Systems and methods for brain activity interpretation

Inventors

Intrator, Nathan

Assignees

Neurosteer Ltd • Neurosteer Inc

Abstract

The present invention provides a computer-implemented method, including: a. obtaining, in real-time, by a specifically programmed processor, electrical signal data representative of brain activity of a particular individual;b. processing, in real-time the electrical signal data representative of brain activity of a particular individual based upon a pre-determined predictor associated with a particular brain state, selected from a library of predictors containing a plurality of pre-determined predictors, wherein each individual pre-determined predictor is associated with a unique brain state, wherein the pre-determined predictor associated with a particular brain state includes: i. a pre-determined mother wavelet,ii. a pre-determined representative set of wavelet packet atoms, created from the pre-determined mother wavelet,iii. a pre-determined ordering of wavelet packet atoms, andiv. a pre-determined set of normalization factors, wherein the processing includes: i. causing, by the specifically programmed processor, the electrical signal data to be deconstructed into a plurality of pre-determined deconstructed wavelet packet atoms, utilizing the pre-determined representative set of wavelet packet atoms, wherein time windows of the electrical signal data are projected onto the pre-determined representative set of wavelet packet atoms wherein the projection is via convolution or inner product, andwherein each pre-determined representative wavelet packet atom corresponds to a particular pre-determined brain activity feature from a library of a plurality of pre-determined brain activity features; ii. storing the plurality of pre-determined deconstructed wavelet packet atoms in at least one computer data object;iii. causing, by the specifically programmed processor, the stored plurality of pre-determined deconstructed wavelet packet atoms to be re-ordered within the computer data object, based on utilizing a pre-determined order;iv. obtaining a statistical measure of the activity of each of the re-ordered plurality of pre-determined deconstructed wavelet packet atoms; andv. normalizing the re-ordered plurality of pre-determined wavelet packet atoms, based on utilizing a pre-determined normalization factor; and c. outputting, a visual indication of at least one personalized mental state of the particular individual, at least one personalized neurological condition of the particular individual, or both, based on the processing, wherein the individual pre-determined predictor associated with a particular brain state from within the plurality of pre-determined predictors is generated by the steps including: i. obtaining the pre-determined representative set of wavelet packet atoms by: a. obtaining from a plurality of individuals, by the specifically programmed processor, at least one plurality of electrical signal data representative of a brain activity of a particular brain state;b. selecting a mother wavelet from a plurality of mother wavelets, wherein mother wavelet is selected from an wavelet family selected from the group consisting of: Haar, Coiflet Daubehies, and Mayer wavelet families;c. causing, by the specifically programmed processor, the at least one plurality electrical signal data to be deconstructed into a plurality of wavelet packet atoms, using the selected mother wavelet;d. storing the plurality of wavelet packet atoms in at least one computer data object;e. determining, an optimal set of wavelet packet atoms using the pre-determined mother wavelet, and storing the optimal set of wavelet packet atoms in at least one computer data object, wherein the determining is via utilizing analysis Best Basis algorithm; andf. applying, by the specifically programmed processor, wavelet denoising to the number of wavelet packet atoms in the optimal set; ii. obtaining the pre-determined ordering of wavelet packet atoms by: a. projecting, by the specifically programmed processor, the at least one plurality of electrical signal data representative of a brain activity for each 4 second window of the data onto the pre-determined representative set of wavelet packet atoms;b. storing the projections in at least one computer data object;c. determining, by the specifically programmed processor, the wire length for every data point in the projection by determining the mean absolute distance of the statistical measure of the projections of different channels from their adjacent channels;d. storing the wire length data in at least one computer data object; ande. re-ordering the stored projections, by the specifically programmed computer to minimize a statistical value of the wire length value across each time window, and across all individuals within the plurality of individuals, and across the projections; and iii. obtaining the pre-determined set of normalization factors by: a. determining, by the specifically programmed computer, the mean and standard deviation of the values of the stored projections.

Core Innovation

The invention addresses detection of at least one abnormal electrical brain activity and determination of a state of pain being felt by a particular individual at a particular state of anesthesia using real-time electroencephalographic (EEG) monitoring. A first EEG monitoring device collects electrical signal data representative of brain activity from a plurality of individuals using a first set of three electrodes including a first recording electrode, a second recording electrode, and a reference electrode. From the recorded electrical signal data, the invention obtains a pre-determined ordering of a denoised optimal set of wavelet packet atoms using a selected mother wavelet and a Coifman-Wickerhauser Best Basis algorithm, and defines pre-determined normalization factors for later use.

The obtained pre-determined ordering and normalization factors are used to process real-time EEG signals from a particular individual with a second EEG monitoring device having a second set of three electrodes including a first recording electrode, a second recording electrode, and a reference electrode. The collected electrical signal data is projected, in real time, onto the pre-determined ordering of the denoised optimal set of wavelet packet atoms to obtain a particular set of projections, where a projection results from a convolution of an electrical signal in each time window and a wavelet packet atom. The particular set of projections is normalized, in real time, using the pre-determined normalization factors to form a particular set of normalized projections.

The normalized projections are processed by applying at least one machine learning algorithm in real time to determine a particular abnormal electrical brain activity and to generate an indication of the at least one abnormal electrical brain activity. In the pain determination embodiment, the machine learning algorithm determines, in real time, at least one particular normalized projection that corresponds to the state of pain being felt by the particular individual at the particular state of anesthesia, and the invention generates, in real time, an output associating the at least one particular normalized projection with the state of pain. A pre-determined minimum-path ordering is obtained based on wire lengths derived from mean or sum of absolute distance of a statistical measure of an energy of projections from adjacent projections, and/or 1 minus correlation of projections onto the wavelet packet atoms, and this ordering is used for the real-time projection stage.

Claims Coverage

The document provides three independent claims directed to detection of at least one abnormal electrical brain activity in real time and determining a state of pain being felt by a particular individual at a particular state of anesthesia. Across the independent claims, the core inventive features include pre-determined wavelet packet atom ordering derived via Best Basis and a minimum-path wire-length metric, real-time projection and normalization, and real-time machine learning-based state determination with a generated indication or output association.

Across the independent claims, the inventive coverage centers on a pre-computed denoised wavelet packet atom ordering, including Best Basis-derived and minimum-path wire-length reordered features, followed by real-time projection and normalization of EEG onto that ordering using pre-determined normalization factors, and real-time machine learning to produce either an indication of abnormal electrical brain activity or an output or indication mapping normalized projections to a state of pain at a particular state of anesthesia.

Stated Advantages

Documented Applications