CN112190261A - Autism electroencephalogram signal classification device based on resting brain network - Google Patents

Autism electroencephalogram signal classification device based on resting brain network Download PDF

Info

Publication number
CN112190261A
CN112190261A CN202010972559.6A CN202010972559A CN112190261A CN 112190261 A CN112190261 A CN 112190261A CN 202010972559 A CN202010972559 A CN 202010972559A CN 112190261 A CN112190261 A CN 112190261A
Authority
CN
China
Prior art keywords
network
autism
calculation module
brain
data
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.)
Pending
Application number
CN202010972559.6A
Other languages
Chinese (zh)
Inventor
李发礼
张树
李存波
尧德中
冯睿
许文明
徐鹏
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.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010972559.6A priority Critical patent/CN112190261A/en
Publication of CN112190261A publication Critical patent/CN112190261A/en
Pending legal-status Critical Current

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
    • A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
    • A61B5/165—Evaluating the state of mind, e.g. depression, anxiety
    • 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

Landscapes

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

Abstract

The invention discloses a device for classifying an autism electroencephalogram signal based on a resting brain network, belongs to the technical field of biomedical information, and particularly relates to a mode classification method in the field of brain-computer interfaces. The invention starts from three aspects of information entropy, power spectrum and brain function network to identify and diagnose the autism, comprises the most common electroencephalogram signal analysis method in the current research, and innovatively uses the SPN filter to diagnose the autism patient, and the accuracy rate is close to 100%. Particularly, by researching resting brain networks of the autism children and the normal children, network topology difference between the autism children and the normal children is explored and compared with other various methods, the pathology of the autism is explored to a certain extent, and a reliable basis is provided for clinical diagnosis of the autism.

Description

Autism electroencephalogram signal classification device based on resting brain network
Technical Field
The invention belongs to the technical field of biomedical information, and particularly relates to a mode classification method in the field of brain-computer interfaces.
Background
Autism, also known as Autism, is one of the Autism Spectrum Disorders (ASD). Most autistic patients begin to develop similar symptoms in infancy and the condition is more common in boys. There are many diagnostic modalities for the clinical and research assessment of autism today. The diagnosis method mainly comprises diagnosis methods of behavior scale, functional nuclear magnetic resonance imaging, magnetoencephalogram and electroencephalogram. In recent years, fMRI has been studied more and more in autism, but its time resolution is low and its specificity and sensitivity are not high in diagnosing a single subject. The Magnetoencephalogram (MEG) is a non-invasive neuroimaging approach, but its technical history is relatively short and expensive, and further innovation is still needed.
Brain-Computer Interface (BCI) is a new technology that can directly control a Computer or start a machine by electroencephalogram signals without depending on peripheral nerve muscles, so as to realize interaction between a human and external information. The feature extraction of the electroencephalogram signals is the key of whether the BCI system can translate brain information. The electroencephalogram, as a non-invasive detection means, can detect the neuroelectrical activity emitted by cerebral neurons by traversing electrodes on the scalp, the frequency range is generally 1-100Hz, and the electroencephalogram is widely used for the research of cerebral neuropathology at present, has many advantages such as no damage, relatively cheap acquisition system and high time resolution, can realize real-time monitoring on a subject, is a powerful tool for research, clinical diagnosis and even functional rehabilitation, and has been widely used for the research of cerebral neuropathology.
Different algorithms are used for extracting and researching the electroencephalogram characteristics of the autism based on different angles. Algorithms such as relative power, absolute power, power density and the like are analyzed from the aspect of nerve oscillation. From the perspective of brain function network connection, there are various entropy algorithms, mutual information algorithms, estimator algorithms, consistency algorithms, synchronization algorithms, and the like. EEG is a generalized stationary random signal which is neither periodic nor square multiplicative and therefore cannot be subjected to Fourier transform, and in order to study the frequency spectrum of EEG and further study the characteristics of the signal, the power spectrum of EEG can be selected to be solved for spectrum analysis. The power spectrum of the EEG reflects the distribution of power energy of each frequency component of the random signal, and can reveal useful information such as periodicity hidden in the signal, spectral peaks close to the periodicity and the like. Currently, there are many studies of power spectra on EEG that reveal the difference between autistic patients and normal persons. Including an increase in the right posterior region theta band, a decrease in the frontal delta band, and an increase in the midline region beta band, and autistic children exhibit decreased connectivity compared to normal. Meanwhile, researches show that the autistic patient has the effects of enhancing the low-frequency power, weakening the high-frequency power and weakening the medium-frequency power.
The previous research report of applying information entropy to electroencephalogram analysis shows that significant differences exist in entropy values between the autism group and the normal group. Studies on multi-scale entropy of the brain electricity have found that the high-risk ASD group has an EEG complexity that is consistently low at all time scales and at all ages. Yet another study showed that the EEG complexity of ASD patients decreased with increasing scale in the temporal occipital region. The multi-scale entropy research on the autism high-risk infants finds that the complexity of the autism high-risk infants is reduced and is most remarkable in the frontal lobe area of the brain. Meanwhile, analysis shows that the ASD group has lower complexity in the forehead and occipital regions compared with the normal group, and the entropy value of the occipital region of the temporalis is obviously weakened.
On the other hand, the brain is a complex network composed of different brain regions that perform various activities and functions together, and there are neuronal communication and coordination activities between the brain regions, and the EEG brain network effectively analyzes communication activities between different regions of the brain. The study of complex brain networks has been widely applied to many aspects, such as the study of cognitive functions of the brain and the study of some neuropsychiatric diseases. The phase-locked chronaxie Analysis method can detect the phase synchronization condition between different EEG leads and further extract the inherent information exchange between brain areas related to the disease of autism patients, and the Coherence Analysis (Coherence Analysis) can effectively and accurately capture the potential interaction between the brain areas related to the spatial distribution and the function. The research on the coherent analysis finds that the ASD adults show enhanced functional connection of frontal cortex areas and reduced long-distance connection of resting brain electricity. The brain electrical signals of the autism patient have weaker network connection between frontal lobe and occipital lobe areas in delta and theta frequency bands.
Although the difference between the power spectrum and the information entropy may explain the difference between the researches in classification accuracy, the information transfer between the brain areas cannot be obtained, and the mutual coordination between different partitions and functions of the brain as a network is ignored. Studies on complex brain networks show that the cognitive function of the brain of autistic children differs from that of normal children under the same conditions, and the correct differentiation between control and autistic children is the first step of providing criteria for the diagnosis of ASD diseases based on brain function analysis, but it is still unknown how ASD-characteristic behaviors are generated by the connection of different regions and functions of the brain. Meanwhile, the age difference of the study subjects can have a remarkable influence on the study results. And the conclusions thereof are greatly different as related studies become more and more. Therefore, the electroencephalogram signals of ASD patients are researched, more effective electroencephalogram characteristics are found out by using various electroencephalogram processing methods, and the multi-method fusion is realized, so that the evaluation and auxiliary diagnosis of autism are facilitated.
Disclosure of Invention
The invention aims to solve the problem that the feature extraction of the autism electroencephalogram signals is difficult in the prior art and realize the purpose of classifying the autism electroencephalogram signals.
The technical scheme adopted by the invention is as follows: a device for classifying an autism electroencephalogram signal based on a resting brain network, the device comprising: the device comprises an electroencephalogram signal acquisition device, a preprocessing module, a sample entropy calculation module, a fuzzy entropy calculation module, a power density calculation module, a weighted undirected connection network calculation module, a network attribute feature extraction module, an SPN filter and an LDA classifier; the electroencephalogram signals obtained by the electroencephalogram signal acquisition device sequentially pass through the preprocessing module, and the output of the preprocessing module is respectively used as: the system comprises a sample entropy calculation module, a fuzzy entropy calculation module, a power density calculation module and an input of a weighted undirected connection network calculation module, wherein the output of the weighted undirected connection network calculation module is used as the input of a network attribute feature extraction module and an SPN filter; finally, the outputs of the sample entropy calculation module, the fuzzy entropy calculation module, the power density calculation module, the network attribute feature extraction module and the SPN filter are used as the inputs of a trained LDA classifier, and classification of the electroencephalogram signals is realized through the trained LDA classifier;
the standard of the electrode placement position of the electroencephalogram signal acquisition device is an international standard 10-20 system, the sampling rate is 500Hz, the band-pass filtering range is 0.5-45Hz, and electroencephalogram signals in a closed-eye resting state are acquired;
the preprocessing module sequentially performs the following steps on the acquired electroencephalogram signals: average reference processing and band-pass filtering processing of frequency bands: delta wave: [1Hz-4Hz), θ wave: [4Hz-8Hz), alpha wave: [8Hz-13Hz) and beta wave: [13Hz-30Hz ], data segmentation processing for 5 seconds, baseline correction processing for segmented data, and ocular artifact removal processing with 70 μ V as a threshold;
the calculation method in the sample entropy calculation module is as follows:
the electroencephalogram signal after the sampling processing is set as { u (i) }, i ═ 0,1,2, …, N }, and the calculation steps of the sample entropy are as follows:
dividing N time sequence points, wherein m data points are one sub-segment, i is more than or equal to 1 and less than or equal to N-m, the m data points are divided into N-m sub-sequence segments, and the sub-sequence segments are marked as X (i):
X(i)＝[x(i),x(i+1),…x(i+m-1)],i＝1,2,…,N-m (1)
calculating the maximum value of the distance between the ith subsequence segment X (i) and the other subsequence segments X (j), wherein N-m-1 times are required to be calculated, and i is taken from 1 to N-m, and the calculation formula is as follows:
Figure BDA0002684619820000031
setting a threshold r, i from 1 to N-m, calculating all d [ X (i), X (j)]Then d [ X (i), X (j)]The number of the distance r is less than r, and the number is divided by the total number of the distances N-m-1 and recorded as
Figure BDA0002684619820000032
Figure BDA0002684619820000033
Calculate out
Figure BDA0002684619820000034
The average values of (a) are as follows:
Figure BDA0002684619820000035
increasing dimension to m +1 dimension, repeating the steps to calculate Bm+1(r); the sample entropy SaEn (m, r) is calculated as follows:
SaEn(m,r)＝-ln[Bm+1(r)/Bm(r)] (5)
the calculation method in the fuzzy entropy calculation module is as follows:
for a given N-dimensional time series u (i), defining a spatial dimension m, m ≦ N-2 and a similarity tolerance r, reconstructing the sequence space:
X(i)＝[u(i),u(i+1),…u(i+m-1)],i＝1,2,…,N+m-1 (6)
Figure BDA0002684619820000041
introducing fuzzy membership function A (x):
Figure BDA0002684619820000042
wherein r is the similarity tolerance;
for i ═ 1,2, …, N-m +1, fuzzy membership is calculated by means of a fuzzy membership function a (x)
Figure BDA0002684619820000043
Figure BDA0002684619820000044
Figure BDA0002684619820000045
Representing the fuzzy degree of membership between the ith and jth sequences when the sequence dimension is m, wherein,
Figure BDA0002684619820000046
is the maximum absolute distance between sequences X (i) and Y (j);
for each i, find
Figure BDA0002684619820000047
To obtain:
Figure BDA0002684619820000048
defining:
Figure BDA0002684619820000049
thus, the fuzzy entropy of the original time series is:
FuzzyEn(m,r)＝limN→∞[lnΦm(r)-lnΦm+1(r)] (13)
the fuzzy entropy estimate for a finite data set is:
FuzzyEn(m,r,N)＝lnΦm(r)-lnΦm+1(r) (14)
the power density calculation module divides data with the length of N into L sections, the length of each section is M, spectral estimation is carried out on each section of data by a periodogram method, and then the L sections are averaged to obtain a power spectrum of the data with the length of N; the calculation method comprises the following steps:
Figure BDA00026846198200000410
wherein, U ═ Σnω (n), ω (n) being a window function, MUL representing the product of M, U, L, xi(n) an nth sample point representing the ith segment of data after segmentation;
the calculation method in the weighted undirected connection network calculation module comprises the following steps: given time series of two nodes, x (t) and y (t), with instantaneous phases phix (t) and phiy (t), wherein a Hilbert transform is used to obtain the corresponding analytic signal Hx(t) and Hy(t)，
Figure BDA0002684619820000051
Where X (t) and Y (t) are real parts of the EEG signals X (t) and y (t) after Hilbert transform, Xh(t) and Yh(t) is the imaginary part of the EEG signal after x (t) and y (t) Hilbert transform, i denotes the imaginary number i2-1, defined as follows:
Figure BDA0002684619820000052
wherein p.v. represents a cauchy principal value; the corresponding analytic signal phases phix (t) and phiy (t) can then be calculated,
Figure BDA0002684619820000053
finally, calculating the connection weight w between the electrodes i, jplv，
Figure BDA0002684619820000054
In the formula, Δ t represents a sampling period, and N represents the number of sampling points; w is aplvIn [0, 1]]In between, 1 represents full phase synchronization and 0 represents no phase synchronization;
the calculation method in the network attribute feature extraction module comprises the following steps:
Figure BDA0002684619820000055
Figure BDA0002684619820000056
c is a clustering coefficient, psi represents the sum of all electrodes, j, h epsilon psi represents a node j and a node h which are adjacent to i in the nodes of the brain network, L is a characteristic path length, and the clustering coefficient C reflects the aggregation degree of all the nodes of the brain network so as to represent the clustering characteristic of the brain network. The characteristic path length L describes the degree of connectivity between different electrodes in the brain network; dijA weighted shortest path length electrode representing the length between two nodes i and j; n represents the total number of electrodes;
the calculation method in the SPN filter is as follows: using three pairs of SPN filters to act on each tested weighted adjacency matrix, and enabling the variance of one type of data to be minimum and the variance of the other type of data to be maximum so as to extract inherent spatial information characteristics; the SPN filter is a projection obtained by maximizing a generalized Rayleigh quotient J (ω), where ω is the projection value, φ1And phi2Weighted adjacency matrix mean, phi, representing normal and autistic children, respectively1And phi2Weighted adjacency matrix phi for normal and autistic children, respectively1And phi2The covariance matrix of (a); the method comprises the following steps:
Figure BDA0002684619820000061
since whether the projection is scaled does not affect the function value, equation (22) is constrained to optimize the equation solution:
Figure BDA0002684619820000062
then introducing a Lagrange multiplier II into the above equation and further rewriting to obtain an equation:
L(ω,λ)＝ωTΦ1ω-λ(ωTΦ2ω-1) (24)
when ω satisfies the condition
Figure BDA0002684619820000063
Equation (24) can be further abbreviated as equation (25), and the projection is estimated using the generalized eigenvalue equation:
Φ1ω＝λΦ2ω (25)
wherein λ represents a eigenvalue of a generalized eigenequation; ω is a feature vector corresponding to the feature value;
since there are multiple filters, the equation is rewritten as follows:
Figure BDA0002684619820000064
wherein W is contained in
Figure BDA0002684619820000065
And ∑ diag (λ)1,λ2,…,λm) Feature vector of (d), diag (λ)1,λ2,…,λm) Each element in the list represents a corresponding singular value; in the invention, three pairs of filters with the most difference are selected to extract difference information in brain network space, and finally 6-dimensional electroencephalogram characteristics are obtained.
The invention has the following advantages:
EEG power spectrum and information entropy are the most common electroencephalogram signal analysis methods, and at present, many studies show the difference of electroencephalogram information between an autism patient and a normal person, so the invention also researches the power spectrum and the information entropy of the autism patient and a control group of children. But at the same time, the brain is a large scale network and has a complex and efficient brain structure, which comprises a plurality of brain areas possibly related to the autism patient. The invention assumes that the communication and connection between the related brain areas of the brain of the autistic patient can not realize normal cognitive function and can not be controlled to be repeated, thereby causing the emotional, social and language barriers of the patient. The method adopts the phase locking time value to construct the rest state brain function network of the autism patient and the normal control group. And through calculating the network attribute, the brain efficiency of the patient under the resting state is quantitatively measured, and the difference between the autism sick child and the normal child can be obtained in a large-scale network. Finally, the SPN adopts a supervised learning method, the extraction of the difference characteristics existing in the brain network topology of the autism patient is realized by emphasizing some important nodes of the brain network and relatively neglecting the importance of other nodes, the SPN filter extracts the SPN characteristics from the brain network topology of the autism group and the normal group, the SPN characteristics have obvious difference, and the remarkable topological difference existing between the brain networks of the autism group and the normal group can be captured in the supervision filtering process.
The invention starts from three aspects of information entropy, power spectrum and brain function network to identify and diagnose the autism, comprises the most common electroencephalogram signal analysis method in the current research, and innovatively uses the SPN filter to diagnose the autism patient, and the accuracy rate is close to 100%. Particularly, by researching resting brain networks of the autism children and the normal children, network topology difference between the autism children and the normal children is explored and compared with other various methods, the pathology of the autism is explored to a certain extent, and a reliable basis is provided for clinical diagnosis of the autism.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention.
FIG. 2 is a sample entropy difference statistical diagram of the present invention.
FIG. 3 is a fuzzy entropy difference statistical chart of the present invention.
FIG. 4 Power Spectrum Difference distribution (p <0.01) between autistic and normal children of the invention.
FIG. 5 the difference in power spectra of the autistic children of the invention and normal children.
FIG. 6 shows the difference in the topology of the brain network according to the present invention.
FIG. 7 shows the network attribute difference statistics of the present invention.
Fig. 8 scalp scores for brain topology obtained for the two most different SPN filters of the present invention.
FIG. 9 is a distribution graph of SPN signatures of autistic and normal children of the present invention.
Detailed Description
The invention is further described with reference to the accompanying drawings and specific embodiments.
S1, collecting experimental data. The invention researches two batches of collected data, wherein the first batch of data is a training set, and the second batch of data is a testing set. The first data included 16 autistic children tested (2-6 years, mean age 3.8 years, 13 women), and 11 normal children (3-5 years, mean age 7.42 years, 8 women). The second data included 11 autistic children trials (2-7 years, mean age 4.82 years, 9 women). The standard of the electrode placement position is an international standard 10-20 system, the sampling rate is 500Hz, the band-pass filtering range is 0.5-45Hz, and a closed-eye resting state of 10 minutes is acquired for each tested sample;
calculating sample entropies of all electroencephalogram channels on four frequency bands of delta, theta, alpha and beta of a training set (first batch of data), and counting statistical differences of normal children and children with all leads, wherein the differences of two channels O1 and O2 on the theta frequency band are found to be the most obvious, and the differences are shown in the figure. And the sample entropy of normal children is higher than that of autistic patients.
S2, preprocessing the acquired electroencephalogram data, wherein the preprocessing process comprises average reference, 1-20Hz band-pass filtering, data segmentation for 5 seconds, baseline correction of segmented data and ocular artifact removal with 70 mu V as a threshold;
and (4) calculating the fuzzy entropy of the training set according to the step of S1, and counting the significance difference, wherein the result is similar to the result of the sample entropy, the difference between the two channels O1 and O2 on the theta frequency band is significant, and the fuzzy entropy of the normal children is higher than that of the autism patients. The information communication of the brain of the self-closing patient occipital lobe is weaker than that of a normal child. Two channel sample entropies O1 and O2 on the electroencephalogram theta frequency band are selected as feature vectors for classification.
Extracting sample entropy of a tested sample; assuming that the acquired original brain electrical signals are { u (i) ═ 0,1,2, …, N }, the calculation steps of the sample entropy are as follows:
dividing N time sequence points, wherein m data points are a sub-segment, m is 2, i is more than or equal to 1 and less than or equal to N-m, and can be divided into (N-m) sub-sequence segments, and the sub-sequence segments are marked as X (i):
X(i)＝[x(i),x(i+1),…x(i+m-1)],i＝1,2,…,N-m (1)
calculating the maximum value of the distance between the ith subsequence segment X (i) and the other subsequence segments X (j), wherein N-m-1 times are required to be calculated, and i is taken from 1 to N-m, and the following is a calculation formula:
Figure BDA0002684619820000081
setting a threshold value r, wherein the r is generally 0.2 times of the standard deviation of the time series; i take N-m from 1, calculate all d [ X (i), X (j)]Then d [ X (i), X (j)]The number of the distance r is less than r, and the number is divided by the total number of the distances N-m-1 and recorded as
Figure BDA0002684619820000082
Figure BDA0002684619820000083
Calculate out
Figure BDA0002684619820000084
The average values of (a) are as follows:
Figure BDA0002684619820000085
increasing dimension to m +1 dimension, repeating the steps to calculate Bm+1(r); the sample entropy SaEn (m, r) is calculated as follows:
SaEn(m,r)＝-ln[Bm+1(r)/Bm(r)] (5)
extracting the tested fuzzy entropy; for a given N-dimensional time series { u (i), i ═ 0,1,2, …, N }, a phase space dimension m (m ≦ N-2) and a similarity tolerance r are defined, the phase space is reconstructed:
x (i) ═ u (i), u (i +1), … u (i + m-1) ], i ═ 1,2, …, N + m-1 (6) wherein,
Figure BDA0002684619820000086
introducing fuzzy membership functions
Figure BDA0002684619820000087
Wherein r is the similarity tolerance;
for i ═ 1,2, …, N-m +1, calculations
Figure BDA0002684619820000091
Wherein the content of the first and second substances,
Figure BDA0002684619820000092
is the maximum absolute distance between window vectors X (i) and Y (j)
For each i, the average value is obtained
Figure BDA0002684619820000093
Definition of
Figure BDA0002684619820000094
Thus, the fuzzy entropy (fuzzy en) of the original time series is
FuzzyEn(m,r)＝limN→∞[lnΦm(r)-lnΦm+1(r)] (13)
For finite data and, the fuzzy entropy is estimated as
FuzzyEn(m,r,N)＝lnΦm(r)-lnΦm+1(r) (14)
S3, calculating the power spectrum density of each frequency band of the autism child and the normal child respectively, and counting the significance difference of the power spectrum density, wherein the PSD of the alpha frequency band of the normal child is found to be remarkably stronger than that of the autism child (p is less than 0.01) in the occipital lobe part, and the figure is shown in figure 4. Fig. 5 shows the difference between the two leads (O1, O2) between the autistic and normal children, and it can be seen that there is a weak energy distribution in the electroencephalogram of the autistic patient. The construction of the classifier is carried out by adopting the average power spectrum characteristics of alpha frequency bands on O1 and O2 with statistical difference (p < 0.01).
A tested Power Spectral Density (PSD) is calculated. A Welch periodic map estimation method is selected, and the classical periodic map algorithm is improved in two aspects: one is in the pair xN(n) when segmenting, allowing the segmented data to have an overlapping part, usually adopting 50% overlap; secondly, selecting a non-rectangular window to segment the data, further improving the influence of poor resolution caused by the rectangular window, and generally selecting a Hamming window to obtain a modified periodogram further:
Figure BDA0002684619820000095
wherein, U ═ Σnω (n), ω (n) is a window function.
Firstly, a phase locking value analysis method is adopted to calculate the weighted undirected connection network of each tested object:
Figure BDA0002684619820000101
the value of PLV is between [0, 1], with 1 representing complete phase synchronization and 0 representing no phase synchronization.
And S4, calculating a brain connection matrix of each frequency band of the training set by adopting a PLV network analysis method, and further constructing a brain network. Spatial information implicit in the weighted connectivity matrix of the brain can be used to represent the interaction between different brain regions, where a topological network of differences in the brain with statistical significance (p <0.05) between the autistic patient group and the normal group was analyzed, as shown in fig. 6. The blue line indicates a significant decrease in the bonding strength, and the red line indicates a significant increase in the bonding strength. As can be seen in the figures, the group of autistic patients exhibited a weakened connection in the temporal lobe and the occipital lobe of the posterior region of the brain in the delta, theta and alpha frequency bands compared to the control group. The brain activated connection condition of the patient group with the autism in the delta frequency band is weaker than that of the normal children, and the patient group with the autism in the theta and alpha frequency bands presents the activated connection enhanced in the frontal lobe-temporal lobe, the frontal lobe-occipital lobe, the frontal lobe-parietal lobe and other areas.
Calculating four common network attribute characteristics, a clustering coefficient C and a characteristic path length L; the clustering coefficient C reflects the aggregation degree of all nodes of the brain network so as to express the clustering characteristic of the brain network; the characteristic path length L describes the degree of connectivity between different electrodes in the brain network;
the calculation steps are respectively as follows:
Figure BDA0002684619820000102
Figure BDA0002684619820000103
wherein wijTo representThe connection strength between the electrodes i, j; dijRepresenting the characteristic path strength between the electrodes i, j; n represents the total number of electrodes, ψ represents the sum of all the electrodes;
and S5, analyzing the difference of brain network attributes of each frequency band of the ASD patient group and the normal group, carrying out quantitative analysis on the network information transmission efficiency between the ASD patient group and the normal group, and carrying out classification identification on the autism children and the normal children by taking the network attributes as a classification characteristic index. The following tables and figures show the clustering coefficients and characteristic path lengths for ASD patients and normal groups. In the resting state, the network attributes of the autistic children and the normal children have statistical significance difference (p <0.01) in the frequency band of delta and the clustering coefficient of the alpha frequency band (p <0.05), and the figure is shown in fig. 7. The clustering coefficient of the autistic group children was significantly lower than that of the normal group, while its characteristic path length was significantly higher than that of the normal group. The reduction of the transmission rate of the brain network of the infant with the autism disease, the reduction of the speed of extracting useful information from the external environment by the brain, and the brain network with lower efficiency exist. And (4) taking the network attribute as a classification characteristic index to perform classification identification on the autism children and the normal children.
TABLE 1 network Attribute characterization of Normal group and ASD patients
Figure BDA0002684619820000111
S6, selecting 3 pairs of SPN filters to extract inherent and potential spatial information existing in the brain adjacency matrix. Fig. 8 shows the two most different SPN filters (filter 1, filter 2), from which the significantly attenuated activation of the occipital region of autistic children can be seen, which is also mutually corroborated with the results of the topological network of differences in brain.
S7, constructing an LDA classifier by adopting the sample entropy, the fuzzy entropy, the power spectrum, the network attribute and the SPN characteristic of the training set to be tested, and evaluating the classification model by using a leave-one-out method, wherein the result is shown in the following table. As shown in the scatter plot of a pair of SPN features from two types of subjects shown in fig. 9, it can be seen that the SPN features extracted from the brain topological network can be used to distinguish between the autistic patient group and the normal group.
TABLE 2 sample entropy Classification accuracy
Figure BDA0002684619820000112
TABLE 3 fuzzy entropy Classification accuracy
Figure BDA0002684619820000113
TABLE 4 PSD Classification accuracy
Figure BDA0002684619820000121
TABLE 5 network Attribute Classification accuracy
Figure BDA0002684619820000122
TABLE 6 SPN Classification accuracy
Figure BDA0002684619820000123
And S8, calculating sample entropy, fuzzy entropy, PSD, network attribute and SPN characteristic of each frequency band of the test set sample, and predicting the test set sample by using the classifier constructed in S7, wherein the result is shown in the following table. When the delta network space topology difference characteristics are used for constructing the LDA classifier, the highest test accuracy can reach 100%. The combination of the PLV method for constructing the brain network and the SPN filter is proved to extract more brain function network characteristics, and the diagnosis accuracy of the autism children is greatly improved.
TABLE 7 SPN Classification accuracy
Figure BDA0002684619820000131

Claims (1)

1. A device for classifying an autism electroencephalogram signal based on a resting brain network, the device comprising: the device comprises an electroencephalogram signal acquisition device, a preprocessing module, a sample entropy calculation module, a fuzzy entropy calculation module, a power density calculation module, a weighted undirected connection network calculation module, a network attribute feature extraction module, an SPN filter and an LDA classifier; the electroencephalogram signals obtained by the electroencephalogram signal acquisition device sequentially pass through the preprocessing module, and the output of the preprocessing module is respectively used as: the system comprises a sample entropy calculation module, a fuzzy entropy calculation module, a power density calculation module and an input of a weighted undirected connection network calculation module, wherein the output of the weighted undirected connection network calculation module is used as the input of a network attribute feature extraction module and an SPN filter; finally, the outputs of the sample entropy calculation module, the fuzzy entropy calculation module, the power density calculation module, the network attribute feature extraction module and the SPN filter are used as the inputs of a trained LDA classifier, and classification of the electroencephalogram signals is realized through the trained LDA classifier;
the standard of the electrode placement position of the electroencephalogram signal acquisition device is an international standard 10-20 system, the sampling rate is 500Hz, the band-pass filtering range is 0.5-45Hz, and electroencephalogram signals in a closed-eye resting state are acquired;
the preprocessing module sequentially performs the following steps on the acquired electroencephalogram signals: average reference processing and band-pass filtering processing of frequency bands: delta wave: [1Hz-4Hz), θ wave: [4Hz-8Hz), alpha wave: [8Hz-13Hz) and beta wave: [13Hz-30Hz ], data segmentation processing for 5 seconds, baseline correction processing for segmented data, and ocular artifact removal processing with 70 μ V as a threshold;
the calculation method in the sample entropy calculation module is as follows:
the electroencephalogram signal after the sampling processing is set as { u (i) }, i ═ 0,1,2, …, N }, and the calculation steps of the sample entropy are as follows:
dividing N time sequence points, wherein m data points are one sub-segment, i is more than or equal to 1 and less than or equal to N-m, the m data points are divided into N-m sub-sequence segments, and the sub-sequence segments are marked as X (i):
X(i)＝[x(i),x(i+1),…x(i+m-1)],i＝1,2,…,N-m (1)
calculating the maximum value of the distance between the ith subsequence segment X (i) and the other subsequence segments X (j), wherein N-m-1 times are required to be calculated, and i is taken from 1 to N-m, and the calculation formula is as follows:
Figure FDA0002684619810000011
setting a threshold r, i from 1 to N-m, calculating all d [ X (i), X (j)]Then d [ X (i), X (j)]The number of the distance r is less than r, and the number is divided by the total number of the distances N-m-1 and recorded as
Figure FDA0002684619810000012
Figure FDA0002684619810000013
Calculate out
Figure FDA0002684619810000014
The average values of (a) are as follows:
Figure FDA0002684619810000015
increasing dimension to m +1 dimension, repeating the steps to calculate Bm+1(r); the sample entropy SaEn (m, r) is calculated as follows:
SaEn(m,r)＝-ln[Bm+1(r)/Bm(r)] (5)
the calculation method in the fuzzy entropy calculation module is as follows:
for a given N-dimensional time series u (i), defining a spatial dimension m, m ≦ N-2 and a similarity tolerance r, reconstructing the sequence space:
X(i)＝[u(i),u(i+1),…u(i+m-1)],i＝1,2,…,N+m-1 (6)
Figure FDA0002684619810000021
introducing fuzzy membership function A (x):
Figure FDA0002684619810000022
wherein r is the similarity tolerance;
for i ═ 1,2, …, N-m +1, fuzzy membership is calculated by means of a fuzzy membership function a (x)
Figure FDA0002684619810000023
Figure FDA0002684619810000024
Figure FDA0002684619810000025
Representing the fuzzy degree of membership between the ith and jth sequences when the sequence dimension is m, wherein,
Figure FDA0002684619810000026
is the maximum absolute distance between sequences X (i) and Y (j);
for each i, find
Figure FDA0002684619810000027
To obtain:
Figure FDA0002684619810000028
defining:
Figure FDA0002684619810000029
thus, the fuzzy entropy of the original time series is:
FuzzyEn(m,r)＝limN→∞[lnΦm(r)-lnΦm+1(r)] (13)
the fuzzy entropy estimate for a finite data set is:
FuzzyEn(m,r,N)＝lnΦm(r)-lnΦm+1(r) (14)
the power density calculation module divides data with the length of N into L sections, the length of each section is M, spectral estimation is carried out on each section of data by a periodogram method, and then the L sections are averaged to obtain a power spectrum of the data with the length of N; the calculation method comprises the following steps:
Figure FDA0002684619810000031
wherein, U ═ Σnω (n), ω (n) being a window function, MUL representing the product of M, U, L, xi(n) an nth sample point representing the ith segment of data after segmentation;
the calculation method in the weighted undirected connection network calculation module comprises the following steps: given time series of two nodes, x (t) and y (t), with instantaneous phases phix (t) and phiy (t), wherein a Hilbert transform is used to obtain the corresponding analytic signal Hx(t) and Hy(t)，
Figure FDA0002684619810000032
Where X (t) and Y (t) are real parts of the EEG signals X (t) and y (t) after Hilbert transform, Xh(t) and Yh(t) is the imaginary part of the EEG signal after x (t) and y (t) Hilbert transform, i denotes the imaginary number i2-1, defined as follows:
Figure FDA0002684619810000033
wherein p.v. represents a cauchy principal value; the corresponding analytic signal phases phix (t) and phiy (t) can then be calculated,
Figure FDA0002684619810000034
finally, calculating the connection weight w between the electrodes i, jplv，
Figure FDA0002684619810000035
In the formula, Δ t represents a sampling period, and N represents the number of sampling points; w is aplvIn [0, 1]]In between, 1 represents full phase synchronization and 0 represents no phase synchronization;
the calculation method in the network attribute feature extraction module comprises the following steps:
Figure FDA0002684619810000036
Figure FDA0002684619810000041
c is a clustering coefficient, psi represents the sum of all electrodes, j, h epsilon psi represents a node j and a node h which are adjacent to i in the nodes of the brain network, L is a characteristic path length, and the clustering coefficient C reflects the aggregation degree of all the nodes of the brain network so as to represent the clustering characteristic of the brain network. The characteristic path length L describes the degree of connectivity between different electrodes in the brain network; dijA weighted shortest path length electrode representing the length between two nodes i and j; n represents the total number of electrodes;
the calculation method in the SPN filter is as follows: using three pairs of SPN filters to act on each tested weighted adjacency matrix, and enabling the variance of one type of data to be minimum and the variance of the other type of data to be maximum so as to extract inherent spatial information characteristics; SPN filter is passA projection obtained by over-maximizing the generalized Rayleigh quotient J (ω), where ω is the projection value, φ1And phi2Weighted adjacency matrix mean, phi, representing normal and autistic children, respectively1And phi2Weighted adjacency matrix phi for normal and autistic children, respectively1And phi2The covariance matrix of (a); the method comprises the following steps:
Figure FDA0002684619810000042
since whether the projection is scaled does not affect the function value, equation (22) is constrained to optimize the equation solution:
Figure FDA0002684619810000043
then introducing a Lagrange multiplier II into the above equation and further rewriting to obtain an equation:
L(ω,λ)＝ωTΦ1ω-λ(ωTΦ2ω-1) (24)
when ω satisfies the condition
Figure FDA0002684619810000044
Equation (24) can be further abbreviated as equation (25), and the projection is estimated using the generalized eigenvalue equation:
Φ1ω＝λΦ2ω (25)
wherein λ represents a eigenvalue of a generalized eigenequation; ω is a feature vector corresponding to the feature value;
since there are multiple filters, the equation is rewritten as follows:
Figure FDA0002684619810000045
wherein W is contained in
Figure FDA0002684619810000046
And ∑ diag (λ)1,λ2,…,λm) Feature vector of (d), diag (λ)1,λ2,…,λm) Each element in the list represents a corresponding singular value; in the invention, three pairs of filters with the most difference are selected to extract difference information in brain network space, and finally 6-dimensional electroencephalogram characteristics are obtained.
CN202010972559.6A 2020-09-16 2020-09-16 Autism electroencephalogram signal classification device based on resting brain network Pending CN112190261A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010972559.6A CN112190261A (en) 2020-09-16 2020-09-16 Autism electroencephalogram signal classification device based on resting brain network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010972559.6A CN112190261A (en) 2020-09-16 2020-09-16 Autism electroencephalogram signal classification device based on resting brain network

Publications (1)

Publication Number Publication Date
CN112190261A true CN112190261A (en) 2021-01-08

Family

ID=74015056

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010972559.6A Pending CN112190261A (en) 2020-09-16 2020-09-16 Autism electroencephalogram signal classification device based on resting brain network

Country Status (1)

Country Link
CN (1) CN112190261A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113143296A (en) * 2021-04-20 2021-07-23 上海外国语大学 Intelligent assessment method and system for communication obstacles
CN113180703A (en) * 2021-05-13 2021-07-30 李天睿 Electroencephalogram signal analysis method based on network topology description
CN113274037A (en) * 2021-06-30 2021-08-20 中国科学院苏州生物医学工程技术研究所 Method, system and equipment for generating dynamic brain function network
CN113425312A (en) * 2021-07-30 2021-09-24 清华大学 Electroencephalogram data processing method and device
CN113576491A (en) * 2021-07-26 2021-11-02 深圳市人民医院 Method and system for automatically analyzing frequency domain characteristics and brain network based on resting EEG
CN113974650A (en) * 2021-06-29 2022-01-28 华南师范大学 Electroencephalogram network function analysis method and device, electronic equipment and storage medium
CN114469095A (en) * 2022-04-19 2022-05-13 之江实验室 Attention deviation training effect evaluation method and system based on electroencephalogram nonlinear characteristics
CN115500790A (en) * 2022-09-23 2022-12-23 燕山大学 Method for predicting and analyzing epileptic seizure based on BI-LSTM network
CN116898401A (en) * 2023-07-17 2023-10-20 燕山大学 Autism spectrum disorder subtype classification method and device based on combined recursion quantification

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107361766A (en) * 2017-07-17 2017-11-21 中国人民解放军信息工程大学 A kind of mood EEG signal identification method based on EMD domains multidimensional information
CN107832709A (en) * 2017-11-10 2018-03-23 辽宁师范大学 The brain electricity emotion identification method of feature extraction is carried out based on experience wavelet transformation
CN108143411A (en) * 2017-12-13 2018-06-12 东南大学 A kind of tranquillization state brain electricity analytical system towards Autism Diagnostic
CN108427929A (en) * 2018-03-19 2018-08-21 兰州大学 A kind of depressed discriminance analysis system based on tranquillization state brain network
CN110059564A (en) * 2019-03-19 2019-07-26 西安交通大学 Feature extracting method based on power spectral density and cross-correlation entropy-spectrum density fusion
CN110897648A (en) * 2019-12-16 2020-03-24 南京医科大学 Emotion recognition classification method based on electroencephalogram signal and LSTM neural network model

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107361766A (en) * 2017-07-17 2017-11-21 中国人民解放军信息工程大学 A kind of mood EEG signal identification method based on EMD domains multidimensional information
CN107832709A (en) * 2017-11-10 2018-03-23 辽宁师范大学 The brain electricity emotion identification method of feature extraction is carried out based on experience wavelet transformation
CN108143411A (en) * 2017-12-13 2018-06-12 东南大学 A kind of tranquillization state brain electricity analytical system towards Autism Diagnostic
CN108427929A (en) * 2018-03-19 2018-08-21 兰州大学 A kind of depressed discriminance analysis system based on tranquillization state brain network
CN110059564A (en) * 2019-03-19 2019-07-26 西安交通大学 Feature extracting method based on power spectral density and cross-correlation entropy-spectrum density fusion
CN110897648A (en) * 2019-12-16 2020-03-24 南京医科大学 Emotion recognition classification method based on electroencephalogram signal and LSTM neural network model

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
JIANNAN KANG: "EEG entropy analysis in autism children", 《JOURNAL OF CLINICAL NEUROSCIENCE》 *
WEITING CHEN: "Characterization of Surface EMG Signal Based on Fuzzy Entroy", 《IEEE TRANSACTION ON NEURAL SYSTEM AND REHABILITATION ENGINEERING》 *
李炎烛: "孤独症儿童脑电特征提取及分类", 《中国优秀硕士学位论文全文数据库》 *
段珂懿: "儿童抽动症、自闭症的静息脑电网络研究", 《中国优秀硕士学位论文全文数据库》 *
雷敏: "基于虚拟开车环境的自闭症儿童脑电样本熵", 《物理学报》 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113143296A (en) * 2021-04-20 2021-07-23 上海外国语大学 Intelligent assessment method and system for communication obstacles
CN113143296B (en) * 2021-04-20 2022-08-30 上海外国语大学 Intelligent assessment method and system for communication disorder
CN113180703A (en) * 2021-05-13 2021-07-30 李天睿 Electroencephalogram signal analysis method based on network topology description
CN113974650A (en) * 2021-06-29 2022-01-28 华南师范大学 Electroencephalogram network function analysis method and device, electronic equipment and storage medium
CN113274037A (en) * 2021-06-30 2021-08-20 中国科学院苏州生物医学工程技术研究所 Method, system and equipment for generating dynamic brain function network
CN113274037B (en) * 2021-06-30 2022-08-26 中国科学院苏州生物医学工程技术研究所 Method, system and equipment for generating dynamic brain function network
CN113576491A (en) * 2021-07-26 2021-11-02 深圳市人民医院 Method and system for automatically analyzing frequency domain characteristics and brain network based on resting EEG
CN113425312A (en) * 2021-07-30 2021-09-24 清华大学 Electroencephalogram data processing method and device
CN114469095A (en) * 2022-04-19 2022-05-13 之江实验室 Attention deviation training effect evaluation method and system based on electroencephalogram nonlinear characteristics
CN115500790A (en) * 2022-09-23 2022-12-23 燕山大学 Method for predicting and analyzing epileptic seizure based on BI-LSTM network
CN116898401A (en) * 2023-07-17 2023-10-20 燕山大学 Autism spectrum disorder subtype classification method and device based on combined recursion quantification
CN116898401B (en) * 2023-07-17 2024-02-02 燕山大学 Autism spectrum disorder subtype classification method and device based on combined recursion quantification

Similar Documents

Publication Publication Date Title
CN112190261A (en) Autism electroencephalogram signal classification device based on resting brain network
Ahmad et al. EEG-based epileptic seizure detection via machine/deep learning approaches: a systematic review
Sangaiah et al. An intelligent learning approach for improving ECG signal classification and arrhythmia analysis
Yu et al. Supervised network-based fuzzy learning of EEG signals for Alzheimer's disease identification
Elhaj et al. Arrhythmia recognition and classification using combined linear and nonlinear features of ECG signals
CN109770924B (en) Fatigue classification method for building brain function network and related vector machine based on generalized consistency
Cong et al. Benefits of multi-domain feature of mismatch negativity extracted by non-negative tensor factorization from EEG collected by low-density array
Alturki et al. Common spatial pattern technique with EEG signals for diagnosis of autism and epilepsy disorders
CN111466876B (en) Alzheimer&#39;s disease auxiliary diagnosis system based on fNIRS and graph neural network
Ahmadi et al. Automated detection of driver fatigue from electroencephalography through wavelet-based connectivity
Kuo et al. Automatic stage scoring of single-channel sleep EEG based on multiscale permutation entropy
Chashmi et al. An efficient and automatic ECG arrhythmia diagnosis system using DWT and HOS features and entropy-based feature selection procedure
Harender et al. DWT based epileptic seizure detection from EEG signal using k-NN classifier
Issa et al. Automatic ECG artefact removal from EEG signals
CN111839506A (en) Mental load detection method and device
Miranda et al. Classification of EEG signals using genetic programming for feature construction
CN113723557A (en) Depression electroencephalogram classification system based on multiband time-space convolution network
Wang et al. Multiband decomposition and spectral discriminative analysis for motor imagery BCI via deep neural network
Saini et al. Discriminatory features based on wavelet energy for effective analysis of electroencephalogram during mental tasks
TaghiBeyglou et al. ADHD diagnosis in children using common spatial pattern and nonlinear analysis of filter banked EEG
Mohi-ud-Din et al. Detection of Autism Spectrum Disorder from EEG signals using pre-trained deep convolution neural networks
Zhang et al. Two-stage intelligent multi-type artifact removal for single-channel EEG settings: A GRU autoencoder based approach
Dehzangi et al. Driver distraction detection using MEL cepstrum representation of galvanic skin responses and convolutional neural networks
CN111671421A (en) Electroencephalogram-based children demand sensing method
Akuthota et al. EEG based Motor Imagery BCI using Four Class Iterative Filtering & Four Class Filter Bank Common Spatial Pattern

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20210108