CN115886735A - Severe encephalitis patient epileptic seizure automatic checkout device based on brain network optimization - Google Patents
Severe encephalitis patient epileptic seizure automatic checkout device based on brain network optimization Download PDFInfo
- Publication number
- CN115886735A CN115886735A CN202211595847.XA CN202211595847A CN115886735A CN 115886735 A CN115886735 A CN 115886735A CN 202211595847 A CN202211595847 A CN 202211595847A CN 115886735 A CN115886735 A CN 115886735A
- Authority
- CN
- China
- Prior art keywords
- network
- layer
- electroencephalogram
- characteristic
- brain
- 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
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention provides a brain network optimization-based automatic detection device for epileptic seizure of a patient with severe encephalitis, which is characterized by comprising an acquisition module, a preprocessing module and a detection module, wherein the acquisition module acquires original electroencephalogram signals to be detected of the patient with encephalitis; the preprocessing module is used for preprocessing the acquired electroencephalogram signal to be detected and inputting the preprocessed electroencephalogram signal to the recognition model in the detection module to obtain a detection result of epileptic seizure; the recognition model in the detection module integrates fusion of a plurality of brain networks and a plurality of characteristics, and the relation among all nodes in the brain networks is reflected more comprehensively from a plurality of relevant dimensions; and the optimization of the network layer and the characteristic layer based on the improved genetic algorithm realizes the optimal determination of a plurality of network weighting coefficients and a plurality of network weighting coefficients, and the optimal network layer and the optimal characteristic layer are used as machine learning input, so that the accuracy of epilepsy recognition is greatly improved.
Description
Technical Field
The invention relates to the field of computers, in particular to a severe encephalitis patient seizure automatic detection device based on brain network optimization.
Background
Encephalitis is a serious disease of neurological dysfunction caused by brain parenchymal inflammation, the incidence rate of encephalitis is about 1.5-10.5/10 ten thousand, wherein the incidence rate of encephalitis accounts for 37% -94.2% of the total incidence rate, the encephalitis is a frequently encountered disease and a common disease of central nervous system infection of children, severe encephalitis is often accompanied by disturbance of consciousness, epileptic seizure and the like, and can endanger life in serious cases. Electroencephalography (EEG) diagnoses epilepsy by tracking the electrical activity of the brain to record brain wave patterns. To detect seizures at this stage, a human expert needs to visually mark long-term electroencephalographic recordings. Which is a very cumbersome, time-consuming and costly task. EEG signals contain very fluctuating information about the functional behaviour of the brain, and their application in epilepsy detection remains a challenging problem. Therefore, it is very necessary to develop an automated technique for detecting epileptic seizures in patients with severe encephalitis.
With the wide application of the complex network theory, the application in the field of electroencephalogram is more and more. A common method is to calculate Mutual Information (MI) between EEG channels. The complex network constructed by the mutual information can further analyze the connectivity, network characteristics and other characteristics of the network, thereby identifying and analyzing mental diseases such as epilepsy and schizophrenia. Besides MI, pearson Correlation Coefficient (PCC) is also a common network Correlation metric. In addition, spearman rank correlation coefficients, normalized alignment correlation information and displacement misalignment Index (PDI) have also been proposed for the construction of brain networks. Because different correlation coefficients reflect different characteristics among network nodes, a network constructed by a single correlation coefficient has unicity, different identification results can be generated when the network is constructed by adopting different correlation coefficients, and the problems of one-sided identification, large limitation, low identification accuracy and the like exist.
Disclosure of Invention
The invention provides the automatic detecting device for the epileptic seizure of the severe encephalitis patient based on brain network optimization, which has high identification accuracy and is used for overcoming the defects of the prior art.
In order to achieve the aim, the invention provides an automatic detecting device for the epileptic seizure of a patient with severe encephalitis based on brain network optimization, which comprises an acquisition module, a preprocessing module and a detection module, wherein the acquisition module acquires original electroencephalogram signals to be detected of the patient with encephalitis; the preprocessing module is used for preprocessing the acquired electroencephalogram signal to be detected and inputting the preprocessed electroencephalogram signal to the recognition model in the detection module to obtain a detection result of epileptic seizure; the recognition model in the detection module is obtained by training in the following way:
acquiring original electroencephalogram signal samples of multiple patients with severe encephalitis and labeling epileptic seizure sections of each channel in each sample;
preprocessing each electroencephalogram signal sample to obtain a plurality of subband signals of different frequency bands;
constructing a multilayer weighted brain network and optimizing the multilayer brain network based on an improved genetic algorithm, the steps comprising: constructing a multilayer weighted brain network based on a plurality of correlation coefficients respectively, wherein each brain network corresponds to a weighting coefficient, and the network weighting coefficients of the plurality of brain networks form a network layer; dividing epileptic seizure segments in an electroencephalogram signal sample into electroencephalogram segments with certain lengths, wherein each electroencephalogram segment is provided with a plurality of sub-band signals correspondingly; extracting network characteristics of a multilayer weighted brain network on different sub-band signals, carrying out weighted average on the characteristics corresponding to all the sub-band signals to obtain a plurality of segment characteristics, wherein the characteristic weighting coefficient of each segment characteristic forms a characteristic layer corresponding to the network layer; respectively and independently optimizing a plurality of network weighting coefficients of a network layer and a plurality of characteristic weighting coefficients of a characteristic layer by adopting an improved genetic algorithm;
multiplying the optimized network layer weighting coefficient by the corresponding characteristic layer weighting coefficient according to the characteristic layer and the network layer position to which each characteristic belongs to construct a characteristic vector of each electroencephalogram segment; and taking the plurality of segment features and the corresponding plurality of feature vectors as input to train a random forest classification model so as to obtain an epilepsy detection model.
According to one embodiment of the invention, the relationship of channels in the electroencephalogram signal is quantified by three correlation coefficients, namely mutual information, pearson correlation coefficients, normalized permutation mutual information and displacement dislocation indexes, respectively, so as to construct a three-layer brain network.
According to an embodiment of the present invention, the plurality of features of each subband signal include weighting degree, clustering coefficient, harmonic centrality, modularity, tight centrality, and feature vector centrality.
According to an embodiment of the invention, when the random forest classification model is trained to obtain the epilepsy recognition result, the feature vector of each electroencephalogram segment is constructed based on the optimized three network layer weighting coefficients and six feature layer weighting coefficients to form the input of the random forest model, and the feature vector of each electroencephalogram segment comprises eighteen vector elements.
According to an embodiment of the present invention, in each layer of brain network, the weighting degree k is i The importance degree of the node i in the brain network is characterized, and the formula is as follows:
wherein M is the set of all nodes in each brain network, and A is an adjacency matrix; a. The i,j Representing the relationship of edges between the node i and the node j for the elements in the adjacency matrix A; omega i,j Is the weight between the i and j nodes;
clustering coefficient c i Reflecting the degree to which nodes tend to cluster together, the formula is as follows:
wherein k is i Is the importance of node i in the network; d is a radical of i Is the degree of node i; the j and h nodes represent the other two vertices of the triplet with node i, d p The degree of the p-th node is used for traversing all the nodes;
harmonious centrality hc i And measuring the difficulty of the node reaching other nodes, wherein the formula is as follows:
where n is the total number of nodes in the network, dist i，j Representing the shortest path between the i and j nodes;
tight centrality clos i The formula (c) is as follows:
where M is the set of all nodes in each brain network, dist i，j Representing the shortest path between the i and j nodes;
the modularity Q represents the strength of the nodes divided into groups and between groups, and its formula:
wherein C is i Is the cluster name of the inode, δ (C) i C j ) When the i node and the j node belong to the same cluster, the node is 1, otherwise, the node is 0;k i is the importance of node i in the network; k is a radical of j Is the importance of node j in the network;
the feature vector centrality reflects the importance of the neighbor nodes, and the formula is as follows:
Ae＝λe
wherein e = [ e = 1 ，…，e i ，…e n ]And a vector composed of the characteristic vectors of all the nodes in a centrality mode, wherein A is an adjacent matrix, and lambda is a corresponding characteristic value of the adjacent matrix.
According to an embodiment of the invention, when the preprocessing module preprocesses the electroencephalogram signal to be detected and the electroencephalogram signal sample:
filtering frequency components above 48Hz and below 1Hz by using a band-pass filter;
and carrying out five-layer wavelet packet decomposition on the filtered electroencephalogram signals to obtain six sub-band signals of 1-4Hz,4-8Hz,8-12Hz,12-16Hz,16-24Hz and 24-32 Hz.
According to an embodiment of the present invention, the step of independently optimizing the network layer and the feature layer based on the improved genetic algorithm comprises:
setting initial population size, evolution algebra, cross probability and mutation probability;
randomly generating a plurality of network weighting coefficients of a network layer and a plurality of characteristic weighting coefficients of a characteristic layer;
performing weighting coefficient optimization of a network layer to select a plurality of network weighting coefficients in a generation with the highest accuracy; setting a constraint condition that a plurality of network weighting coefficients in a network layer are added to be 1, wherein a fitness function is the accuracy of a classification test set of a maximized random forest classification model, and optimizing the plurality of network weighting coefficients through crossing, recombination and variation operations of a genetic algorithm;
performing coefficient optimization of the characteristic layer to select a weighting coefficient in a generation with the highest accuracy; setting a constraint condition that a plurality of characteristic weighting coefficients in a characteristic layer are added to be 1, setting a fitness function to maximize the accuracy of a random forest classification model classification test set, and performing weight optimization through crossing, recombination and variation operations of a genetic algorithm;
and repeating the steps until the classification accuracy of the test set is not changed any more, and finishing the optimization of the genetic algorithm.
According to one embodiment of the invention, when the acquisition module acquires electroencephalogram signals of a plurality of patients with severe encephalitis, the sampling frequency is 256Hz, the electrode distribution adopts the international 10-20 electroencephalogram acquisition standard, and 20 channels of electroencephalogram data are acquired in total.
In conclusion, the device for automatically detecting the epileptic seizure of the severe encephalitis patient based on brain network optimization, provided by the invention, has the advantages that the recognition model in the detection module integrates fusion of a plurality of brain networks and a plurality of characteristics, and the relation among all nodes in the brain networks is reflected more comprehensively from a plurality of relevant dimensions; and the optimization of the network layer and the characteristic layer based on the improved genetic algorithm realizes the optimal determination of a plurality of network weighting coefficients and a plurality of network weighting coefficients, and the optimal network layer and the optimal characteristic layer are used as machine learning input, so that the accuracy of epilepsy recognition is greatly improved.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
Fig. 1 is a schematic structural diagram of an automatic detecting device for detecting epileptic seizure of a severe encephalitis patient based on brain network optimization according to an embodiment of the present invention.
FIG. 2 is a schematic flow chart illustrating training of recognition models in the detection module of FIG. 1.
Fig. 3 is a schematic flow chart of step S30 in fig. 2.
FIG. 4 is a schematic flow chart showing optimization of network layer and feature layer based on improved genetic algorithm.
Detailed Description
As shown in fig. 1, the present embodiment provides an automatic detecting apparatus for detecting epileptic seizure of a severe encephalitis patient based on brain network optimization, which includes an acquisition module 10, a preprocessing module 20, and a detection module 30. The acquisition module 10 acquires an original electroencephalogram signal to be detected of a encephalitis patient; specifically, the acquisition module 10 samples electroencephalogram signals of a patient with severe encephalitis based on electrode distribution by adopting an international 10-20 electroencephalogram acquisition standard, the sampling frequency is 256Hz, and 20 channels of electroencephalogram data are acquired in total. After acquiring the original electroencephalogram signals of the encephalitis patient, preprocessing the signals by a preprocessing module 20; specifically, the preprocessing module 20 filters frequency components above 48Hz and below 1Hz in the original electroencephalogram signal by using a band-pass filter; carrying out five-layer wavelet packet decomposition on the EEG signals obtained after filtering to respectively obtain six sub-band signals of 1-4Hz,4-8Hz,8-12Hz,12-16Hz,16-24Hz and 24-32 Hz; however, the present invention does not limit the division of the preprocessed subband signals. The six subband signals are input to the detection module 30 to obtain the detection result of the epileptic seizure.
In this embodiment, a recognition model for automatically detecting an epileptic seizure is stored in the detection module 30, and the recognition model is obtained by training, and the specific training steps include: obtaining a plurality of original electroencephalogram signal samples of patients with severe encephalitis and labeling the epileptic seizure section of each channel in each sample (step S10). Each electroencephalogram signal sample is preprocessed to obtain a plurality of subband signals of different frequency bands (step S20). A multi-layer weighted brain network is constructed and optimized based on the improved genetic algorithm (step S30). Training a random forest model based on the network layer weighting coefficient and the characteristic layer weighting coefficient of the multilayer network optimized by the improved genetic algorithm to realize epileptic seizure result classification (step S40); specifically, according to the position of a feature layer and a network layer to which each feature belongs, multiplying an optimized network layer weighting coefficient by a corresponding feature layer weighting coefficient to construct a feature vector of each electroencephalogram segment; and taking the plurality of segment features and the corresponding plurality of feature vectors as input to train a random forest classification model so as to obtain an epilepsy detection model.
Training of the recognition model begins at step S10, in which long-range monitored raw brain signals of a large number of patients, each of which is different in gender and age, are collected as brain signal samples using a multi-lead electroencephalograph. Wherein, the sampling frequency is 256Hz, the electrode distribution adopts the international 10-20 electroencephalogram acquisition standard, and 20 channels of electroencephalogram data are acquired in total. And marking the epileptic seizure segments of each channel on a large number of acquired electroencephalogram signal samples to form a sample database. And then, executing the step S20, and preprocessing the original electroencephalogram signal sample marked in the sample database, wherein the preprocessing step is the same as that in the preprocessing module 20, namely, a band-pass filter is adopted to filter interference signals and then frequency domain sub-band signal division is carried out to obtain six sub-band signals.
After the six subband signals are obtained, step S30 is executed to construct a multilayer weighted brain network and perform feature extraction on each layer of the network based on the six subband signals to form a plurality of feature weighting parameters, and then the weighting coefficients of the network layer and the feature layer are optimized based on a genetic optimization algorithm. Specifically, the method comprises the following steps:
step S31: and constructing a multilayer weighted brain network based on the plurality of correlation coefficients respectively, wherein each brain network is provided with a weighting coefficient correspondingly, and the network weighting coefficients of the plurality of brain networks form a network layer. Specifically, the relationship of channels in the electroencephalogram signals is quantified by three correlation coefficients, namely mutual information, pearson correlation coefficients, standardized arrangement mutual information and displacement dislocation indexes, so as to construct a three-layer weighted brain network. Each layer of network corresponds to a network weighting coefficient which is respectively represented by K1, K2 and K3, and the three network weighting coefficients form the network layer weighting coefficient.
And then, executing a step S32, and dividing the epileptic seizure segments in the electroencephalogram signal sample into electroencephalogram segments with certain lengths, wherein each electroencephalogram segment is corresponding to a plurality of sub-band signals. Specifically, dividing an epileptic seizure segment marked in an electroencephalogram signal sample into a plurality of 4s electroencephalogram segments by using a sliding window method, and marking the electroencephalogram segments obtained by segmentation as x (N), wherein N =1,2, \8230, and N; in this embodiment, the sampling frequency of the electroencephalogram signal samples is 256Hz, so that N =1024. Based on the preprocessing of step S20, there will also be six subband signals within each computer slice.
And step S33, extracting network characteristics of the multilayer weighted brain network on different sub-band signals, carrying out weighted average on the characteristics corresponding to all the sub-band signals to obtain a plurality of segment characteristics, and forming a characteristic layer corresponding to the network layer by the characteristic weighting coefficient of each segment characteristic. In this embodiment, the plurality of characteristics of each subband signal includes a weighting k i Cluster coefficient c i Harmonizing center property hc i Modularity Q, tight centrality close i And six network features of feature vector centrality e. For the first sub-band of 1-4Hz, the weighting degree k of the first sub-band is obtained after extraction i First sub-band clustering coefficient c i First subband harmonic centrality hc i A first sub-band modularity Q, a first sub-band tight centrality close i And a first subband feature vector centrality e; for the second sub-band of 4-8Hz, the weighting degree k of the second sub-band is obtained after extraction i Second subband clustering coefficient c i Second subband harmonic centrality hc i Second sub-band modularity Q, second sub-band tight centrality close i And a second subband feature vector centrality e. Correspondingly, six features corresponding to each sub-band can be obtained in sequence.
Weighting the first sub-band by a weight k i、 Second sub-band weighting degree k i 823060, 823080, sixth weighting degree k i Obtaining the weighting degree k of the electroencephalogram segment after weighted average i And a corresponding characteristic weighting coefficient H1; clustering the first sub-band into coefficients c i Second sub-band clustering coefficient c i 8230a sixth sub-band clustering coefficient c i Obtaining the clustering coefficient c of the electroencephalogram segment after weighted average i And a corresponding characteristic weighting coefficient H2; the concordance centrality hc of the electroencephalogram segment can be obtained by analogy i (corresponding feature weighting factor H3), modularity Q (corresponding feature weighting factor H4), closeness to centrality close i (corresponding feature weighting factor H5) and feature vector centrality e (corresponding feature weighting factor H6). The multilayer weighted brain network corresponds to six characteristic weighting coefficients (H1, H2, H3, H4, H5 and H6) on each electroencephalogram segment.
For six network features, in each layer of brain network, the weighting degree k i The importance degree of the node i in the brain network is characterized, and the formula is as follows:
wherein the content of the first and second substances,m is the set of all nodes in each brain network, A is an adjacent matrix; a. The i,j Representing the relationship of edges between the node i and the node j for the elements in the adjacency matrix A; omega i,j Is the weight between the i and j nodes;
clustering coefficient c i Reflecting the degree to which nodes tend to cluster together, the formula is as follows:
wherein k is i Is the importance of node i in the network; d i Is the degree of node i; the j and h nodes represent the other two vertices of the triplet with node i, d p The degree of the p-th node is used for traversing all the nodes;
harmonious centrality hc i And measuring the difficulty of the node reaching other nodes, wherein the formula is as follows:
where n is the total number of nodes in the network, dist i,j Representing the shortest path between the i and j nodes;
tight centrality clos i The formula of (1) is as follows:
where M is the set of all nodes in each brain network, dist i,j Representing the shortest path between the i and j nodes;
the modularity Q represents the strength of the nodes divided into groups and between groups, and its formula:
wherein C i Is the cluster name of the inode, δ (C) i C j ) When the i and j nodes belong to the same cluster, the node is 1, otherwise the node is 0; k is a radical of i Is the importance of node i in the network; k is a radical of j Is the importance of node j in the network;
the feature vector centrality reflects the importance of the neighbor nodes, and the formula is as follows:
Ae＝λe
wherein e = [ e = 1 ，…，e i ，…e n ]And a vector composed of the characteristic vectors of all the nodes in a centrality mode, wherein A is an adjacent matrix, and lambda is a corresponding characteristic value of the adjacent matrix.
The recognition result of the random forest classification model is related to the extracted features on each segment and the weighting coefficient of the segment. As expressed by the above equation, the computation of the features is related to the adjacency matrix of the brain network, which is determined by the features of the brain electrical signals. In order to improve the identification accuracy, related weighting coefficients need to be determined; the weighting coefficients of the segments are divided into network layers according to different network structures, and the network layer weighting coefficients have three network weighting coefficients; the characteristic layer is divided into characteristic layers according to different characteristics, and the characteristic layers are provided with six characteristic weighting coefficients.
In this embodiment, in step S34, an improved genetic algorithm is used to independently optimize the network layer weighting coefficient and the feature layer weighting coefficient, and the optimal network layer weighting coefficient and the optimal feature layer weighting coefficient are obtained by combining the prediction of the random forest model on the test set in step S40. Specifically, the method comprises the following steps:
step S341: setting initial population size, evolution algebra, cross probability and mutation probability;
step S342: randomly generating a plurality of network weighting coefficients of a network layer and a plurality of characteristic weighting coefficients of a characteristic layer;
step S343: and performing weighting coefficient optimization of the network layer to select a plurality of network weighting coefficients in the generation with the highest accuracy. Setting a constraint condition that a plurality of network weighting coefficients in a network layer are added to be 1, wherein a fitness function is the accuracy of the random forest model classification test set in the maximization step S40, and optimizing the plurality of network weighting coefficients through the crossing, recombination and variation operations of a genetic algorithm;
step S344: coefficient optimization of the feature layer is performed to select the weighting coefficients in the generation with the highest accuracy. Setting a constraint condition that a plurality of characteristic weighting coefficients in the characteristic layer are added to be 1, wherein a fitness function is the accuracy of the random forest model classification test set in the maximization step S40, and carrying out weight optimization through crossing, recombination and variation operations of a genetic algorithm;
and repeating the steps S342 to S344 until the accuracy of the test set does not change any more, finishing the optimization of the genetic algorithm and obtaining the optimized network layer weighting coefficient and the optimized characteristic layer weighting coefficient.
After obtaining the network layer weighting coefficients and the feature layer weighting coefficients of the multilayer network optimized based on the improved genetic algorithm in step S30, executing step S40: training a random forest model to realize epileptic seizure result classification. Specifically, the method comprises the following steps:
step S41: and multiplying the optimized network layer weighting coefficient by the corresponding characteristic layer weighting coefficient according to the characteristic layer and the network layer position to which each characteristic belongs to construct the characteristic vector of each electroencephalogram segment. Specifically, a feature vector of each electroencephalogram segment is constructed based on the optimized three network layer weighting coefficients and six feature layer weighting coefficients, and the feature vector KH of each electroencephalogram segment contains eighteen vector elements, KH = [ K1 × H1, K1 × H2, K1 × H3, K1 × H4, K1 × H5, K1 × H6, K2 × H1, K2 × H2, K2 × H3, K2 × H4, K2 × H5, K2 × H6, K3 × H1, K3 × H2, K3 × H3, K3 × H4, K3 × H5, K3 × H6].
Step S42: randomly dividing a plurality of electroencephalogram segments of a plurality of electroencephalogram signal samples into a training set and a testing set, multiplying a plurality of network characteristics of each electroencephalogram segment in the training set with corresponding characteristic vectors, and inputting the multiplied network characteristics into a random forest classifier to train a plurality of decision trees in the random forest classifier, so as to form a random forest model. And multiplying a plurality of network characteristics of each electroencephalogram segment in the test set by corresponding characteristic vectors, inputting the multiplied network characteristics into the trained random forest model to obtain a test result of epileptic seizure detection, and determining the trained random forest model based on the test result. The random forest classifier adopts integration of a plurality of weak classifiers and obtains classification results through voting, a sub data set with the size equal to that of an original data set is constructed by adopting a sampling structure with a back-put function, and then a sub decision tree is constructed by utilizing the sub data set, namely each decision tree obtains one classification result according to the sub data set. And finally, voting all results of the sub-decision tree to obtain the output of the random forest classifier. Where each decision tree has the same distribution but the classification capability depends on the data entered.
To this end, the recognition model based on the multi-layer network structure and the random forest model in the detection module 30 is trained. When the encephalitis patient epileptic seizure is automatically detected, the electroencephalogram signals processed by the preprocessing module 20 are input to the detection module 30. And after signal segmentation, segment feature extraction is carried out, the extracted segment features are associated with the optimized network layer correlation coefficient and the feature layer correlation coefficient and then input into a random forest model, and after the random forest model is classified, a seizure detection result is obtained so as to realize automatic detection.
In this embodiment, the apparatus for automatically detecting epileptic seizure of a severe encephalitis patient based on brain network optimization further comprises an optimization training module 40, wherein the optimization training module performs multi-layer network construction, double-layer weighting coefficient optimization and random forest model training to obtain an identification model in the detection module 30; the training module 40 includes an improved genetic algorithm optimization multi-layer network unit 43 and a model training unit 44. The training steps S10 and S20 can be implemented based on the acquisition module 10 and the preprocessing module 20 in the automatic detecting device for epileptic seizure of a severe encephalitis patient based on brain network optimization provided in this embodiment. The improved genetic algorithm optimized multi-layer network unit 43 executes S30 in the training step, and the model training unit 44 executes S30 in the training step, which is not described in detail herein. However, the present invention is not limited thereto. In other embodiments, the recognition model in the detection module can also be obtained by training through an external computer device.
In conclusion, the device for automatically detecting the epileptic seizure of the severe encephalitis patient based on brain network optimization, provided by the invention, has the advantages that the recognition model in the detection module integrates fusion of a plurality of brain networks and a plurality of characteristics, and the relation among all nodes in the brain networks is reflected more comprehensively from a plurality of relevant dimensions; and the optimization of the network layer and the characteristic layer based on the improved genetic algorithm realizes the optimal determination of a plurality of network weighting coefficients and a plurality of network weighting coefficients, and the optimal network layer and the optimal characteristic layer are used as machine learning input, so that the accuracy of epilepsy recognition is greatly improved.
Although the present invention has been described with reference to the preferred embodiments, it should be understood that various changes and modifications can be made therein by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (8)
1. An automatic detecting device for the epileptic seizure of a severe encephalitis patient based on brain network optimization is characterized by comprising an acquisition module, a preprocessing module and a detection module, wherein the acquisition module acquires original electroencephalogram signals to be detected of the encephalitis patient; the preprocessing module preprocesses the acquired electroencephalogram signals to be detected and inputs the preprocessed electroencephalogram signals to the recognition model in the detection module to obtain a detection result of epileptic seizure; the recognition model in the detection module is obtained by training in the following way:
acquiring original electroencephalogram signal samples of multiple patients with severe encephalitis and labeling epileptic seizure sections of each channel in each sample;
preprocessing each electroencephalogram signal sample to obtain a plurality of subband signals of different frequency bands;
constructing a multilayer weighted brain network and optimizing the multilayer brain network based on an improved genetic algorithm, wherein the steps comprise: constructing a multilayer weighted brain network based on a plurality of correlation coefficients respectively, wherein each brain network corresponds to a weighting coefficient, and the network weighting coefficients of the plurality of brain networks form a network layer; dividing epileptic seizure sections in the electroencephalogram signal sample into electroencephalogram segments with certain lengths, wherein each electroencephalogram segment corresponds to a plurality of sub-band signals; extracting network characteristics of a multilayer weighted brain network on different sub-band signals, carrying out weighted average on the characteristics corresponding to all the sub-band signals to obtain a plurality of segment characteristics, wherein the characteristic weighting coefficient of each segment characteristic forms a characteristic layer corresponding to the network layer; respectively and independently optimizing a plurality of network weighting coefficients of a network layer and a plurality of characteristic weighting coefficients of a characteristic layer by adopting an improved genetic algorithm;
multiplying the optimized network layer weighting coefficient by the corresponding characteristic layer weighting coefficient according to the characteristic layer and the network layer position to which each characteristic belongs to construct a characteristic vector of each electroencephalogram segment; and taking the plurality of segment features and the corresponding plurality of feature vectors as input to train a random forest classification model so as to obtain an epilepsy detection model.
2. The brain network optimization-based automatic detecting device for epileptic seizure of patients with severe encephalitis according to claim 1, characterized in that the relationship of the channels in the brain electrical signals is quantified by three correlation coefficients, namely mutual information, pearson correlation coefficient, normalized permutation mutual information and displacement dislocation index, respectively, to construct a three-layer brain network.
3. The automatic severe encephalitis patient seizure detection apparatus based on brain network optimization according to claim 1 or 2, wherein the plurality of features of each subband signal includes weighting degree, clustering coefficient, harmonic centrality, modularity, close centrality and feature vector centrality.
4. The brain network optimization-based automatic severe encephalitis patient seizure detection device according to claim 3, wherein when training the random forest classification model to obtain the seizure recognition result, the feature vector of each electroencephalogram segment is constructed based on the optimized three network layer weighting coefficients and six feature layer weighting coefficients to form the input of the random forest model, and the feature vector of each electroencephalogram segment contains eighteen vector elements.
5. The brain network optimization-based automatic detecting device for epileptic seizure of severe encephalitis patients according to claim 3, wherein said weighting degree k is in each layer of brain network i The importance degree of the node i in the brain network is characterized, and the formula is as follows:
wherein M is the set of all nodes in each brain network, and A is an adjacency matrix; ai, j are elements in the adjacency matrix A and represent the relationship of edges between the node i and the node j; ω i, j is the weight between the i and j nodes;
clustering coefficient c i Reflecting the degree to which nodes tend to cluster together, the formula is as follows:
wherein k is i Is the importance of node i in the network; d is a radical of i Degree of node i; the j and h nodes represent the other two vertices of the triplet with node i, d p The degree of the p-th node is used for traversing all nodes;
mediation centerSex hc i And measuring the difficulty of the node to reach other nodes, wherein the formula is as follows:
where n is the total number of nodes in the network, dist i,j Representing the shortest path between the i and j nodes;
tight centrality clos i The formula of (1) is as follows:
where M is the set of all nodes in each brain network, dist i,j Representing the shortest path between the i and j nodes;
the modularity Q represents the strength of the nodes divided into groups and between the groups, and the formula is as follows:
wherein C i Is the cluster name of the inode, δ (C) i C j ) When the i and j nodes belong to the same cluster, the node is 1, otherwise the node is 0; k is a radical of formula i Is the importance of node i in the network; k is a radical of j Is the importance of node j in the network;
the feature vector centrality reflects the importance of the neighbor nodes, and the formula is as follows:
Ae＝λe
wherein e = [ e = [ e ] 1 ，…，e i ，…e n ]And a vector composed of the characteristic vectors of all the nodes in a centrality mode, wherein A is an adjacent matrix, and lambda is a corresponding characteristic value of the adjacent matrix.
6. The brain network optimization-based automatic detecting device for epileptic seizure of patients with severe encephalitis according to claim 1, wherein the preprocessing module, when preprocessing the electroencephalogram signal and electroencephalogram signal samples to be detected:
filtering frequency components above 48Hz and below 1Hz by using a band-pass filter;
and carrying out five-layer wavelet packet decomposition on the filtered electroencephalogram signals to obtain six sub-band signals of 1-4Hz,4-8Hz,8-12Hz,12-16Hz,16-24Hz and 24-32 Hz.
7. The brain network optimization-based automatic detecting device for epileptic seizure of a severe encephalitis patient according to claim 1, wherein the step of optimizing the network layer and the feature layer independently based on the improved genetic algorithm respectively comprises:
setting initial population size, evolution algebra, cross probability and mutation probability;
randomly generating a plurality of network weighting coefficients of a network layer and a plurality of characteristic weighting coefficients of a characteristic layer;
performing weighting coefficient optimization of a network layer to select a plurality of network weighting coefficients in a generation with the highest accuracy; setting a constraint condition that a plurality of network weighting coefficients in a network layer are added to be 1, wherein a fitness function is used for maximizing the accuracy of a random forest classification model classification test set, and optimizing the plurality of network weighting coefficients through the crossing, recombination and variation operations of a genetic algorithm;
performing coefficient optimization of the characteristic layer to select a weighting coefficient in a generation with the highest accuracy; setting a constraint condition that a plurality of characteristic weighting coefficients in a characteristic layer are added to be 1, wherein a fitness function is used for maximizing the accuracy of a random forest classification model classification test set, and weight optimization is carried out through crossover, recombination and variation operations of a genetic algorithm;
and repeating the steps until the classification accuracy of the test set is not changed any more, and finishing the optimization of the genetic algorithm.
8. The device for automatically detecting the epileptic seizure of the patients with severe encephalitis based on brain network optimization according to claim 1, wherein the sampling frequency of the acquisition module when acquiring the electroencephalogram signals of a plurality of patients with severe encephalitis is 256Hz, the electrode distribution adopts the international 10-20 electroencephalogram acquisition standard, and 20 channels of electroencephalogram data are acquired in total.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211595847.XA CN115886735A (en) | 2022-12-12 | 2022-12-12 | Severe encephalitis patient epileptic seizure automatic checkout device based on brain network optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211595847.XA CN115886735A (en) | 2022-12-12 | 2022-12-12 | Severe encephalitis patient epileptic seizure automatic checkout device based on brain network optimization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115886735A true CN115886735A (en) | 2023-04-04 |
Family
ID=86477859
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211595847.XA Pending CN115886735A (en) | 2022-12-12 | 2022-12-12 | Severe encephalitis patient epileptic seizure automatic checkout device based on brain network optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115886735A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116522210A (en) * | 2023-07-03 | 2023-08-01 | 中国医学科学院生物医学工程研究所 | Motor imagery electroencephalogram signal classification method based on brain network difference analysis |
-
2022
- 2022-12-12 CN CN202211595847.XA patent/CN115886735A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116522210A (en) * | 2023-07-03 | 2023-08-01 | 中国医学科学院生物医学工程研究所 | Motor imagery electroencephalogram signal classification method based on brain network difference analysis |
| CN116522210B (en) * | 2023-07-03 | 2023-09-01 | 中国医学科学院生物医学工程研究所 | Motor imagery electroencephalogram signal classification method based on brain network difference analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106886792B (en) | Electroencephalogram emotion recognition method for constructing multi-classifier fusion model based on layering mechanism | |
| CN111956221B (en) | Temporal lobe epilepsy classification method based on wavelet scattering factor and LSTM neural network model | |
| Costa et al. | Epileptic seizure classification using neural networks with 14 features | |
| CN112641451B (en) | Multi-scale residual error network sleep staging method and system based on single-channel electroencephalogram signal | |
| Supakar et al. | A deep learning based model using RNN-LSTM for the Detection of Schizophrenia from EEG data | |
| CN110584596B (en) | Sleep stage classification method based on dual-input convolutional neural network and application thereof | |
| CN115919330A (en) | EEG Emotional State Classification Method Based on Multi-level SE Attention and Graph Convolution | |
| US20230108916A1 (en) | Method and system for forecasting non-stationary time-series | |
| CN113712571A (en) | Abnormal electroencephalogram signal detection method based on Rinyi phase transfer entropy and lightweight convolutional neural network | |
| CN115886735A (en) | Severe encephalitis patient epileptic seizure automatic checkout device based on brain network optimization | |
| CN107045624B (en) | Electroencephalogram signal preprocessing and classifying method based on maximum weighted cluster | |
| CN116226710A (en) | Electroencephalogram signal classification method and parkinsonism detection device | |
| Liu et al. | Automatic sleep arousals detection from polysomnography using multi-convolution neural network and random forest | |
| Gnana Rajesh | Analysis of MFCC features for EEG signal classification | |
| KR102298709B1 (en) | Device and method for learning connectivity | |
| CN108021873B (en) | Electroencephalogram signal epilepsy classification method and system for clustering asymmetric mutual information | |
| CN110443276A (en) | Time series classification method based on depth convolutional network Yu the map analysis of gray scale recurrence | |
| CN115270847A (en) | Design decision electroencephalogram recognition method based on wavelet packet decomposition and convolutional neural network | |
| Ulhaq et al. | Epilepsy seizures classification with EEG signals: A machine learning approach | |
| Abenna et al. | Alcohol use disorders automatic detection based BCI systems: a novel EEG classification based on machine learning and optimization algorithms | |
| Iscan | Mlsp competition, 2010: description of second place method | |
| Nagwanshi et al. | Detection of Epilepsy patients using coot optimization based feed forward multilayer neural network | |
| CN114287908A (en) | Brain connection classification method with multiple band convolution fusion | |
| Kaur et al. | Digital Fiat Currency (DFC): A Taxonomy for Automatic Sleep Stage Classification | |
| CN115146666A (en) | EEG epileptic seizure detection algorithm based on automatic machine learning |
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 |