Identifying and extracting electroencephalogram signals

Patent metadata

Field	Value
Publication number	US-12478308-B2
Application number	US-201917598139-A
Country	US
Kind code	B2
Filing date	Mar 29, 2019
Priority date	Mar 29, 2019
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

Disclosed is a process for identifying and extracting pain-related electroencephalogram (EEG) signals. The process comprises receiving, from one or more trials, EEG data for each trial; determining a current density for each signal; estimating the current density for a set of neural activity regions of interest, based on the computed current density; and computing at least one spectrum characteristic for each trial based on the estimated current density. Thus mean and variance of changes in the EEG data between EEG data labeled as being indicative of a pain state and EEG data labeled as being indicative of a non-pain state, for each neural activity region of interest can be calculated, and pain-related EEG signals can be identified based on at least one a region of interest at which the variance is below a predetermined threshold.

First claim

The invention claimed is:

1 . A computer process for identifying and extracting pain-related electroencephalogram (EEG) signals, comprising:
receiving, from a plurality of trials, EEG data for each trial, the EEG data comprising one or more signals each associated with a respective coordinate vector, the EEG data being labelled as being indicative of a pain state or a non-pain state; computing a first current density for each of the one or more signals, in the respective coordinate vector; estimating a second current density for a set of neural activity regions of interest, based on the computed first current density; computing at least one spectrum characteristic for each trial based on the estimated second current density; computing, based on the at least one spectrum characteristic, a mean and a variance of changes in the EEG data between the EEG data labelled as being indicative of the pain state and the EEG data labelled as being indicative of the non-pain state, for each neural activity region of interest; and identifying pain-related EEG signals based on at least one said neural activity region of interest at which the respective variance is below a predetermined threshold.

2 . The computer process according to claim 1 , wherein each of the first current density and the second current density is a radial current density.

3 . The computer process according to claim 1 , wherein computing the mean and the variance of changes in the EEG data involves computing at least one of a between-class scatter matrix and a total scatter matrix, based on the at least one spectrum characteristic, to determine a co-variance between at least two pain-related EEG classes.

4 . The computer process according to claim 1 , wherein computing the mean and the variance of changes in the EEG data comprises computing at least one of a between-class scatter matrix and a total scatter matrix.

5 . The computer process according to claim 4 , wherein both the between-class scatter matrix and the total scatter matrix are computed, and wherein identifying pain-related EEG signals based on at least one neural activity region of interest comprises satisfying an equation:
argmin L f(L) where f(L) is a multi-variate Fisher score for the neural activity region of interest L, where f(L) is:
f ( L )=trace { S b ( S t +γ I ) −1 }

where S b is the between-class scatter matrix, S t is the total scatter matrix, γ is a positive regularisation parameter, and I is an identity matrix.

6 . The computer process according to claim 1 , wherein the neural activity regions of interest are cortex locations of a subject.

7 . The computer process according to claim 1 , wherein the neural activity regions of interest are defined by EEG coordinates on a scalp of a subject.

8 . The computer process according to claim 3 , wherein the at least two pain-related EEG classes are a pain EEG class and a pain-free EEG class.

9 . The computer process according to claim 1 , further comprising:
building a binary classification machine for receiving current density or current density activity from further EEG data and outputting a scalar indicator; receiving the further EEG data; applying the binary classification machine to the further EEG data; and classifying the further EEG data as being indicative of the pain state or the non-pain state based on the scalar indicator associated with the further EEG data.

10 . The computer process according to claim 1 , comprising:
collecting the EEG data from the plurality of trials, the plurality of trials comprising:
at least one trial performed by a first subject susceptible to experience pain; and
at least one trial performed by a second, healthy subject,
wherein the EEG data for each trial comprises the one or more signals; setting an initial coordinate vector for each signal; building a binary classification machine to discriminate between the at least two pain-related EEG states, based on EEG measurements from the at least one neural activity region of interest; receiving further EEG data; applying the binary classification machine to the further EEG data; and receiving a scalar indicator from the binary classification machine and classifying, based on the scalar indicator, the further EEG data as being representative of either a pain EEG class or a pain-free EEG class.

11 . The computer process according to claim 10 , wherein the binary classification machine is at least one of:
a support vector machine; a multi-layer neural network; and a generalised linear discriminant analyser.

12 . The computer process according to claim 10 , wherein computing the mean and the variance comprises computing at least one of a between-class scatter matrix and a total scatter matrix, based on the at least one spectrum characteristic, to determine a co-variance between the pain EEG class and the pain-free EEG class.

13 . The computer process according to claim 12 , wherein computing the between-class scatter matrix and the total scatter matrix comprises:
generating spline interpolation matrices for at least two electrical properties of the EEG data; transforming the EEG data into a current density estimate using the spline interpolation matrices; computing band-powers for the current density estimate; and calculating the between-class scatter matrix and the total scatter matrix using band-powers for the current density estimate.

14 . The computer process according to claim 13 , wherein computing the band-powers for the current density estimate comprises:
determining a current density time series for the current density estimate; computing a vector of the band-powers for the current density time series using at least one of Fourier-based decomposition and band-pass filtering followed by energy computation.

15 . The computer process according to claim 13 , wherein classifying the further EEG sample using the binary classification machine comprises:
transforming the further EEG data into a further current density estimate using the spline interpolation matrices; computing band-powers for the further current density estimate; and applying the binary classification machine to the band-powers, wherein a positive output of the binary classification machine is indicative of a first state of the pain EEG state and the pain-free EEG state of a source of the further EEG data, and wherein a negative output of the binary classification machine is indicative of a second state of a pain EEG state and a pain-free EEG state of the source that is different from the first state.

16 . The computer process according to claim 10 , further comprising receiving an initial input comprising a numerical pain level assessment, and wherein receiving the further EEG data comprises receiving successive EEG epochs from continuously recorded EEG data, the computer process further comprising incrementing or decrementing the numerical pain level assessment based on whether each successive EEG epoch is classified as being representative of the pain EEG class or the pain-free EEG class.

17 . A system for identifying and extracting pain-related EEG signals in accordance with the computer process of claim 1 , comprising:
an EEG signal source; memory; and at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the computer process according to claim 1 .

18 . The system according to claim 17 , wherein the EEG signal source is arranged to

Assignees

• Agency Science Tech & Res
• Tan Tock Seng Hospital Pte Ltd

Inventors

• Phua Kok Soon
• Zhang hai hong
• Yang su-yin
• Loy Fong Ling
• Tang Ka Yin Christina
• Ng Soon Huat
• Wang Chuan Chu
• Guan Cuntai
• Tjan Soon Yin

Classifications

• A61B5/7267

  involving training the classification device · CPC title

• A61B5/4824 Primary

  Touch or pain perception evaluation (A61B5/4821 takes precedence) · CPC title

• G16H50/30

  for calculating health indices; for individual health risk assessment · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title