Artificial Intelligence Enables Real-Time and Intuitive Control of Prostheses via Nerve Interface.

OBJECTIVE: The next generation prosthetic hand that moves and feels like a real hand requires a robust neural interconnection between the human minds and machines. METHODS: Here we present a neuroprosthetic system to demonstrate that principle by employing an artificial intelligence (AI) agent to translate the amputee's movement intent through a peripheral nerve interface. The AI agent is designed based on the recurrent neural network (RNN) and could simultaneously decode six degree-of-freedom (DOF) from multichannel nerve data in real-time. The decoder's performance is characterized in motor decoding experiments with three human amputees. RESULTS: First, we show the AI agent enables amputees to intuitively control a prosthetic hand with individual finger and wrist movements up to 97-98% accuracy. Second, we demonstrate the AI agent's real-time performance by measuring the reaction time and information throughput in a hand gesture matching task. Third, we investigate the AI agent's long-term uses and show the decoder's robust predictive performance over a 16-month implant duration. Conclusion & significance: Our study demonstrates the potential of AI-enabled nerve technology, underling the next generation of dexterous and intuitive prosthetic hands.

A portable, self-contained neuroprosthetic hand with deep learning-based finger control.

Objective.Deep learning-based neural decoders have emerged as the prominent approach to enable dexterous and intuitive control of neuroprosthetic hands. Yet few studies have materialized the use of deep learning in clinical settings due to its high computational requirements.Approach.Recent advancements of edge computing devices bring the potential to alleviate this problem. Here we present the implementation of a neuroprosthetic hand with embedded deep learning-based control. The neural decoder is designed based on the recurrent neural network architecture and deployed on the NVIDIA Jetson Nano-a compacted yet powerful edge computing platform for deep learning inference. This enables the implementation of the neuroprosthetic hand as a portable and self-contained unit with real-time control of individual finger movements.Main results.A pilot study with a transradial amputee is conducted to evaluate the proposed system using peripheral nerve signals acquired from implanted intrafascicular microelectrodes. The preliminary experiment results show the system's capabilities of providing robust, high-accuracy (95%-99%) and low-latency (50-120 ms) control of individual finger movements in various laboratory and real-world environments.Conclusion.This work is a technological demonstration of modern edge computing platforms to enable the effective use of deep learning-based neural decoders for neuroprosthesis control as an autonomous system.Significance.The proposed system helps pioneer the deployment of deep neural networks in clinical applications underlying a new class of wearable biomedical devices with embedded artificial intelligence.Clinical trial registration: DExterous Hand Control Through Fascicular Targeting (DEFT). Identifier: NCT02994160.

Deep Learning-Based Approaches for Decoding Motor Intent From Peripheral Nerve Signals.

Previous literature shows that deep learning is an effective tool to decode the motor intent from neural signals obtained from different parts of the nervous system. However, deep neural networks are often computationally complex and not feasible to work in real-time. Here we investigate different approaches' advantages and disadvantages to enhance the deep learning-based motor decoding paradigm's efficiency and inform its future implementation in real-time. Our data are recorded from the amputee's residual peripheral nerves. While the primary analysis is offline, the nerve data is cut using a sliding window to create a "pseudo-online" dataset that resembles the conditions in a real-time paradigm. First, a comprehensive collection of feature extraction techniques is applied to reduce the input data dimensionality, which later helps substantially lower the motor decoder's complexity, making it feasible for translation to a real-time paradigm. Next, we investigate two different strategies for deploying deep learning models: a one-step (1S) approach when big input data are available and a two-step (2S) when input data are limited. This research predicts five individual finger movements and four combinations of the fingers. The 1S approach using a recurrent neural network (RNN) to concurrently predict all fingers' trajectories generally gives better prediction results than all the machine learning algorithms that do the same task. This result reaffirms that deep learning is more advantageous than classic machine learning methods for handling a large dataset. However, when training on a smaller input data set in the 2S approach, which includes a classification stage to identify active fingers before predicting their trajectories, machine learning techniques offer a simpler implementation while ensuring comparably good decoding outcomes to the deep learning ones. In the classification step, either machine learning or deep learning models achieve the accuracy and F1 score of 0.99. Thanks to the classification step, in the regression step, both types of models result in a comparable mean squared error (MSE) and variance accounted for (VAF) scores as those of the 1S approach. Our study outlines the trade-offs to inform the future implementation of real-time, low-latency, and high accuracy deep learning-based motor decoder for clinical applications.

A bioelectric neural interface towards intuitive prosthetic control for amputees.

Objective. While prosthetic hands with independently actuated digits have become commercially available, state-of-the-art human-machine interfaces (HMI) only permit control over a limited set of grasp patterns, which does not enable amputees to experience sufficient improvement in their daily activities to make an active prosthesis useful.Approach. Here we present a technology platform combining fully-integrated bioelectronics, implantable intrafascicular microelectrodes and deep learning-based artificial intelligence (AI) to facilitate this missing bridge by tapping into the intricate motor control signals of peripheral nerves. The bioelectric neural interface includes an ultra-low-noise neural recording system to sense electroneurography (ENG) signals from microelectrode arrays implanted in the residual nerves, and AI models employing the recurrent neural network (RNN) architecture to decode the subject's motor intention.Main results. A pilot human study has been carried out on a transradial amputee. We demonstrate that the information channel established by the proposed neural interface is sufficient to provide high accuracy control of a prosthetic hand up to 15 degrees of freedom (DOF). The interface is intuitive as it directly maps complex prosthesis movements to the patient's true intention.Significance. Our study layouts the foundation towards not only a robust and dexterous control strategy for modern neuroprostheses at a near-natural level approaching that of the able hand, but also an intuitive conduit for connecting human minds and machines through the peripheral neural pathways.Clinical trial: DExterous Hand Control Through Fascicular Targeting (DEFT). Identifier: NCT02994160.

Fascicle specific targeting for selective peripheral nerve stimulation.

OBJECTIVE: Electrical stimulation is a blunt tool for evoking neural activity. Neurons are naturally activated asynchronously and non-uniformly, whereas stimulation drives simultaneous activity within a population of cells. These differences in activation pattern can result in unintended side effects, including muddled sensory percepts and undesirable muscle contractions. These effects can be mitigated by the placement of electrodes in close approximation to nerve fibers and careful selection of the neural interface's location. This work describes the benefits of placing electrodes within specific fascicles of peripheral nerve to form selective neural interfaces for bidirectional neuroprosthetic devices. APPROACH: Chronic electrodes were targeted to individual fascicles of the ulnar and median nerves in the forearm of four human subjects. During the surgical implant procedure, fascicles were dissected from each nerve, and functional testing was used to identify the relative composition of sensory and motor fibers within each. FAST-LIFE arrays, composed of longitudinal intrafascicular arrays and fascicular cuff electrodes, were implanted in each fascicle. The location, quality, and stimulation parameters associated with sensations evoked by electrical stimulation on these electrodes were characterized throughout the 90-180 d implant period. MAIN RESULTS: FAST-LIFE arrays enable selective and chronic electrical stimulation of individual peripheral nerve fascicles. The quality of sensations evoked by stimulation in each fascicle is predictable and distinct; subjects reported tactile and cutaneous sensations during stimulation of sensory fascicles and deeper proprioceptive sensations during stimulation of motor fascicles. Stimulation thresholds and strength-duration time constants were typically higher within sensory fascicles. SIGNIFICANCE: Highly selective, stable neural interfaces can be created by placing electrodes within and around single fascicles of peripheral nerves. This method enables targeting electrodes to nerve fibers that innervate a specific body region or have specific functions. Fascicle-specific interfacing techniques have broad potential to maximize the therapeutic effects of electrical stimulation in many neuromodulation applications. (Clinical Trial ID NCT02994160.).

Botulinum toxin suppression of CNS network activity in vitro.

The botulinum toxins are potent agents which disrupt synaptic transmission. While the standard method for BoNT detection and quantification is based on the mouse lethality assay, we have examined whether alterations in cultured neuronal network activity can be used to detect the functional effects of BoNT. Murine spinal cord and frontal cortex networks cultured on substrate integrated microelectrode arrays allowed monitoring of spontaneous spike and burst activity with exposure to BoNT serotype A (BoNT-A). Exposure to BoNT-A inhibited spike activity in cultured neuronal networks where, after a delay due to toxin internalization, the rate of activity loss depended on toxin concentration. Over a 30 hr exposure to BoNT-A, the minimum concentration detected was 2 ng/mL, a level consistent with mouse lethality studies. A small proportion of spinal cord networks, but not frontal cortex networks, showed a transient increase in spike and burst activity with exposure to BoNT-A, an effect likely due to preferential inhibition of inhibitory synapses expressed in this tissue. Lastly, prior exposure to human-derived antisera containing neutralizing antibodies prevented BoNT-A induced inhibition of network spike activity. These observations suggest that the extracellular recording from cultured neuronal networks can be used to detect and quantify functional BoNT effects.

Thiol-ene/acrylate substrates for softening intracortical electrodes.

Neural interfaces have traditionally been fabricated on rigid and planar substrates, including silicon and engineering thermoplastics. However, the neural tissue with which these devices interact is both 3D and highly compliant. The mechanical mismatch at the biotic-abiotic interface is expected to contribute to the tissue response that limits chronic signal recording and stimulation. In this work, novel ternary thiol-ene/acrylate polymer networks are used to create softening substrates for neural recording electrodes. Thermomechanical properties of the substrates are studied through differential scanning calorimetry and dynamic mechanical analysis both before and after exposure physiological conditions. This substrate system softens from more than 1 GPa to 18 MPa on exposure to physiological conditions: reaching body temperature and taking up less than 3% fluid. The impedance of 177 µm(2) gold electrodes electroplated with platinum black fabricated on these substrates is measured to be 206 kΩ at 1 kHz. Specifically, intracortical electrodes are fabricated, implanted, and used to record driven neural activity. This work describes the first substrate system that can use the full capabilities of photolithography, respond to physiological conditions by softening markedly after insertion, and record driven neural activity for 4 weeks.

Effects of carbon nanotube and conducting polymer coated microelectrodes on single-unit recordings in vitro.

Neuronal networks cultured on microelectrode arrays (MEAs) have been utilized as biosensors that can detect all or nothing extracellular action potentials, or spikes. Coating the microelectrodes with carbon nanotubes (CNTs), either pristine or conjugated with a conductive polymer, has been previously reported to improve extracellular recordings presumably via reduction in microelectrode impedance. The goal of this work was to examine the basis of such improvement in vitro. Every other microelectrode of in vitro MEAs was electrochemically modified with either conducting polymer, poly-3,4-ethylenedioxythiophene (PEDOT) or a blend of CNT and PEDOT. Mouse cortical tissue was dissociated and cultured on the MEAs to form functional neuronal networks. The performance of the modified and unmodified microelectrodes was evaluated by activity measures such as spike rate, spike amplitude, burst duration and burst rate. We observed that the yield, defined as percentage of microelectrodes with neuronal activity, was significantly higher by 55% for modified microelectrodes compared to the unmodified sites. However, the spike rate and burst parameters were similar for modified and unmodified microelectrodes suggesting that neuronal networks were not physiologically altered by presence of PEDOT or PEDOT-CNT. Our observations from immunocytochemistry indicated that neuronal cells were more abundant in proximity to modified microelectrodes.

The use of a novel carbon nanotube coated microelectrode array for chronic intracortical recording and microstimulation.

Micro-electrode arrays (MEAs) have been used in a variety of intracortical neural prostheses. While intracortical MEAs have demonstrated their utility in neural prostheses, in many cases MEA performance declines after several months to years of in vivo implantation. The application of carbon nanotubes (CNTs) may increase the functional longevity of intracortical MEAs through enhanced biocompatibility and charge injection properties. An MEA metalized with platinum (Pt) on all electrodes had a CNT coating applied to the electrodes on half of the array. This Pt/Pt-CNT MEA was implanted into feline motor cortex for >1 year. Recordings of action potentials and 1 kHz impedance measurements were made on all electrodes to evaluate device functionality. Additionally, electromyogram (EMG) responses were evoked using micro-stimulation via the MEA to measure device performance. These metrics were compared between Pt and Pt-CNT electrodes. There was no significant difference in the data acquisition or micro-stimulation performance of Pt and the Pt-CNT electrodes. However, impedances were lower on the Pt-CNT electrodes. These results demonstrate the functionality of CNT coatings during chronic in vivo implantation. The lower impedances suggest that for microstimulation applications CNT coatings may impart enhanced interface properties.

Normal molecular repair mechanisms in regenerative peripheral nerve interfaces allow recording of early spike activity despite immature myelination.

Clinical use of neurally controlled prosthetics has advanced in recent years, but limitations still remain, including lacking fine motor control and sensory feedback. Indwelling multi-electrode arrays, cuff electrodes, and regenerative sieve electrodes have been reported to serve as peripheral neural interfaces, though long-term stability of the nerve-electrode interface has remained a formidable challenge. We recently developed a regenerative multi-electrode interface (REMI) that is able to record neural activity as early as seven days post-implantation. While this activity might represent normal neural depolarization during axonal regrowth, it can also be the result of altered nerve regeneration around the REMI. This study evaluated high-throughput expression levels of 84 genes involved in nerve injury and repair, and the histological changes that occur in parallel to this early neural activity. Animals exhibiting spike activity increased from 29% to 57% from 7 to 14 days following REMI implantation with a corresponding increase in firing rate of 113%. Two weeks after implantation, numbers of neurofilament-positive axons in the control and REMI implanted nerves were comparable, and in both cases the number of myelinated axons was low. During this time, expression levels of genes related to nerve injury and repair were similar in regenerated nerves, both in the presence or absence of the electrode array. Together, these results indicate that the early neural activity is intrinsic to the regenerating axons, and not induced by the REMI neurointerface.

Carbon nanotube coating improves neuronal recordings.

Implanting electrical devices in the nervous system to treat neural diseases is becoming very common. The success of these brain-machine interfaces depends on the electrodes that come into contact with the neural tissue. Here we show that conventional tungsten and stainless steel wire electrodes can be coated with carbon nanotubes using electrochemical techniques under ambient conditions. The carbon nanotube coating enhanced both recording and electrical stimulation of neurons in culture, rats and monkeys by decreasing the electrode impedance and increasing charge transfer. Carbon nanotube-coated electrodes are expected to improve current electrophysiological techniques and to facilitate the development of long-lasting brain-machine interface devices.

The neuro-glial properties of adipose-derived adult stromal (ADAS) cells are not regulated by Notch 1 and are not derived from neural crest lineage.

We investigated whether adipose-derived adult stromal (ADAS) are of neural crest origin and the extent to which Notch 1 regulates their growth and differentiation. Mouse ADAS cells cultured in media formulated for neural stem cells (NSC) displayed limited capacity for self-renewal, clonogenicity, and neurosphere formation compared to NSC from the subventricular zone in the hippocampus. Although ADAS cells expressed Nestin, GFAP, NSE and Tuj1 in vitro, exposure to NSC differentiation supplements did not induce mature neuronal marker expression. In contrast, in mesenchymal stem cell (MSC) media, ADAS cells retained their ability to proliferate and differentiate beyond 20 passages and expressed high levels of Nestin. In neuritizing cocktails, ADAS cells extended processes, downregulated Nestin expression, and displayed depolarization-induced Ca(2+) transients but no spontaneous or evoked neural network activity on Multi-Electrode Arrays. Deletion of Notch 1 in ADAS cell cultures grown in NSC proliferation medium did not significantly alter their proliferative potential in vitro or the differentiation-induced downregulation of Nestin. Co-culture of ADAS cells with fibroblasts that stably expressed the Notch ligand Jagged 1 or overexpression of the Notch intracellular domain (NICD) did not alter ADAS cell growth, morphology, or cellular marker expression. ADAS cells did not display robust expression of neural crest transcription factors or genes (Sox, CRABP2, and TH); and lineage tracing analyses using Wnt1-Cre;Rosa26R-lacZ or -EYFP reporter mice confirmed that fewer than 2% of the ADAS cell population derived from a Wnt1-positive population during development. In summary, although media formulations optimized for MSCs or NSCs enable expansion of mouse ADAS cells in vitro, we find no evidence that these cells are of neural crest origin, that they can undergo robust terminal differentiation into functionally mature neurons, and that Notch 1 is likely to be a key regulator of their cellular and molecular characteristics.

Autologous transplants of Adipose-Derived Adult Stromal (ADAS) cells afford dopaminergic neuroprotection in a model of Parkinson's disease.

Adult adipose contains stromal progenitor cells with neurogenic potential. However, the stability of neuronal phenotypes adopted by Adipose-Derived Adult Stromal (ADAS) cells and whether terminal neuronal differentiation is required for their consideration as alternatives in cell replacement strategies to treat neurological disorders is largely unknown. We investigated whether in vitro neural induction of ADAS cells determined their ability to neuroprotect or restore function in a lesioned dopaminergic pathway. In vitro-expanded naïve or differentiated ADAS cells were autologously transplanted into substantia nigra 1 week after an intrastriatal 6-hydroxydopamine injection. Neurochemical and behavioral measures demonstrated neuroprotective effects of both ADAS grafts against 6-hydroxydopamine-induced dopaminergic neuron death, suggesting that pre-transplantation differentiation of the cells does not determine their ability to survive or neuroprotect in vivo. Therefore, we investigated whether equivalent protection by naïve and neurally-induced ADAS grafts resulted from robust in situ differentiation of both graft types into dopaminergic fates. Immunohistological analyses revealed that ADAS cells did not adopt dopaminergic cell fates in situ, consistent with the limited ability of these cells to undergo terminal differentiation into electrically active neurons in vitro. Moreover, re-exposure of neurally-differentiated ADAS cells to serum-containing medium in vitro confirmed ADAS cell phenotypic instability (plasticity). Lastly, given that gene expression analyses of in vitro-expanded ADAS cells revealed that both naïve and differentiated ADAS cells express potent dopaminergic survival factors, ADAS transplants may have exerted neuroprotective effects by production of trophic factors at the lesion site. ADAS cells may be ideal for ex vivo gene transfer therapies in Parkinson's disease treatment.

Robust cell migration and neuronal growth on pristine carbon nanotube sheets and yarns.

Carbon nanotubes (CNTs) have unique chemical and physical properties anticipated to enable broad novel biomedical applications. Yet the question concerning their biocompatibility remains controversial. We recently reported a method for rapidly preparing strong, highly electrically conducting sheets and yarns from multi-walled CNTs. The present studies demonstrate that highly oriented 50-nm-thick semi-transparent CNT sheets and yarns, produced with a minimal residual content of catalytic transition materials, support the long-term growth of a variety of cell types ranging from skin fibroblasts and Schwann cells, to postnatal cortical and cerebellar neurons. We show that CNT sheets stimulate fibroblast cell migration compared to plastic and glass culture substrates; entice neuronal growth to the level of those achieved on polyornithine-coated glass and can be used for directed cellular growth. These findings have positive implications for the use of CNTs in applications such as tissue engineering, wound healing, neural interfaces and biosensors.

A segment of the Mecp2 promoter is sufficient to drive expression in neurons.

Rett syndrome (RTT) is caused by mutations in the gene encoding methyl CpG-binding protein 2 (MeCP2). Although MeCP2 shows widespread expression in both neuronal and non-neuronal tissues, the symptoms of RTT are largely neurological. Herein, we have identified the regulatory region of the mouse Mecp2 gene that is sufficient for its restricted expression in neurons. A segment of the Mecp2 gene (-677/+56) exhibited strong promoter activity in neuronal cell lines and cortical neurons, but was inactive in non-neuronal cells and glia. The region necessary for neuronal-specific promoter activity was located within a 19 bp region (-63/-45). Several nuclear factors were found to bind to this region and some of these factors were enriched in nuclear extracts prepared from the brain. To examine the activity of the Mecp2 promoter in vivo, we generated transgenic mice expressing the LacZ reporter driven by the -677/+56 region of the Mecp2 gene. The transgene was expressed in the mesencephalon as early as embryonic day 10 and in the hindbrain and spinal cord by E12. Interestingly, a marked induction of transgene expression was observed postnatally throughout the brain, similar to that of endogenous MeCP2. However, expression of the transgene was absent in non-neuronal tissues that are known to express Mecp2. Taken together, these data indicate that the -677/+56 region of the Mecp2 promoter partially recapitulates the native expression pattern of the Mecp2 gene, which possesses restricted expression in neurons of the central nervous system.

A cultural renaissance: in vitro cell biology embraces three-dimensional context.

Increasingly, researchers are recognizing the limitations of two-dimensional (2-D), monolayer cell culture and embracing more realistic three-dimensional (3-D) cell culture systems. Currently, 3-D culture techniques are being employed by neuroscientists to grow cells from the central nervous system. From this work, it has become clear that 3-D cell culture offers a more realistic milieu in which the functional properties of neurons can be observed and manipulated in a manner that is not possible in vivo. The implications of this technical renaissance in cell culture for both clinical and basic neuroscience are significant and far-reaching.

Cultured rat hippocampal neural progenitors generate spontaneously active neural networks.

We previously demonstrated that the neural cell adhesion molecule (N-CAM) inhibited the proliferation of cultured rat hippocampal progenitor cells and increased the number of neurons generated. We demonstrate here that the continued presence of fibroblast growth factor 2 along with N-CAM or brain-derived neurotrophic factor over 12 days of culture greatly increased the number of both progenitors and neurons. These progenitor-derived neurons expressed neurotransmitters, neurotransmitter receptors, and synaptic proteins in vitro consistent with those expressed in the mature hippocampus. Progenitor cells cultured on microelectrode plates formed elaborate neural networks that exhibited spontaneously generated action potentials after 21 days. This activity was observed only in cultures grown in the presence of fibroblast growth factor 2 and either N-CAM or brain-derived neurotrophic factor. Analysis of neuronal activity after various pharmacological treatments indicated that the networks formed functional GABAergic and glutamatergic synapses. We conclude that mitogenic growth factors can synergize with N-CAM or neurotrophins to generate spontaneously active neural networks from neural progenitors.