CN112957014A - Pain detection and positioning method and system based on brain waves and neural network - Google Patents

Pain detection and positioning method and system based on brain waves and neural network Download PDF

Info

Publication number
CN112957014A
CN112957014A CN202110167532.4A CN202110167532A CN112957014A CN 112957014 A CN112957014 A CN 112957014A CN 202110167532 A CN202110167532 A CN 202110167532A CN 112957014 A CN112957014 A CN 112957014A
Authority
CN
China
Prior art keywords
pain
neural network
lstm
cnn
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110167532.4A
Other languages
Chinese (zh)
Other versions
CN112957014B (en
Inventor
伍冯洁
麦伟健
唐一晟
刘庆焜
向宇涵
罗文俊
郭子芊
刘根生
钟键
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou University
Original Assignee
Guangzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangzhou University filed Critical Guangzhou University
Priority to CN202110167532.4A priority Critical patent/CN112957014B/en
Publication of CN112957014A publication Critical patent/CN112957014A/en
Application granted granted Critical
Publication of CN112957014B publication Critical patent/CN112957014B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/48—Other medical applications
    • A61B5/4824—Touch or pain perception evaluation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Signal Processing (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Psychiatry (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physiology (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Hospice & Palliative Care (AREA)
  • Pain & Pain Management (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a pain detection and positioning method and system based on brain waves and a neural network, wherein the method comprises the following steps: removing noise related to an original electroencephalogram signal through independent component analysis, carrying out pain grade division on the electroencephalogram signal, utilizing signal time division window processing, and establishing a pain grade data set after integrating pain grade data; respectively generating spectral topographic maps of time windows of Theta, Alpha and Beta frequency bands related to pain through Fourier transform, azimuth equidistant projection and a Cloughtocher interpolation algorithm, and combining the spectral topographic maps into a multichannel electroencephalogram sequence; obtaining a time-space feature vector of brain wave sequences related to the pain degree and the pain position through a CNN-LSTM-AM neural network; the brain wave pain features learned by the CNN-LSTM-AM neural network are input into a pain classifier model to assess pain level and pain location. The invention can accurately and efficiently extract and process the pain degree change and position change characteristics of the brain waves and automatically identify the pain grade and the pain position.

Description

Pain detection and positioning method and system based on brain waves and neural network
Technical Field
The invention relates to the technical field of pain detection, in particular to a pain detection positioning method and system based on brain waves and a neural network.
Background
Recent studies have shown that pain is accompanied by significant physiological changes, mainly in autonomic nervous system response parameters such as brain waves, heart rate, skin level, etc. Measuring pain levels using electrophysiological signals associated with pain is an effective way to achieve objective assessment of pain. Among pain-related electrophysiological signals, electroencephalograms (EEGs) are an ideal physiological index for objectively evaluating pain, have a high time resolution of the order of milliseconds and a relatively low data acquisition cost, directly reflect electrical activities of neurons, and acquire a large amount of information reflecting human physiology, psychology and diseases by performing feature extraction, pattern recognition and the like on the EEGs.
At present, the standard method for analyzing electroencephalogram data is to aggregate spectral measurement data of all electrodes into a feature vector. However, this approach obviously ignores the inherent structure of the data in space, frequency and time.
Disclosure of Invention
In order to overcome the defects and shortcomings of the prior art, the invention provides a pain detection and positioning method and system based on brain waves and a neural network, which can realize automatic identification of four pain levels of painless, light pain, medium pain and heavy pain, position the pain, accurately and efficiently extract and process the brain wave change characteristics during pain, and automatically evaluate and feed back the pain level and the pain position.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a pain detection and positioning method based on brain waves and a neural network, which comprises the following steps:
removing original EEG signal noise by adopting an independent component analysis algorithm, carrying out EEG signal pain grade segmentation, segmenting each pain grade into a plurality of equal-length time windows, obtaining a multi-channel EEG time sequence, and obtaining a preprocessed pain data set;
respectively generating a spectral topographic map of each time window of Theta, Alpha and Beta frequency bands related to pain through Fourier transform, azimuth equidistant projection and a Cloughtocher interpolation algorithm, and combining the spectral topographic maps into a multichannel brain wave sequence which is used as the input of a CNN-LSTM-AM neural network;
constructing a CNN-LSTM-AM neural network, training the CNN-LSTM-AM neural network, and obtaining time-space feature vectors of brain wave (EEG) sequences related to pain degree and pain position through the CNN-LSTM-AM neural network;
and constructing a Softmax pain classifier model, inputting the brain wave pain characteristics learned by the CNN-LSTM-AM neural network into the pain classifier model, and matching and identifying the pain level and the pain position.
As a preferred technical scheme, the method for removing the noise of the original electroencephalogram signal by adopting the independent component analysis algorithm comprises the following specific steps:
eliminating power frequency interference by adopting a 50Hz wave trap;
filtering EEG signal high frequency component by band-pass filter;
the artifact signal mixed with the EEG signal is processed using an independent component analysis algorithm.
As a preferred technical solution, the brain wave sequences combined into multiple channels specifically include:
calculating the power spectral density of each electrode Theta, Alpha and Beta frequency band of the electroencephalogram signals through FFT;
generating a spectral topographic map of each time window of the corresponding frequency band through azimuth equidistant projection and Clough-Tocher interpolation;
and combining the spectral topographic map sequence into a three-channel electroencephalogram image sequence as the input of the CNN-LSTM-AM network.
As a preferred technical scheme, the power spectral density of each electrode Theta, Alpha and Beta frequency band of the electroencephalogram signal is calculated through FFT, and the specific calculation formula is as follows:
Figure BDA0002937932560000031
Figure BDA0002937932560000032
Figure BDA0002937932560000033
wherein S isxx(ω) represents the requested power spectral density, T represents the signal interval,
Figure BDA0002937932560000034
represents [0, T]Fourier transform of the brain electrical signal x (t) within the interval,
Figure BDA0002937932560000035
representing the expected value of the squared amplitude spectral density.
As a preferred technical scheme, the CNN-LSTM-AM neural network construction method specifically comprises the following steps:
constructing a convolutional neural network layer: ConvNet is arranged on the convolutional neural network layer, the ConvNet is adopted to extract spatial features in an electroencephalogram sequence, all sub-networks of the ConvNet share parameters among frames, and the output is reshaped into sequence frames for researching time features in an electroencephalogram topographic map;
constructing a long-short term memory network layer: the long-short term memory network layer captures the time evolution of the ConvNet output sequence by using a BilSTM;
constructing an attention layer: the Attention layer uses an Attention Model to simulate the Attention Model of the human brain for applying different Attention weights to different EEG pain variation characteristics in pain level identification and different EEG position variation characteristics in pain localization.
As a preferred technical solution, the specific steps of constructing the Softmax pain classifier model include:
given m training samples:
{(x(1),y(1)),(x(2),y(2)),…,(x(m),y(m))}
setting a hypothesis function h:
Figure BDA0002937932560000036
wherein x is(i)Representing input features, y(i)Representing a labeled sample, theta representing an introduced hypothesis parameter, and p () representing a probability computation function;
the loss function was constructed as:
Figure BDA0002937932560000041
wherein λ represents a random constant, and J (θ) represents a loss function;
and substituting the derivative of the loss function into a gradient descent algorithm to obtain the pain multi-classification model.
The invention also provides a pain detection and positioning system based on brain waves and a neural network, which comprises: the system comprises a data preprocessing module, a brain wave sequence building module, a CNN-LSTM-AM neural network training module, a Softmax pain classifier model building module and a matching and identifying module;
the data preprocessing module is used for preprocessing the original electroencephalogram signals, removing noise of the original electroencephalogram signals by adopting an independent component analysis algorithm, carrying out electroencephalogram signal pain grade segmentation, segmenting each pain grade into a plurality of equal-length time windows, obtaining a multi-channel electroencephalogram time sequence and obtaining a preprocessed pain data set;
the brain wave sequence building module is used for building a multichannel brain wave sequence, generating a spectrum topographic map of each time window of Theta, Alpha and Beta frequency bands related to pain through Fourier transform, azimuth equidistant projection and a Clough interpolation algorithm, and combining the spectrum topographic maps into the multichannel brain wave sequence which is used as the input of a CNN-LSTM-AM neural network;
the CNN-LSTM-AM neural network construction module is used for constructing a CNN-LSTM-AM neural network;
the CNN-LSTM-AM neural network training module is used for training a CNN-LSTM-AM neural network, and obtaining time-space feature vectors of brain wave (EEG) sequences related to pain degree and pain position through the CNN-LSTM-AM neural network;
the Softmax pain classifier model building module is used for building a Softmax pain classifier model;
the matching and identifying module is used for inputting the brain wave pain characteristics learned by the CNN-LSTM-AM neural network into the pain classifier model, and matching and identifying the pain level and the pain position.
As a preferred technical scheme, the data preprocessing module is provided with a power frequency interference elimination unit, a high-frequency component filtering unit and an artifact signal processing unit;
the power frequency interference elimination unit is used for eliminating power frequency interference by adopting a 50Hz wave trap;
the high-frequency component filtering unit is used for filtering the EEG signal high-frequency component by adopting a band-pass filter;
the artifact signal processing unit is used for processing an artifact signal mixed with the EEG signal by adopting an independent component analysis algorithm.
As a preferred technical solution, the brain wave sequence construction module includes a power spectral density calculation unit, a spectral topographic map generation unit and a synthesis unit;
the power spectral density calculating unit is used for calculating the power spectral density of each electrode Theta, Alpha and Beta frequency band of the electroencephalogram signals through FFT;
the spectral topographic map generating unit is used for generating a spectral topographic map of each time window of the corresponding frequency band through azimuth equidistant projection and Clough-Tocher interpolation;
the synthesis unit is used for combining the spectral topographic map sequence into a three-channel electroencephalogram image sequence as the input of the CNN-LSTM-AM network.
As a preferred technical solution, the CNN-LSTM-AM neural network includes a convolutional neural network layer, a long-short term memory network layer, and an attention layer;
the convolutional neural network layer is provided with a ConvNet architecture, the ConvNet architecture comprises an input layer, a hidden layer and an output layer, the hidden layer comprises a convolutional layer, a pooling layer and a full-connection layer, and the full-connection layer converts the characteristic diagram into a characteristic vector as output;
the long-short term memory network layer is in a repetitive module chain form of a neural network, and each module is a memory block and comprises a forgetting gate, an input gate, an output gate and a memory unit;
the attention layer adopts a neural network attention model to simulate an attention model of a human brain and is used for adding different attention weights to different EEG pain change characteristics in pain grade identification and different EEG position change characteristics in pain positioning.
Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) the standard electroencephalogram data analysis method for aggregating the spectral measurement data of all the electrodes into the feature vectors in the prior art ignores the inherent structures of the data in space, frequency domain and time domain, converts the electroencephalogram activity into a series of multi-channel spectral images, reserves the topological structure of EEG data, is beneficial to extraction of pain features, reserves more useful information than brain waves in the prior art, and thus obviously improves pain identification efficiency and pain positioning accuracy.
(2) The invention constructs a depth recursive convolutional network (CNN-LSTM-AM) giving attention to learn robust features of pain data in space, frequency domain and time domain from an image sequence, so that the accuracy of pain identification is obviously improved.
(3) The invention is very extensive and can be used not only for pain classification and localization tasks, but also for any EEG-based classification task, showing the potential advantages of the invention.
Drawings
Fig. 1 is a flowchart illustrating a pain detection and localization method based on brain waves and neural networks according to embodiment 1;
FIG. 2 is a graph showing the effects of ICA use in this example 1;
FIG. 3 is a schematic diagram of signal segmentation based on pain level in this example 1;
fig. 4 is a flow chart of electroencephalogram topographic map generation in the present embodiment 1;
fig. 5 is a schematic network structure diagram of the ConvNet model in this embodiment 1;
FIG. 6 is a ConvNet parameter chart of the present example 1;
FIG. 7 is a diagram of an LSTM memory cell model of the present embodiment 1;
FIG. 8 is a detailed diagram of BiLSTM + Attention in the present embodiment 1;
FIG. 9 is a general structural diagram of the CNN-LSTM-AM-Conv according to the present embodiment 1;
FIG. 10 is a diagram showing the results of pain classification and localization in the present example 1;
fig. 11 is a diagram comparing the method proposed in this embodiment 1 with other methods.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Example 1
As shown in fig. 1, the present embodiment provides a pain detection and localization method based on brain waves and neural networks, which includes the following steps:
s1: removing noise related to an original electroencephalogram (EEG) through an Independent Component Analysis (ICA), carrying out pain grade segmentation on the EEG, processing recorded electroencephalogram data by utilizing a signal time segmentation window, and establishing a pain grade data set;
in the embodiment, the brain wave pain level data set acquisition method specifically comprises the following steps:
s11: as shown in fig. 2, a filter and an Independent Component Analysis (ICA) algorithm are used to remove noises such as electro-oculogram, electrocardiogram and myoelectricity in the original electroencephalogram signal, so as to obtain electroencephalogram data with high signal-to-noise ratio;
in this embodiment, filtering and Independent Component Analysis (ICA) algorithm is used to filter EEG noise, and the specific steps include:
eliminating power frequency interference by adopting a 50Hz wave trap;
filtering out high-frequency components of the EEG signal by adopting a band-pass filter (0-50 Hz);
processing the artifact signal mixed with the EEG signal by an Independent Component Analysis (ICA) algorithm, specifically comprising the following steps:
in EEG analysis, the rows of the input matrix X are EEG signals recorded on different electrodes, and the columns are measurements recorded at different points in time. ICA finds a "decomposition" matrix W that linearly decomposes the multi-channel scalp into sums of temporally independent and spatially fixed components. Row U of the output data matrix is WX, which is the time course of the activation of the ICA component. The column inv (w) of the inverse matrix gives the relative projected intensity of the individual components on each scalp sensor. These scalp weights give the scalp topography of each ingredient and provide evidence for the physiological origin of these ingredients, specifically:
eye movements should be projected mainly to the frontal lobe site with low-pass time course;
blinking movements should be projected to the frontal lobe and have a large punctate activation;
temporal muscle activity should be projected to the temporal position with spectral peaks above 20 Hz.
S12: as shown in fig. 3, the electroencephalogram data is divided into four pain levels;
s13: dividing each pain grade into a plurality of equal-length time windows to obtain a multichannel electroencephalogram time sequence to obtain preprocessed pain data;
s14: establishing classifiers, wherein each classifier divides the obtained pain data set into a special training set, a verification set and a test set by a leave-subject-out cross-validation method;
s2: as shown in fig. 4, spectral topograms of time windows of Theta, Alpha and Beta bands related to pain are generated by fourier transform (FFT), Azimuthal Equidistant Projection (AEP) and CloughTocher scheme interpolation algorithm, respectively, and then merged into 3-channel brain wave (EEG) sequences as input to CNN;
in the embodiment, Theta, Alpha and Beta are selected as main frequency bands for pain feature extraction, and a large amount of experimental evidence shows that when pain occurs, the power of the Alpha and Beta frequency bands is increased, and the power of the Theta frequency band is decreased, so that the Theta frequency band is an important physiological feature for realizing pain intensity detection by EEG;
in this embodiment, the specific construction steps of the brain wave (EEG) sequence of 3 channels include:
s21: the Power Spectral Density (PSD) of each electrode Theta (4-8 Hz), Alpha (8-13 Hz) and Beta (13-40 Hz) frequency band of the electroencephalogram signals is calculated through FFT, and the specific calculation formula is as follows:
Figure BDA0002937932560000081
wherein T represents a signal interval, Sxx(ω) represents the requested power spectral density,
Figure BDA0002937932560000091
represents [0, T]Fourier transform of the electroencephalogram signal x (t) in the interval;
Figure BDA0002937932560000092
Figure BDA0002937932560000093
represents the expected value of the squared amplitude spectral density:
Figure BDA0002937932560000094
pair of Parceval's theorem
Figure BDA0002937932560000095
Is developed in the formula
Figure BDA0002937932560000096
Is xTAnd (t) inverting the x axis by the conjugate complex number, and performing the subsequent steps of only using Fourier transform to deform the formula.
Figure BDA0002937932560000097
Is x*(t) an inverse Fourier transform of the (t),
Figure BDA0002937932560000098
is the Fourier transform of x (t').
S22: generating a spectral topographic map (32x32) of each time window of the corresponding frequency band by Azimuth Equidistant Projection (AEP) and interpolation of a Clough-Tocher scheme;
in the embodiment, the Azimuthal Equidistant Projection (AEP) uses a mapping application for reference, and a projection plane of the AEP is generally tangent to the earth at two poles, equator or any middle point, so that the distance from the projection center to any other point is kept unchanged, and therefore, an activity map generated by all electrodes distributed in a scalp three-dimensional space is converted into a two-dimensional electroencephalogram image;
in the embodiment, Bicubic Interpolation based on the Clough-Tocher scheme interpolates the measured values of the scattering power of the electrodes in the two-dimensional electroencephalogram image, and estimates the values between the electrodes on a 32x32 grid to form an electroencephalogram topographic map;
s23: combining the spectral topographic map sequence into a three-channel electroencephalogram image sequence (EEG images) as input of a CNN-LSTM-AM network;
s3: obtaining a time-space feature vector of brain wave (EEG) sequence related to pain degree and pain position through a CNN-LSTM-AM neural network;
in this embodiment, the specific construction steps of the CNN-LSTM-AM neural network include:
s31: constructing a convolutional neural network Layer (CNN Layer):
the CNN Layer is composed of ConvNet, ConvNet is used for extracting spatial features in an electroencephalogram sequence, a ConvNet architecture with the best performance is adopted for each frame, all sub-networks of ConvNet share parameters among the frames in order to reduce the number of parameters in the network, and the output is reshaped into a sequence frame for researching time features in an electroencephalogram spectral topographic map;
as shown in fig. 5, each ConvNet is composed of three major parts, i.e., an input layer, a hidden layer and an output layer, and is a convolutional neural network similar to the architecture of a VGG network, which uses stacked convolutional layers with small receiving fields, so that the network has a highly scalable architecture, and the superposition of multiple convolutional layers results in an effective high-dimensional receiving field, and requires much fewer parameters.
As shown in fig. 6, the specific network architecture and parameters of ConvNet are:
(1) an input layer:
the input layer of ConvNet can process multidimensional data, specifically, the input of the network is an RGB three-channel EEG picture with a picture dimension of 32x 3.
(2) Hidden layer:
the hidden layers of convNet comprise convolutional layers, pooling layers and fully-connected layers, wherein the convolutional layers and the pooling layers are specific to the convolutional neural network, convolutional cores in the convolutional layers comprise weight coefficients, and the pooling layers do not comprise weight coefficients.
In this embodiment, the convolutional layer functions to perform feature extraction on input data, and includes a plurality of convolutional kernels therein, and each element constituting a convolutional kernel corresponds to a weight coefficient and a bias vector (bias vector), and is similar to a neuron of a feedforward neural network. Each neuron in the convolution layer is connected to a plurality of neurons in a closely located region in the previous layer, the size of the region being dependent on the size of the convolution kernel, i.e., the receptive field. When the convolution kernel works, the convolution kernel regularly sweeps the input characteristic, matrix element multiplication summation is carried out on the input characteristic in a receptive field, deviation amount is superposed, meanwhile, an excitation function is used for assisting in expressing the complex characteristic, and the specific calculation formula is as follows:
Figure BDA0002937932560000111
the summation in the formula is equivalent to solving a cross-correlation (cross-correlation), b is the deviation, zlAnd zl+1Represents the convolutional input and output of the l-th layer, also called feature map, ωl+1Is a weight matrix of layer L +1, Ll+1Is zl+1The feature pattern length and width are assumed to be the same. Z (i, j) corresponds to the pixels of the feature map, K is the number of channels of the feature map,
Figure BDA0002937932560000112
is the kth channel l +
Weight matrix of 1 layer, f, s0And p is a convolutional layer parameter, corresponding to the convolutional kernel size, convolutional step length (stride), and number of padding (padding) layers;
in this embodiment, the convolutional layer parameters include convolutional kernel size, step size and padding, which together determine the size of the convolutional layer output feature map, and are hyper-parameters of the convolutional neural network.
In this embodiment, the convolution kernel size can be specified as an arbitrary value smaller than the input image size, and the larger the convolution kernel is, the more complicated the extractable input feature is;
in this embodiment, a convolution step (stride) defines a distance between positions of convolution kernels adjacent to and sweeping through the feature map twice, when the convolution step is 1, the convolution kernels sweep through elements of the feature map one by one, and when the step is n, n-1 pixels are skipped in the next scanning;
in this embodiment, padding (padding) is a method of artificially increasing the size of the feature map before it passes through the convolution kernel to offset the effect of size shrinkage in the computation, where the Same padding (Same padding) keeps the feature maps of the output and input the Same size by doing enough padding.
In this embodiment, the stimulus function is behind the convolutional layer to help express complex features, which is expressed as follows:
Figure BDA0002937932560000121
wherein
Figure BDA0002937932560000122
Is the output of the convolutional layer after passing through the excitation function as
Figure BDA0002937932560000123
Convolutional neural networks typically use a Linear rectification function (RecU), which is calculated as:
f(x)＝max(0,x)
in a neural network, linear rectification is used as an activation function of a neuron, and the neuron is defined in linear transformation wTThe nonlinear output result after x + b, i.e. for the input vector x from the neural network of the upper layer entering the neuron, the neuron using the linear rectification activation function will output:
max(0,wTx+b)
where w is a weight matrix, wTIs the transposition of w;
output to the next layer of neurons or as an entire neural network;
in this embodiment, after the feature extraction is performed on the convolutional layer, the pooling layer performs feature selection and information filtering on the output feature map, which includes a preset pooling function, and functions to replace the result of a single point in the feature map with the feature map statistics of its neighboring area. The step of selecting a pooling area by the pooling layer is the same as the step of scanning the characteristic diagram by the convolution kernel, and the pooling size, the step length and the filling are controlled;
in this embodiment, the pooling scheme L is commonly usedPPooling (Lp Pooling) is a type of Pooling model that is inspired by the hierarchical structure within the visual cortex and is generally establishedThe representation form is:
Figure BDA0002937932560000124
step length s in the formula0Pixel (i, j) has the same meaning as the convolution layer, p is a pre-specified parameter,
Figure BDA0002937932560000125
is the output of the convolutional layer after excitation; lp pooling is averaged over the pooling area when p is 1, called mean-pooling (Average-Pool); when p → ∞, Lp pooling takes maxima within a region, called Max-Pool (Max-Pool); both mean pooling and maximum pooling preserve the background and texture information of the image at the expense of losing part of the information or size of the feature map;
in this embodiment, the fully-connected layer is located at the last part of the hidden layer of the convolutional neural network, and only signals are transmitted to other fully-connected layers. The characteristic diagram loses a space topological structure in a full connection layer, is expanded into a characteristic vector and is used as the input of the next stage through an excitation function;
in this embodiment, specific parameters of the ConvNet hidden layer are as follows:
convolution layers all use convolution kernels of 3 x3 size, 1 pixel step, Same padding and ReLU excitation functions;
a plurality of convolutional layers are stacked together, followed by a Max-Pool layer (Max-Pool), which is performed over a 2 × 2 window, with a step size of 2 pixels;
for layers located in deeper stacks, the number of cores in each convolutional Layer increases by a factor of two, where Layer1 is 32 cores, Layer2 is 64 cores, and Layer3 is 128 cores.
The feature map is converted into a feature vector as output using a full connectivity layer (FC-512) of 512 nodes.
(3) Output layer
The output layer is an LSTM network at the next stage, and the output of ConvNet of each time frame is reshaped into sequence frames for researching the time characteristics in an EEG sequence.
S32: constructing long and short term memory network Layer (LSTM Layer)
In this embodiment, LSTM Layer employs a cell number of 128 of BiLSTM to capture the temporal evolution of the ConvNets output sequence, since brain activity is a time-dynamic process, and changes between frames may contain additional information about the underlying mental state;
in the embodiment, the BilSTM is a bidirectional LSTM model, so that the time characteristic learning capability of the LSTM is further enhanced, because the current characteristics are not only related to the past time, but also related to the future characteristics, and the recognition accuracy can be greatly improved by considering the past and future information;
as shown in fig. 7, LSTM is a special Recurrent Neural Network (RNN) which can effectively solve the long-term dependence problem of EEG sequence, and the network has a repetitive module chain form of a neural network, each module is a memory block, and mainly includes three gates (forgetting gate, input gate and output gate) and a memory cell (cell), and the core of the memory cell is a cell state (cell state);
each memory block of the LSTM controls the state of a cell through three gates, and information can be added or deleted to the cell, and the specific steps are as follows:
in this embodiment, the first step is to determine which information needs to be discarded from the cell state, and this part of the operation is handled by a sigmoid cell called forget gate. It looks through ht-1And xtThe information is used to output a vector between 0 and 1, where 0 to 1 value in the vector represents the cell state Ct-1Which information is retained or how much is discarded. 0 means no reservation and 1 means both reservations, and the calculation formula is as follows:
ft＝σ(Wf·[ht-1,xt]+bf)
in this embodiment, the next step is to decide which new information to add to the cell state. This step is again divided into two steps, first of all, by means of ht-1And xtThe decision as to which information to update is made by an operation called the input gate. Then use ht-1And xtObtaining new candidate unit information through a tanh layer
Figure BDA0002937932560000141
This information may be updated into the unit information. These two steps are described below:
it＝σ(Wi·[ht-1,xt]+bf)
Figure BDA0002937932560000142
in the present embodiment, the old cell information C will be updated as followst-1Becomes new unit information Ct. The updated rule is that a part of the old cell information is forgotten through forgetting gate selection, and candidate cell information is added through input gate selection
Figure BDA0002937932560000143
Part of which obtains new cell information Ct. The update operation is as follows:
Figure BDA0002937932560000144
in this embodiment, h is input after the cell state is updatedt-1And xtThe state characteristics of the output cell are judged, the input needs to pass through a sigmoid layer called an output gate to obtain a judgment condition, then the cell state passes through a tanh layer to obtain a vector with a value between-1 and 1, and the vector is multiplied by the judgment condition obtained by the output gate to obtain the final output of the cell. The method comprises the following steps:
ot＝σ(Wo[ht-1,xt]+bo)
ht＝ot*tanh(Ct)
s33: build Attention Layer (Attention Layer)
As shown in fig. 8, the Attention Layer uses the Attention Model (AM) to simulate the Attention Model of the human brain, and adds different Attention weights to different EEG pain variation characteristics during pain level identification and different EEG position variation characteristics during pain localization, so as to further improve the accuracy of pain identification and pain localization.
The specific calculation formula is as follows:
this AM input is the EEG pain perception characteristic information learned by BilSTM:
Figure BDA0002937932560000151
H＝[h1,h2,…,hT]
wherein is hiThe output vector is a combination of forward and backward outputs of the BilSTM, H is a matrix formed by each output vector, and T is the number of time frames;
the next processing utilizes the AM structure to process the input information:
M＝tan(h)
α＝softmax(wTM)
r＝HαT
wherein
Figure BDA0002937932560000152
dwIs a vector dimension, w is a training parameter vector, wTIs the transposed matrix of w; the dimensions of w, alpha and r are respectively dw,T,dw；
The final output is:
h*＝tanh(r)
s4: as shown in fig. 8 and 9, a Softmax pain classifier model is constructed, and brain wave (EEG) pain features learned by the CNN-LSTM-AM neural network are input into the pain classifier model, so that automatic matching and recognition of pain levels and pain positions are realized.
In this embodiment, the specific steps of constructing the Softmax pain classifier model include:
the Softmax regression is an expansion of a Logistic regression algorithm in multiple classifications, and the problem that the Logistic regression can only be applied to two classifications is solved. When the pain category label y > 2, given m training samples:
{(x(1),y(1)),(x(2),y(2)),…,(x(m),y(m))}
for the Softmax regression algorithm, the input features are
Figure BDA0002937932560000161
(dimension n +1) and labeled sample y(i)E {0,1, … k }; setting a hypothesis function h:
Figure BDA0002937932560000162
where, theta denotes the assumed parameter introduced,
Figure BDA0002937932560000163
the classification task is converted into a calculation of probability:
Figure BDA0002937932560000164
in implementing Softmax regression, a matrix of k × (n +1) is used to represent θ:
Figure BDA0002937932560000165
defining a loss function J to judge the advantages and disadvantages of the classifier and minimizing an optimized loss function, wherein the loss function of Softmax regression is as follows:
Figure BDA0002937932560000166
and (3) solving by Softmax regression:
after the definition of the loss function is obtained, a gradient descent method is used for iterating an optimization algorithm to minimize the loss function J (theta); by derivation, one can obtain:
Figure BDA0002937932560000171
wherein
Figure BDA0002937932560000172
Is a vector whose first element
Figure BDA0002937932560000173
Is J (theta) is thetajThe partial derivative of the l component of (a);
after the derived loss function is obtained, it is brought into a gradient descent algorithm to minimize J (θ), resulting in a pain classifier model.
Introducing a weight attenuation term into the loss function to solve the problem that Softmax is easy to generate multiple solutions, and adding the weight attenuation term
Figure BDA0002937932560000174
Modify the corresponding loss function:
Figure BDA0002937932560000175
wherein λ is a random constant in the range of 0 to 1;
after the attenuation term is introduced, the loss function J (θ) becomes a strict convex function, so that a unique optimal solution can be ensured, and the derivative of the new loss function can be written as:
Figure BDA0002937932560000176
similarly, the derivative of the new loss function is substituted into the gradient descent algorithm to minimize J (θ), resulting in an efficient pain multi-classification model.
In this embodiment, the accuracy obtained by each classifier is averaged to obtain the accuracy of the pain multi-classification model;
in this embodiment, the pain multi-classification model includes 4 pain grade labels, which are respectively pain-free (0), pain-slight (1), pain-moderate (2) and pain-severe (3), and if the pain grade is detected to be 1-3, the pain position is continuously detected through the model, and the pain position is divided into four body parts, namely a left hand, a right hand, a left foot and a right foot, so as to verify the effectiveness of pain positioning;
in this embodiment, the accuracy of classifying the pain recognition model is improved by adjusting the hyper-parameters of the network, and the specific steps are as follows:
training by optimizing a cross entropy loss function;
and (3) training a recursive convolution network by adopting an Adam algorithm with a learning factor of 10-3, wherein the first-order moment attenuation rate and the second-order moment attenuation rate are respectively 0.9 and 0.999. Batch set to 20, where Adam proved to have a fast convergence rate when training ConvNets and multi-layer neural networks;
dropout with a probability of 0.5 is used in all fully connected layers, where Dropout regularization has proven to be an effective method to reduce overfitting in deep neural networks and neuroimaging applications with millions of parameters;
early stopping is used by monitoring the performance of the model on a randomly selected validation set, which is also an effective way to prevent overfitting of the model.
In this embodiment, the model demonstrates the feasibility and efficiency of the algorithm through the detection of four pain levels and four pain locations, and is easily extended to any other EEG-based classification task.
In order to prove the high efficiency and the innovation of the invention, the method is used for evaluating the pain detection effect, and realizing the automatic identification of four pain grades of painless, slight pain, medium pain and heavy pain and the pain positioning of the left hand, the right hand, the left foot and the right foot by combining with the graph shown in FIG. 10; due to the electroencephalogram difference among individuals, the accuracy of each classifier changes, but the average accuracy of pain classification of the model is 92%, and the average accuracy of pain localization is 95%; as shown in fig. 11, compared with the pain detection algorithm published before, the method obtains electroencephalogram pain information of three frequency bands Theta, Alpha and Beta, and then realizes classification of four pain levels and pain positions through a deep neural network model, and compared with the existing basic algorithms such as Support Vector Machine (SVM), Naive Bayes (Naive Bayes), K-nearest neighbor (KNN), Random Forest (RF), and the like, it can be seen that the completely new neural network algorithm model used by the method greatly improves the accuracy of pain level evaluation, and the first-attempted four pain localizations also have higher accuracy.
Example 2
The embodiment also provides a pain detection and localization system based on brain waves and a neural network, which includes: the system comprises a data preprocessing module, a brain wave sequence building module, a CNN-LSTM-AM neural network training module, a Softmax pain classifier model building module and a matching and identifying module;
in the embodiment, the data preprocessing module is used for preprocessing the original electroencephalogram signals, removing noise of the original electroencephalogram signals by adopting an independent component analysis algorithm, carrying out electroencephalogram pain grade segmentation, segmenting each pain grade into a plurality of equal-length time windows, obtaining a multi-channel electroencephalogram time sequence, and obtaining a preprocessed pain data set;
in this embodiment, the brain wave sequence building module is configured to build a multichannel brain wave sequence, generate spectral topographic maps of time windows of Theta, Alpha and Beta bands related to pain through fourier transform, azimuthal equidistant projection and a clougthocher interpolation algorithm, respectively, and merge the spectral topographic maps into the multichannel brain wave sequence, which is used as an input of the CNN-LSTM-AM neural network;
in this embodiment, the CNN-LSTM-AM neural network building module is used to build a CNN-LSTM-AM neural network;
in this embodiment, the CNN-LSTM-AM neural network training module is configured to train a CNN-LSTM-AM neural network, and obtain a time-space feature vector of a brain wave (EEG) sequence related to a pain degree and a pain location through the CNN-LSTM-AM neural network;
in this embodiment, the Softmax pain classifier model building module is used to build a Softmax pain classifier model;
in this embodiment, the matching and recognition module is used to input the brain wave pain features learned by the CNN-LSTM-AM neural network into the pain classifier model, and match and recognize the pain level and pain location.
In this embodiment, the data preprocessing module is provided with a power frequency interference elimination unit, a high-frequency component filtering unit and an artifact signal processing unit;
in this embodiment, the power frequency interference elimination unit is configured to eliminate power frequency interference by using a 50Hz notch filter; the high-frequency component filtering unit is used for filtering the EEG signal high-frequency component by adopting a band-pass filter; the artifact signal processing unit is used for processing an artifact signal mixed with the EEG signal by adopting an independent component analysis algorithm.
In this embodiment, the brain wave sequence construction module includes a power spectral density calculation unit, a spectral topographic map generation unit, and a synthesis unit;
in the embodiment, the power spectral density calculating unit is used for calculating the power spectral density of each electrode Theta, Alpha and Beta frequency band of the electroencephalogram signal through FFT; the spectral topographic map generating unit is used for generating a spectral topographic map of each time window of the corresponding frequency band through azimuth equidistant projection and Clough-Tocher interpolation; the synthesis unit is used for combining the spectral topographic map sequence into a three-channel electroencephalogram image sequence as the input of the CNN-LSTM-AM network.
In this embodiment, the CNN-LSTM-AM neural network includes a convolutional neural network layer, a long-short term memory network layer, and an attention layer;
in this embodiment, the convolutional neural network layer is provided with a ConvNets architecture, the ConvNets architecture includes an input layer, a hidden layer and an output layer, the hidden layer includes a convolutional layer, a pooling layer and a full-connection layer, and the full-connection layer converts the characteristic diagram into a characteristic vector as an output;
in this embodiment, the long-short term memory network layer has a repetitive module chain form of a neural network, and each module is a memory block and comprises a forgetting gate, an input gate, an output gate and a memory unit;
in this embodiment, the attention layer uses a neural network attention model to simulate the attention model of the human brain for applying different attention weights to different EEG pain variation characteristics in pain level recognition and different EEG position variation characteristics in pain localization.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (10)

1. A pain detection and positioning method based on brain waves and a neural network is characterized by comprising the following steps:
removing original EEG signal noise by adopting an independent component analysis algorithm, carrying out EEG signal pain grade segmentation, segmenting each pain grade into a plurality of equal-length time windows, obtaining a multi-channel EEG time sequence, and obtaining a preprocessed pain data set;
respectively generating a spectral topographic map of each time window of Theta, Alpha and Beta frequency bands related to pain through Fourier transform, azimuth equidistant projection and a Cloughtocher interpolation algorithm, and combining the spectral topographic maps into a multichannel brain wave sequence which is used as the input of a CNN-LSTM-AM neural network;
constructing a CNN-LSTM-AM neural network, training the CNN-LSTM-AM neural network, and obtaining time-space feature vectors of brain wave (EEG) sequences related to pain degree and pain position through the CNN-LSTM-AM neural network;
and constructing a Softmax pain classifier model, inputting the brain wave pain characteristics learned by the CNN-LSTM-AM neural network into the pain classifier model, and matching and identifying the pain level and the pain position.
2. The pain detection and localization method based on brain waves and neural networks as claimed in claim 1, wherein the method for removing noise of original brain electrical signals by using independent component analysis algorithm comprises the following steps:
eliminating power frequency interference by adopting a 50Hz wave trap;
filtering EEG signal high frequency component by band-pass filter;
the artifact signal mixed with the EEG signal is processed using an independent component analysis algorithm.
3. The method for detecting and locating pain based on brain waves and neural networks according to claim 1, wherein the merging into a multichannel brain wave sequence comprises the following specific steps:
calculating the power spectral density of each electrode Theta, Alpha and Beta frequency band of the electroencephalogram signals through FFT;
generating a spectral topographic map of each time window of the corresponding frequency band through azimuth equidistant projection and Clough-Tocher interpolation;
and combining the spectral topographic map sequence into a three-channel electroencephalogram image sequence as the input of the CNN-LSTM-AM network.
4. The method for detecting and locating pain based on brain waves and neural network of claim 3, wherein the power spectral density of each electrode Theta, Alpha and Beta frequency band of the brain electrical signals is calculated by FFT, and the specific calculation formula is:
Figure FDA0002937932550000021
Figure FDA0002937932550000022
Figure FDA0002937932550000023
wherein S isxx(ω) represents the requested power spectral density, T represents the signal interval,
Figure FDA0002937932550000024
represents [0, T]Fourier transform of the brain electrical signal x (t) within the interval,
Figure FDA0002937932550000025
representing the expected value of the squared amplitude spectral density.
5. The method for detecting and locating pain based on brain waves and neural network of claim 1, wherein the step of constructing the CNN-LSTM-AM neural network comprises the following steps:
constructing a convolutional neural network layer: ConvNet is arranged on the convolutional neural network layer, the ConvNet is adopted to extract spatial features in an electroencephalogram sequence, all sub-networks of the ConvNet share parameters among frames, and the output is reshaped into sequence frames for researching time features in an electroencephalogram topographic map;
constructing a long-short term memory network layer: the long-short term memory network layer captures the time evolution of the ConvNet output sequence by using a BilSTM;
constructing an attention layer: the Attention layer uses an Attention Model to simulate the Attention Model of the human brain for applying different Attention weights to different EEG pain variation characteristics in pain level identification and different EEG position variation characteristics in pain localization.
6. The brain wave and neural network-based pain detection and localization method according to claim 1, wherein the specific step of constructing the Softmax pain classifier model comprises:
given m training samples:
{(x(1),y(1)),(x(2),y(2)),…,(x(m),y(m))}
setting a hypothesis function h:
Figure FDA0002937932550000031
wherein x is(i)Representing input features, y(i)Representing a labeled sample, theta representing an introduced hypothesis parameter, and p () representing a probability computation function;
the loss function was constructed as:
Figure FDA0002937932550000032
wherein λ represents a random constant, and J (θ) represents a loss function;
and substituting the derivative of the loss function into a gradient descent algorithm to obtain the pain multi-classification model.
7. A pain detection and localization system based on brain waves and a neural network is characterized by comprising: the system comprises a data preprocessing module, a brain wave sequence building module, a CNN-LSTM-AM neural network training module, a Softmax pain classifier model building module and a matching and identifying module;
the data preprocessing module is used for preprocessing the original electroencephalogram signals, removing noise of the original electroencephalogram signals by adopting an independent component analysis algorithm, carrying out electroencephalogram signal pain grade segmentation, segmenting each pain grade into a plurality of equal-length time windows, obtaining a multi-channel electroencephalogram time sequence and obtaining a preprocessed pain data set;
the brain wave sequence building module is used for building a multichannel brain wave sequence, generating a spectrum topographic map of each time window of Theta, Alpha and Beta frequency bands related to pain through Fourier transform, azimuth equidistant projection and a Clough interpolation algorithm, and combining the spectrum topographic maps into the multichannel brain wave sequence which is used as the input of a CNN-LSTM-AM neural network;
the CNN-LSTM-AM neural network construction module is used for constructing a CNN-LSTM-AM neural network;
the CNN-LSTM-AM neural network training module is used for training a CNN-LSTM-AM neural network, and obtaining time-space feature vectors of brain wave (EEG) sequences related to pain degree and pain position through the CNN-LSTM-AM neural network;
the Softmax pain classifier model building module is used for building a Softmax pain classifier model;
the matching and identifying module is used for inputting the brain wave pain characteristics learned by the CNN-LSTM-AM neural network into the pain classifier model, and matching and identifying the pain level and the pain position.
8. The pain detection and localization system based on brain waves and neural network of claim 7, wherein the data preprocessing module is provided with a power frequency interference elimination unit, a high frequency component filtering unit and an artifact signal processing unit;
the power frequency interference elimination unit is used for eliminating power frequency interference by adopting a 50Hz wave trap;
the high-frequency component filtering unit is used for filtering the EEG signal high-frequency component by adopting a band-pass filter;
the artifact signal processing unit is used for processing an artifact signal mixed with the EEG signal by adopting an independent component analysis algorithm.
9. The brain wave and neural network based pain detection and localization system according to claim 7, wherein the brain wave sequence construction module includes a power spectral density calculation unit, a spectral topographic map generation unit, and a synthesis unit;
the power spectral density calculating unit is used for calculating the power spectral density of each electrode Theta, Alpha and Beta frequency band of the electroencephalogram signals through FFT;
the spectral topographic map generating unit is used for generating a spectral topographic map of each time window of the corresponding frequency band through azimuth equidistant projection and Clough-Tocher interpolation;
the synthesis unit is used for combining the spectral topographic map sequence into a three-channel electroencephalogram image sequence as the input of the CNN-LSTM-AM network.
10. The brain wave and neural network based pain detection and localization system according to claim 7, wherein the CNN-LSTM-AM neural network comprises a convolutional neural network layer, a long-short term memory network layer and an attention layer;
the convolutional neural network layer is provided with a ConvNet architecture, the ConvNet architecture comprises an input layer, a hidden layer and an output layer, the hidden layer comprises a convolutional layer, a pooling layer and a full-connection layer, and the full-connection layer converts the characteristic diagram into a characteristic vector as output;
the long-short term memory network layer is in a repetitive module chain form of a neural network, and each module is a memory block and comprises a forgetting gate, an input gate, an output gate and a memory unit;
the attention layer adopts a neural network attention model to simulate an attention model of a human brain and is used for adding different attention weights to different EEG pain change characteristics in pain grade identification and different EEG position change characteristics in pain positioning.
CN202110167532.4A 2021-02-07 2021-02-07 Pain detection and positioning method and system based on brain waves and neural network Active CN112957014B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110167532.4A CN112957014B (en) 2021-02-07 2021-02-07 Pain detection and positioning method and system based on brain waves and neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110167532.4A CN112957014B (en) 2021-02-07 2021-02-07 Pain detection and positioning method and system based on brain waves and neural network

Publications (2)

Publication Number Publication Date
CN112957014A true CN112957014A (en) 2021-06-15
CN112957014B CN112957014B (en) 2023-06-20

Family

ID=76275108

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110167532.4A Active CN112957014B (en) 2021-02-07 2021-02-07 Pain detection and positioning method and system based on brain waves and neural network

Country Status (1)

Country Link
CN (1) CN112957014B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113598794A (en) * 2021-08-12 2021-11-05 中南民族大学 Training method and system for detection model of ice drug addict
CN113723206A (en) * 2021-08-04 2021-11-30 三峡大学 Brain wave identification method based on quantum neural network algorithm
CN113812933A (en) * 2021-09-18 2021-12-21 重庆大学 Acute myocardial infarction real-time early warning system based on wearable equipment
CN115990108A (en) * 2022-11-01 2023-04-21 山东海天智能工程有限公司 Acupuncture control system based on brain electrical signals, readable storage medium and electronic equipment
CN116831595A (en) * 2023-06-21 2023-10-03 中国人民解放军西部战区总医院 Pain grade assessment method
CN117075741A (en) * 2023-10-17 2023-11-17 首都医科大学附属北京天坛医院 Consciousness interaction communication method and system
CN117894477A (en) * 2024-03-13 2024-04-16 西安工会医院 Construction method of predictive assessment model based on postherpetic neuralgia

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103405225A (en) * 2013-08-28 2013-11-27 深圳先进技术研究院 Method, apparatus and device for obtaining pain feeling evaluation indexes
US20140066739A1 (en) * 2012-08-29 2014-03-06 Bin He System and method for quantifying or imaging pain using electrophysiological measurements
US20150025334A1 (en) * 2014-09-09 2015-01-22 Lakshya JAIN Method and system for monitoring pain of patients
CN107184204A (en) * 2017-04-28 2017-09-22 北京易飞华通科技开发有限公司 The extraction of nociceptive component and expression way in a kind of brain wave
CN109222906A (en) * 2018-09-13 2019-01-18 复旦大学 A method of pain status prediction model is constructed based on brain electrical signals
CN109271964A (en) * 2018-10-11 2019-01-25 刘仕琪 Emotion identification method and system based on deep learning model Yu length memory network
CN110515456A (en) * 2019-08-14 2019-11-29 东南大学 EEG signals emotion method of discrimination and device based on attention mechanism
CN110610168A (en) * 2019-09-20 2019-12-24 合肥工业大学 Electroencephalogram emotion recognition method based on attention mechanism
CN111126263A (en) * 2019-12-24 2020-05-08 东南大学 Electroencephalogram emotion recognition method and device based on double-hemisphere difference model
CN111951958A (en) * 2020-08-21 2020-11-17 深圳大学 Pain data evaluation method based on self-coding and related components

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140066739A1 (en) * 2012-08-29 2014-03-06 Bin He System and method for quantifying or imaging pain using electrophysiological measurements
CN103405225A (en) * 2013-08-28 2013-11-27 深圳先进技术研究院 Method, apparatus and device for obtaining pain feeling evaluation indexes
US20150025334A1 (en) * 2014-09-09 2015-01-22 Lakshya JAIN Method and system for monitoring pain of patients
CN107184204A (en) * 2017-04-28 2017-09-22 北京易飞华通科技开发有限公司 The extraction of nociceptive component and expression way in a kind of brain wave
CN109222906A (en) * 2018-09-13 2019-01-18 复旦大学 A method of pain status prediction model is constructed based on brain electrical signals
CN109271964A (en) * 2018-10-11 2019-01-25 刘仕琪 Emotion identification method and system based on deep learning model Yu length memory network
CN110515456A (en) * 2019-08-14 2019-11-29 东南大学 EEG signals emotion method of discrimination and device based on attention mechanism
CN110610168A (en) * 2019-09-20 2019-12-24 合肥工业大学 Electroencephalogram emotion recognition method based on attention mechanism
CN111126263A (en) * 2019-12-24 2020-05-08 东南大学 Electroencephalogram emotion recognition method and device based on double-hemisphere difference model
CN111951958A (en) * 2020-08-21 2020-11-17 深圳大学 Pain data evaluation method based on self-coding and related components

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113723206A (en) * 2021-08-04 2021-11-30 三峡大学 Brain wave identification method based on quantum neural network algorithm
CN113723206B (en) * 2021-08-04 2023-10-31 三峡大学 Brain wave identification method based on quantum neural network algorithm
CN113598794A (en) * 2021-08-12 2021-11-05 中南民族大学 Training method and system for detection model of ice drug addict
CN113812933A (en) * 2021-09-18 2021-12-21 重庆大学 Acute myocardial infarction real-time early warning system based on wearable equipment
CN115990108A (en) * 2022-11-01 2023-04-21 山东海天智能工程有限公司 Acupuncture control system based on brain electrical signals, readable storage medium and electronic equipment
CN115990108B (en) * 2022-11-01 2024-03-26 山东海天智能工程有限公司 Acupuncture control system based on brain electrical signals, readable storage medium and electronic equipment
CN116831595A (en) * 2023-06-21 2023-10-03 中国人民解放军西部战区总医院 Pain grade assessment method
CN116831595B (en) * 2023-06-21 2024-04-12 中国人民解放军西部战区总医院 Pain grade assessment method
CN117075741A (en) * 2023-10-17 2023-11-17 首都医科大学附属北京天坛医院 Consciousness interaction communication method and system
CN117075741B (en) * 2023-10-17 2023-12-12 首都医科大学附属北京天坛医院 Consciousness interaction communication method and system
CN117894477A (en) * 2024-03-13 2024-04-16 西安工会医院 Construction method of predictive assessment model based on postherpetic neuralgia
CN117894477B (en) * 2024-03-13 2024-05-31 西安工会医院 Construction method of predictive assessment model based on postherpetic neuralgia

Also Published As

Publication number Publication date
CN112957014B (en) 2023-06-20

Similar Documents

Publication Publication Date Title
CN112957014B (en) Pain detection and positioning method and system based on brain waves and neural network
CN110399857B (en) Electroencephalogram emotion recognition method based on graph convolution neural network
Aslan et al. Automatic Detection of Schizophrenia by Applying Deep Learning over Spectrogram Images of EEG Signals.
Kuanar et al. Cognitive analysis of working memory load from EEG, by a deep recurrent neural network
CN112932502B (en) Electroencephalogram emotion recognition method combining mutual information channel selection and hybrid neural network
CN110969108B (en) Limb action recognition method based on autonomic motor imagery electroencephalogram
Ghosh et al. Automatic eyeblink and muscular artifact detection and removal from EEG signals using k-nearest neighbor classifier and long short-term memory networks
CN110522412B (en) Method for classifying electroencephalogram signals based on multi-scale brain function network
Zeng et al. GRP-DNet: A gray recurrence plot-based densely connected convolutional network for classification of epileptiform EEG
CN111544256A (en) Brain-controlled intelligent full limb rehabilitation method based on graph convolution and transfer learning
De Martino et al. Multimodal imaging: an evaluation of univariate and multivariate methods for simultaneous EEG/fMRI
Chen et al. Two-dimensional phase lag index image representation of electroencephalography for automated recognition of driver fatigue using convolutional neural network
Agarwal et al. Fusion of pattern-based and statistical features for Schizophrenia detection from EEG signals
Hwaidi et al. A noise removal approach from eeg recordings based on variational autoencoders
Alkanhal et al. Automatic seizure detection via an optimized image-based deep feature learning
Wang et al. Multiband decomposition and spectral discriminative analysis for motor imagery BCI via deep neural network
Behnam et al. Periodogram pattern feature-based seizure detection algorithm using optimized hybrid model of MLP and ant colony
Zhang et al. DCT based multi-head attention-BiGRU model for EEG source location
Chen et al. A transfer learning-based CNN deep learning model for unfavorable driving state recognition
CN111931578B (en) Electroencephalogram identification method based on minimum spanning tree and regional double-layer network
Havugimana et al. Predicting cognitive load using parameter-optimized cnn from spatial-spectral representation of eeg recordings
CN116662736A (en) Human body state assessment method based on deep learning hybrid model
CN115813409A (en) Ultra-low-delay moving image electroencephalogram decoding method
CN113995422B (en) Transient brain power source positioning method and system based on nonnegative block sparse Bayesian learning
Wang EEG-based Anxious Personality Prediction

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant