Method and device for determining a subject's sleep state by processing an electroencephalographic signal

• US 5,154,180 A
• Filed: 01/24/1991
• Issued: 10/13/1992
• Est. Priority Date: 01/24/1990
• Status: Expired due to Fees

• Current Assignee: Orange S.A.
• Original Assignee: Orange S.A.
• Inventors: Blanchet, Gerard, Prado, Jacques
• Primary Examiner(s): Kamm, William E.
• Application Number:
  US07/645,561
• Time in Patent Office: 628 Days
• Field of Search: 128/731, 364/413.05
• US Class Current: 600/544
• CPC Class Codes: A61B 5/374 Detecting the frequency distribution of signals, e.g. detecting delta, theta, alpha, beta or gamma waves

Abstract

An electroencephalographic signal processing method allows the stages of sleep to be determined in real time as an EEG signal is being picked up. The EEG signal is divided into time segments of the same length; each signal segment is sampled digitally to obtain successive blocks of N samples; for each block i, M standardized correlation coefficients R_i,k are determined, M being much far smaller than N; for each block i, the distances D(i,j) between the coefficients R_i,k of the current block and the coefficients R_j,k of the centers of gravity of existing classes j are calculated; the minimum D(i,p) of distances D(i,j) is selected; block i is assigned to class p and the center of gravity of the class is updated if D(i,p) is below a given threshold, while a new class is created if D(i,p) is above the given threshold; and each class is assigned to the closest stage from the standpoint of energy distribution beween frequency bands.

Claims

1. Device for processing an encephalographic signal and for determining sleep stages during a time period, comprising:

(a) means for acquiring an EEG signal;

(b) a sampler and an analog-digital converter for digitally sampling said EEG signal during a plurality of successive EEG signal time segments of a predetermined length and delivering said successive blocks of N samples each corresponding to one of the EEG signal time segments;

(c) a preprocessing microprocessor programmed to calculate M standardized correlation coefficients R_i,k for each of said blocks of said EEG signal, M being an integer smaller than N, and

(d) a computer for storing said signal blocks and correlation coefficients, and,

for the first block i=1, creating a first EEG signal distribution class defined by the correlation coefficients R₁,k of said first block, and

for each later block i;

computing the coefficients R_j,k of the centers of gravity of already existing EEG signal distribution class or classes j, calculating distances D(i,j) between the coefficients R_i,k of a current block and the coefficients R_j,k of the centers of gravity of existing EEG signal distribution classes j, selecting a minimum D(i,p) of the distances D(i,j) and assigning block i to EEG signal distribution class p and updating the center of gravity of the EEG signal distribution class if D(i,p) is less than a given threshold, or creating a new EEG signal distribution class if D(i,p) is greater than said given threshold to automatically classify sleep stages based on energy distribution of said EEG signal distribution classes.

2. Device according to claim 1, wherein said acquiring means includes a filter which has a passband of from 1 to 40 Hz and said sampler has a sampling frequency of 100 Hz.

3. Device according to claim 1, wherein said acquiring means includes a bandpass filter having a predetermined bandpass frequency range and said sampler has a sampling frequency of at least twice an upper limit of the bandpass frequency range.

4. Method for processing an electroencephalographic signal for determining sleep stages during a signal pickup period, comprising the steps of:

(a) detecting an EEG signal during said pickup period;

(b) dividing the EEG signal into EEG signal time segments having a same predetermining length;

(c) digitally sampling each EEG signal time segment to obtain successive blocks i of N samples of said EEG signal time segments, N being an integer greater than 1;

(d) for each block i, determining M standardized correlation coefficients R_i,k, with M being an integer smaller than N;

(e) for a first block i=1, storing self-correlation coefficients R₁,k as representative of a first EEG signal distribution class and as being centers of gravity for the first class;

(f) for each block i, calculating distances D(i,j) between the coefficients R_i,k of a current block and coefficients R_j,k of centers of gravity of existing EEG signal distribution classes, said coefficients R_j,k being only self-correlation coefficients of the first class for a second block;

(g) selecting a minimum D(i,p) of the distances D(i,j);

(h) assigning block i to an EEG signal distribution class p and updating the center of gravity of class p if D(i,p) is less than a given threshold, or creating a new EEG signal distribution class if D(i,p) is greater than the given threshold; and

(i) assigning each EEG signal distribution class to a closest sleep stage based on energy distribution between frequency bands to automatically classify sleep stages based on characteristics of the EEG signal.

5. Method according to claim 4, wherein step (i) includes the steps of:

generating an autoregressive parametric model for each EEG signal distribution class;

deriving a frequency spectrum from said model,

computing, for a plurality of frequency bands significant for assignment to a sleep stage, the percentage B_m of the total energy of the signal contained in the band;

computing, for each EEG signal distribution class m, coefficients c_s in a number equal to that of the previously defined sleep stages; and

assigning each of the classes obtained by automatic frequency analysis to a sleep stage most likely to be representative of the EEG signal distribution class.

6. Method according to claim 5, wherein said frequency bands consist of three bands of 1 to 4 Hz, 4 to 7 Hz, and 12 to 14 Hz.