Systems and methods for multichannel neuromuscular electrical stimulation

Abstract

Systems and methods for multichannel neuromuscular electrical stimulation in accordance with embodiments of the invention are illustrated. One embodiment includes a Multichannel Neuromuscular Electrical Stimulation (NMES) system, including multiple NMES Units, each NMES Unit including a processor, an input / output (I / O) interface capable of wireless communication, an inertial measurement unit (IMU), stimulation circuitry, and a memory storing a gait application, wherein each NMES Unit is configured by the gait application to collect IMU data using its on-board IMU, wirelessly distribute the IMU data to other NMES Units in the multiple NMES Units, identify gait events using a machine learning model, and deliver stimulation, if appropriate, in response to the identified gait events based on the position of the NMES Unit on a user's body.

Description

Systems and Methods for Multichannel Neuromuscular Electrical StimulationCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The current application claims the benefit of and priority under 35 U.S.C. § 119(e) to U.S. Provisional Patent Application No. 63 / 581 ,600 entitled “Multichannel Neuromuscular Electrical Stimulation Systems” filed September 8, 2023. The disclosure of U.S. Provisional Patent Application Ns. 63 / 581 ,600 is hereby incorporated by reference in its entirety for all purposes.FIELD OF THE INVENTION

[0002] The present invention generally relates to multichannel neuromuscular electrical stimulation, namely wearable sensors with distributed computing and stimulation capabilities.BACKGROUND

[0003] Neuromuscular electrical stimulation (NMES) involves using electrodes to deliver electrical impulses to muscles and or nerves that cause targeted muscles to contract. NMES is often used to train particular muscle functions.SUMMARY OF THE INVENTION

[0004] Systems and methods for multichannel neuromuscular electrical stimulation in accordance with embodiments of the invention are illustrated. One embodiment includes a method for Multichannel Neuromuscular Electrical Stimulation (NMES), including collecting inertial measurement unit (IMU) data independently by each of multiple NMES Units using their own on-board IMUs, wirelessly distributing the IMU data from each NMES Unit to the other NMES Units in the multiple NMES Units in near real time, identifying gait events by each individual NMES Unit using a machine learning model, and delivering stimulation, if appropriate, by each NMES Unit in response to the identified gait events and the position of the NMES Unit on a user’s body.

[0005] In a further embodiment, the wireless distribution of IMU data occurs in less than 5 milliseconds.

[0006] In still another embodiment, the machine learning model is a neural network.

[0007] In a still further embodiment, the neural network includes an input layer accepting a window of IMU data, multiple convolutional layers, a global average pooling layer, multiple dense layers, and an output layer outputting probabilities for each gait event.

[0008] In yet another embodiment, the method further includes steps for fine-tuning the machine learning model for the user.

[0009] In a yet further embodiment, the delivered stimulation is between 30V and 36V.

[0010] One embodiment includes a Multichannel Neuromuscular Electrical Stimulation (NMES) system, including multiple NMES Units, each NMES Unit including a processor, an input / output (I / O) interface capable of wireless communication, an inertial measurement unit (IMU), stimulation circuitry, and a memory storing a gait application, wherein each NMES Unit is configured by the gait application to collect IMU data using its on-board IMU, wirelessly distribute the IMU data to other NMES Units in the multiple NMES Units, identify gait events using a machine learning model, and deliver stimulation, if appropriate, in response to the identified gait events based on the position of the NMES Unit on a user’s body.

[0011] In another additional embodiment, the method further includes steps for a NMES Controller communicatively coupled to the multiple NMES Units.

[0012] In a further additional embodiment, the NMES Controller is configured to provide commands to the NMES units including starting gait tracking, stopping gait tracking, applying corrective stimulation, and starting data collection.

[0013] In another embodiment again, each NMES Unit is configured to utilize low voltage stimulation between 30V and 36V.

[0014] In a further embodiment again, the machine learning model is a neural network trained to recognize patterns in the IMU data that correspond to different events of the gait cycle.

[0015] In still yet another embodiment, the machine learning model is trained to identify a gait of the user.

[0016] One embodiment includes a NMES Unit, including a processor, an I / O interface capable of wireless communication, an IMU, stimulation circuitry, and a memory storinga gait application, wherein the gait application configures the processor to obtain local IMU data from the IMU, wirelessly transmit the local IMU data to other NMES Units in a swarm using the I / O interface, obtain multiple remote IMU data from other NMES Units in the swarm using the I / O interface, provide the local IMU data and multiple IMU data to a machine learning model, identify a gait event of a gait cycle using the machine learning model, and provide stimulation in response to the identified gait event using the stimulation circuitry.

[0017] In a still yet further embodiment, the gait application further directs the processor to communicate with an NMES Controller.

[0018] In still another additional embodiment, the NMES Controller is configured to provide commands to the NMES units including starting gait tracking, stopping gait tracking, applying corrective stimulation, and starting data collection.

[0019] In a still further additional embodiment, the stimulation circuitry is configured to utilize low voltage stimulation between 30V and 36V.

[0020] In still another embodiment again, the machine learning model is a neural network trained to recognize patterns in IMU data that correspond to different events of the gait cycle.

[0021] In a still further embodiment again, the machine learning model is trained to identify a gait of a specific user.

[0022] Additional embodiments and features are set forth in part in the description that follows, and in part will become apparent to those skilled in the art upon examination of the specification or may be learned by the practice of the invention. A further understanding of the nature and advantages of the present invention may be realized by reference to the remaining portions of the specification and the drawings, which forms a part of this disclosure.BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The description and claims will be more fully understood with reference to the following figures and data graphs, which are presented as exemplary embodiments of the invention and should not be construed as a complete recitation of the scope of the invention.

[0024] FIG. 1 illustrates a multichannel neuromuscular electrical stimulation system (NMES) in accordance with an embodiment of the invention.

[0025] FIG. 2 is a block diagram for a NMES Unit in accordance with an embodiment of the invention.

[0026] FIG. 3 is a block diagram for an NMES Controller in accordance with an embodiment of the invention.

[0027] FIG. 4 is a flow chart for a multichannel NMES process in accordance with an embodiment of the invention.DETAILED DESCRIPTION

[0028] There is a significant gap in current gait correction treatment for individuals with walking disabilities, including children and adults affected by conditions such as cerebral palsy, spina bifida, stroke, incomplete spinal cord injury, and muscular dystrophy. These conditions often lead to fatigue, limited mobility, and progressive deterioration of functional independence. A difficulty in treatment using NMES devices is that synchronized stimulation is required across a number of muscle groups. In order to reduce latency, conventionally a centralized computer is used to identify gait and deliver stimulation via electrodes connected via wires. However, this approach is cumbersome and not suitable for continuous use. When moving to a wireless communication scheme, complexity and latency become an issue. In order to achieve real-time stimulation, extremely low latency, robust communication is needed.

[0029] The present invention addresses these challenges by introducing a wearable multichannel Neuromuscular Electrical Stimulation (NMES) system designed for both exercise and gait applications, with a particular emphasis on pediatric use while maintaining applicability to adult patients. In many embodiments, the device provides biomechanical assistance during gait (neuroprosthetic improvement) for individuals with gait abnormalities due to weakness by providing stimulation to lower limb muscles during the normal muscle timing, including the gluteal, quadriceps, soleus, and tibialis anterior muscles, while the user walks. In numerous embodiments, NMES stimulation levels are adjustable based on the individual’s comfort level. Furthermore, in many embodiments, NMES Units utilize low voltage (~33V), substantially lower than current available devicesvoltages, significantly improving safety, especially in a pediatric population.

[0030] In multichannel NMES systems described herein, latency is reduced to a level that real-time wireless operation is possible. Multichannel NMES systems described herein utilize a number of wireless NMES Units attached to a patient. As opposed to a centralized processor which would need to receive all information, perform processing, and then transmit back commands, a swarm architecture where each NMES Unit performs its own processing requires only transmission of data between NMES Units. In many embodiments, specialized communications protocols enable robust, sub-5 millisecond communications between NMES Units so that data can be shared between them. This enables synchronized stimulation of appropriate muscle groups by specific NMES units in the swarm in order to correct gait.

[0031] In a variety of embodiments, NMES Units each store a machine learning model that is trained to detect gait events based on the swarm’s collected measurements. The machine learning model can be trained on specific patient data in order to continuously adapt and improve performance for said patient. In a variety of embodiments, a NMES controller is in communication with the swarm (either via a single primary NMES Unit which distributes commands, or by direct connection to each NMES Unit) which can be used to control the system. In various embodiments, the NMES Units can identify safety events such as a fall or stumble, which can in turn trigger the controller to distribute an alert. Further, while NMES Units in many embodiments are capable of providing neuromuscular stimulation, they can also be used to merely record a user’s gait without providing any corrective stimulation.

[0032] Overall, various embodiments of systems and methods described herein provide some or all of a number of key advantages, including (but not limited to): a) Improved safety, especially for pediatric use, through the utilization of low voltage (~33V) stimulation, significantly lower than current available devices. b) Enhanced adaptability to individual gait patterns via on-board machine learning models that can be fine-tuned to each user's specific needs. c) Efficient distributed processing approach, where each NMES Unit performs its own gait event detection and stimulation decision-making, reducing overall system latency.d) Real-time wireless operation enabled by specialized communication protocols that allow for robust, sub-5 millisecond communications between NMES Units. e) Versatility in application, from providing biomechanical assistance during gait for individuals with walking disabilities to merely recording a user's gait without corrective stimulation.

[0033] In a variety of embodiments, NMES Units can be produced without stimulation circuitry without departing from the scope or spirit of the invention. A description of multichannel NMES systems below is followed by a description of methods of their use.Multichannel NMES Systems

[0034] Multichannel NMES systems utilize a number of NMES Units that are each capable of independent gait detection. All of the deployed NMES Units on a given user are connected to each other via wireless communication. In many embodiments, a user wears two or more NMES Units, although applications of 8 or 10+ NMES Units are possible. A NMES controller is connected to the NMES Units and can be used to provide a user interface to the system.

[0035] Turning now to FIG. 1, a multichannel NMES system in accordance with an embodiment of the invention is illustrated. Multichannel NMES system 100 includes a number of individual NMES Units 120. NMES Units are attachable to a user. In many embodiments, NMES Units have a housing with an attachment mechanism such as, but not limited to, an adhesive or strap that is used to secure the NMES Unit to a particular muscle group on the user. In many embodiments, the muscle groups targeted include gluteal, quadriceps, soleus, and tibialis anterior muscles. However, additional muscles can be targeted as appropriate to the requirements of specific applications of embodiments of the invention. While 4 NMES Units are shown on the user in FIG. 1 , as can readily be appreciated, fewer or additional NMES Units can be used without departing from the scope or spirit of the invention.

[0036] The NMES Units 120 are communicatively coupled to NMES Controller 120. In numerous embodiments, NMES Controllers connect to a primary NMES Unit which relays commands to the rest of the NMES Units (“secondary units”). In various embodiments,the NMES Controller is configured to communicate with all of the NMES Units. In many embodiments, the NMES Controller is a computing device such as a personal computer, a smart phone, a purpose-built computing unit, or any other computing device as appropriate to the requirements of specific applications of embodiments of the invention. In numerous embodiments, the NMES Controller is communicatively coupled and / or contains a display that can be used to view data from the NMES Units. NMES Controllers can be used to start and stop Gait tracking, corrective stimulation, and / or data collection. In various embodiments, NMES Controllers enable stimulation delivered by the NMES Units to be modified, e.g. type of correction, voltage, etc.

[0037] Turning now to FIG. 2, a block diagram for NMES Unit architecture in accordance with an embodiment of the invention is illustrated. NMES Unit 200 includes a processor 210. In numerous embodiments, the processor is a microprocessor. However, any number of different computing circuits such as (but not limited to) central processing units, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), and or any other processing circuitry can be used as appropriate to the requirements of specific applications of embodiments of the invention. The NMES Unit 200 also includes an input / output (I / O) interface.

[0038] In numerous embodiments, the I / O interface is capable of wireless communication with other NMES Units and / or a NMES Controller. In various embodiments, the I / O interface is capable of transmitting and receiving wireless signals with very low latency. In many embodiments, the wireless communication protocol implemented by the I / O interface employs a channel hopping mechanism to minimize interference between NMES Units. The wireless communications protocol can also utilize time division multiple access (TDMA) modulation to prevent data collision. Clock synchronization packets can also be frequently transmitted by the I / O interface to ensure synchronization across the NMES Units. In many embodiments, the communication protocol includes priority queuing for time-sensitive data, such as (but not limited to) commands to start and stop stimulation in the transmission queue. I / O interfaces can also enable connection intervals with the above under a millisecond.

[0039] NMES Unit 200 further includes an inertial measurement unit (IMU) 230 which is used to record IMU data for tracking the movement of the NMES Unit, and a stimulationcircuitry 240 which is used to deliver neuromuscular electrical stimulation to the muscle group to which the NMES Unit is attached. The NMES Unit 200 additionally has a Memory 250. Memory can be implemented as volatile memory, non-volatile memory, or a combination thereof. Memory 250 stores a gait application 252 which can configure the processor to carry out various gait tracking and gait correction processes such as those discussed herein. In many embodiments, NMES Units further include a power source such as (but not limited to) a battery.

[0040] While NMES Unit 200 shows a particular architecture, as can be readily appreciated, additional device architectures are possible such as (but not limited to) NMES Units with no stimulation circuitry, those which include hardwired instructions not stored in a software application, and / or any other modification as appropriate to the requirements of specific applications of embodiments of the invention.

[0041] Turning now to FIG. 3, a block diagram showing NMES Controller architecture in accordance with an embodiment of the invention is illustrated. NMES Controller 300 includes a processor 310. Processors for NMES Controllers can be any number of different logic processing circuitries including (but not limited to) central processing units, graphics processing units, microprocessors, FPGAs, ASICs, and / or any other computing circuitry as appropriate to the requirements of specific applications of embodiments of the invention. The NMES Controller 300 further includes an I / O Interface capable of interfacing with I / O interfaces of NMES Units. However, the I / O interface of a NMES Controller does not have to be identical to that of a NMES Unit. The NMES Controller 300 further includes a memory 320 which can be made of volatile memory, non-volatile memory, or a combination thereof. The memory 320 includes a NMES Control Application which can configure the processor to carry out various processes as described herein.

[0042] While a particular architecture is shown in FIG. 3, any number of different modules may be added without departing from the scope or spirit of the invention. For example, in numerous embodiments, the NMES Controller is a smart phone that is programmed via an app stored in memory. Indeed, any number of different implementations using different computing hardware can be used as appropriate to the requirements of specific applications of embodiments of the invention. NMES processes that utilize NMES Units and NMES Controllers are discussed further below.Multichannel NMES

[0043] A significant advantage of a lightweight, wireless system as described above is that the user is untethered from a central computer. However, when removed from a single orchestrator, new challenges arise. Continuing to use a central computer via wireless communication is possible but introduces challenges around introduction of latency. This is because data would need to be aggregated at the central computer, processed, and then commands would need to be sent back. In contrast, systems described herein utilize NMES Units that while collectively controlled from a central controller, also act as independent agents. In this way, each NMES Unit need only collect data from other NMES Units and can generate their own “commands” internally and independently.

[0044] In many embodiments, this is achieved using low latency communication between NMES Units coupled with a machine learning model onboard each NMES Unit. In various embodiments, the neural network in each Unit is trained to recognize patterns in the IMU data that correspond to different events of the gait cycle, such as initial contact (IC) and foot off (FO). This allows the device to automatically detect when the user is walking and to trigger the appropriate stimulation pattern for that phase of the gait cycle. The neural network can be fine-tuned to adapt to the user's individual gait patterns and adjust the stimulation pattern accordingly. As each NMES Unit is aware of which muscle group it is attached to, each individual Unit can make its own determination as to when and how it should deliver stimulation to that muscle group based on the gait cycle phase. In many embodiments, the NMES Units are programmed by the NMES Controller to know which muscle group they are attached to. In some embodiments, the NMES Units are hardcoded for a specific muscle group.

[0045] In order to provide stimulation at the appropriate time in the gait cycle, processing must occur very quickly. Processing efficiency is achieved using intelligent management of computational resources. In various embodiments, by caching previously calculated values, redundant processing is avoided. Every new IMU data point does not start the inference from scratch; instead, the model builds on the previous computations. This approach significantly reduces the processing requirements, ensuring low-latency processing even on embedded low-power integrated circuits.

[0046] Turning now to FIG. 4, a process for NMES using multichannel NMES systems in accordance with an embodiment of the invention is illustrated. Process 400 includes collecting (400) IMU data. These data are collected independently by each NMES Unit using their own on-board IMUs. The IMU data from each NMES Unit is wirelessly distributed (420) to the other NMES Units in near real time (less than 5 milliseconds). Each individual NMES Unit identifies (430) gait events using a machine learning model. Each NMES Unit subsequently delivers stimulation (440), if appropriate, in response to the identified gait events.

[0047] Detection of gait events as noted above occurs using a machine learning model. In numerous embodiments, the machine learning model is a neural network. In a variety of embodiments, the neural network's architecture uses 1 D convolutional layers, which are particularly adept at capturing spatial hierarchies and patterns in sequential data like IMU readings. By keeping all intermediate results of the layers for the last X IMU data points, the processing requirements are significantly reduced for each new data point from the IMU. This efficient implementation allows for real-time processing, which is useful for accurate and timely gait event detection and stimulation delivery.

[0048] An example neural network architecture includes:

[0049] 1. Input layer: The input layer accepts a window of IMU data (e.g., 100 time steps x 6 channels for 3-axis accelerometer and 3-axis gyroscope)

[0050] 2. Convolutional layers: 5 layers with 32, 64, 128, 256, and 512 filters respectively, each with a kernel size of 3 and ReLU activation

[0051] 3. Global Average Pooling layer: This reduces spatial dimensions and number of parameters

[0052] 4. Dense layers: 2 fully connected layers with 256 and 128 neurons, usingReLU activation

[0053] 5. Output layer: Dense layer with softmax activation, outputting probabilities for each gait event (e.g., heel strike, toe off, mid-stance, mid-swing)

[0054] The neural network can be trained using a dataset of labeled IMU data collected from various gait patterns. Transfer learning techniques can be employed to fine-tune the network for individual users, enhancing its adaptability and accuracy. While a specific network architecture is enumerated above, as can be readily appreciated,layers can be removed or added to achieve similar effect without departing from the scope or spirit of the invention.

[0055] Although specifics are discussed above, many different system architectures and methods can be implemented in accordance with many different embodiments of the invention. It is therefore to be understood that the present invention may be practiced in ways other than specifically described, without departing from the scope and spirit of the present invention. Thus, embodiments of the present invention should be considered in all respects as illustrative and not restrictive. Accordingly, the scope of the invention should be determined not by the embodiments illustrated, but by the appended claims and their equivalents.

Claims

WHAT IS CLAIMED IS:

1. A method for Multichannel Neuromuscular Electrical Stimulation (NMES), comprising: collecting inertial measurement unit (IMU) data independently by each of a plurality of NMES Units using their own on-board IMUs; wirelessly distributing the IMU data from each NMES Unit to the other NMES Units in the plurality of NMES Units in near real time; identifying gait events by each individual NMES Unit using a machine learning model; and delivering stimulation, if appropriate, by each NMES Unit in response to the identified gait events and the position of the NMES Unit on a user’s body.

2. The method of claim 1 , wherein the wireless distribution of IMU data occurs in less than 5 milliseconds.

3. The method of claim 1 , wherein the machine learning model is a neural network.

4. The method of claim 3, wherein the neural network comprises: an input layer accepting a window of IMU data; multiple convolutional layers; a global average pooling layer; multiple dense layers; and an output layer outputting probabilities for each gait event.

5. The method of claim 1 , further comprising fine-tuning the machine learning model for the user.

6. The method of claim 1 , wherein the delivered stimulation is between 30V and 36V.

7. A Multichannel Neuromuscular Electrical Stimulation (NMES) system, comprising: a plurality of NMES Units, each NMES Unit comprising: a processor; an input / output (I / O) interface capable of wireless communication; an inertial measurement unit (IMU); stimulation circuitry; and a memory storing a gait application; wherein each NMES Unit is configured by the gait application to: collect IMU data using its on-board IMU; wirelessly distribute the IMU data to other NMES Units in the plurality of NMES Units; identify gait events using a machine learning model; and deliver stimulation, if appropriate, in response to the identified gait events based on the position of the NMES Unit on a user’s body.

8. The system of claim 7, further comprising a NMES Controller communicatively coupled to the plurality of NMES Units.

9. The system of claim 8, wherein the NMES Controller is configured to provide commands to the NMES units comprising: starting gait tracking, stopping gait tracking, applying corrective stimulation, and starting data collection.

10. The system of claim 7, wherein each NMES Unit is configured to utilize low voltage stimulation between 30V and 36V.

11. The system of claim 7, wherein the machine learning model is a neural network trained to recognize patterns in the IMU data that correspond to different events of the gait cycle.

12. The system of claim 7, wherein the machine learning model is trained to identify a gait of the user.

13. A Multichannel Neuromuscular Electrical Stimulation (NMES) Unit, comprising: a processor; an input / output (I / O) interface capable of wireless communication; an inertial measurement unit (IMU); stimulation circuitry; and a memory storing a gait application, wherein the gait application configures the processor to: obtain local IMU data from the IMU; wirelessly transmit the local IMU data to other NMES Units in a swarm using the I / O interface; obtain a plurality of remote IMU data from other NMES Units in the swarm using the I / O interface; provide the local IMU data and plurality of remote IMU data to a machine learning model; identify a gait event of a gait cycle using the machine learning model; and provide stimulation in response to the identified gait event using the stimulation circuitry.

14. The NMES Unit of claim 13, wherein the gait application further directs the processor to communicate with an NMES Controller.

15. The NMES Unit of claim 14, wherein the NMES Controller is configured to provide commands to the NMES units comprising: starting gait tracking, stopping gait tracking, applying corrective stimulation, and starting data collection.

16. The NMES Unit of claim 13, wherein the stimulation circuitry is configured to utilize low voltage stimulation between 30V and 36V.

17. The NMES Unit of claim 13, wherein the machine learning model is a neural network trained to recognize patterns in IMU data that correspond to different events of the gait cycle.

18. The NMES Unit of claim 13, wherein the machine learning model is trained to identify a gait of a specific user.

Images