Method and a system for classifying neural signals, and a method of selecting electrodes for direct neural control

1. A classification method for classifying neural signals, the method comprising the following steps:

a) acquiring a plurality of neural signal samples (x 1 (t)-x 16 (t)) over a determined period of time using a plurality of electrodes (EL 1 -EL 16 ); and using a data processor device:

b) estimating a covariance matrix (P i ) of said neural signals associated with the determined period of time and considered as a point of the Riemann manifold of symmetric positive definite matrices of dimensions equal to the dimensions of said covariance matrix; and

c) classifying said neural signals, the classification being performed:

either in the Riemann manifold of symmetric positive definite matrices of dimensions equal to the dimensions of said covariance matrix;

or else in a tangent space of said Riemann manifold.

2. A classification method according to claim 1 , wherein said classification is performed using a criterion based:

either on a Riemann distance defined on the Riemann manifold of symmetric positive definite matrices of dimensions equal to the dimensions of said covariance matrix; or else

on a Euclidean distance defined on a tangent space of said Riemann manifold.

3. A classification method according to claim 2 , wherein said classification criterion is a criterion of least Riemann distance between said covariance matrix (P i ) of the signals to be classified and the symmetric positive definite matrices (P 1 , P 2 ) representing centers of a plurality of classes.

4. A classification method according to claim 3 , wherein each of said symmetric positive definite matrices representing the center of a class corresponds to the geometric mean or the median in said Riemann manifold of covariance matrices of training data associated with said class.

5. A classification method according to claim 2 , wherein said classification criterion is a k nearest neighbors criterion in said Riemann manifold, between said covariance matrix of the signals that are to be classified and the covariance matrices of training data associated with the classes.

6. A classification method according to claim 1 , including, prior to said classification step c):

determining a tangent space of said Riemann manifold relative to a symmetric positive matrix corresponding to a geometric mean or a median in said Riemann manifold of a set of covariance matrices of training data associated with classes;

projecting said covariance matrix that needs to be classified together with said covariance matrices of training data into said tangent space;

applying dimensional filtering in said tangent space; and

back-projecting the filtered data into said Riemann manifold in order to perform said classification step c).

7. A classification method according to claim 1 , wherein said classification step c) is implemented in a tangent space of said Riemann manifold, determined relative to a symmetric positive matrix (P) corresponding to a geometric mean or to a median in said Riemann manifold of a set of covariance matrices of training data associated with classes.

8. A method according to claim 7 , wherein said classification step c) is implemented by Fisher linear discriminant analysis in said tangent space.

9. A classification method according to claim 7 , comprising:

prior to said classification step c):

determining a tangent space of said Riemann manifold relative to a symmetric positive matrix corresponding to a geometric mean or to a median in said Riemann manifold of a set of covariance matrices of training data associated with classes;

projecting said covariance matrix that is to be classified together with said covariance matrices of training data into said tangent space; and

applying dimensional filtering in said tangent space; and then

implementing said classification step c) on the filtered data in said tangent space.

10. A method according to claim 1 , including a step of preprocessing said neural signals by frequency filtering and centering.

11. A method according to claim 1 , wherein said neural signals are electroencephalographic signals.

12. A direct neural control method comprising:

a step of classifying neural signals in accordance with claim 1 ; and

a step of generating a signal for controlling an apparatus as a function of the result of said classification step.

13. A direct neural control method according to claim 12 , including a prior operation of selecting neural signal acquisition electrodes comprising selecting a subset of electrodes or electrode locations from a set having a greater number thereof, so as to maximize the sum of the Riemann distances between covariance matrices representative of signals of different classes coming from said electrodes.

14. A system comprising:

a set of neural signal acquisition electrodes (EL 1 -EL 16 ); and

a data processor device (PR) adapted to receive said signals and implement a method according to claim 1 .