Signal processing for decoding intended movements from electromyographic signals

Patent metadata

Field	Value
Publication number	US-11596346-B2
Application number	US-201716322855-A
Country	US
Kind code	B2
Filing date	Aug 1, 2017
Priority date	Aug 1, 2016
Publication date	Mar 7, 2023
Grant date	Mar 7, 2023

Abstract

A technology is described for determining an intended movement from neuromuscular signals. An example method (800) includes receiving electromyography (EMG) data corresponding to single-ended channels of an electrode array (810), where EMG signals are detected by electrodes comprising the single-ended channels of the electrode array and the EMG signals are converted to the EMG data. Determining differential channel pairs for the single-ended channels of the electrode array (820) and extracting feature data from the EMG data of the differential channel pairs (830). Thereafter a feature data set is selected from the feature data of the differential channel pairs (840) and the feature data set is input to a decode model configured to correlate the feature data set to an intended movement (850). Decode output is received from the decode model indicating the intended movement (860) and the decode output is provided to a device (870).

First claim

What is claimed is:

1. A system for determining an intended movement from neuromuscular signals, comprising:
an electrode array;
at least one processor;
a memory device including instructions that, when executed by the at least one processor, cause the system to:
receive electromyography (EMG) data via single-ended channels of the electrode array,
wherein the EMG data is associated with a training movement performed by a user in response to receiving a training movement cue;
determine differential channel pairs for the single-ended channels of the electrode array associated with the EMG data by computing a number of possible different channel pairs for an n number of the single-ended channels using a matrix multiplication method or by looping through possible pairs and calculating a difference;
extract feature data from the EMG data of the differential channel pairs;
align the feature data to movement cue data of the training movement cue;
select a feature data set from the feature data as training input to a decode model configured to correlate the feature data set to an intended movement;
configure the decode model using the feature data set;
receive decode output from the decode model indicating the intended movement; and
provide the decode output to a device.

2. The system as in claim 1 , wherein the electrode array comprises a surface electrode array.

3. The system as in claim 1 , wherein the memory device includes instructions that, when executed by the processor, cause the system to further extract the feature data from the EMG data of the single-ended channels of the electrode array.

4. The system as in claim 1 , wherein the feature data is aligned to the training movement cue using at least one of: a global shift correlation technique, a global shift recursive regression technique, a trial by trial correlation technique, or a trial by trial recursive regression technique.

5. The system as in claim 1 , wherein the feature data set is selected from the feature data of differential channel pairs using a Gram-Schmidt or a recursive regression technique.

6. The system as in claim 1 , wherein the feature data set is selected from the feature data of differential channel pairs having a higher correlation with training movement data as compared to the feature data of other differential channel pairs.

7. The system as in claim 1 , wherein the feature data set is manually selected by a user.

8. The system as in claim 1 , wherein the memory device includes instructions that, when executed by the processor, cause the system to further modify decode output data set using a gain parameter and a threshold parameter to control sensitivity and stability of decode output provided by the decode model.

9. The system as in claim 1 , wherein the memory device includes instructions that, when executed by the processor, cause the system to further update decode model coefficients using a new feature data set obtained via new training movement cue data.

10. The system as in claim 1 , wherein the decode model comprises any one or more of: a Kalman filter model, a direct weighted ridge regression Kalman filter model, a Recalibrated Feedback Intention-Trained (Re-FIT) Kalman filter model, a linear regression model, a support vector regression model, a non-linear support vector regression model, or a neural network model.

11. The system as in claim 1 , wherein the EMG data is generated by a user that is an amputee, the device is a robotic prosthetic, and the decode output is employed to control the robotic prosthetic.

12. The system as in claim 1 , wherein the device is a remote robotic arm and hand used for surgical, hazardous environment, space, or underwater applications, a joystick, a game controller, or a computer input device.

13. A computer implemented method for determining an intended movement from neuromuscular signals, comprising:
receiving electromyography (EMG) data corresponding to single-ended channels of an electrode array, wherein EMG signals are detected by electrodes comprising the single-ended channels of the electrode array and the EMG signals are converted to the EMG data;
determining differential channel pairs for the single-ended channels of the electrode array by computing a number of possible different channel pairs for an n number of the single-ended channels using a matrix multiplication method or by looping through possible pairs and calculating a difference;
extracting feature data from the EMG data of the differential channel pairs;
selecting a feature data set from the feature data of the differential channel pairs;
inputting the feature data set to a decode model configured to correlate the feature data set to an intended movement;
receiving decode output from the decode model indicating the intended movement; and
providing the decode output to a device.

14. The method as in claim 13 , wherein selecting the feature data set from the feature data of the differential channel pairs further comprises calculating EMG amplitudes for the differential channel pairs using the EMG data and selecting the feature data set based in part on the EMG amplitudes.

15. The method as in claim 13 , wherein inputting the feature data set to the decode model further comprises inputting regression coefficient parameters and degrees of freedom (DOF) parameters.

16. The method as in claim 15 , further comprising updating the coefficient parameters with each cycle of input to the decode model.

17. The method as in claim 13 , further comprising:
receiving DOF gains parameters and DOF threshold parameters from a user; and
applying the DOF gains parameters and DOF threshold parameters to the decode output.

18. The method as in claim 13 , further comprising receiving a movement mode parameter from a user, wherein the movement mode parameter specifies: a position control mode comprising output for a neutral baseline position associated with non-activity, a latching mode comprising velocity control, and a leaky mode comprising a combination of the position control mode and the latching mode.

19. The method as in claim 13 , wherein providing the decode output to the device further comprises providing the decode output to a prosthesis device.

20. The method as in claim 13 , wherein providing the decode output to the device further comprises providing the decode output to a control device.

21. A non-transitory machine readable storage medium having instructions embodied thereon, the instructions when executed by a processor:
receive electromyography (EMG) data corresponding to single-ended channels of an electrode array;
determine differential channel pairs for the single-ended channels of the electrode array by computing a number of possible different channel pairs for an n number of the single-ended channels using a matrix multiplication method or by looping through each possible pair and calculating a difference;
calculate EMG amplitudes for the differential channel pairs and the single-ended channels of the electrode array using the EMG data;
select a feature data set from the EMG data of the differential channel pairs and the single-ended channels of the electrode array based in part on the EMG amplitudes;
input the feature data set to a decode model configured to estimate an intended movement based on the feature data set;
receive decode output from the decode model indicating the intended movement; and
provide the decode output to a device.

22. The non-transitor

Assignees

• Univ Utah Res Found

Inventors

• Wendelken Suzanne
• Davis Tyler
• Page David
• Kluger David
• Warren David
• Clark Gregory
• Duncan Christopher
• Nieveen Jacob

Classifications

• A61B5/6824

  Arm or wrist · CPC title

• A61B5/7246

  using correlation, e.g. template matching or determination of similarity · CPC title

• A61B2505/09

  Rehabilitation or training · CPC title

• A61B2562/0252

  Load cells · CPC title

• A61B5/389 Primary

  Electromyography [EMG] · CPC title