Neuroprosthesis system and method for restoring sensorimotor function

Abstract

The present disclosure relates to a neuromorphic prosthetic system for facilitating sensorimotor function in a subject in need thereof. The system includes a plurality of stimulators each having one or more channels configured to be disposed at different locations of the subject to provide electrical stimulation to muscle groups, peripheral nerves, nerve plexus, and / or spinal cord. The system further includes a plurality of sensors configured to detect and communicate continuous or periodic distribution of body weight, joint angular motion, kinematics, and / or electrophysiological parameter data, and a plurality of controllers configured to receive combined data from the plurality of sensors. The controllers are further configured to process the combined data and communicate the combined data to a fMEP-based artificial circuitry device configured for implementing topological restructuring of a central pattern generator (CPG). The fMEP-based artificial circuitry device is configured to apply biologically plausible neuronal topologies to motor function patterns and coordinate application of electrical signals to the subject in need through the plurality of stimulators to perform the motor function patterns.

Description

Technical Field

[0001] This invention generally relates to systems for activating sensorimotor functions. More specifically, this invention relates to a system for the recovery of motor dysfunction after spinal cord injury based on real-time assessment of functional motor evoked potentials (fMEP) and reconstruction of neuronal circuit topology based on fMEP, and a method for reconstructing neuronal circuit topology. Background Technology

[0002] Spinal cord injury (SCI), brain injury, stroke, and other neurological disorders disrupt communication between the spinal cord and the rest of the body, thus affecting voluntary control of functions below the level of injury. Animal and human studies have shown that epidural and / or intraspinal electrical stimulation can activate spinal motor circuits following SCI. For example, epidural electrical stimulation (EES) restored coordinated movement in animal models of SCI and isolated leg movement in patients with motor paralysis.

[0003] It has been demonstrated that the structure of the spinal cord comprises various microcircuits. These microcircuits process sensory information from limb movement and input from various brain regions to generate motor responses and coordinate complex motor patterns. This pattern forms a circuit responsible for rhythmic motor outcomes, such as movement, and is expected to include a group of mutually inhibitory nuclei with periodic inhibitory and / or excitatory activity. The evidence suggests that simulating microcircuit patterns can drive the development of neuromodulation programs at different levels of the central nervous system. In particular, implementing motor-responsible microcircuit patterns into neuroprosthetic systems to compensate for lost motor function can restore motor function in patients with spinal cord injury (SCI). Summary of the Invention

[0004] In one aspect, the present invention provides a neuromorphic prosthesis system for facilitating sensorimotor function in a desired subject. The system includes multiple stimulators, each having one or more channels, configured at different locations on the subject to provide coordinated electrical stimulation of muscle groups, peripheral nerves, nerve plexuses, and the spinal cord. The system also includes multiple sensors configured to detect and transmit continuous or periodic distributions of body weight, joint angle motion, kinesiology, and electrophysiological parameters. Furthermore, the system includes multiple controllers configured to receive combined data from the multiple sensors and to process and transmit the combined data to an fMEP-based artificial circuit device configured to reconstruct the topology of a central pattern generator (CPG). The fMEP-based artificial circuit device applies a biologically reliable neuronal topology to the motor function pattern and coordinates the application of electrical signals matched to the biological neuronal stimulation through the multiple stimulators to execute the motor function pattern.

[0005] In another aspect, the present invention provides a method for activating motor function of a limb, targeting several levels of a subject, such as central circuits at the muscle, peripheral nerve, nerve plexus, and spinal cord levels. The method includes providing multiple stimulators, each having one or more channels, configured at different locations on the subject, to provide coordinated electrical stimulation of muscle groups, peripheral nerves, nerve plexuses, and the spinal cord to control multiple targets. The method further includes providing multiple sensors configured to detect and transmit continuous or periodically distributed data on body weight, angled joint movements, kinematics, and electrophysiological parameters from the multiple sensors. Furthermore, the method includes providing a controller configured to receive data from the multiple sensors to process and transmit the combined data to an fMEP-based artificial circuit device configured to reconstruct the topology of a central pattern generator (CPG). Further, the method includes applying electrical signals to the subject via the multiple stimulators to stimulate the subject to perform motor functions. The fMEP-based artificial circuit device applies a biologically reliable neuronal topology to the motor function pattern and coordinates the application of electrical signals.

[0006] In another aspect, the present invention provides a method for reconstructing the neuronal topology responsible for activating limb motor function based on the assessment of functional motor evoked potentials. The method includes detecting neuronal response patterns associated with input stimulus parameters, analyzing the neuronal response patterns to obtain stable neuronal response patterns, which are determined based on multiple peaks, each with a maximum and minimum value, and each peak having amplitude and duration. The method further includes assessing the neuronal circuit topology by comparing the neuronal response patterns with known neuronal biological responses, the neuronal circuit topology forming the output signal that activates limb motor function.

[0007] In another aspect of the invention, oscillator motifs (OMs) are used to generate neuronal response patterns, wherein the OMs include mutually stimulating components and feedback inhibition components.

[0008] In another aspect of the invention, a multilevel spinal motor circuit (mSLC) model is used to generate neuronal response patterns, the mSLC model comprising monosynaptic layers and multisynaptic layers. Attached Figure Description

[0009] To facilitate understanding of the invention, a more detailed description of the invention, briefly described above, will be provided with reference to the specific embodiments presented in the accompanying drawings. It should be understood that these drawings depict only exemplary embodiments of the invention and should not be construed as limiting its scope. Various aspects of the invention will be described and explained with additional features and details using the drawings.

[0010] Figure 1 (A)-1(C) depicts an oscillator motif (OM) according to an embodiment of the present invention;

[0011] Figure 2 A multi-level spinal motor circuit (mSLC) model according to an embodiment of the present invention is described;

[0012] Figure 3 Electromyography depicts the functional motor evoked potential (fMEP) patterns of extensor and flexor muscles in response to epidural electrical stimulation (EES) pulses according to an embodiment of the present invention.

[0013] Figure 4 A neuronal response pattern based on the peak concept for detection according to an embodiment of the present invention is described;

[0014] Figure 5 A pyramidal multisynaptic fMEPs pattern with peak density is depicted according to an embodiment of the present invention;

[0015] Figure 6 This illustrates that, according to an embodiment of the invention, the delay is identified as the time (distance) between monosynapses and multisynaptic fMEPs;

[0016] Figure 7 A neuromorphic prosthesis system according to an embodiment of the present invention is described;

[0017] Figure 8 A digital neuromorphic prosthesis system according to an embodiment of the present invention is described;

[0018] Figure 9 A pulsed (simulated) neuromorphic prosthesis system according to an embodiment of the present invention is described;

[0019] Figure 10 Various pulse activities according to embodiments of the present invention are described;

[0020] Figure 11 This illustrates the pulse shapes in (a1) the Izhikevich model, (a2) the Leaky Integrate-and-Fire model, and (a3) ​​the simple digital neuron.

[0021] Figure 12 A simplified digital neuron model according to an embodiment of the present invention is described. Detailed Implementation

[0022] In this specification, references to "specific embodiments" or similar expressions mean that a specific feature, structure, or characteristic described in a specific embodiment is included in at least one specific embodiment of the invention. Therefore, "in a specific embodiment" or similar expressions in this specification do not necessarily refer to the same specific embodiment.

[0023] In the following description, various embodiments of the invention will be described in more detail with reference to the accompanying drawings. Nevertheless, it should be understood that those skilled in the art can make modifications to the invention based on the following description to achieve the superior effects of the invention. Therefore, the following description should be considered as a general and explanatory description relating to the invention to those skilled in the art, and not as a limitation of the claims.

[0024] In this specification, references to "an embodiment," "a particular embodiment," or similar expressions mean that the relevant features, structures, or characteristics described in that embodiment are included in at least one embodiment of the invention. Therefore, "in one embodiment," "in a particular embodiment," or similar expressions in this specification do not necessarily refer to the same specific embodiment.

[0025] A system for recovering motor dysfunction following spinal cord injury (SCI) is provided based on reconstructing the topology of neuronal circuits. The structure of the spinal cord comprises various neuronal microcircuits. These neuronal microcircuits process sensory information, such as sensory information from limb movement, and input from different brain regions to generate motor responses. Motor pattern formation is introduced by a group of nuclei that typically exhibit mutual and periodic inhibition and / or excitatory projection and activity. The reconstruction and / or simulation of neuronal microcircuit patterns can advance the development of neuromodulation solutions to achieve motor control in SCI patients.

[0026] The topology of the neuronal circuits responsible for generating complex rhythmic patterns, namely the central pattern generator (CPG), is largely indeterminate. However, it has been demonstrated that epidural electrical stimulation (EES) can facilitate complex movements in SCI animals and patients by providing sensory (e.g., electrical) input to the CPG circuits. Embodiments of this invention specify that, since the topology of the CPG circuits remains largely unknown, the reconstruction (i.e., reverse engineering) of the spinal CPG neuronal circuits based on fMEP, responsible for the formation of movement patterns, can generate a computational model (in silico) that can be implemented in a single-board computer (in cyberico). The resulting computational model of the neuronal topology can then be implemented in a CPG implementation device that allows the generation of electrical signals adapted to the biological neuronal stimulation of the SCI patient's muscles to activate motor function in the SCI patient's limbs.

[0027] To reconstruct the neuronal topology responsible for motor patterns in patients with spastic ileus (SCI), according to embodiments of the present invention, neuronal response patterns are detected based on known inputs (i.e., sensory modulation and stimulation parameters) and electrophysiological outputs associated with specific patterns. The relationship between neuronal response patterns and temporal and amplitude modulation of the response is then analyzed. Based on the analysis of neuronal response patterns, the neuronal microcircuit topology is reconstructed. This neuronal microcircuit topology is then realized in a pattern-forming circuit that can generate electrical signals adapted to biological neuronal stimulation to activate limb motor function.

[0028] Detecting neuronal response patterns associated with input stimuli

[0029] According to embodiments of the invention, the detection of neuronal response patterns is based on known or partially known stimulation parameters, such as the amplitude, frequency, and duty cycle of isolated neuronal circuits. The following in vivo models can be used to induce motor responses: air stepping (i.e., a biological model in which a subject walks in the air without contact with a treadmill, fully supported by a body weight support (BWS) system), a model of a subject walking with minimal contact with the treadmill (TOE), and a biological model of a subject walking bipedally with BWS support to maintain a vertical posture (PLT). Furthermore, biological models with different treadmill speeds, models incorporating the pharmacological effects of quinoperazine (QPZ) and strychnine (STR), and models with various known epidural stimuli can be used to detect neuronal response patterns. The diversity of inputs and outputs assessed in an in vivo experimental setting is described in detail elsewhere [Islam et al., 2019]. We have implemented this approach to assess the potential topology and functional organization of circuits based on variations in motor outputs from various sensory inputs. To record the evoked motor response, any known method can be used, for example, by placing wires on the following muscles to capture electromyography (EMG): gluteus maximus (GLU), rectus femoris (RF), lateral thigh muscles (VL); tibialis anterior (TA), soleus (SOL), and medial gastrocnemius (MG).

[0030] Furthermore, computer simulation models can be used to detect and simulate the response patterns of neurons to known input parameters, such as the NEURON, NEST, and / or GRAS simulators. According to one implementation, the Hodgkin-Huxley model of neurons can be used with the NEURON simulator. See Hodgkin, AL, Huxley, AF: Quantitative description of membrane currents and their application in neural conduction and excitation, *Journal of Physiology*, 1952, Vol. 117, No. 4, pp. 500-544. The NEST simulator can also be used. NEST is a simulator for spiking neural network models that focuses on the dynamics, size, and structure of the nervous system rather than on the exact morphology of individual neurons. See Jordan, J., Ippen, T., Helias, M., Kitayama, I., Sato, M., Igarashi, J., Diesmann, M., Kunkel, S.: Highly scalable spiking neural network simulation programs: from laptops to very large-scale computers, *Frontiers in Neuroinformatics*, February 2018, Vol. 12.

[0031] The GRAS simulator uses the C++ programming language and Nvidia CUDA technology to perform GPU processing on biologically reliable neural networks. More specifically, the GRAS simulator is a real-time simulator of neuronal activity with predefined neuronal topologies. It can support three different neuron models: (1) the Simpler Real-Time Model of Neurons (ESRN), see Leukhin, A., Talanov, M., Suleimanova, A., Toschev, A. & Lavrov, I. Simpler Real-Time Model of Neurons, BioNanoscience, 2020, pp. 1-4; (2) Hodgkin-Huxley neurons, see Hodgkin, AL & Huxley, AF: Quantitative Description of Membrane Currents and Their Application in Neural Conduction and Excitation, Journal of Physiology, Vol. 117, pp. 500-544; (3) Izhikevich neurons, see Izhikevich, EM Multitemporalization: Computation with Impulses, Neural Networks, 2006. GRAS focuses on the entire neuronal topology model, contrasting with individual neurons in NEST. Due to its high efficiency, GRAS also uses GPUs to handle its computations.

[0032] According to an embodiment of the present invention, to detect and simulate neuronal response patterns, a simplified neuron simulator model (a simpler real-time neuron model (ESRN)) can be used to achieve real-time processing in embedded biocompatible devices. A key issue in biomimetic neural simulation is computation time; wearable neural prostheses need to process the topology of thousands of neurons and hundreds of thousands of synapses in real time on a single-board computer, which is lightweight and typically not connected to an external network. The simplified digital neuron model meets the following requirements: real-time processing on a single-board computer, considering only biologically reliable pulse duration, refractory period, inhibitory effects, and threshold-based input pulse processing. According to one embodiment of the present invention, the simplified simulator model includes the following assumptions to optimize the neuron model code for real-time processing of a large number of neurons (thousands or more) and a large number of synapses (hundreds of thousands or more): modeling only motor neurons and interneurons, disregarding pulse amplitude, and utilizing biologically reliable pulse and refractory period durations. For real-time computation, the following simplified equation for membrane potential is used.

[0033] L=ΣW±Leakage+Noise

[0034] The level is the sum of projection weight, leakage, and noise.

[0035] According to an embodiment of the present invention, in neuronal response pattern detection and analysis, such as Figure 1 The oscillator motifs (OMs) shown and described in (A)-1(C) are used as the basis (building blocks) of neuronal circuits and their microcircuits, for example, found in the spinal cord. OMs comprise four common motifs of the central nervous system (CNS): (i) feedback inhibition; (ii) rhythmic excitation; (iii) divergence; and (iv) convergence. More specifically, an OM consists of two functional parts: (i) mutual excitation (1, 2) (as shown in (A)-1(C)). Figure 1 (B) and (ii) feedback inhibition 1, 3 (as shown in Figure 1) and 3 Figure 1 (As shown in (C)). OM can generate neuronal activity of varying durations. For example... Figure 1 As shown in (A)-1(C), the duration of neuronal activity depends on the weight balance between excitatory and inhibitory nuclei. Input nucleus 1 receives signals, such as those from afferent and / or EES, and triggers mutual excitation with the second nucleus 2. The first and second excitatory nuclei 1 and 2 have strong reciprocal projections and produce output activity, which is terminated by inhibitory projections from the third nucleus 3, which projects weakly to the excitatory nuclei 1 and 2. Furthermore, output activity can be terminated by external excitatory stimuli to the third nucleus (e.g., external excitatory stimuli). Figure 1(As shown in A). In other words, according to an embodiment of the invention, OM generates neuronal activity with a predetermined duration and amplitude through a balance between inhibitory and excitatory projections.

[0036] like Figure 1 As shown in (A)-1(C), OM generates the neuronal activity dynamics of functional motor evoked potentials (fMEP). fMEP can be measured by, for example, (i) the delay after each EES pulse; (ii) the duration; (iii) the amplitude; and (iv) the number of peaks between a pair of maximum and minimum extrema (e.g. Figure 4 (As shown).

[0037] Figure 2 A detailed illustration of a multi-level spinal motor circuit (mSLC) model according to an embodiment of the present invention is shown. During walking, the generation of fMEPs is triggered by the sequential activation of monosynaptic and multisynaptic layers via combined input from ventral skin (CV) sensory afferents and EES pulses, thereby amplifying the neuronal activity of CV1 through CV5 (e.g., Figure 2 (As shown in the diagram). mSLC includes polysynaptic level 20 (associated with multiple fMEPs) and monosynaptic level 30 (associated with single fMEPs). Monosynaptic level 30 includes motor neurons (MN-E, MN-F), afferent fibers (Iaf-E, Iaf-F), interneurons associated with Ia-afferents, antagonistic muscles that inhibit motor neurons (Ia-E, Ia-F), and Lanshoe cells (RE, RF), which protect muscles from strong contractions that create negative feedback. Figure 2 As shown, this level is triggered by Ia-induction, such as EES activation. Multisynaptic levels 20 form multisynaptic responses, typically longer than monosynaptic responses, but with lower maximum amplitudes, thus forming complex neuronal structures. The multisynaptic layer 20 is composed of… Figure 2 It consists of five layers as shown.

[0038] The number of multisynaptic layers 20 corresponds to the number of multisynaptic fMEP delays. Each of the five layers contains its own OM (such as...). Figure 1 As shown), and formed through two motifs: mutual excitation ( Figure 1 (B) and feedback inhibition Figure 1 (C)), this motif also includes an inhibitory nucleus, which modulates the length of neuronal activity in the OM and ensures timely neuronal output. In other words, the multisynaptic level 20 includes, for example... Figure 2As shown, CVs 1 to 5, OMs 1 to 5, and the excitatory (e) and inhibitory (i) portions of the interneuronal cisterns: IP-E, IP-F, where the capital letter E represents the extensor nucleus and F represents the flexor nucleus. Each CV is a group of ascending fibers, which includes fibers activated by CV inputs and forming the baseline activity of the corresponding CV, as well as fibers enhanced by EES activation, for example, the fiber electrical activity (e) nucleus.

[0039] The increased delay between monosynaptic and polysynaptic responses is due to inhibition of the lower layer by the upper layer (e.g., Figure 2 (As shown), excitatory projections onto inhibitory nuclei are formed. Higher layers CV3 to 5 inhibit lower layers OM1 to 3. Each layer is triggered by sensory afferents or ventral skin CV1 to 5 and / or EES.

[0040] The multisynaptic level extensor mode has two modes. The first is the standing mode, where each OM is triggered by co-activation from CVs 1 to 5 and EES. Cutaneous afferents are triggered from the toes to the heel according to the distribution of body weight. In the second mode (swinging mode), the working sequence of the multisynaptic layer 20 is independent of the activation of CVs, because CVs are not activated when one foot is in the air. Thus, OM1 (such as...) Figure 2 As shown, it was triggered by the EES and CV1 of the other leg.

[0041] Figure 3 Electromyography (EMG) of the fMEP pattern in response to EES pulses in the extensor (Soleus) and flexor (Tibialis anterior) muscles is shown, with a significant delay in each slice (the period between EES pulse stimulations is 40 Hz). In the standing mode, the first EES pulse and the thalamic cutaneous afferent nerve trigger OM1 (e.g., OM1) via the CV1 neuronal group. Figure 2 (As shown). This generated neuronal activity in the first slice (13 to 15 milliseconds (ms)), as... Figure 3 As shown. The next EES pulse and weight transfer from the toes to the heel activate the neuronal populations of CV1 and CV2, triggering OM1 and OM2 (via projections from CV), resulting in an extended fMEP with higher amplitude for the next slice (13 to 21 ms). The third EES pulse and weight transfer activate CV3, triggering OM2 and later OM3, resulting in a long fMEP (15 to 23 ms) and an fMEP (0 to 3 ms) for the fourth slice. CV3 inhibits OM1, implementing feedback inhibition.

[0042] The fourth and fifth EES pulses, along with weight transfer, near the toe cutaneous afferent nerves projecting to CV4-OM3 and OM4 (19 ms), produce short, dense activity in the fifth and sixth slices of slices four and five (19-25 ms and 0-3 ms, respectively), inhibiting OM1 and OM2. The sixth EES pulse has a special function because it triggers OM5 via CV5 in conjunction with weight transfer to the toe and inhibits OM1-OM3 (e.g., ...). Figure 2 (as shown), thus effectively preventing them from generating early responses and projecting short-term neuronal activity with relatively high amplitude.

[0043] During the swing mode, given the lack of skin input, Figure 3 The image shows three delays: 17 milliseconds, 12 milliseconds, and 20 milliseconds, with durations of 17 milliseconds, 44 milliseconds, and 5 milliseconds, respectively. Figure 2 The five-layer structure shown produced Figure 3 The flexor fMEP pattern shown in the electromyography. More specifically, the first EES pulse of CV1 triggers the OMs in the first and second layers. First, OM1 is triggered, with the 0-preceding-1 excitation nuclei of OM1 (such as...). Figure 2 (As shown) triggers the second nucleus of OM2. This produces strong fMEP activity, starting at 17 ms in the seventh slice and continuing until 3 ms in the eighth slice (as shown). Figure 3 (As shown in TA). The second EES pulse triggers CV2 and CV1, effectively boosting the 17 ms of the 8th slice to 25 ms, up to the 3 ms output fMEP of the 9th slice (as shown in TA). Figure 3 (As shown in TA). The third and fourth EES pulses trigger OM3 via CV3 and CV4, while nucleus 2 of OM3 triggers OM4 (as shown in TA). Figure 2 (As shown). This pair produces long and dense fMEPs from 12ms of the ninth slice to 6ms of the eleventh slice (as shown). Figure 3 (As shown in TA). Then, the fifth EES pulse triggered CV5 and the projection from nucleus 0 of OM 4 to nucleus 0 of OM 5, and activated OM 5 (as shown in TA). Figure 2 (As shown). This generates 20 to 25 millisecond neuronal pulses in the eleventh slice.

[0044] When the second nucleus triggers the next OM, the flexor pattern is organized in a sequential OM activation manner (e.g., Figure 2 (As shown). Inhibitory presynaptic projections are present on the right and left IP-E, inhibiting excitatory projections of the second nucleus of the OM during standing mode. These projections are activated via CV1 to CV5. During swinging mode, these projections play a crucial role in the formation of flexor patterns.

[0045] According to an embodiment of the invention, there are two main input stimulus mSLC models for detecting neuronal response patterns EES and skin. The main effect of skin input applies to a monosynaptic level 30, which operates simultaneously on both legs. The balance between the two legs has a diagonal nature; when the extensors of the right leg are activated due to skin input, the flexors of the left leg are also activated. On the other hand, when the extensors of the left leg are active, the flexors of the right leg are also activated, while the rest are inhibited.

[0046] Analysis of detected neuronal response patterns

[0047] According to an embodiment of the present invention, the analysis of the detected neuronal response patterns is based on the assessment of functional motor evoked potentials (fMEPs) and the analysis of complex patterns generated by the neuronal network. Specifically, peaks are considered as a pair of extreme values: a first maximum and a subsequent minimum, such as... Figure 4 As described. The increment (Δ) between the maximum and minimum values ​​is the amplitude of the peak, and the duration of the peak value = t. min -t max (like Figure 4 (As shown in dA). According to an embodiment of the present invention, for a peak to be valid, the peak duration should be between 0.15 ms and 4 ms, and the relative amplitude should be greater than 3% of the maximum fMEP amplitude, such as... Figure 4 As shown. The minimum time and amplitude thresholds can be determined by taking the 15th and 25th percentiles of all peaks and comparing different time and amplitude thresholds. The ranges described herein allow for the avoidance of long electromyographic waves that are not peaks or "noise-like" peaks.

[0048] According to embodiments of the present invention, electromyography is analyzed using any suitable method, such as kinematic density estimation (KDE), to determine stable patterns. Figure 3 and Figure 5 As shown, stable modes can be determined by using in vivo and computer data from various models, such as AIR, TOE, PLT, QPZ, STR, NEURON an / or GRAS, through a specific number of peaks, relative aggregation amplitudes, and delays.

[0049] According to embodiments of the invention, the data can be validated using the KDE test. In particular, validating computer data via in vivo electromyography requires a uniform representation of the dataset. In vivo data, or myogram packages, are recorded during movement in a biological model (animal), and may inherently produce unwanted artifacts. For example, zero-point shift is unacceptable in the context of computer data validation because it is not included in the simulation data, as the computer model does not experience electrode movement during movement. To compensate for zero-point shift, PCA eigenvectors can be used to rotate the center point of the electromyography package within the computer model.

[0050] According to embodiments of the present invention, given the different properties of multisynaptic and monosynaptic fMEPs, two different analytical procedures for neuronal response modes can be used. First, monosynaptic fMEPs demonstrate the regularity of peaks in the range of 3 to 8 milliseconds across each slice of in vivo or computer data (e.g., Figure 3 (As shown). On the other hand, multisynaptic fMEPs modes exhibit pyramidal peak densities, for example, in the PLT mode, within the range of 8 ms to 25 ms and 0 ms to 3 ms in consecutive slices (represented by KDE, as shown). Figure 5 (As shown). By analyzing the peak density that does not belong to the monosynaptic fMEP interval, for example, in the range of 8 ms to 25 ms, the boundary line between monosynaptic and multisynaptic fMEPs is selected.

[0051] Furthermore, considering a synaptic delay of 2 ms, the delay Δ between responses determines the presence of more than one nucleus generating the response, and its value determines the number of synapses between activations of each nucleus. The number of peaks determines the distribution of delay and projection weights within a nucleus (e.g., Figure 5 As shown). The relative amplitude Δ indicates the presence of more than one nucleus with different neuronal activities (e.g., ...). Figure 5 (As shown). The response duration is related to the number of peaks and determines the distribution of delays. If the overall duration is longer than a synaptic delay, there exists an internal nucleus topology with multiple synaptic connections.

[0052] like Figure 6 As shown, the delay was defined as the time (distance) between monosynaptic and multisynaptic fMEPs within a time range of 5 ms to 20 ms, where there is overlap between monosynaptic fMEPs (3 ms to 8 ms) and guaranteed multisynaptic fMEPs (8 ms to 25 ms). To detect and measure the overall delay value of each fMEP package in each in vivo model under each mode, a three-dimensional KDE mapping method comparing volumes below a threshold line can be used. The threshold for the QPZ / STR model is determined by the following formula:

[0053] isoline th = 2 / 3(max + min)

[0054] For the AIR / TOE / PLT model:

[0055] isoline th = 1 / 2(max + min)

[0056] Where max is the maximum peak density in the current fMEP packet, and min is the minimum peak density in the current fMEP packet.

[0057] According to embodiments of the present invention, the amount of delay between monosynaptic and multisynaptic fMEPs, the number of peaks, the amplitude in monosynaptic and multisynaptic fMEPs, and the shape of the multisynaptic mode (e.g.) are taken into consideration. Figure 3 The low-level details shown can be further verified using any suitable method, such as the KDE test, to validate the neuron's response pattern.

[0058] Reconstruction of neuronal circuit patterns / topology

[0059] According to embodiments of the invention, the above-mentioned results, such as (i) the approximate number of nuclei, (ii) the approximate delay between activations of each nucleus, (iii) the distribution range of weights and delays for each projection (synaptic connection), and (iv) the type of neurons in each nucleus, are used to reconstruct the neuronal circuit topology using known types of neurons that form output signals. According to embodiments of the invention, two levels of abstraction can be used to reconstruct the neuronal circuit topology: a high-level topology and a low-level topology. In particular, the high-level topology of neurons includes delays between neuronal responses, the relative amplitude of each response, and the number of peaks. Within a synaptic delay, the duration of the response, the aggregate amplitude and number of peaks, the distribution of projection weights and delays are determined to reconstruct the low-level topology. For example, if the duration exceeds one synaptic delay, the neuronal topology within the nucleus is assumed, taking into account the number of neurons in each nucleus and the distribution of their connection parameters. Then, the activation time of the nucleus is used to determine the input activity, taking into account the common activation projections of the nuclei. Furthermore, in the case of high sampling rates of neuronal activity recordings, it is possible to use clustering methods to identify neurons that generate pulses. When applying the above methods, known motifs of neurons in the CNS, for example, can be used to reconstruct neuronal topology. The neuronal microcircuit topology can then be realized in a CPG device that generates electrical signals adapted to biological neuronal stimulation to activate limb motor function.

[0060] Figure 7A neuromorphic prosthetic system 100 is described. According to an embodiment of the invention, the neuromorphic prosthetic system 100 is configured to facilitate walking in patients with complete or partial SCI and leg paralysis. The neuromorphic prosthetic system 100 includes an ankle controller 110 and a hip controller 115 for each limb, a central pattern generator device 130, and a stimulator 150. Controllers 110 and 115 are data processing devices that receive input data from pressure sensors 160, which may be disposed on an insole, and flexion sensors 170, which may be disposed around the ankle and / or knee joints. The ankle controller 110 may be disposed around the ankle and transmits data to the hip controller 115. The hip controller 115 preprocesses the data received from the ankle controller 110 and combines the data received from the ankle controller 110 with the data received from the flexion sensor 170. The hip controller 115 then transmits the combined data to a central CPG implementation device 130.

[0061] The central CPG device 130 can be implemented as a group of synchronized single-board digital computers (e.g., PINE A 64) (as shown). Figure 8 (as shown) or pulse (analog) schematic diagram (e.g. Figure 9 As shown), a central pattern generator or an application-specific integrated circuit (ASIC) chip with the same function is used to realize a segment of the spinal cord topology. The CPG 130 device realizes a bioreliable topology for the walking pattern generator, capable of generating electrical signals that are adapted to the biological neuronal stimulation of the lower limb muscles. Figure 2 The text details the neuronal topology of the CPG, specifically the two-layer topology (monosynaptic and multisynaptic) of the motor pattern generator and the five-layer multisynaptic layer of the motor pattern generator. The CPG device 130 can also balance flexor and extensor pairs and the left and right legs.

[0062] Stimulator 150 may include a back stimulator 158, a left ankle stimulator 151, a right ankle stimulator 153, a right hip stimulator 157, and a left hip stimulator 159. Stimulator 150 may be an electrode positioned on a muscle, nerve, nerve plexus, or for percutaneous spinal cord stimulation. Stimulators 150 and 250 may be single-channel or multi-channel stimulators, suitable for stimulation according to embodiments of the present invention.

[0063] The output of the CPG triggers stimulators 151, 153, 157, 158, and 159, which transmit the walking pattern individually to each flexor-extensor pair via ascending nerves or direct innervation of the hindlimb muscles (hips and feet). The muscles, in turn, balance the weight toward the direction of movement, thereby altering the weight distribution recorded by pressure sensor 160 and the joint angle recorded by flexion sensor 170.

[0064] System 100 can be an open-loop or closed-loop feedback system. In some embodiments, controllers 110, 115 can communicate with an external computing system (not shown). For example, a user can define a stimulation program on the external computing system, and then the program can be uploaded to controllers 110, 115, or to the central CPG implementation device 130.

[0065] According to another embodiment of the present invention, Figure 8 The image shows a digital neuromorphic prosthesis system 200. The digital neuromorphic prosthesis system 200 includes an ankle controller 210 and an insole controller 225, a CPG device 230, a stimulator 250, a pressure sensor 260, a digital-to-analog converter (DAC) 220, and a universal asynchronous transmitter (UART) serial port 240. According to this embodiment, the CPG device 230 may be a single-board computer.

[0066] Here, the input signal is formed by a weight distribution “from toe to heel” and processed by the insole controller 225. Pressure sensor data is transmitted from the insole controller 225 to the central pattern generator device 230 (single-board computer) via UART serial port 240, where the weighted Δ “from toe to heel” of the sensor data is used as input for the CPG device 230. The central pattern generator device 230 generates digital motion patterns for the hind limb muscles in real time and triggers the stimulator 250 via DAC 220. That is, DAC 220 converts the voltage signal from the stimulator 250 into a safe current signal within the range of human neuromuscular stimulation, for example, from -100 to +100 mA.

[0067] According to yet another embodiment of the present invention, Figure 9 The image shows a pulsed (analog) neuromorphic prosthesis system 300. The pulsed neuromorphic prosthesis system 300 includes an ankle controller 310 and an insole controller 325, a pulsed central pattern generator (CPG) device 330, a stimulator 350, a pressure sensor 360, and a digital-to-analog converter (DAC) 340.

[0068] Here, the input signal is formed by weight distribution "from toes to heel" and processed by the insole controller 325. Pressure sensor data is transmitted from the insole controller 325 to the pulsed CPG device 330 via the DAC 340. The accelerated CPG device generates a walking pattern with a defined frequency and duration, corresponding to walking speed. This walking pattern is generated by a plate simulating the CPG device 330 and transmitted to a stimulator 350, which stimulates descending nerves to trigger muscles.

[0069] Example #1 (Oscillator Sequence)

[0070] summary

[0071] To reconstruct the fundamental circuit motifs of more complex neural networks in the central nervous system (CNS), we introduce a novel method for oscillator motif simulation. The microcircuits of the oscillator motif are constructed using four frequently occurring motifs in CNS: feedback inhibition, rhythmic excitation, divergence, and convergence. Simulation results reveal four basic patterns of neural activity generation: minimal connection weights between the four nuclei, maximum weights between nuclei, weights between maximum and minimum, and the involvement of inhibitory nuclei that effectively suppress output neuron activity. To test the oscillator motif by comparing the generated activities, two neural simulators, NEURON and GRAS, were used.

[0072] Introduction

[0073] We used one of the possible solutions to reconstruct the microcircuits of the basic oscillator motif (OM) that can be used in the spinal central pattern generator (CPG), and demonstrated and validated it using neural simulations from two simulators: NEURON (see Hines, M., Carnevale, N.: Neuron Simulation Environment, 2nd Edition, 2003; NEST (Jordan, J., Ippen, T., Helias, M., Kitayama, I., Sato, M., Igarashi, J., Diesmann, M., Kunkel, S.: Highly Scalable Spiking Neural Network Simulation Program: From Laptops to Very Large Scale Computers, Frontiers in Neuroinformatics, February 2018, Vol. 12) and a new C++ programming language simulator, GRAS.

[0074] Theme Review

[0075] Several articles have been published describing the spinal cord structure of microcircuits (see Ampatzis, K., Song, J., Ausborn, J., El Manira, A.: Independent microcircuit modules of different V2a interneurons and motor neurons control the speed of movement, *Neuro Journal*, 2014, Vol. 83, No. 4, pp. 934-943; Chopek, JW, Nascimento, F., Beato, M., Brownstone, RM, Zhang, Y.: Subpopulations of spinal cord V3 neurons form focal modules of stratified premotor microcircuits, *Cell Reports*, 2018, Vol. 25, No. 1, pp. 146-156.e3; Deska-Gauthier, D., Zhang, Y.: Functional diversity and motor control of spinal cord interneurons, *New Perspectives on Biology*, 2019, Vol. 8, pp. 99-108), starting from basic research describing reflex arcs (see Ilya A. Rybak, Kimberly J. Dougherty, Natalia A. Shevtsova: Organization of motor CpGs in mammals: A review of computational models and circuit structures of spinal interneurons based on genetic recognition, *eNeuro*, 2015; Markin, SN, Klishko, AN, Shevtsova, NA, Lemay, MA, Prilutsky, BI, Rybak LA: Afferent control of motor CpGs: Insights from a simple neuromechanical model: Afferent control of motor CpGs, *Annals of the New York Academy of Sciences*, 2010, Vol. 1198, No. 1, pp. 21–34; Rybak, LA, Shevtsova, NA, Lafrenier-Roula, M., McCrea, DA: Modeling of spinal circuits involved in motor pattern generation: Implications from the absence of hypothetical movement, *Journal of Physiology*, 2016, Vol. 577, No. 2, pp. 617–639) to state-of-the-art methods, including multi-layered models of rhythmic pattern generators (see, for example, van den Brand, R., Heutschi, J., Barraud, Q., DiGiovanna, J., Bartholdi, K., Huerlimann, M., Friedli, L., Vollenweider, I., Moraud, E.M., Duis, S., Dominici, N., Micera, S., Musienko, P., Courtine, G.: Restore voluntary motor control after paralytic spinal cord injury, Science, 2012, Vol. 336, No. 6085, pp. 1182-1185; Capogrosso, M.Wenger, N., Raspopovic, S., Musienko, P., Beauparlant, J., Bassi Luciani, L., Courtine, G., Micera, S.: Computational Model of Spinal Sensorimotor Circuit Based on Epidural Stimulation, Journal of Neuroscience, 2013, Vol. 33, No. 49, pp. 19326-19340; Moraud, EM, Capogrosso, M., Formento, E., Wenger, N., DiGiovanna, J., Courtine, G., Micera, S.: Neural regulatory mechanisms of spinal cord circuits correcting gait and balance deficits after spinal cord injury, Neuron, 2016, Vol. 89, No. 4, pp. 814-828; Wagner, FB, Mignardot, JB, Go-Mignardot, CGL, Demesmaeker, R., Komi, S., Capogrosso, M., Rowald, A., Señonez, I., Caban, M., Pirondini, E., Vat, M., McCracken, LA, Heimgartner, R., Fodor, I., Watrin, A., Seguin, P., Paoles, E., Keybus, KVD, Eberle, G., Schurch, B., Pralong, E., Becce, F., Prior, J., Buse, N., Buschman, R., Neufeld, E., Kuster, N., Carda, S., Zitzewitz, Jv, Delattre, V., Denison, T., Lambert, H., Minassian, K., Bloch, J., Courtine, G.: Targeted neurotechnology enables patients with spinal cord injuries to regain walking ability, *Nature*, 2018, Vol. 563, No. 7729, p. 65. Historically, the formation of motor patterns has been explained as a group of nuclei that typically exhibit mutual and periodic inhibition and / or excitatory projection and activity (see, for example, Ampatzis, K., Song, J., Ausborn, J., El Manira, A.: Independent microcircuit modules of different V2a interneurons and motor neurons control the speed of movement, *Neuro Journal*, 2014, Vol. 83, No. 4, pp. 934-943); Chopek, JW, Nascimento, F., Beato, M., Brownstone, RM., Zhang, Y.Subpopulations of spinal V3 neurons form focal modules of stratified premotor microcircuits, *Cell Reports*, 2018, Vol. 25, No. 1, pp. 146-156.e3. Meanwhile, the pattern formation and organization of microcircuits capable of generating motor patterns remain unclear (see, for example, Gad, P., Lavrov, I., Shah, P., Zhong, H., Roy, RR, Edgerton, VR, Gerasimenko, Y.: Neuromodulation of spinal motor evoked potentials in stepping, *Journal of Neurophysiology*, 2013, Vol. 110, No. 6, pp. 1311-1322). A recurring, stable motif of neuronal microcircuits (see English, DF, McKenzie). I.E.S., Evans, T., Kim, K., Yoon, E., Buzsaki, G.: Structure and dynamics of pyramidal cell-interneuron circuits in hippocampal networks, Neuron, 2017, Vol. 96, No. 2, pp. 505-520. Found in the brain and / or spinal cord (see, e.g., Chopek, JW, Nascimento, F., Beato, M., Brownstone, RM, Zhang, Y.: Focal modules of subpopulations of spinal V3 neurons forming stratified premotor microcircuits, Cell Reports, 2018, Vol. 25, No. 1, pp. 146-156.e3). In the analysis, the following criteria were used: (1) generators of mutual excitation of at least two nuclei (2) after an active period of neuronal activity inhibiting the generator, the activity should be managed by the third nucleus (3) the entire motif can learn self-regulation and self-organization through pulse time-dependent plasticity (STDP) (see, e.g., Evans, T., Kim, K., Yoon, E., Buzsaki, G.: Structure and dynamics of pyramidal cell-interneuron circuits in hippocampal networks, Neuron, 2017, Vol. 96, No. 2, pp. 505-520). denBrand, R., Heutschi, J., Barraud, Q., DiGiovanna, J., Bartholdi, K., Huerlimann, M., Friedli, L., Vollenweider, I., Moraud, E.M., Duis, S., Dominici, N., Micera, S., Musienko, P., Courtine, G.: Restoration of voluntary motor control after paralytic spinal cord injury, *Science*, 2012, Vol. 336, No. 6085, pp. 1182-1185.

[0076] method

[0077] Considering the specific circumstances of neuronal activity, a key low-level component for complex circuits is proposed, here named the oscillator motif (OM). Figure 1a). The initial design of OM was inspired by the work of Paz (Paz, JT, Huguenard, JR: Microcircuits and their interactions in epilepsy: Is the focus off? Nature Neuroscience, 2015, Vol. 18, No. 3, pp. 35-359; Womelsdorf (Womelsdorf, T., Valiante, TA, Sahin, NT, Miller, KJ, Tiesinga, P.: Dynamic circuits under rhythmic gain control, gating and integration). Preface, *Nature Neuroscience*, 2014, Vol. 17, No. 8, pp. 1031-1039) and English (English, DF, McKenzie, S., Evans, T., Kim, K., Yoon, E., Buzsaki, G.: Structure and dynamics of pyramidal cell-interneuron circuits in the hippocampal network, *Neuron*, 2017, Vol. 96, No. 2, pp. 505-520), and with repetitive or rhythmic excitation of two nuclei ( Figure 1 b) Feedback inhibition Figure 1 c) Divergence ( Figure 1 d) and convergence ( Figure 1 e) is the basis. The basic function of OM is to generate neuronal activity within a specified period of 2 to 15 milliseconds. For this purpose, rhythmic excitatory motifs are used, comprising two nuclei with mutual excitatory projections (e). Figure 1 b nuclei (1, 2) generate extended neuronal activity. For activity termination, two feedback inhibitory motifs ( Figure 1 Nuclei c (1, 3) are used in conjunction with weak excitatory and strong inhibitory projections that determine the duration of neuronal activity. Input electrical activity, as indicated by the arrow in nucleus 1, triggers mutual excitation of nuclei 1 and 2, which in turn provides output neuronal activity and weakens excitation in nucleus 3. When neurons in nucleus 3 reach their threshold, these neurons strongly inhibit the activity of nuclei 1 and 2. The balance between weak input and strong output in nucleus 3 determines the duration of output neuronal activity in OM.

[0078] result

[0079] Using the NEURON neural simulator and the Hodgkin-Huxley model (see Hodgkin, AL, Huxley, AF: Quantitative description of membrane currents and their application in neural conduction and excitation, *Journal of Physiology*, 1952, Vol. 117, No. 4, pp. 500-544), the following neuronal parameters were used: membrane capacitance of 1 μF / cm², resistivity of 100 Ω / cm, and conductances of 0:2 S / cm² and 0:04 S / cm² for sodium and potassium channels (see Dougherty, K., Kiehn, O.: firing and cellular characteristics of V2a interneurons in rodent spinal cord, *Journal of Neuroscience*: Official Journal of the Society for Neuroscience, 2010, Vol. 30, pp. 24-37). The model is multicompartmental, consisting of somatic cells, dendrites, and axons, each part presented as a compartment. The diameter of neuronal somatic cells ranges from 3 to 8 micrometers (see Chen, S., Yang, G., Zhu, Y., Liu, Z., Wang, W., Wei, J., Li, K., Wu, J., Chen, Z., Li, Y., Mu, S., OuYang, L., Lei, W.: Comparative Study of Three Interneurons in the Rat Spinal Cord, PLOS ONE, 2016, Vol. 11, No. 9, ISSN e0162969). The OM consists of two main parts: two nuclei that trigger each other to generate extended activity (…). Figure 1 b) and three nuclei that form negative feedback ( Figure 1 c) and divergence ( Figure 1 d) and convergence ( Figure 1 e). Each nucleus contains 50 neurons, with synapses ranging from 30 to 50 per projection. The balance between excitatory and inhibitory weights provides for a variety of impulse activities ( Figure 10 OM generates a series of pulses that have weak inhibitory and weak excitatory projections onto the third nucleus. Figure 10 a) When the third nucleus is in a subthreshold state, the output signal frequency of the second nucleus is approximately 120 Hz. The activity lasts for 50 milliseconds without decreasing. When inhibitory and excitatory projections are strong, neuronal activity decreases significantly because the first nucleus triggers the second nucleus, producing an output signal, and then the third nucleus simultaneously and strongly inhibits both excitatory nuclei (e.g., ...). Figure 10 (As shown in b). When the connection between the excitatory nuclei (1 and 2) and the third nucleus is weak, the number of pulses is adjusted by the difference between the excitatory and inhibitory weights. Although the inhibitory projection is strong, the output nucleus produces two average peaks ranging from 7 ms to 20 ms. The excitatory nuclei gradually increase the potential of the third nucleus. When it exceeds a threshold, the third nucleus is activated and inhibits the first and second nuclei (as shown in b). Figure 10(As shown in c). The duration of output neuron activity also depends on the external signal that triggers the third nucleus. If the inhibitory connections have sufficient weight, output activity will cease ( Figure 10 d).

[0080] The second series of experiments ( Figure 10 a2, b2, c2, d2) are specifically used to test the OM in the GRAS simulator. GRAS was created for GPU processing of biologically believable neural networks using the C++ programming language and Nvidia CUDA technology. The GRAS neural simulator uses the Hodgkin-Huxley neuron model with the following parameters: membrane capacitance of 1 μF / cm² and resistance of 100 Ω / cm. This model is a compartment.

[0081] The third series of experiments ( Figure 10 (a3, b3, c3, d3) were specifically used to test the OM simulator in the NEST neural simulator (see Jordan, J., Ippen, T., Helias, M., Kitayama, I., Sato, M., Igarashi, J., Diesmann, M., Kunkel, S.: A Highly Scalable Simulation Program for Spiking Neuronal Networks: From Laptops to Very Large Scale Computers, Frontiers in Neuroinformatics, Vol. 12, February 2018). A Hodgkin-Huxley neuronal model with the following parameters was used: membrane capacitance of 1 μF / cm², resistance of 100 Ω / cm, and default conductance for sodium and potassium channels. The model was a single-compartment model.

[0082] discuss

[0083] The weaker first mode is projected between nuclei 1, 2, and 3. Figure 10 A comparison of a1 and a2 shows that the synaptic delay used in GRAS is lower, while the pulse amplitude is larger in the NEURON simulator. The second mode, which has the highest projection weight among nuclei 1, 2, and 3, ( Figure 10 b1, b2) indicate that the pulses and refractory periods used in GRAS are short. Although similar to the NEURON simulation, nuclei 1 and 2 generate a pulse before the third nucleus strongly inhibits them.

[0084] In the third mode ( Figure 10 In c1, c2), when the weights are between the first and second modes, NEURON simulates nucleus 3 generating suppression impulse activity later than in the second mode, which is similar to GRAS.

[0085] Overall, the NEST and GRAS simulation results show high similarity. Comparison with low-weight OM activities of nuclei 1, 2, and 3 ( Figure 10a1, a2, a3) indicate that, due to the short refractory period of 50 ms throughout the simulation, NEST and GRAS have 7 higher pulse rates, while NEURON has 6. The highest weight used in the second experiment ( Figure 10 b1, b2, b3) indicates that the NEST results are close to the GRAS simulator, with the same equidistant pulses within 15ms, which is explained by the fact that the close model has less detail than the NEURON simulator. Figure 10 c1, c2, and c3 represent the simulation results of OM, with weights in the middle range between experiments 1 and 2, where short-series pulses 2 in NEURON and 3 in GRAS were observed, and the amplitude decreased during the 20 ms period in the NEST simulator. Projection to nucleus 3 increased the inhibitory effects of nuclei 1 and 2, resulting in low-amplitude subthreshold neuronal activity. Figure 10 d1, d2, d3).

[0086] in conclusion

[0087] This Example 1 demonstrates that more complex OMs than previously proposed models can serve as fundamental building blocks for complex neural circuits, such as pattern generators in mammals. The proposal to organize OMs using the four motifs of the central nervous system previously described: rhythmic excitation, feedback inhibition, divergence, and convergence has been supported by data (see [link to example 1]). Figure 10 Bioreliable simulation results using three neural simulators have been demonstrated. Simulation results show that the proposed OM can be used for networks responsible for generating neuronal activity, i.e., for simulating motion.

[0088] Example #2 (A simpler real-time model of neurons)

[0089] summary

[0090] This embodiment #2 demonstrates a simplified model of neurons for real-time processing in embedded biocompatible devices. Biocompatible neuronal models include those by Izhikevich (see Izhikevich, E.: A Simple Model of Spiking Neurons. IEEE Transactions on Neural Networks, 2003, Vol. 14, No. 6, pp. 1569-1572) and Rozenblatt (see Rosenblatt, F.: Perceptrons: A Probabilistic Model of Information Storage and Organization in the Brain. Psychological Review, 1958, Vol. 65, No. 6, p. 386), and provides a block diagram and key principles for signal processing and learning. Simplified neuronal models using other models close to the proposed Leakage Integral Trigger (LIF) and Izhikevich (IZH) models have been validated. Furthermore, the current simplified model shows better performance than the IZH and LIF models and is biocompatible from a temporal perspective.

[0091] introduction

[0092] Several previously published neuron models exist (see, for example, Hodgkin, AL, Huxley, AF: Quantitative description of membrane currents and their application in neural conduction and excitation, *Journal of Physiology*, 1952, Vol. 117, No. 4, pp. 500-544; McCulloch, WS, Pitts, W.: Logical calculus of the inner thoughts in neural activity, *Bulletin of Mathematical Biophysics*, 1943, Vol. 5, No. 4, pp. 115-133). Yale University's "Model DB" database of neuronal models now contains thousands of published models (see McDougal, RA, Morse, TM, Carnevale, T., Marenco, L., Wang, R., Migliore, M., Miller, PL, Shepherd, GM, Hines, ML: ModelDB Two Decades and Beyond: Building Fundamental Modeling Tools for the Future of Neuroscience, *Journal of Computational Neuroscience*, 2017, Vol. 42, No. 1, pp. 1-10). One of the key issues in biological simulation is computational time. This is especially important considering that the main influence on computational time is exerted by synaptic and neuron models.In one example, autonomous robots (see Lobov, S., Kazantsev, V., Makarov, VA: Spiking neurons as general building blocks of hybrid systems, *Advanced Science Letters*, 2016, Vol. 22, No. 10, pp. 2633-2637) or wearable neural prostheses (see, for example, Deska-Gauthier, D., Zhang, Y.: Functional diversity and motor control of spinal interneurons, *Modern Physiological Perspectives*, 2019, Vol. 8, pp. 99-108; Gill, ML, Grahn, PJ, Calvert, JS, Linde, MB, Lavrov, IA, Strommen, JA, Beck, LA, Sayenko, DG, Straaten, MGV, Drubach, DI, Veith, DD, Thoreson, AR, Lopez, C., Gerasimenko, YP, Edgerton, VR, Lee, KH, Zhao, KD: Neural modulation of the lumbosacral spinal cord network enables independent walking after complete paraplegia, *Nature Medicine*, 2018; Wagner, FB, Mignardot, JB, Go-Mignardot, CGL, Demesmaeker,R.,Komi,S.,Capogrosso,M.,Rowald,A.,Senez,I.,Caban,M.,Pirondini,E.,Vat,M.,McCracken,LA,Heimgartner,R. ,Fodor,I.,Watrin,A.,Seguin,P.,Paoles,E.,Keybus,KVD,Eberle,G.,Schurch,B.,Pralong,E.,Becce,F.,Prior,J.,Buse,N.,Bus (chman, R., Neufeld, E., Kuster, N., Carda, S., Zitzewitz, Jv., Delattre, V., Denison, T., Lambert, H., Minassian, K., Bloch, J., Courtine, G.: Targeted neurotechnology enables spinal cord injury patients to regain walking, Nature, 2018, Vol. 563, No. 7729, p. 65). This requires real-time processing of the topology of thousands of neurons and hundreds of thousands of synapses on a single-board computer that is finite in weight and typically does not require network connectivity.

[0093] A simplified neuron model is proposed that meets the following requirements: real-time processing on a single-board computer, taking into account biologically reliable pulse duration, refractory period, inhibitory effects, and threshold-based input pulse processing.

[0094] Model Comparison

[0095] For performance comparison, three closely related spiking neuron models were used: the LIF model, the IZH model, and the Simplified Digital Neuron (SDN) model presented below. These three models are close in performance, with the IZH and SDN models specifically designed to save computation time.

[0096] IZH is a simplification of the Hodgkin-Huxley model, incorporating biologically reliable parameters such as pulse timing and amplitude. Figure 11 a1), but the maximum pulse amplitude is used as a threshold. Thus, the pulse width can be large (e.g., 5 milliseconds) when the input synaptic weights are low. Furthermore, the IZH model can operate in different modes depending on the type of neuron.

[0097] The LIF model uses input current to charge the capacitor to update the membrane potential. This model is considered simpler than the IZH model. The parameters of the LIF model, such as pulse rate or refractory period, correspond to the parameters of biological neurons (…). Figure 11 a2). For the SDN model, the time parameter is the most important factor for real-time processing, while the amplitude is used as the unit pulse to reduce the computation time of multiplying the synaptic weights by the corresponding pulse amplitude. Therefore, in the SDN model, the biologically reliable refractory period and pulse duration of neurons are used for real-time processing. In the SDN model, the pulse shape is similar to that of the LIF model, but the refractory period curve to the resting state is not smooth ( Figure 11 a3). In the SDN model, current is not used to calculate the level (membrane potential). Instead, current is represented by the sum of projected weights, leakage, and noise to improve model performance.

[0098] SDN

[0099] A simplified neuron model capable of real-time processing of topologies involving thousands of neurons is presented, as computation time is a critical issue for biologically reliable neuron models. This simplified digital neuron (SDN) model meets the requirements of real-time processing on a single-board computer, taking into account biologically reliable pulse times, refractory periods, inhibitory effects, and threshold-based input pulse processing. Figure 12As shown, signals entering excitatory synapses increase the probability of a neuron generating an ejaculate pulse, while inputs entering inhibitory synapses decrease the pulse. During the refractory period, these inputs cannot affect the pulse level (membrane potential). In the SDN model, the following assumptions are taken into account to optimize the code for a neuron model used to process thousands of neurons and hundreds of thousands of synapses in real time:

[0100] 1. Only consider spinal cord neurons (motor neurons and interneurons);

[0101] 2. The pulse amplitude in the calculation effect has been excluded;

[0102] 3. Inhibition reduces the probability of neurons generating output impulses; and

[0103] 4. For the time parameters of the output neuron response, biologically reliable impulse and refractory periods were used.

[0104] L= ΣW±leakage+noise(1)

[0105] For real-time calculation purposes, a simplified equation (Equation 1 above) has been provided to represent the level, similar to the membrane potential, where the level is the sum of the projected weights, leakage, and noise. When the level exceeds a specified threshold, the model generates a signal represented by a unit impulse. A block diagram of the SDN is shown below. Figure 12 In the SDN, the input unit signal (pulse) is processed by the router, which transmits it to the corresponding neuron and its synapse. All synapses in the SDN store their weights, as each pulse is represented as "1", so it is not necessary to multiply the synaptic weight by the pulse before input. Somatic dynamics are maintained by leakage parameters representing the sodium / potassium pump and potassium leakage through the semi-permissive membrane. It has been found that noise is important even in this simplified model due to the randomness of all neuronal channels, even in the simple circuitry of the spinal cord. The leakage and noise parameters are set individually for each neuron. When the level reaches a threshold, a threshold function triggers an output pulse generator for a preset duration. The refractory period is implemented as negative feedback to the SOMA integrator, causing each cell to reduce the level to the current value within a set period. The synapse table stores the neuron and its synapses with axonal and synaptic delays to resolve the output pulses through the router.

[0106] To reduce RAM consumption and computation time, the C++ unsigned short integer (uint16_t) type (2 bytes) has been used for level variables because floating-point operations are slower than integer operations (e.g., see Limare, N.: Speed ​​and Precision of Integer and Floating-Point Operations (2014)). The use of unsigned short integers allows for a balance between the possible states of the level (in abstract unsigned units from 0 to 65,535) and storage in RAM. Therefore, it is possible to allocate additional SDN neurons at twice the speed of LIF and four times the speed of the IZH model (4x and 8x respectively if doubled). This is important for the single-board computer implementation of the network.

[0107] SDN models allow for low computational cost of varying parameters for individual neurons while maintaining real-time processing of computations for neurons with hundreds of thousands of synapses.

[0108] result

[0109] To compare the three models, a standalone computer with 96GB of RAM was used. The processor was an E5-2650 v2 (8 cores, 2:60GHz), and the computer featured a parallel (OpenMP) implementation of the same neural network size, with different neuron models on the CPU. During the simulation, all neuron models had constant parameters except for membrane potentials (and U_m for the IZH model). These parameters were implemented as arrays. Two simulation modes were tested (resting and pulsed, affecting neuron state). In the resting mode simulation, no neuron had any input current. Conversely, in the pulsed mode, all neurons were triggered with pulsed activity every 7.5 milliseconds (the pulse period plus the refractory period to ensure the resting potential was reached). To minimize the side effects of computational time measurements of the neuron models, synapses were not used, and pulse or potential recorders were not employed.

[0110] The simulations consist of 10 tests per neuron, with network sizes ranging from 1,000 to 20,000 neurons and a step size of 1,000. A total of 200 simulations are performed on each model.

[0111] The results show that Figure 11 In the middle, and presented as a trend line (using the polyfit function of the NumPy package). In resting mode ( Figure 11 (b) SDN and LIF models are faster than real-time with 20,000 neurons. If the number of neurons is less than 13,000, the IZH model shows a time below the real-time threshold. In spiking mode ( Figure 11 c) SDN and LIF meet real-time requirements, while IZH exceeds this metric. Figure 11In b and c, all three test models have linear time dependencies with different slopes, due to additional arithmetic operations and different numbers of variables.

[0112] in conclusion

[0113] Lightweight neuron models for real-time computation are available, for example, on limited hardware and wearable devices. Performance comparisons of the IZH and LIF models in two modes (resting and spiking) show that the SDN model performs better.

[0114] The foregoing detailed description of the embodiments serves to further clarify the features and spirit of the invention. The foregoing description of each embodiment is not intended to limit the scope of the invention. Various modifications to the foregoing embodiments and equivalent arrangements should fall within the scope of protection of this invention. Therefore, the scope of the invention should be interpreted most broadly based on the claims described below in conjunction with specific embodiments, and should cover all possible equivalent variations and arrangements.

[0115] This invention can be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium (or medium) having computer-readable program instructions thereon for causing a processor to perform various aspects of the invention.

[0116] Computer-readable storage media can be tangible means capable of retaining and storing instructions for use by an instruction execution device. Computer-readable storage media can be, but is not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital versatile optical disc (DVD), memory sticks, floppy disks, mechanical encoding devices such as punched cards or raised structures in recesses containing instructions, and any suitable combination of the foregoing. The term "computer-readable storage media" as used herein should not be construed as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through optical fibers), or electrical signals transmitted through wires.

[0117] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to their respective computing / processing devices, or downloaded via a network to an external computer or external storage device, such as the Internet, a local area network (LAN), a wide area network (WAN), and / or a wireless network. This network may include copper transmission cables, optical fibers, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the respective computing / processing device.

[0118] Computer-readable program instructions used to perform the operations of this invention may be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, or similar languages, and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or wide area network (WAN), or connected to an external computer (e.g., using an internet service provider via the internet). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may execute the computer-readable program instructions by utilizing state information from the computer-readable program instructions to personalize the electronic circuitry and thereby perform various aspects of the invention.

[0119] Various aspects of the present invention will be described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0120] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine, thereby creating means for implementing the functions / behaviors of one or more blocks specified in the flowchart and / or block diagram, as executed by the processor of the computer or other programmable data processing apparatus. These computer-readable program instructions may also be stored in a computer-readable storage medium that can instruct a computer, programmable data processing apparatus, and / or other apparatus to operate in a particular manner, such that a computer-readable storage medium having instructions stored therein includes an article of manufacture comprising instructions for implementing aspects of the functions / behaviors specified in the flowchart and / or block diagram blocks.

[0121] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device, thereby producing a computer-implemented process, such that the instructions that execute on the computer, other programmable apparatus or other device implement the functions / behaviors specified in the flowchart and / or block diagram blocks.

[0122] The flowcharts and block diagrams in the figures illustrate the structure, function, and operation of possible embodiments of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions, comprising one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions indicated in a block may not appear in the order shown in the figures. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or, depending on the functions involved, these blocks may sometimes be executed in reverse order. It will also be noted that each block in the block diagrams and / or flowchart descriptions, and combinations of blocks in the block diagrams and / or flowchart descriptions, may be implemented by a system based on special-purpose hardware that performs the specified function or behavior or executes a combination of special-purpose hardware and computer instructions.

[0123] The terminology used herein is for describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprising” and / or “including,” when used in this specification, specify the presence of the stated feature, integer, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0124] The corresponding structures, materials, behaviors, and equivalents of all means or steps plus functional elements in the following claims are intended to include any structure, material, or behavior used in combination with elements of other specific claims to perform functions. The description of the invention is for illustrative and descriptive purposes only and is not intended to be exhaustive or limited to the forms described. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the invention. The embodiments were chosen and described to best explain the principles and practical application of the invention and to enable others skilled in the art to understand the various embodiments of the invention, as well as various modifications suitable for the particular intended use.

Claims

1. A neuromorphic prosthesis system for facilitating sensorimotor function in a desired individual, the system comprising: Multiple stimulators having one or more channels, the stimulators being configured to provide electrical stimulation to muscle groups, peripheral nerves, nerve plexuses and / or the spinal cord; Multiple sensors are configured to detect and transmit continuous or periodic distributions of body weight, angular joint motion, kinetic and / or electrophysiological parameter data; Multiple controllers are configured to receive combined data from multiple sensors, process the combined data, and transmit the combined data to an fMEP-based artificial loop device. The fMEP-based artificial loop device is configured to perform topology reconstruction of a central pattern generator (CPG), comprising: Based on known input stimulus parameters and detected electrophysiological output, neuronal response patterns associated with the input stimulus are identified, wherein the neuronal response patterns are determined by identifying a pair of extreme peaks defined by a maximum and a minimum value based on an evaluation of fMEP. The neuron response pattern was analyzed, and a stable neuron response pattern was obtained by evaluating the number, amplitude, and delay of peaks. Based on the analysis of the neuronal response patterns, the neuronal microcircuit topology is reconstructed, wherein the topology includes monosynaptic layers and multisynaptic layers, and the multisynaptic layers are composed of multiple oscillator motifs (OMs), each of which includes mutually stimulating components and feedback inhibition components.

2. The system according to claim 1, characterized in that, The fMEP-based artificial loop device includes a set of synchronized single-board digital computers.

3. The system according to claim 1, characterized in that, The fMEP-based artificial circuit device includes a pulse simulation diagram of a central pattern generator that realizes a segment of spinal cord topology or an application-specific integrated circuit (ASIC) chip.

4. The system according to claim 2, further comprising: Universal Asynchronous Receiver / Transmitter (UART) serial port; A digital-to-analog converter (DAC), wherein the DAC is configured to convert voltage signals received from multiple stimulators into current signals.

5. The system according to claim 3, further comprising: A digital-to-analog converter (DAC), wherein the DAC is configured to convert voltage signals received by the plurality of stimulators into current signals.

6. The system according to claim 1, characterized in that, The plurality of sensors include pressure sensors located around the insole area of ​​the object and flexion sensors located in the ankle area of ​​the object.

7. The system according to claim 1, characterized in that, The plurality of stimulators includes a back stimulator, a left ankle stimulator, a right ankle stimulator, a right hip stimulator, and a left hip stimulator.

Images