Scoring method based on improved signals analysis
Disclosed is a method for scoring in real time neural signals of a subject with respect to a reference state characterized by k=1 . . . K reference covariance matrices, the method including the following steps: (i) obtaining neural signals from the subject; (ii) computing a covariance matrix of the neural signals; (iii) computing the Riemannian distances between the covariance matrix and k=1 . . . K reference covariance matrices; (iv) computing a continuous score s in real time based on at least one of the distances computed in step (iii). Also disclosed is a system and method for self-paced modulation or external modulation of neural activity of a subject.
1. A system for self-paced modulation or external modulation of neural activity of a subject comprising:
acquisition device for acquiring at least one neural signal from a subject;
computing device programmed for:
receiving the at least one neural signal from the subject acquired with said acquisition device;
filtering the at least one neural signal in at least one frequency band;
computing a covariance matrix of said at least one filtered neural signal;
computing the Riemannian distances d k between said covariance matrix and k=1 . . . K reference covariance matrices characteristic of a reference state; and
computing a real time score s, based on at least one of the Riemannian distances d k and on
at least one of Riemannian distances d r , so as to be bounded between two predetermined values, wherein said Riemannian distances d r are obtained by:
obtaining r=1 . . . R baseline covariance matrices which are characteristic of a baseline state, said baseline state being different from the reference state; and
computing the Riemannian distances d r between at least one of said baseline covariance matrices, for r=1 . . . R, and the k=1 . . . K reference covariance matrices;
at least one modulation device configured to receive the score s and provide at least one output on the base of said score s to the subject so as to modulate and modify the neural activity of the subject towards the reference state.
2. The system according to claim 1 , wherein the score s is based on a z-score function of the distances d k between the covariance matrix and the k=1 . . . K reference covariance matrices, the geometric means of the distances d r and the geometric standard deviation of the distances d r , for r=1 . . . R.
3. The system according to claim 1 , wherein the covariance matrix is a spatiofrequential covariance matrix.
4. The system according to claim 1 , wherein the Riemannian distances are estimated in the Riemannian manifold of symmetric positive definite matrices of dimensions equal to the dimensions of said covariance matrix.
5. The system according to claim 1 , wherein the k=1 . . . K reference covariance matrices are obtained by mean-shift as follows:
i. obtaining p=1 . . . P covariance matrices X p of neural signals from a database and defining the initialization of Q=P modes, such as for q=1 . . . Q, p=q and M q =X p ;
ii. defining the hyper-parameter h and the kernel window g;
iii. for each mode M q :
a. computes the P distances d(M q , X p ) between M q and each matrix X p ;
b. estimating the mean shift vector m h (M q ) as the weighted sum of tangent vectors
m
h
(
M
q
)
=
∑
p
=
1
P
g
(
d
2
(
M
q
,
X
p
)
h
2
)
Log
M
q
(
X
p
)
∑
p
=
1
P
g
(
d
2
(
M
q
,
X
p
)
h
2
)
c. then, computing M q =Exp M q (m h (M q )), with Exp M q the exponential map with links a point on the tangent space to a point on the Riemannian manifold;
d. while the mean shift vector m h (M q ) is superior to a threshold value repeat steps a. to c.;
e. if the mean shift vector is inferior to a threshold value, retain M q as a local mode;
iv. obtain K distinct local modes by fusion of modes with a distance inferior to h.
6. The system according to claim 1 , wherein the k=1 . . . K reference covariance matrices are obtained by k-means as follows:
i. defining the hyper-parameter K;
ii. obtaining p=1 . . . P covariance matrices X p of neural signals from a database and defining the initialization of the k=1 . . . K references M k (random or arbitrary attribution from the database);
iii. for each matrix X p :
a. computing the K distances d(M k , X p ) between X p and each matrix M k ;
b. attributing matrix X p to the cluster k p giving the minimum distance
k p =arg min k=1 . . . K d ( M k ,X p )
iv. updating the K reference matrices as the Riemannian means of the matrices attributed to their respective clusters:
M k =Riemannian_Average ({ X p s.t.k p =k} p=1 P )
v. repeat steps iii and iv until the K reference matrices change no more from an iteration to another.
7. A method for self-paced or external modulation of neural activity of a subject in order to reach a reference state, said method comprising continuously:
acquiring at least one neural signal from the subject with an acquisition device; and
in at least one computing device:
filtering the at least one neural signal in at least one frequency band;
computing a covariance matrix of said at least one filtered neural signal;
computing the Riemannian distances d k between said covariance matrix and k=1 . . . K reference covariance matrices characteristic of the reference state; and
computing a real time score s, based on at least one of the Riemannian distances d k and
at least one of Riemannian distances d r , so as to be bounded between two predetermined values, wherein said Riemannian distances d r are obtained by:
obtaining r=1 . . . R baseline covariance matrices which are characteristic of a baseline state, said baseline state being different from the reference state, and
computing the Riemannian distances d r between at least one of said baseline covariance matrices, for r=1 . . . R, and the k=1 . . . K reference covariance matrices;
and
receiving in a modulation device the score s and provide in real time with said modulation device at least one output on the base of said score s to the subject so as to modulate and modify the neural activity of the subject towards the reference state.
8. The method according to claim 7 , wherein said method further comprises applying external modulation to the subject in order to modulate the score s towards a targeted score defined by the reference state.
9. The method according to claim 8 wherein the external modulation is applied by deep brain stimulation, indirect brain stimulation, electroconvulsive therapy, magnetic seizure therapy, transcranial direct current stimulation, transcranial magnetic stimulation, repetitive transcranial magnetic stimulation or vagus nerve stimulation.
10. The method according to claim 7 , wherein the at least one signal filtered in at least one frequency band is concatenated.
11. The method according to claim 7 , wherein the score s is based on a z-score function of the distances d k between the covariance matrix and the k=1 . . . K reference covariance matrices, the geometric means of the distances d r and the geometric standard deviation of the distances d r , for r=1 . . . R.
12. The method according to claim 7 , wherein at least two neural signals are obtained, filtered in at least two frequency bands and concatenated.
13. The method according to claim 7 , wherein the covariance matrix is a spatiofrequential covariance matrix.
14. The method according to claim 7 , wherein the Riemannian distances are estimated in the Riemannian manifold of symmetric positive definite matrices of dimensions equal to the dimensions of said covariance matrix.
15. The method according to claim 7 , wherein the k=1 . . . K reference covariance matrices are obtained by a Riemannian clustering method from P covariance matrices of neural signals characteristics of the reference state from a database.
16. The method according to claim 7 , wherein the Riemannian clustering method is selected from mean-shift, k-means, average or principal geodesic analysis.
17. The method according to claim 7 , wherein the k=l . . . K reference covariance matrices are obtained by mean-shift as follows:
i. obtaining p=1 . . . P covariance matrices X p of neural signals from a database and defining the initialization of Q=P modes, such as for q=1 . . . Q, p=q and M q =X p ;
ii. defining the hyper-parameter h and the kernel window g;
iii. for each mode M q :
a. computes the P distances d(M q ,X p ) between M q and each matrix X p ;
b. estimating the mean shift vector m h (M q ) as the weighted sum of tangent vectors
m
h
(
M
q
)
=
∑
p
=
1
P
g
(
d
2
(
M
q
,
X
p
)
h
2
)
Log
M
q
(
X
p
)
∑
p
=
1
P
g
(
d
2
(
M
q
,
X
p
)
h
2
)
c. then, computing M q =Exp M q (m h (M q )), with Exp M q the exponential map with links a point on the tangent space to a point on the Riemannian manifold;
d. while the mean shift vector m h (M q ) is superior to a threshold value repeat steps a. to c.;
e. if the mean shift vector is inferior to a threshold value, retain M q as a local mode;
iv. obtain K distinct local modes by fusion of modes with a distance inferior to h.
18. The method according to claim 7 , wherein the k=1 . . . K reference covariance matrices are obtained by k-means as follows:
i. defining the hyper-parameter K;
ii. obtaining p=1 . . . P covariance matrices X p of neural signals from a database and defining the initialization of the k=1 . . . K references M k (random or arbitrary attribution from the database);
iii. for each matrix X p :
a. computing the K distances d(M k ,X p ) between X p and each matrix M k ;
b. attributing matrix X p to the cluster k p giving the minimum distance
k p =arg min k=1 . . . K d ( M k ,X p )
iv. updating the K reference matrices as the Riemannian means of the matrices attributed to their respective clusters:
M k =Riemannian_Average ({ X p s.t.k p =k} p=1 P )
v. repeat steps iii and iv until the K reference matrices change no more from an iteration to another.
19. The method according to claim 7 , wherein the k=l . . . K reference covariance matrices further comprises at least one subject-specific covariance matrix.
20. The method according to claim 7 , wherein the at least one neural signal is obtained by electrocorticography, electroencephalography, magnetoencephalography, magnetic resonance imaging, near-infrared spectroscopy, positron emission tomography or stereoelectroencephalography.