Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, completeSite preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based onEmbodiment in the present invention, those skilled in the art's every other implementation obtained without creative effortsExample, shall fall within the protection scope of the present invention.
The embodiment of the present application provides a kind of eeg signal detection method, relevant device (such as eeg signal detection deviceWith electronic equipment etc.) and storage medium.The eeg signal detection device can integrate in the electronic device, which canTo be server, it is also possible to the Medical Devices such as terminal, such as electroencephalograph, personal computer (PC, PersonalComputer), in the equipment such as tablet computer or laptop.
For example, as shown in Figure 1, electronic equipment can acquire brain wave of the object to be detected under interference signal discrete stimulusSignal (for example including the electroencephalogram of eeg signal) then obtains the test object dry according to the eeg signal respectivelyThe brain area link information under stimulation and the brain area link information under non-interference stimulation are disturbed, and calculating should be under interference stimulationBrain area link information and non-interference stimulation under brain area link information between similarity, subsequently, according to the similarityIt determines histological type belonging to the eeg signal, for example determines whether to belong to anxiety disorder, or determine whether to belong to melancholia,Etc., and then the testing result of the test object can be generated based on the histological type.
Wherein, object to be detected, which can be, any has large-brained biology, such as people, monkey, cat or dog etc.；In order to retouchIt states conveniently, in the embodiment of the present application, will be illustrated taking human as example.
In addition, in the embodiment of the present application, so-called interference signal refers to can be to biology (object to be detected or detection sampleThis etc.) the extraneous things that has an impact of brain wave, and stimulate and then refer to being biology (object to be detected or detection hereinSample etc.) feel and its cranial nerve cell is caused to react.Wherein, cranial nerve cell can generate accordingly when reactingElectric wave, these electric waves are known as brain wave, are acquired to brain wave and express in the form of a signal, and as brain wave is believedNumber.
For example, the test topic of some preset kinds can be provided for object to be detected by phased manner, and it is required to answer,Wherein these test topics can be considered interference signal, and collected " the object to be detected generated brain when answering of instituteElectric wave signal " can be considered the generated eeg signal in the case where the interference signal stimulates (abbreviation interference stimulation).It needs to illustrate, since the interference signal is discontinuity, so, when interference signal is not present, brain caused by the object to be detectedElectric wave signal can be considered the generated eeg signal under non-interference signal stimulus (stimulation of abbreviation non-interference).
It is described in detail separately below.It should be noted that the following description sequence is not as excellent to embodimentThe restriction of choosing sequence.
The present embodiment will be described from the angle of eeg signal detection device, and the eeg signal detection device is specificIt can integrate in the electronic equipments such as server or terminal, which may include the Medical Devices such as electroencephalograph, PC, plate electricityBrain or laptop etc..
A kind of eeg signal detection method, comprising: acquire brain electricity of the object to be detected under interference signal discrete stimulusWave signal obtains brain area link information of the test object under interference stimulation according to the eeg signal and non-respectivelyBrain area link information under interference stimulation calculates this in the brain area link information under interference stimulation and the brain under non-interference stimulationSimilarity between area's link information, according to the similarity determine the eeg signal belonging to histological type, be based on the pathologyType generates the testing result of the test object.
As shown in Fig. 2, the detailed process of the eeg signal detection method can be such that
101, eeg signal of the object to be detected under interference signal discrete stimulus is acquired.
For example, can specifically acquire object to be detected under interference signal discrete stimulus electroencephalogram (EEG,Electroencephalogram), wherein the electroencephalogram includes eeg signal.
For example, if the eeg signal detection device is integrated in electroencephalograph, at this point, can specifically pass through electroencephalogramInstrument acquires the electroencephalogram；And if the eeg signal detection device is integrated in other equipment that can not directly acquire electroencephalogramIn, then at this point it is possible to receive the electroencephalogram that other equipment such as electroencephalograph is sent, alternatively, also can receive the brain of user's inputElectrograph (electroencephalogram can acquire equipment by other electroencephalograms and be acquired), etc..
Wherein, electroencephalogram is to be put the spontaneous biotic potential of brain from scalp by accurate electronic instrumentThe figure for recording greatly and obtaining is the spontaneity and section of the cranial nerve cell group recorded by electrode (electrophysiological index)Rule property electrical activity.Electroencephalogram can recorde the variation of generated eeg signal when brain activity, be the electricity of cranial nerve cellOverall reflection of the physiological activity in cerebral cortex or scalp surface.
Optionally, acquire electroencephalogram mode can there are many, for example, Aeeg Monitoring can be passed through(ambulatoryEEG monitoring) or Video-EEG Monitoring (VEEG, Video-EEG, also known as video feedback prisonSurvey) etc. modes acquire the electroencephalogram.Wherein, Aeeg Monitoring mainly by Electroencephalogram device such as electroencephalograph come intoRow record, usually can continuously record 24 hours or so, and Video-EEG Monitoring is increased on the basis of Electroencephalogram deviceSynchronization video equipment, monitoring time can flexibly be grasped according to appointed condition and actual demand, not from a few hours to a couple of daysDeng.
Optionally, in specifically acquisition eeg signal, corresponding channel can be respectively set for different brain areas, theseChannel is known as brain area channel, it is then possible to be carried out respectively to electric wave caused by corresponding brain area by each brain area channelAcquisition, the signal in acquired multiple brain area channels is eeg signal, that is to say, that passes through this kind of mode brain electricity collectedWave signal may include the signal in multiple brain area channels.
Wherein, brain area (ER, Encephalic Region) is the abbreviation of " brain function subregion ", can be by multiple brainsPower dimension is composed, this to combine the scope that can break through medicine, that is to say, that the division of brain area is not drawing for physical feelingPoint, but division functionally, specific division mode can be depending on the demands of practical application, therefore not to repeat here.
102, according to the eeg signal obtain respectively brain area link information of the test object under interference stimulation andBrain area link information under non-interference stimulation.
For biology in activity, brain also will do it corresponding information processing simultaneously, and these information processings need differenceTherefore collaborative work and dynamic interaction between brain area will necessarily generate connection relationship between Different brain region, and these can reflectThe information of connection relationship between Different brain region is known as brain area link information in the embodiment of the present application.
Wherein, " according to the eeg signal obtain respectively brain area link information of the test object under interference stimulation, withAnd non-interference stimulation under brain area link information " mode can there are many, for example, specifically can be such that
(1) eeg signal is divided into the first kind signal segment comprising interference stimulation and is stimulated comprising non-interferenceThe second class signal segment.
For example, corresponding label can also be recorded in electroencephalogram collected other than it can recorde eeg signal,To indicate that this section of eeg signal is the signal collected under interference signal stimulation, or adopted under non-interference signal stimulusThe signal of collection；I.e. the eeg signal " is divided into the first kind signal segment comprising interference stimulation and comprising non-interference by stepStimulation the second class signal segment " may include:
The label of each signal segment in the eeg signal is obtained, if the label indication signal section is to pierce in interference signalSignal collected under swashing, then be classified as first kind signal segment for signal segment, if the label indication signal section is in non-interference signalSignal segment is then classified as the second class signal segment by signal collected under stimulation.
Optionally, for simplicity, the label that can only identify the signal collected under interference signal stimulation, and if notThere are labels, it is determined that for the signal collected under non-interference signal stimulus；For example, referring to Fig. 3, in the electroencephalogram,C342, Y88, d464 and N99 are the label (being detailed in black vertical line part) of the signal collected under interference signal stimulation.
Similarly, the label of the signal collected under non-interference signal stimulus can also be only identified, and is marked if it does not existLabel, it is determined that for signal collected under interference signal stimulation, etc..
Wherein, the notation methods of label specifically can be depending on the demand of practical application, and therefore not to repeat here.
(2) it determines the correlation in the first kind signal segment between the signal in Different brain region channel, obtains the test objectBrain area link information under interference stimulation；For example, specifically can be such that
Calculate the Pearson correlation coefficients (Pearson in first kind signal segment between the signal in Different brain region channelCorrelation coefficient), the first related coefficient is obtained, which is generated dry according to the first related coefficientDisturb the brain area link information under stimulation.
Wherein, Pearson correlation coefficients are mainly used for whether on one wire measuring two datasets conjunction, i.e. measurement spacingLinear relationship between variable.It therefore, can be by each signal (i.e. brain area electrode signal, in the application implementation in brain area channelAbbreviation signal in example) regard a data acquisition system as, for example, including 128 signals with a brain area channel, and every referring to Fig. 3For a signal may include 100,000 or more time point datas again, then at this point, each brain area channel can correspond to 128 dataTo gather, again may include 100,000 or more elements in each data acquisition system, i.e. each signal can be used as a data acquisition system,And each time point data can be used as an element.
For example, the two are different so that the signal in two Different brain region channel respectively corresponds two datasets conjunction x and y as an example(i.e. the calculation formula of Pearson correlation coefficients r between data acquisition system x and data acquisition system y) can be as the signal in brain area channelUnder:
Wherein, xiFor the element in data acquisition system x, yiElement in data acquisition system y, N are data acquisition system x and data acquisition system yIn element quantity (length of the time point data sampled), such as N can be 100,000 or more, etc..
According to the mode of above-mentioned calculating Pearson correlation coefficients r, for every two brain area channel, can obtain and its signalThe consistent Pearson correlation coefficients of quantity, such as 128 Pearson correlation coefficients, so, it can use these related coefficients constructionCorresponding correlation matrix is handled so as to subsequent；Wherein, for convenience, in the embodiment of the present application, by the benefitIt is known as the first correlation matrix with the correlation matrix that the first related coefficient constructs.I.e. step is " according to the first related coefficientGenerate brain area link information of the test object under interference stimulation " may include:
The first correlation matrix is constructed using the first related coefficient, which is converted to multiple oneThe feature vector of dimension obtains fisrt feature set.
Wherein, the feature vector in the fisrt feature set is used to reflect that brain area of the test object under interference stimulation to connectConnect information.
For example, by taking each data acquisition system includes 128 signals as an example, then at this point it is possible to construct one 128*128's of buildingThen first correlation matrix is converted to multiple one-dimensional feature vectors by the first correlation matrix, obtain the first spyCollection is closed.
Optionally, since the first correlation matrix is diagonal line symmetrical shape, in order to improve treatment effeciency, subtractFirst correlation matrix, can be divided into symmetrical two regions by the consumption of few computing resource, then, only to one of themThe element in region carries out vector stretching, is converted to one-dimensional feature vector to convert element therein.I.e. step is " by first phaseRelationship matrix number is converted to multiple one-dimensional feature vectors, obtains fisrt feature collection " may include:
First correlation matrix is divided into symmetrical two regions, is selected from two regions according to preset strategyElement in the region of selection, is converted to one-dimensional feature vector by one region, obtains fisrt feature set.
For example, as shown in figure 4, if shown in Fig. 4 being the first correlation matrix, at this point it is possible to by the first correlationCoefficient matrix is diagonally divided into symmetrical two regions, and then, the element in the region in Fig. 4 in broken line triangle is turnedIt is changed to one-dimensional feature vector, obtains fisrt feature set, etc..
(3) it determines the correlation in the second class signal segment between the signal in Different brain region channel, obtains the test objectBrain area link information under non-interference stimulation；For example, specifically can be such that
The Pearson correlation coefficients in the second class signal segment between the signal in Different brain region channel are calculated, the second correlation is obtainedCoefficient generates brain area link information of the test object under non-interference stimulation according to the second related coefficient.
Wherein, calculate Pearson correlation coefficients formula can for details, reference can be made to the calculating of the first related coefficient in (2), hereinIt does not repeat.
According to the mode of above-mentioned calculating Pearson correlation coefficients r, for every two brain area channel, can obtain and its signalThe consistent Pearson correlation coefficients of quantity, such as 128 Pearson correlation coefficients, so, it can use these related coefficients constructionCorresponding correlation matrix is handled so as to subsequent；Wherein, for convenience, in the embodiment of the present application, by the benefitIt is known as the second correlation matrix with the correlation matrix that the second related coefficient constructs.I.e. step is " according to the second related coefficientGenerate brain area link information of the test object under non-interference stimulation " may include:
The second correlation matrix is constructed using the second related coefficient, which is converted to multiple oneThe feature vector of dimension obtains second feature set.
Wherein, the feature vector in the second feature set is used to reflect brain area of the test object under non-interference stimulationLink information.
For example, still by taking each data acquisition system includes 128 signals as an example, then at this point it is possible to construct one 128* of buildingThen second correlation matrix is converted to multiple one-dimensional feature vectors, obtained by 128 the second correlation matrixSecond feature set.
It is similar with to the progress vector stretching of the first correlation matrix, optionally, since the second correlation matrix is pairTherefore linea angulata symmetrical shape in order to improve treatment effeciency, reduces the consumption of computing resource, can be by the second correlation matrixSymmetrical two regions are divided into, then, vector stretching only are carried out to the element in one of region, to convert element thereinBe converted to one-dimensional feature vector.I.e. second correlation matrix " is converted to multiple one-dimensional feature vectors, obtained by stepSecond feature set " may include:
Second correlation matrix is divided into symmetrical two regions, is selected from two regions according to the preset strategyA region is selected, and the element in the region of selection is converted to one-dimensional feature vector, obtains second feature set.
For example, as shown in figure 4, if shown in Fig. 4 being the second correlation matrix, at this point it is possible to by the second correlationCoefficient matrix is diagonally divided into symmetrical two regions, and then, the element in the region in Fig. 4 in broken line triangle is turnedIt is changed to one-dimensional feature vector, obtains second feature set, etc..
103, the object to be detected is calculated in the brain area link information under interference stimulation and the brain area under non-interference stimulationThen similarity between link information executes step 104.
For example, if reflecting the brain area link information of the test object with feature vector, at this point, can specifically calculateThe similarity between feature vector in feature vector in one characteristic set and second feature set, such as cosine similarity(Cosine similarity) can obtain brain area link information of the object to be detected under interference stimulation and in non-interferenceThe similarity between brain area link information under stimulation.
Wherein, the calculation formula of cosine similarity can be such that
Wherein, A is fisrt feature collection resultant, and B is second feature set；AiFor the feature vector in fisrt feature set；BiFor the feature vector in second feature set, n is that the quantity of feature vector in fisrt feature set (or is also possible to the second spyThe quantity of feature vector in collection conjunction, i.e., the quantity of feature vector and feature vector in second feature set in fisrt feature setQuantity it is identical).
104, according to the similarity determine the eeg signal belonging to histological type.
Wherein, the histological type can depending on the demand of practical application, for example, can be anxiety disorder, depression orObsessive-compulsive disorder etc., it is of course also possible to indicate that the eeg signal is " without exception ", such as non-anxiety disorder, non-depressed or unforcedDisease etc..
For example, by taking anxiety disorder as an example, since normal person and anxiety patient brain area connection under interference signal stimulation are possibleDifferent situations are presented, i.e., most of normal person connect in the brain area connection under interference stimulation with the brain area under non-interference stimulationThe similarity of situation is relatively low, and brain area of the anxiety patient under the brain area connection and non-interference stimulation under interference stimulation connectsIt is not shown significantly different (i.e. similarity is higher) to connect situation, therefore, can by calculate interference stimulation under feature vector with it is non-The cosine similarity between feature vector under interference stimulation characterizes whether the object to be detected belongs to high Jiao Renqun.
Specifically, detecting mould after the similarity being calculated in step 103 such as cosine similarity can be input to trainingIn type, predicted using histological type belonging to detection model eeg signal after the training.For example, specifically can be asUnder:
Cosine similarity matrix is constructed according to the cosine similarity being calculated, by detection model after training to the cosineType belonging to similarity matrix is predicted, histological type belonging to the eeg signal is obtained.
For example, if in step 103, the feature vector in fisrt feature set and second feature set is 128, then128 cosine similarities can be calculated, then at this point, cosine similarity can be constructed based on this 128 cosine similaritiesThen matrix predicts type belonging to the cosine similarity matrix by detection model after training, obtains the brain waveHistological type belonging to signal.
Wherein, detection model is formed by multiple eeg signal sample trainings for being labeled with histological type after the training, i.e.,Detection model after the training can be obtained by the mode of machine learning (ML, Machine Learning).
So-called machine learning is a multi-field cross discipline, is related to probability theory, statistics, Approximation Theory, convextiry analysis, algorithmThe multiple subjects such as complexity computation.The learning behavior that the mankind were simulated or realized to computer how is specialized in, is known so that acquisition is newKnowledge or technical ability, reorganize the existing structure of knowledge and are allowed to constantly improve the performance of itself.Machine learning be artificial intelligence (AI,Artificial Intelligence) core, be make computer have intelligence fundamental way, application spread artificial intelligenceThe every field of energy.Machine learning and deep learning generally include artificial neural network, confidence network, intensified learning, migrationThe technologies such as habit and inductive learning.Wherein, artificial intelligence is the machine mould controlled using digital computer or digital computerIt is quasi-, extend and the intelligence of extension people, environment can be perceived, obtain knowledge and using the theory of Knowledge Acquirement optimum, method,Technology and application system.Artificial intelligence technology is an interdisciplinary study, is related to that field is extensive, and the technology of existing hardware view also hasThe technology of software view.Artificial intelligence basic technology is generally comprised such as sensor, Special artificial intelligent chip, cloud computing, distributionThe technologies such as formula storage, big data processing technique, operation/interactive system, electromechanical integration, therefore not to repeat here.
Optionally, detection model is supplied to eeg signal detection dress after the training after being trained by other equipmentIt sets, alternatively, can also be formed by the eeg signal detection device is voluntarily trained, i.e., " according to the similarity, passes through instruction in stepDetection model predicts histological type belonging to the eeg signal after white silk " before, the eeg signal detection method is alsoIt may include step S1~S5, as follows:
The eeg signal sample of S1, the multiple detection samples of acquisition under interference signal discrete stimulus, the detection sample markIt is marked with histological type.
For example, specifically can with electroencephalogram of the acquisition testing sample under interference signal discrete stimulus, for convenience,In the embodiment of the present application, the electroencephalogram of the detection sample is known as electroencephalogram sample, wherein the electroencephalogram sample includes brain waveSample of signal, and the eeg signal sample is labeled with histological type.
Wherein, acquire electroencephalogram mode can there are many, for example, can pass through receive other equipment for example electroencephalograph send outThe electroencephalogram sent, alternatively, also can receive the electroencephalogram of user's input, or, it can also directly be examined by the eeg signalIt surveys device and the electroencephalogram, etc. is acquired by the electroencephalogram acquisition component of itself.
Optionally, the detection sample can be it is any have large-brained biology, such as people, monkey, cat or dog etc., specificallyIt can be selected according to actual needs；For example, needing to choose at this time if the test object of detection model is people after training" people " as detection sample, and if the test object of detection model is cat after training, need to choose " cat " at this time as detectionSample, etc..
In addition, in order to improve the training effectiveness of the detection model, when selection detects sample can also targetedly intoRow selection, for example, can choose multiple anxiety patient conducts at this time if the histological type of required detection is " anxiety disorder "Sample is detected, certainly, in order to improve the detection accuracy of the detection model, in addition to can choose multiple anxiety patients as justExcept sample, it is also an option that multiple normal persons are as negative sample.For another example, if the histological type of required detection is " depressionDisease ", then can choose the positive sample of multiple patients with depression detection samples at this time, and select multiple normal persons as detection sampleThis negative sample, etc..
S2, obtained respectively according to the eeg signal sample brain area link information of the detection sample under interference stimulation,And the brain area link information under non-interference stimulation.For example, specifically can be such that
A, the eeg signal sample is divided into the first kind signal segment sample comprising interference stimulation and is done comprising non-Disturb the second class signal segment sample of stimulation.
For example, the label of each signal segment sample in the eeg signal sample is obtained, if the label indication signal sectionFor signal collected in the case where interference signal stimulates, then signal segment is classified as first kind signal segment sample, if label instruction letterNumber section is the signal collected under non-interference signal stimulus, then signal segment is classified as the second class signal segment sample.
B, it determines the correlation in the first kind signal segment sample between the signal in Different brain region channel, obtains the detection sampleOriginally the brain area link information under interference stimulation；For example, specifically can be such that
The Pearson correlation coefficients in first kind signal segment sample between the signal in Different brain region channel are calculated, obtain firstRelated coefficient sample generates brain area link information of the detection sample under interference stimulation according to the first related coefficient sample；ThanSuch as, it specifically can use first related coefficient sample architecture the first related coefficient sample matrix, then, by first related coefficientSample matrix is converted to multiple one-dimensional feature vectors, obtains first sample characteristic set.
Wherein, the feature vector in the first sample characteristic set is for reflecting the brain of the detection sample under interference stimulationArea's link information.
Wherein, the calculation of Pearson correlation coefficients is for details, reference can be made to the embodiment of front, and therefore not to repeat here.
Optionally, since the first related coefficient sample matrix is diagonal line symmetrical shape, in order to improve processing effectRate reduces the consumption of computing resource, and the first related coefficient sample matrix can be divided into symmetrical two regions, then, onlyVector stretching is carried out to the element in one of region, is converted to one-dimensional feature vector to convert element therein.That is step" the first related coefficient sample matrix being converted to multiple one-dimensional feature vectors, obtain first sample feature set " can wrapIt includes:
The first related coefficient sample matrix is divided into symmetrical two regions, according to preset strategy from two regionsA region is selected, the element in the region of selection is converted to one-dimensional feature vector, obtains first sample characteristic set.
C, it determines the correlation in the second class signal segment sample between the signal in Different brain region channel, obtains the detection sampleOriginally the brain area link information under non-interference stimulation；For example, specifically can be such that
The Pearson correlation coefficients in the second class signal segment sample between the signal in Different brain region channel are calculated, obtain secondRelated coefficient sample generates brain area link information of the detection sample under non-interference stimulation according to the second related coefficient sample,For example, specifically can use second related coefficient sample architecture the second related coefficient sample matrix, by the second related coefficient sampleThis matrix conversion is multiple one-dimensional feature vectors, obtains the second sample feature set.
Wherein, the feature vector in second sample feature set is for reflecting the detection sample under non-interference stimulationBrain area link information.
Wherein, the calculation of Pearson correlation coefficients is for details, reference can be made to the embodiment of front, and therefore not to repeat here.
Optionally, since the second related coefficient sample matrix is diagonal line symmetrical shape, in order to improve processing effectRate reduces the consumption of computing resource, and the second related coefficient sample matrix can be divided into symmetrical two regions, then, onlyVector stretching is carried out to the element in one of region, is converted to one-dimensional feature vector to convert element therein.That is step" the second related coefficient sample matrix being converted to multiple one-dimensional feature vectors, obtain the second sample characteristics collection " can wrapIt includes:
The second related coefficient sample matrix is divided into symmetrical two regions, according to preset strategy from two regionsA region is selected, the element in the region of selection is converted to one-dimensional feature vector, obtains the second sample feature set.
Wherein, the execution of step B and C can be in no particular order.
S3, calculate the detection sample under interference stimulation brain area link information with non-interference stimulation under brain area connectSimilarity between information.
For example, can specifically calculate the feature vector in first sample characteristic set and the spy in the second sample feature setThe cosine similarity between vector is levied, which is the detection sample under interference stimulationBrain area link information and the similarity between the brain area link information under non-interference stimulation.
Wherein, the calculation formula of the cosine similarity can be such that
Wherein, C is first sample feature set resultant, and D is the second sample feature set；CiFor in first sample characteristic setFeature vector；DiFor the feature vector in the second sample feature set, n is the number of feature vector in first sample characteristic setAmount (or is also possible to the quantity of feature vector in the second sample feature set, i.e. feature vector in first sample characteristic setQuantity it is identical as the quantity of feature vector in the second sample feature set).
S4, according to the similarity, histological type belonging to the eeg signal sample is predicted by detection model；For example, specifically can be such that
Cosine similarity matrix is constructed according to the cosine similarity being calculated, by detection model to the cosine similarityType belonging to matrix is predicted, histological type belonging to the eeg signal sample is obtained.
Wherein, the network structure and parameter of the detection model can be depending on the demands of practical application, for example, specifically may be usedReferring to Fig. 5 a, Fig. 5 b and Fig. 5 c, which can be convolutional neural networks, residual error network (ResNet, ResidualNeural Network) or image segmentation network (VGG, Visual Geometry Group, i.e. view-based access control model geometry intoA kind of convolutional neural networks of row image recognition and segmentation) network etc..
Wherein, Fig. 5 a is convolutional neural networks topology example figure, which may include multilayer convolutional layer, pondChange the network layers such as layer and full articulamentum；The ginseng such as the quantity of convolutional layer, convolution kernel size, step-length and the size of each convolutional layerNumber can be configured according to the demand of practical application, and for example, the size of first time convolutional layer can be set to 64, volumeProduct core is dimensioned to " 7 × 7 ", and it is 64,128,256 and 512 etc. that the size of subsequent convolutional layer, which can be set gradually,Convolution kernel size is disposed as " 3 × 3 ".In this way, when needing to predict histological type belonging to eeg signal sample, it canUsing cosine similarity matrix as input, it is directed into the convolutional neural networks, then successively by each of the convolutional neural networksA network layer carries out process of convolution, pondization processing and full connection processing, final convolutional neural networks to the cosine similarity matrixThe result of output is the histological type of prediction.
Similar, Fig. 5 b is the topology example figure of ResNet, and Fig. 5 c is the topology example figure of VGG network, when needing to predictWhen histological type belonging to eeg signal sample, can using cosine similarity matrix as input, be directed into ResNet orIn VGG network, convolution successively then is carried out to the cosine similarity matrix by each network layer of ResNet or VGG networkThe result of reason, pondization processing and full connection processing, final ResNet or VGG network output is the histological type of prediction, otherModel is similar, and therefore not to repeat here.
S5, the detection model is restrained according to the histological type of mark and the histological type of prediction, after being trainedDetection model.
For example, specifically can by preset loss function according to the histological type of mark and the histological type of prediction to thisDetection model is restrained, detection model after being trained.
That is, need to be adjusted the network parameter of detection model at this time, so that the histological type of mark and pre-The histological type of survey can be approached ad infinitum, and every adjustment is primary, i.e., it is believed that have carried out primary training to the detection model (i.e. completeAt primary study).And so on, after repeatedly training (all detection samples are all passed through with the processing of step S2~S5),Detection model after being trained.
Wherein, which can be depending on the demand of practical application, for example, specifically can be cross entropy (crossEntropy loss) loss function etc..
105, the testing result of the test object is generated based on the histological type.
For example, then available preset indicating template carries out the histological type according to the format of the indicating templateIt has been shown that, obtains the testing result of the test object.
Optionally, name, gender, age, address, the correspondent party of object to be detected can also be shown in the testing resultThe information such as formula and/or occupation.
From the foregoing, it will be observed that the present embodiment can acquire eeg signal of the object to be detected under interference signal discrete stimulus,Brain area link information of the test object under interference stimulation is obtained respectively according to the eeg signal and is stimulated in non-interferenceUnder brain area link information, then, calculate under interference stimulation brain area link information with non-interference stimulation under brain area connectConnect the similarity between information, and according to the similarity determine the eeg signal belonging to histological type, to generate the detectionThe testing result of object；Since the program can be by obtaining test object under interference stimulation and under non-interference stimulation respectivelyBrain area link information, and corresponding histological type is determined by similarity between the two, to achieve the purpose that interpret automatically,For the existing scheme that can only be relied on and manually be interpreted, treatment effeciency can be greatly improved, moreover, becauseThe program is without relying on human interpretation, it can thus be avoided being conducive to there is a situation where the erroneous judgement caused by human factorImprove the accuracy of detection.
Citing, is described in further detail by the method according to described in preceding embodiment below.
In the present embodiment, it will specifically be integrated with the eeg signal detection device in the electronic device, histological typeIncluding being illustrated for anxiety disorder and non-anxiety disorder.
(1) training of detection model.
Firstly, electronic equipment can acquire electroencephalogram of multiple detection samples under interference signal discrete stimulus, wherein shouldElectroencephalogram includes the information such as eeg signal sample；For example, can choose multiple ordinary peoples and multiple anxiety patients as inspectionThen test sample sheet allows these detection samples to connect EEG signal electrodes, to detect its be interfered signal stimulus or non-interference letterNumber stimulation under brain wave variation, these brain waves are acquired, can obtain include eeg signal sample brainElectrograph.For example, specific testing process can be such that
One screen is set, interference signal is presented on the screen, and requires detection sample that the interference signal is being watched to occurWhen complete corresponding task, for example referring to Fig. 6, detection sample can be allowed first to watch the central point 800m of Fig. 6 (a) attentively, screen is aobvious laterShow an interference signal, detection sample is needed when the interference signal occurs, and judgement is water such as the line segment in Fig. 6 (b) in diamond shapeFlat or vertical.After interference signal, screen content switches to Fig. 6 (c), and detection sample continues to watch attentively in Fig. 6 (c)Heart point (watches duration about 1600-2000ms attentively), and next interference signal is waited to occur, and so on, until detection finishes.
Optionally, other than discontinuity display interference signal, which can also be presented interference signal according to preset strategyIt can show such as the interference signal in Fig. 7 (a), and detection sample is required to need at this with non-interference signal for example, referring to Fig. 7When interference signal occurs, judge to be horizontal or vertical such as the line segment in Fig. 6 (b) in diamond shape, after interference signal, screenCurtain content switches to Fig. 7 (b), and detection sample continues to watch screen attentively, and next interference signal is waited to occur, and so on, untilDetection finishes.
It should be noted that other than including eeg signal sample, also will record in these electroencephalograms of acquisitionCorresponding label, to indicate that this section of eeg signal sample is the signal collected under interference signal stimulation, or non-drySignal collected under signal stimulus is disturbed, for example, in the electroencephalogram, c342, Y88, d464 and N99 are dry referring to Fig. 3Disturb the label of signal collected under signal stimulus.
Detection sample is being collected after the electroencephalogram under interference signal discrete stimulus, it can be to the pathology of each electroencephalogramType is labeled, for example, histological type is labeled as " anxiety disorder ", if test samples if test samples are anxiety patientFor normal population, then histological type is labeled as " non-anxiety disorder ", etc., hereafter, these can be labelled with histological typeTraining sample of the electroencephalogram as detection model, i.e. electroencephalogram sample.
Secondly, after obtaining electroencephalogram sample, as shown in figure 8, electronic equipment can be by the brain in the electroencephalogram sampleElectric wave signal sample is divided into the first kind signal segment sample comprising interference stimulation and the second class letter comprising non-interference stimulationNumber section sample；For example, in the available eeg signal sample of electronic equipment each signal segment sample label, if the markSigning indication signal section is the signal collected under interference signal stimulation, then signal segment is classified as first kind signal segment sample, ifThe label indication signal section is the signal collected under non-interference signal stimulus, then signal segment is classified as the second class signal segment sampleThis, etc..
Eeg signal sample in electroencephalogram sample is being divided into first kind signal segment sample and the second class signal segmentAfter sample, electronic equipment can determine the brain area link information detected sample under interference stimulation and accordingly in non-interferenceBrain area link information under stimulation.For example, specifically can be such that
A) the brain area link information under interference stimulation；
Electronic equipment calculates the Pearson correlation coefficients in first kind signal segment sample between the signal in Different brain region channel(referring to the embodiment of front) obtains the first related coefficient sample, then, related using the first related coefficient sample architecture firstCoefficient sample matrix, and the first related coefficient sample matrix is converted into multiple one-dimensional feature vectors, obtain first sampleCharacteristic set, wherein the feature vector in the first sample characteristic set is for reflecting the detection sample under interference stimulationBrain area link information.
B) the brain area link information under non-interference stimulation；
Electronic equipment calculates the Pearson correlation coefficients in the second class signal segment sample between the signal in Different brain region channel(referring to the embodiment of front) obtains the second related coefficient sample, then, related using the second related coefficient sample architecture secondCoefficient sample matrix, and the second related coefficient sample matrix is converted into multiple one-dimensional feature vectors, obtain the second sampleCharacteristic set, wherein the feature vector in second sample feature set is for reflecting the detection sample under non-interference stimulationBrain area link information.
Optionally, due to correlation matrix (such as the first related coefficient sample matrix or the second related coefficient sample momentBattle array) it is therefore diagonal line symmetrical shape in order to improve treatment effeciency, reduces the consumption of computing resource, it can be by these phase relationsMatrix number is divided into symmetrical two regions, then, only vector stretching is carried out to the element in one of region, to convert whereinElement be converted to one-dimensional feature vector.
It should be noted that only provide for convenience, in Fig. 8 a correlation matrix and one-dimensional feature toThe schematic diagram of amount, it should be appreciated that Fig. 8 is only example, and the correlation matrix in Fig. 8 may include the first related coefficientSample matrix and the second related coefficient sample matrix, and the one-dimensional feature vector in Fig. 8 also may include first sample featureThe feature vector in feature vector and the second sample feature set in set, therefore not to repeat here.
Hereafter, electronic equipment can calculate feature vector and the second sample feature set in first sample characteristic setIn feature vector between cosine similarity, and based on these cosine similarities construct a two-dimensional similarity matrix (oneA dimension is " interference stimulation ", another dimension is " non-interference stimulation "), cosine similarity matrix is obtained, for example, referring to Fig. 8.
Since different situations may be presented in brain area connection under interference signal stimulation for normal person and anxiety patient, i.e., big portionDivide the cosine similarity of brain area connection of the normal person under the brain area connection and non-interference stimulation under interference stimulation inclinedIt is low, and anxiety patient is in the cosine phase of brain area connection and the brain area connection under non-interference stimulation under interference stimulationIt is higher like spending, for example, therefore, can use these cosine similarity matrixes referring to Fig. 9 and carry out machine to preset detection modelStudy, and then obtain one and can be used for detecting whether eeg signal belongs to detection model after the training of " anxiety disorder ".
For example, electronic equipment is obtaining the cosine similarity matrix of current eeg signal sample (i.e. under interference stimulationBrain area connection and the brain area connection under non-interference stimulation cosine similarity constructed by matrix) after, can be withType belonging to the cosine similarity matrix is predicted by detection model, obtains disease belonging to the eeg signal sampleThen reason type restrains the detection model according to the histological type of mark and the histological type of prediction, convergence finishesAfterwards, it can be another eeg signal sample by current eeg signal Sample Refreshment, similarly, updated work as obtaining thisThe cosine similarity of preceding eeg signal sample, and after constructing cosine similarity matrix, it can be remaining to this by detection modelType belonging to string similarity matrix is predicted, histological type belonging to current eeg signal sample is obtained, then, according toThe histological type of mark and the histological type of prediction restrain the detection model, and so on, until all brain wavesSample of signal training finishes, then detection model after being trained
(2) by detection model after the training, the eeg signal that can treat test object is detected.
For example, as shown in Figure 10, the detailed process of text recognition methods can be such that
201, electronic equipment acquires electroencephalogram of the object to be detected under interference signal discrete stimulus, wherein the electroencephalogramIncluding eeg signal.
For example, then specifically can be such that so that object to be detected is Zhang San as an example
A screen can be set, interference signal is presented on the screen, and require Zhang San that the interference signal is being watched to occurWhen complete corresponding task, such as the central point 800m that Zhang San can be allowed first to watch attentively in Fig. 6 (a), screen shows primary interference laterSignal, Zhang San need when the interference signal occurs, and the judgement such as line segment in Fig. 6 (b) in diamond shape is horizontal or vertical.?After interference signal, screen content switches to Fig. 6 (c), and Zhang San continues to watch attentively the central point in Fig. 6 (c), waits next drySignal appearance is disturbed, and so on, until detection finishes.
To in the detection process, brain wave caused by Zhang San is acquired, and can be obtained Zhang San and is interrupted in interference signalElectroencephalogram under stimulation.
Optionally, corresponding label can also be recorded other than it can recorde eeg signal in the electroencephalogram, to refer toShow this section of eeg signal be interference signal stimulation under signal collected, or under non-interference signal stimulus it is collectedSignal；Wherein, the notation methods of the label specifically can be depending on the demand of practical application, and therefore not to repeat here.
202, electronic equipment by the eeg signal in the electroencephalogram be divided into the first kind signal segment comprising interference stimulation,And the second class signal segment comprising non-interference stimulation.
For example, as shown in figure 11, if the electroencephalographic record of acquisition has corresponding label, at this point, electronic equipment can obtainThe label of each signal segment in the eeg signal is taken, if the label indication signal section is is acquired under interference signal stimulationSignal, then signal segment is classified as first kind signal segment, if the label indication signal section is is adopted under non-interference signal stimulusSignal segment is then classified as second class signal segment, etc. by the signal of collection.
203, electronic equipment calculates the Pearson correlation coefficients in first kind signal segment between the signal in Different brain region channel,The first related coefficient is obtained, then, executes step 204.
For example, the two are different so that the signal in two Different brain region channel respectively corresponds two datasets conjunction x and y as an example(i.e. the calculation formula of Pearson correlation coefficients r between data acquisition system x and data acquisition system y) can be as the signal in brain area channelUnder:
Wherein, xiFor the element in data acquisition system x, yiElement in data acquisition system y, N are data acquisition system x and data acquisition system yIn element quantity (length of the time point data sampled), such as N can be 100,000 or more, etc..
204, electronic equipment constructs the first correlation matrix using the first related coefficient, by first correlation matrixMultiple one-dimensional feature vectors are converted to, fisrt feature set is obtained.
Wherein, the feature vector in the fisrt feature set is used to reflect that brain area of the test object under interference stimulation to connectConnect information.
For example, by taking each data acquisition system includes 128 signals as an example, then at this point it is possible to construct one 128*128's of buildingThen first correlation matrix is converted to multiple one-dimensional feature vectors by the first correlation matrix, obtain the first spyCollection is closed.
Optionally, as shown in figure 11, since the first correlation matrix is diagonal line symmetrical shape, in order to improveTreatment effeciency reduces the consumption of computing resource, and the first correlation matrix can be divided into symmetrical two regions, then,Vector stretching only is carried out to the element in one of region (region in broken line triangle in such as Figure 11), it is therein to convertElement is converted to one-dimensional feature vector, is detailed in the embodiment of front, therefore not to repeat here.
It should be noted that only providing a correlation matrix and one-dimensional feature for convenience, in Figure 11The schematic diagram of vector, it should be appreciated that the correlation matrix in Figure 11 may include the first related coefficient sample matrix andSecond related coefficient sample matrix, and the one-dimensional feature vector in Figure 11 also may include the spy in first sample characteristic setThe feature vector in vector and the second sample feature set is levied, therefore not to repeat here.
205, electronic equipment calculates the Pearson correlation coefficients in the second class signal segment between the signal in Different brain region channel,The second related coefficient is obtained, step 206 is then executed.
Wherein, for details, reference can be made to steps 203 for the calculation of Pearson correlation coefficients, and therefore not to repeat here.
It should be noted that step 203 and 205 execution sequence can be in no particular order.
206, electronic equipment constructs the second correlation matrix using the second related coefficient, by second correlation matrixMultiple one-dimensional feature vectors are converted to, second feature set is obtained.
Wherein, the feature vector in the second feature set is used to reflect brain area of the test object under non-interference stimulationLink information.
For example, still by taking each data acquisition system includes 128 signals as an example, then at this point it is possible to construct one 128* of buildingThen second correlation matrix is converted to multiple one-dimensional feature vectors, obtained by 128 the second correlation matrixSecond feature set.
It is similar with to the progress vector stretching of the first correlation matrix, optionally, as shown in figure 11, due to the second phase relationMatrix number is therefore diagonal line symmetrical shape in order to improve treatment effeciency, reduces the consumption of computing resource, can be by the second phaseRelationship matrix number is divided into symmetrical two regions, then, only to one of region (in the broken line triangle in such as Figure 11Region) element carry out vector stretching, be converted to one-dimensional feature vector to convert element therein, for details, reference can be made to frontsEmbodiment, therefore not to repeat here.
207, electronic equipment calculate fisrt feature set in feature vector and second feature set in feature vector itBetween cosine similarity.Wherein, the calculation formula of cosine similarity can be such that
Wherein, A is fisrt feature collection resultant, and B is second feature set；AiFor the feature vector in fisrt feature set；BiFor the feature vector in second feature set, n is that the quantity of feature vector in fisrt feature set (or is also possible to the second spyThe quantity of feature vector in collection conjunction, i.e., the quantity of feature vector and feature vector in second feature set in fisrt feature setQuantity it is identical).
208, electronic equipment is according to the cosine similarity building cosine similarity matrix being calculated, and detects after passing through trainingModel predicts type belonging to the cosine similarity matrix, obtains histological type belonging to the eeg signal.
For example, can be calculated at this time if the feature vector in fisrt feature set and second feature set is 128128 cosine similarities out, then at this point, cosine similarity matrix can be constructed based on this 128 cosine similarities, thenType belonging to the cosine similarity matrix is predicted by detection model after training, is obtained belonging to the eeg signalHistological type.
For example, still by taking the eeg signal of Zhang San as an example, it, can be remaining by this after constructing cosine similarity matrixString similarity matrix imports detection model after the training, carries out feature to the cosine similarity matrix by detection model after the trainingIt extracts, and the feature of extraction is identified, if recognition result meets " anxiety disorder " feature, it is determined that the eeg signal of Zhang SanHistological type be " anxiety disorder "；Otherwise, if recognition result does not meet " anxiety disorder " feature, it is determined that the eeg signal of Zhang SanHistological type be " non-anxiety disorder ".
209, electronic equipment generates the testing result of the test object based on the histological type.
For example, then available preset indicating template carries out the histological type according to the format of the indicating templateIt has been shown that, obtains the testing result of the test object.
Optionally, name, gender, age, address, the correspondent party of object to be detected can also be shown in the testing resultThe information such as formula and/or occupation.
From the foregoing, it will be observed that the present embodiment is largely labeled with the electroencephalogram sample of histological type by acquisition, suffered from using anxiety disorderWhen receiving interference information stimulation, the difference of brain area connection is trained detection model, and benefit for person and normal populationIdentified with the histological type that detection model after training treats test object eeg signal, accordingly, with respect to it is existing can onlyFor relying on the scheme manually interpreted, the efficiency to anxiety disorder detection can be greatly improved, moreover, because the program is not necessarily toHuman interpretation is relied on, it can thus be avoided being conducive to improve detection there is a situation where the erroneous judgement caused by human factorAccuracy.
In order to better implement above method, the embodiment of the present invention also provides a kind of eeg signal detection device, the brainElectric wave signal detection device specifically can integrate in the electronic equipments such as server or terminal.
For example, as shown in figure 12, the eeg signal detection device may include acquisition unit 301, acquiring unit 302,Computing unit 303, determination unit 304 and generation unit 305 are as follows:
(1) acquisition unit 301；
Acquisition unit 301, for acquiring eeg signal of the object to be detected under interference signal discrete stimulus.
For example, the acquisition unit 301, specifically can be used for acquiring brain of the object to be detected under interference signal discrete stimulusElectrograph, wherein the electroencephalogram includes eeg signal.
(2) acquiring unit 302；
Acquiring unit 302, for obtaining brain area of the test object under interference stimulation respectively according to the eeg signalLink information and the brain area link information under non-interference stimulation.
For example, the acquiring unit 302 may include dividing subelement, the first determining subelement and the second determining subelement,It is as follows:
Subelement is divided, for the eeg signal to be divided into the first kind signal segment comprising interference stimulation and packetSecond class signal segment of the stimulation containing non-interference.
First determines subelement, for determining the correlation in the first kind signal segment between the signal in Different brain region channelProperty, obtain brain area link information of the test object under interference stimulation；
Second determines subelement, for determining the correlation in the second class signal segment between the signal in Different brain region channelProperty, obtain brain area link information of the test object under non-interference stimulation.
For example, corresponding label can also be recorded when recording eeg signal in electroencephalogram, to indicate this section of brain waveSignal is the signal collected under interference signal stimulation, or the signal collected under non-interference signal stimulus, then, thisWhen divide subelement eeg signal can be divided based on the label, it may be assumed that
Subelement is divided, specifically can be used for obtaining the label of each signal segment in the eeg signal, if the labelIndication signal section is the signal collected under interference signal stimulation, then signal segment is classified as first kind signal segment；If the labelIndication signal section is the signal collected under non-interference signal stimulus, then signal segment is classified as the second class signal segment.
Optionally, determine the correlation between the signal in Different brain region channel mode can there are many, for example, can be asUnder:
The first determining subelement, specifically can be used for calculating in first kind signal segment between the signal in Different brain region channelPearson correlation coefficients, obtain the first related coefficient, which generated under interference stimulation according to the first related coefficientBrain area link information；For example, specifically can use the first related coefficient constructs the first correlation matrix, by first correlationCoefficient matrix is converted to multiple one-dimensional feature vectors, obtains fisrt feature set.
The second determining subelement, specifically can be used for calculating in the second class signal segment between the signal in Different brain region channelPearson correlation coefficients, obtain the second related coefficient, according to the second related coefficient generate the test object non-interference stimulateUnder brain area link information；For example, specifically can use the second related coefficient constructs the second correlation matrix, by second phaseRelationship matrix number is converted to multiple one-dimensional feature vectors, obtains second feature set, the feature in the second feature set toMeasure the brain area link information for reflecting the test object under non-interference stimulation
Wherein, the feature vector in the fisrt feature set is used to reflect that brain area of the test object under interference stimulation to connectInformation is connect, and the feature vector in the second feature set is used to reflect brain area connection of the test object under non-interference stimulationInformation.
Optionally, since correlation matrix (the first correlation matrix and the second correlation matrix) is diagonal line pairClaim form therefore in order to improve treatment effeciency, to reduce the consumption of computing resource, correlation matrix can be divided into symmetricallyTwo regions vector stretching only then is carried out to the element in one of region, with convert element therein be converted to it is one-dimensionalFeature vector.That is:
The first determining subelement, specifically can be used for for first correlation matrix being divided into the symmetrical area Liang GeDomain selects a region according to preset strategy from two regions, the element in the region of selection is converted to one-dimensional featureVector obtains fisrt feature set.
The second determining subelement, specifically can be used for for second correlation matrix being divided into the symmetrical area Liang GeDomain selects a region according to the preset strategy from two regions, and the element in the region of selection is converted to one-dimensionalFeature vector obtains second feature set.
(3) computing unit 303；
Computing unit 303, for calculating this in the brain area link information under interference stimulation and the brain under non-interference stimulationSimilarity between area's link information；
For example, the computing unit 303, specifically can be used for calculating the feature vector and second feature in fisrt feature setThe cosine similarity between feature vector in set.
(4) determination unit 304；
Determination unit 304, for according to the similarity determine the eeg signal belonging to histological type.
For example, the determination unit 304, is specifically used for according to the similarity, by detection model after training to the brain waveHistological type belonging to signal is predicted, for example, can be according to the cosine similarity, by detection model after training to the brainHistological type belonging to electric wave signal is predicted, specific as follows:
The determination unit 304 specifically can be used for according to the cosine similarity building cosine similarity matrix being calculated,Type belonging to the cosine similarity matrix is predicted by detection model after training, is obtained belonging to the eeg signalHistological type.
(5) generation unit 305；
Generation unit 305, for generating the testing result of the test object based on the histological type.
For example, the generation unit 305, specifically can be used for obtaining preset indicating template, then, according to the indicating templateFormat the histological type is shown, obtain the testing result of the test object.
Optionally, name, gender, age, address, the correspondent party of object to be detected can also be shown in the testing resultThe information such as formula and/or occupation.
Wherein, detection model is formed by multiple eeg signal sample trainings for being labeled with histological type after the training.It shouldDetection model is supplied to the eeg signal detection device after training after being trained by other equipment, alternatively, can also be by thisVoluntarily training forms eeg signal detection device, i.e., as shown in figure 13, which can also include instructionPractice unit 306, as follows:
The acquisition unit 301 can be also used for acquiring brain wave of multiple detection samples under interference signal discrete stimulusSample of signal, the eeg signal sample are labeled with histological type.
The acquiring unit 302 can be also used for obtaining the detection sample respectively according to the eeg signal sample and interfereBrain area link information under stimulation and the brain area link information under non-interference stimulation.
For example, the acquiring unit 302, specifically can be used for for the eeg signal sample being divided into comprising interference stimulationFirst kind signal segment sample and the second class signal segment sample stimulated comprising non-interference, then, it is determined that the first kind signal segmentCorrelation in sample between the signal in Different brain region channel obtains brain area connection letter of the detection sample under interference stimulationBreath, and determine the correlation in the second class signal segment sample between the signal in Different brain region channel, obtain the detection sampleBrain area link information under non-interference stimulation.For example, specifically can be such that
The Pearson correlation coefficients in first kind signal segment sample between the signal in Different brain region channel are calculated, obtain firstRelated coefficient sample, using first related coefficient sample architecture the first related coefficient sample matrix, by the first related coefficient sampleThis matrix conversion is multiple one-dimensional feature vectors, obtains first sample characteristic set, the spy in the first sample characteristic setSign vector is for reflecting brain area link information of the detection sample under interference stimulation；And calculate the second class signal segment samplePearson correlation coefficients between the signal in middle Different brain region channel obtain the second related coefficient sample, utilize the second phase relationNumerical example construct the second related coefficient sample matrix, by the second related coefficient sample matrix be converted to multiple one-dimensional features toAmount, obtains the second sample feature set, and the feature vector in second sample feature set is for reflecting the detection sample non-Brain area link information under interference stimulation.
The computing unit 303, can be also used for calculating brain area link information of the detection sample under interference stimulation withThe similarity between brain area link information under non-interference stimulation；For example, the feature in first sample characteristic set can be calculatedThe cosine similarity between feature vector in vector and the second sample feature set, it is to be checked which can be used as thisTest sample sheet is in the brain area link information under interference stimulation and the similarity between the brain area link information under non-interference stimulation.
The training unit 306 is used for according to the similarity, by detection model to disease belonging to the eeg signal sampleReason type is predicted, is restrained, is instructed to the detection model according to the histological type of mark and the histological type of predictionDetection model after white silk.
For example, the training unit 306, specifically can be used for according to the cosine similarity building cosine similarity being calculatedMatrix predicts type belonging to the cosine similarity matrix by detection model, obtains the eeg signal sample instituteThen the histological type of category restrains the detection model according to the histological type of mark and the histological type of prediction, obtainsDetection model after training.