Systems and methods for intracortical brain machine interface decoding

Patent metadata

Field	Value
Publication number	US-12393826-B2
Application number	US-202418641245-A
Country	US
Kind code	B2
Filing date	Apr 19, 2024
Priority date	Aug 14, 2014
Publication date	Aug 19, 2025
Grant date	Aug 19, 2025

Abstract

A brain machine interface (BMI) to control a device is provided. The BMI has a neural decoder, which is a neural to kinematic mapping function with neural signals as input to the neural decoder and kinematics to control the device as output of the neural decoder. The neural decoder is based on a continuous-time multiplicative recurrent neural network, which has been trained as a neural to kinematic mapping function. An advantage of the invention is the robustness of the decoder to perturbations in the neural data; its performance degrades less—or not at all in some circumstances—in comparison to the current state decoders. These perturbations make the current use of BMI in a clinical setting extremely challenging. This invention helps to ameliorate this problem. The robustness of the neural decoder does not come at the cost of some performance, in fact an improvement in performance is observed.

First claim

The invention claimed is:

1. A brain-machine interface system, comprising:
at least one multi-electrode array implanted into a user's brain; and
a neural signal decoder, the neural signal decoder implemented using a computing device in communication with the at least one multi-electrode array, where the neural signal decoder comprises a neural network trained by:
obtaining a plurality of neural signals from the user using the at least one multi-electrode array;
modifying the plurality of neural signals by randomly adding and removing neural spikes from the plurality of neural signals such that a mean number of neural spikes across the plurality of neural signals is preserved;
instantiating the neural network, where the recurrent neural network is configured to map neural signals to kinematic functions of a prosthetic device; and
training the neural network using the plurality of modified neural signals; and

where the neural signal decoder is configured to:
obtain at least one neural signal from the user using the at least one multi-electrode array;
decode an intended movement of the prosthetic device from the obtained neural signal using the neural network; and
transmit a control signal to the prosthetic device that commands the prosthetic device to perform the intended movement.

2. The brain-machine interface system of claim 1 , wherein the at least one multi-electrode array is implanted into the motor cortex of the user.

3. The brain-machine interface system of claim 1 , wherein the neural network is a recurrent neural network.

4. The brain-machine interface system of claim 3 , wherein the recurrent neural network is a multiplicative neural network.

5. The brain-machine interface system of claim 1 , wherein the neural network is retrained using a second plurality of neural signals.

6. The brain-machine interface system of claim 1 , wherein each signal in the plurality of neural signals is obtained from a different channel of the at least one multielectrode array; and wherein modifying the plurality of neural signals by randomly adding and removing neural spikes from the plurality of neural signals such that a mean number of neural spikes across the plurality of neural signals is preserved, comprises randomly adding and removing neural spikes from each neural signal in the plurality of neural signals such that a mean number of neural spikes in a given neural signal from the plurality of neural signals is preserved.

7. The brain-machine interface system of claim 1 , wherein the prosthetic device is a robotic arm.

8. The brain-machine interface system of claim 1 , wherein the prosthetic device is a virtual cursor.

9. The brain-machine interface system of claim 1 , further comprising the prosthetic device.

10. The brain-machine interface system of claim 1 , wherein the neural network is trained to decode intended prosthetic device position from neural signals; wherein the neural signal decoder comprises a second recurrent neural network trained to decode intended prosthetic device velocity from neural signals; and wherein the neural signal decoder is further configured to decode the intended movement of the prosthetic device from the obtained neural signal using the recurrent neural network to determine the intended prosthetic device position and using the second recurrent neural network to determine the intended prosthetic device velocity, wherein the intended movement is a result of blending the intended prosthetic device velocity and the intended prosthetic device position.

11. A brain-machine interfacing method, comprising:
training a neural network of a neural signal decoder by:
obtaining a plurality of neural signals from a user using at least one multi-electrode array implanted into the user's brain;
modifying the plurality of neural signals by randomly adding and removing neural spikes from the plurality of neural signals such that a mean number of neural spikes across the plurality of neural signals is preserved;
instantiating the neural network, where the neural network is configured to map neural signals to kinematic functions of a prosthetic device; and
training the neural network using the plurality of modified neural signals;

integrating the trained neural network into a brain machine interface, wherein the brain machine interface is in communication with the prosthetic device and the at least one multi-electrode array;
obtaining at least one neural signal from the user's brain using the at least one multi-electrode array;
decoding an intended movement of the prosthetic device from the obtained neural signal using the neural network; and
transmitting a control signal to the prosthetic device that commands the prosthetic device to perform the intended movement.

12. The brain-machine interfacing method of claim 11 , wherein the at least one multi-electrode array is implanted into the motor cortex of the user.

13. The brain-machine interfacing method of claim 11 , wherein the neural network is a recurrent neural network.

14. The brain-machine interfacing method of claim 13 , wherein the recurrent neural network is a multiplicative neural network.

15. The brain-machine interfacing method of claim 11 , wherein the neural network is retrained using a second plurality of neural signals.

16. The brain-machine interfacing method of claim 11 , wherein each signal in the plurality of neural signals is obtained from a different channel of the at least one multielectrode array; and wherein modifying the plurality of neural signals by randomly adding and removing neural spikes from the plurality of neural signals such that a mean number of neural spikes across the plurality of neural signals is preserved, comprises randomly adding and removing neural spikes from each neural signal in the plurality of neural signals such that a mean number of neural spikes in a given neural signal from the plurality of neural signals is preserved.

17. The brain-machine interfacing method of claim 11 , wherein the prosthetic device is a robotic arm.

18. The brain-machine interfacing method of claim 11 , wherein the prosthetic device is a virtual cursor.

19. The brain-machine interfacing method of claim 11 , further comprising the prosthetic device.

20. The brain-machine interfacing method of claim 11 , wherein the neural network is trained to decode intended prosthetic device position from neural signals; and wherein the method further comprises:
decoding the intended movement of the prosthetic device from the obtained neural signal using the neural network to determine the intended prosthetic device position; and
using a second neural network to determine intended prosthetic device velocity, wherein the intended movement is a result of blending the intended prosthetic device velocity and the intended prosthetic device position, and wherein the second neural network is trained to decode intended prosthetic device velocity from neural signals.

Assignees

• Univ Leland Stanford Junior

Inventors

• Sussillo David
• Kao Jonathan C
• Stavisky Sergey
• Shenoy Krishna V

Classifications

• G06N3/08

  Learning methods · CPC title

• G06N3/044 Primary

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• A61B5/24 Primary

  Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title