves and stained with different techniques. The model leverages a pretrained encoder, reducing the need for extensive training datasets and allowing us to generalize to nerve types and histological stains previously unseen during training. The quality of the resulting segmentation has been evaluated using both the Dice coefficient and domain-specific metrics tailored to assess the quality of the reconstructed fascicle topography. Furthermore, we employed automatically segmented nerve sections to build computational models of peripheral nerve stimulation and assess the impact of segmentation on the accuracy of fascicle-wise recruitment predictions. Our results highlight that automated segmentation can reliably inform the modeling of neuromodulation applications, with minimal error in predicting recruitment thresholds. This approach paves the way for harnessing the large quantities of histological data that can be extracted from cadaveric nerve samples for use in computational models of neural interfaces, potentially advancing the design of next generation neuroprosthetic and bioelectronic medicine applications.

Unmyelinated fibers account for a remarkable fraction of the peripheral nervous system and their activity is linked to many autonomic and somatic functions. While electrical recording of such activity from human-sized peripheral nerves holds significant potential for neuroengineering applications, it has been shown only in acute settings via microneurography. This leaves unclear whether current implantable electrodes could achieve the same outcome. To address this matter, we simulated recordings from the human vagus nerve through a transverse intrafascicular multichannel electrode (TIME), a microneurographic (μNG) needle, and a commercial cuff electrode. Recording signals were studied fiber-wise across relevant electrode insertions, revealing that the possibility of recording unmyelinated activity is shared by the TIME but unlikely by the cuff. These results suggest that no physical limitations of implantable electrodes underlie the missing evidence of recordings from unmyelinated fibers, and draw attention to experimental design choices that may have concealed this capability thus far.

BACKGROUND:Emerging research increasingly supports that epidural spinal cord electrical stimulation (EES) combined with neurorehabilitation can improve motor recovery in spinal cord injury (SCI) subjects. Patients with lesions involving the medullary cone may be challenging to treat with this approach, probably due to potential peripheral nervous system damage, leaving the open question of whether this large population may benefit from EES. METHODS:A T11-T12 SCI patient, with medullary cone involvement, underwent EES implant in a clinical trial (NCT05926843). During three months of testing, we determined optimal stimulation protocols for improving isolated movements and integrated them to reinstate independent walking with a walker. FINDINGS:EES substantially boosted hip flexor, spinal erector, and abdominal muscle contraction, improving the patient's performance in isolated movements. Over three months of combining continuous subthreshold EES with personalized rehabilitation, the patient progressed from being unable to walk to overground ambulation using a two-wheeled walker and bilateral knee and foot orthoses. At the time of hospital discharge, the patient managed to cover 58 m in the 6-min walking test and completed the 10-meter walking test in 40.29 s. Six months after EES implant, the patient was able to walk independently for 1 km with a walker. CONCLUSIONS:These results underscore the potential of neurorehabilitation protocols integrating EES also for patients with medullary cone lesions and pave the way for new rehabilitation prospects. FUNDING:This work was funded by Università Vita-Salute San Raffaele, Boston Scientific Spa, Fondazione Cariplo, Bertarelli Foundation, and the Ministry of University and Research (MUR).

Restoring the ability to walk is a priority for individuals with neurological disorders or neurotraumatic injuries, given its significant impact on independence and quality of life. Multimodal closed-loop strategies that integrate robotic assistance and neuromodulation present promising avenues for personalized and physiological gait recovery. These approaches capitalize on residual motor activity, fostering neuroplasticity and motor relearning. This narrative review emphasizes the importance of mobile brain/body imaging (MoBI) for guiding the development of closed-loop systems that integrate volitional brain signals with residual motor activity in stroke and spinal cord injury patients. We explore the potential of rehabilitative and assistive interventional strategies based on robotic devices, such as exoskeletons and powered orthoses, and neuromodulation techniques like functional electrical stimulation and spinal cord stimulation. We highlight the limitations of the single interventional strategies and the potential of the synergistic combination of MoBI, robotics, and neuromodulation for gait recovery. By leveraging residual motor functions and integrating multimodal data from the different domains involved in motor recovery (i.e., brain, muscle, and biomechanics), the complementarity of these interventional strategies has the potential to enable dynamic patient-specific interventions. We outline a perspective framework on how future directions can exploit such integration to promote physiological recovery of lower limb functions and personalized therapies that are both challenging and feasible. Advancing along this path holds the promise of enhancing rehabilitative strategies, ultimately promoting functional recovery and long-term independence for individuals with neuromotor disorders.

Spinal cord injury (SCI) causes severe motor and sensory deficits, and there are currently no approved treatments for recovery. Nearly 70% of patients with SCI experience pathological muscle cocontraction and spasticity, accompanied by clinical signs such as patellar hyperreflexia and ankle clonus. The integration of epidural electrical stimulation (EES) of the spinal cord with rehabilitation has substantial potential to improve recovery of motor functions; however, abnormal muscle cocontraction and spasticity may limit the benefit of these interventions and hinder the effectiveness of EES in promoting functional movements. High-frequency excitation block introduced in peripheral nerve stimulation could reduce abnormal activity and lead to more physiological activation patterns. Here, we evaluated the application of high-frequency EES (HF-EES) in alleviating undesired muscular cocontraction and spasticity in two patients with motor incomplete SCI implanted with a commercial 32-channel EES paddle commonly used for pain therapy. To design custom HF-EES protocols, we first mapped the muscles targeted by different EES configurations. Our results showed that HF-EES substantially reduced patellar reflex in one participant and eliminated both patellar reflex and ankle clonus in the other participant. By combining HF-EES and low-frequency EES (LF-EES) to enhance functional movements with intensive rehabilitation, we observed notable improvements in lower limb kinematics, muscle strength, and clinical lower limb motor assessments over the trial period. This study suggests that HF-EES could be an important supplementary tool in SCI treatment, emphasizing the importance of personalized rehabilitation approaches and advanced tools to optimize EES treatments and offering hope for individuals with SCI-related motor deficits.

While electrical stimulation can restore functional movements in paralyzed patients, stimulation parameter tuning is mostly performed by clinicians through trial-and-error. Computational models may enable automatic parameter optimization on digital twins, substantially reducing the required testing. Nonetheless, current computational models of electrical stimulation do not consider how neural activation is converted into movement through the musculoskeletal system. Musculoskeletal models may fill this gap, but exploring a large set of candidate stimulation protocols requires solving the forward dynamics problem for a very large number of degrees of freedom. Here, we show that traditional machine learning can be used to train surrogates of musculoskeletal models to predict the steady-state value of a specific joint angle given a set of tetanic muscle activations. We show that such machine learningbased surrogate models enable accurate angle prediction across different joints and musculoskeletal models, while providing an acceleration of more than three orders of magnitude. Additionally, these models are interpretable, and feature importance analysis correctly identifies the agonist-antagonist muscle groups maximally influencing the movement of a specific joint. Machine learning-based acceleration of musculoskeletal models paves the way for their integration into computational models to predict and optimize the effects of electrical stimulation in motor neuroprostheses.

Vagus nerve stimulation (VNS) holds promise for a wide range of clinical applications, as the vagus nerve innervates numerous vital organs. Achieving stable and reliable recordings from the vagus nerve is therefore crucial for enabling closed-loop VNS strategies, which may enhance therapeutic efficacy compared to open-loop approaches. However, a major challenge lies in the fact that a large proportion of the vagus nerve consists of small unmyelinated fibers, whose electrogenic activity is typically too weak to rise above the noise floor in neural recordings. Yet, leveraging the spectral properties of such fibers, it may be possible to identify a more reliable proxy of their activity. In this study, we explore in silico the potential of the spiking band power (SBP), originally introduced in the field of brain-computer interfaces, for this purpose. We show that SBP strongly correlates with the activity of unmyelinated fibers in intraneural recordings, even under low signal-to-noise ratio conditions. Furthermore, when fed as a feature to a machine learning regressor, SBP enables accurate prediction of the firing rate of fibers, outperforming traditional threshold-crossing methods. Experimental validation of our in-silico findings could pave the way for more effective decoding of unmyelinated activity in future closed-loop VNS approaches. Clinical Relevance– Our in-silico analysis highlights potential strategies to improve the decoding of feedback signals for closed-loop vagus nerve stimulation.

Implantable neural devices can restore functional movement in patients with motor impairments through electrical stimulation. However, determining optimal stimulation parameters after the implantation remains a key challenge. Selectivity metrics, which quantify the balance between intended and unintended muscle activation, are commonly employed to guide parameter optimization. Each selectivity measure exhibits specific mathematical properties that affect their suitability as objective functions for optimization algorithms. In this study, we conducted a comparison of different selectivity definitions using computational modeling of the median nerve derived from human cadaveric histological data and implanted with two transverse intraneural multichannel electrodes (TIMEs). We evaluated four selectivity metrics and analyzed their effects on optimization outcomes. Our results show that the choice of the target selectivity measure influenced optimization performance, even when assessed using other selectivity definitions. Our findings underscore the importance of carefully selecting selectivity criteria to improve clinical optimization of neural stimulation protocols in patients with motor disabilities.

Electrical stimulation of peripheral nerves via implanted electrodes has been shown to be a promising approach to restore sensation, movement, and autonomic functions across a wide range of illnesses and injuries. While in principle computational models of neuromodulation can allow the exploration of large parameter spaces and the automatic optimization of stimulation devices and strategies, their high time complexity hinders their use on a large scale. We recently proposed the use of machine learning-based surrogate models to estimate the activation of nerve fibers under electrical stimulation, producing a considerable speed-up with respect to biophysically accurate models of fiber excitation while retaining good predictivity. Here, we characterize the performance of four frequently employed machine learning algorithms and provide an illustrative example of their ability to generalize to unseen stimulation protocols, stimulating sites, and nerve sections. We then discuss how the ability to generalize to such scenarios is relevant to different optimization protocols, paving the way for the automatic optimization of neuromodulation applications.

Electrical stimulation of the visual nervous system could improve the quality of life of patients affected by acquired blindness by restoring some visual sensations, but requires careful optimization of stimulation parameters to produce useful perceptions. Neural correlates of elicited perceptions could be used for fast automatic optimization, with electroencephalography as a natural choice as it can be acquired non-invasively. Nonetheless, its low signal-to-noise ratio may hinder discrimination of similar visual patterns, preventing its use in the optimization of electrical stimulation. Our work investigates for the first time the discriminability of the electroencephalographic responses to visual stimuli compatible with electrical stimulation, employing a newly acquired dataset whose stimuli encompass the concurrent variation of several features, while neuroscience research tends to study the neural correlates of single visual features. We then performed above-chance single-trial decoding of multiple features of our newly crafted visual stimuli using relatively simple machine learning algorithms. A decoding scheme employing the information from multiple stimulus presentations was implemented, substantially improving our decoding performance, suggesting that such methods should be used systematically in future applications. The significance of the present work relies in the determination of which visual features can be decoded from electroencephalographic responses to electrical stimulation-compatible stimuli and at which granularity they can be discriminated. Our methods pave the way to using electroencephalographic correlates to optimize electrical stimulation parameters, thus increasing the effectiveness of current visual neuroprostheses.

Epidural electrical stimulation (EES) has been employed to restore motor functions in the lower limbs of patients with spinal cord injuries (SCIs). However, non-invasive methods for controlling this stimulation have not yet been explored. This study aims to evaluate various classification approaches to distinguish movements as a preparatory step for controlling walking in SCI patients with an epidural electrical implant. Four different classification techniques were tested on electroencephalographic (EEG) data analyzed in both time and frequency domains to decode movements from rest condition. The classification algorithm with the highest accuracy (79 ± 3 ± 4 ± 3

Retinal stimulation (RS) allows restoring vision in blind patients, but it covers only a narrow region of the visual field. Optic nerve stimulation (ONS) has the potential to produce visual perceptions spanning the whole visual field, but it produces very irregular phosphenes. We introduced a geometrical model converting retinal and optic nerve firing rates into visual perceptions and vice versa and a method to estimate the best perceptions elicitable through an electrode configuration. We then compared in silico ONS and RS through simulated prosthetic vision of static and dynamic visual scenes. Both simulations and SPV experiments showed that it might be possible to reconstruct natural visual scenes with ONS and RS, and that ONS wide field-of-view allows the perception of more detail in dynamic scenarios than RS. Our findings suggest that ONS could represent an interesting approach for vision restoration and that our model can be used to optimize it.

Vagus nerve stimulation (VNS) is a promising application of bioelectronic medicine to treat many pathologies, ranging from epilepsy and depression to cardiovascular diseases. Conventional VNS is not optimized taking into account the topographic organization of the vagus nerve, resulting in suboptimal stimulation protocols, which can lead to severe adverse effects. The development of in vivo methods to determine topographic organization would allow more selective stimulation protocols and is thus pivotal in the development of future therapies. Here, we show that it may be possible to reverse-engineer vagus nerve topographic organization starting from experimental evoked compound action potentials (ECAPs). The parameters of a biophysical model of ECAP generation are varied until the optimal matching between synthetic and experimental ECAPs is determined. In the present work, we managed to match an experimental ECAP obtained from swine VNS, and to replicate the shape of an unseen experimental ECAP through the optimized parameters. This work paves the way to the automatic optimization of selective VNS protocols, opening to new important therapeutic opportunities.