Methods for closed-loop neural-machine interface systems for the control of wearable exoskeletons and prosthetic devices
Brain-Machine Interface (BMI) systems or movement-assist systems may be utilized to aid users with paraplegia or tetraplegia in ambulation or other movement or in rehabilitation of motor function after brain injury or neurological disease, such as stroke, Parkinson's disease or cerebral palsy. The BMI may translate one or more neural signals into a movement type, a discrete movement or gesture or a series of movements, performed by an actuator. System and methods of decoding a locomotion-impaired and/or an upper-arm impaired subject's intent with the BMI may utilize non-invasive methods to provide the subject the ability to make the desired motion using an actuator or command a virtual avatar.
1. A method for decoding user intent from brain activity, the method comprising:
obtaining brain activity data with an electroencephalography (EEG);
transmitting the brain activity data from the EEG to a brain-machine interface (BMI); and
using the BMI:
arranging the brain activity data into a feature matrix, wherein the feature matrix provides a collection of feature vectors over time, and a feature vector represents an amplitude modulation of the brain activity data while a user conducts a specific type of motion with a robotic exoskeleton;
calculating probabilities of a current feature vector from the collection feature vectors belonging to a class selected from a plurality of classes, wherein the selected class represents a desired movement or posture by a user;
identifying a class label in accordance with the probabilities calculated, wherein the class label identified is associated with a maximum probability;
utilizing dimensionality reduction to reduce dimensionality of the brain activity data in the feature matrix;
identifying movement-related cortical potentials (MRCP) channels;
finding negative peak values for a spatially averaged MRCP signal computed from the MRCP channels;
extracting a ‘Go’ window starting at a negative peak of the spatially averaged MRCP signal occurring within a set period of time from voluntary movement onset, wherein the ‘Go’ window traverses into a past a predetermined amount of time from the negative peak;
mapping states of the robotic exoskeleton to the feature matrix; and
transmitting commands to the robotic exoskeleton, wherein the commands correspond to the mapped states.
2. The method of claim 1 , further comprising filtering the brain activity data received by the BMI to obtain delta band brain activity data.
3. The method of claim 2 , further comprising standardizing the delta band brain activity data.
4. The method of claim 1 , wherein the dimensionality reduction is performed utilizing a Local Fisher's Discriminant Analysis (LFDA).
5. The method of claim 4 , wherein the dimensionality reduction is performed in real-time.
6. The method of claim 1 , wherein the mapping of the feature matrix is performed utilizing a Gaussian Mixture Model (GMM).
7. The method of claim 1 , wherein a specified window size of the brain activity data is received by the BMI.
8. The method of claim 1 , wherein separate channels of the brain activity data are used to create the feature matrix.
9. The method of claim 1 , further comprising converting the class label identified into an associated motion of the robotic exoskeleton.
10. A neural interface system comprising:
a robotic exoskeleton;
an electroencephalography (EEG) for obtaining brain activity data;
a brain-machine interface (BMI) coupled to the robotic exoskeleton and EEG, wherein the BMI arranges the brain activity data into a feature matrix, the feature matrix provides a collection of feature vectors over time, and a feature vector represents an amplitude modulation of the brain activity data while a user conducts a specific type of motion with the robotic exoskeleton,
wherein further the BMI calculates probabilities of a current feature vector from the collection of feature vectors belonging to a class selected from a plurality of classes, wherein the selected class represents a desired movement or posture by a user, and the BMI identifies a class label in accordance with the probabilities calculated, wherein the class label identified is associated with a maximum probability, the BMI reduces dimensionality of the brain activity data in the feature matrix, and wherein
wherein further the BMI identifies movement-related cortical potentials (MRCP) channels, finds negative peak values for a spatially averaged MRCP signal computed from the MRCP channels, extracts a ‘Go’ window starting at a negative peak of the spatially averaged MRCP signal occurring within a set period of time from voluntary movement onset, and the ‘Go’ window traverses into a past a predetermined amount of time from the negative peak, and
the BMI maps states of a robotic exoskeleton to the feature matrix, and the BMI transmits commands corresponding to the mapped states to the robotic exoskeleton.
11. The system of claim 10 , further comprising a filter for filtering the brain activity data received by the BMI to obtain delta band brain activity data.
12. The system of claim 11 , wherein the BMI standardizes the delta band brain activity data.
13. The system of claim 10 , wherein the dimensionality reduction is performed utilizing a Local Fisher's Discriminant Analysis (LFDA).
14. The system of claim 13 , wherein the dimensionality reduction is performed in real-time.
15. The system of claim 10 , wherein the feature matrix is mapped utilizing a Gaussian Mixture Model (GMM).
16. The system of claim 10 , wherein a specified window size of the brain activity data is provided to the BMI from the EEG.
17. The system of claim 10 , wherein separate channels of the brain activity data are used to create the feature matrix.
18. The system of claim 10 , wherein the BMI converts the class label identified into an associated motion of the robotic exoskeleton.