CN110338786A - A kind of identification of epileptiform discharges and classification method, system, device and medium - Google Patents

A kind of identification of epileptiform discharges and classification method, system, device and medium Download PDF

Info

Publication number
CN110338786A
CN110338786A CN201910579663.6A CN201910579663A CN110338786A CN 110338786 A CN110338786 A CN 110338786A CN 201910579663 A CN201910579663 A CN 201910579663A CN 110338786 A CN110338786 A CN 110338786A
Authority
CN
China
Prior art keywords
lead
time series
lead time
model
eeg
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910579663.6A
Other languages
Chinese (zh)
Other versions
CN110338786B (en
Inventor
李琼
张子闻
高剑波
黄淇
吴原
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Normal University
Original Assignee
Beijing Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Normal University filed Critical Beijing Normal University
Priority to CN201910579663.6A priority Critical patent/CN110338786B/en
Publication of CN110338786A publication Critical patent/CN110338786A/en
Application granted granted Critical
Publication of CN110338786B publication Critical patent/CN110338786B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/369—Electroencephalography [EEG]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device

Landscapes

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

Abstract

The present invention relates to a kind of identification of epileptiform discharges and classification method, system, device and medium, method includes obtaining multiple original multi-lead eeg datas；Each original multi-lead eeg data is pre-processed respectively, is obtained and each original multi-lead eeg data multiple lead models correspondingly；Liapunov exponent method using power spectral density method and dependent on scale carries out feature extraction to each lead model respectively, obtains and each lead model multiple characteristic index set correspondingly；Target random forest grader is constructed according to all characteristic index set；EEG signals to be detected are identified and classified according to target random forest grader, obtain testing result.The present invention can overcome the disadvantages that deficiency of traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis, realize the classification to normal brain electric signal and epileptiform discharges, and the accuracy rate of identification and classification is high.

Description

A kind of identification of epileptiform discharges and classification method, system, device and medium
Technical field
The present invention relates to digitlization EEG Processing and analysis technical fields, more particularly to one kind to be based on epileptiform discharges Identification and classification method, system, device and medium.
Background technique
Brain, as the most important organ of human body, structure and function are all sufficiently complex, with Electrophysiology technology It continues to develop, the research of human brain function has been developed to from the synthesis electrical activity of brain that scalp electrode records with Patch-clamp techniques cell Film single channel current potential, and combined with Protocols in Molecular Biology, further illustrate the movable secret of human brain.Clinical application is come It says, electroencephalogram (Electroencephalogram, EEG) is the spontaneity of the brain cell group recorded by electrode, the rhythm and pace of moving things Property electrical activity, be to detect the most sensitive method of brain function, be the important means of auxiliary diagnosis and treatment the nervous system disease, especially It is in the qualitative and orientation problem for solving the paroxysmals brain function exceptions such as epilepsy, electroencephalogram has irreplaceable role.
Epilepsy is a kind of common chronic syndromes, using epileptic attack as Clinical symptoms.Clinically there is typical epileptic attack Patient in, about 80% can EEG inspection in find epileptiform discharges.And the waveform of epileptiform discharges mainly have it is following several Kind:
1. spike
Spike is one of most typical epileptic chracter waveform.Its phase majority is minus phase, some are positive phase.Spike Potential difference wave amplitude size generally more than 100 μ V, minority less than 50 μ V, the period is within 80ms.Generally in irritating venereal disease When stove, the just generation of meeting spike.When appearing in normal background or slow wave by observing spike, disease can be inferred that The position of stove, and increase situation suddenly if there is spike, generally imply that epilepsy will break out.Spike is found in various insane Epilepsy.
2. sharp wave
Sharp wave be also a kind of typical epileptic chracter waveform.The ascending branch gradient of sharp wave is more steep, and in decent But seem more gentle.The period of sharp wave is generally higher than 80ms, and is less than 200ms, potential difference wave amplitude 100~200 μ V it Between, generally in minus phase from phase.Its lesion is generally large, and discharge process compare it is slow.In various epilepsies In, visible sharp wave.
3. the slow complex wave of spine-
The slow complex wave of spine-is made of, from phase two kinds of wave components the slow wave of spike and period between 200~500ms See to be essentially all minus phase on position.Potential difference wave amplitude is much greater compared with first two waveform, reaches as high as 500 μ V or more.Spine-is slow Complex wave belongs to complex wave, and main component is slow wave, and regularity is especially strong, and spike wave amplitude is lower compared with slow wave.And with before Two kinds of waveform differences, its not various epilepsy all have, and have small range.
4. point-slow complex wave
The slow complex wave of point-is by the two kinds of wave components of sharp wave and slow wave.Appearance form is all kinds of, and epileptic attack is not mostly at advising Then asynchronous appearance.The lesion of the slow complex wave of point-is normally at brain tissue depths.
Spike more than 5.
More spikes are generally multiple spikes and connect together appearance, and occasional is plus one to several slow waves.When spike continuously goes out When existing, it can generally indicate that epilepsy will break out, the breaking-out incipient stage also will appear more spikes.
The slow complex wave of spine-more than 6.
The slow complex wave of more spine-is made of multiple spikes plus a slow wave, and when breaking-out occurs irregularly, and wave amplitude is not also united One.Often occur in the episode process of lafora's disease disease.
7. paroxysmal or outburst sexuality
Also known as ictal rhythm and pace of moving things wave.Paroxysmal or the feature of explosive activity are that occur suddenly, have no the disappearance of sign.Battle array Hair property or explosive movable lesion are normally at brain centres system derived from centrencephalic system, and wave amplitude is especially high when breaking-out, break Bad power is especially strong.
It is particularly important whether to there are epileptiform discharges in clinical EEG inspection, in identification EEG.Currently, usually being led to by expert The a large amount of EEG signal of inspectional analysis is crossed, from the EEG signal under a cloud with epilepsy or epileptic, identifies irregular appearance , relevant to epilepsy transient state characteristic waveform.The complexity as existing for EEG signal and uncertain feature are also difficult at present To rely on instrument automatic identification and classification, lead to the mass data of the EEG signal generated long lasting for monitoring, is only capable of by special Industry personnel manually estimate and read figure, and not only workload is very big, but also are difficult to real-time judge and feed back to clinic and do timely intervene.Cause This, people are seeking always the method for carrying out digital assay, identification and classification to EEG signal.
At this stage, the method applied to EEG signal analysis is based on time-domain analysis more or frequency-domain analysis is theoretical.Time-domain analysis is anti- What is reflected is dynamic change of the voltage in time scale, the property of EEG signal waveform is mainly utilized, such as wave amplitude, mean value, side Poor, skewed degree and kurtosis etc. carry out observation analysis to EEG signal；But since part abnormal electrical activity is only based on frequency shift, It is difficult to relevant auto Analysis, therefore such method easily causes false negative result.Frequency-domain analysis is to assume that EEG believes Premised on number waveform has stationarity, based on Fourier transformation, recognition detection is carried out on frequency domain to EEG signal waveform； But it since EEG signals have randomness and non-stationary, simply uses time domain or frequency-domain analysis method is easily lost in EEG Information, cannot to EEG situation of change carry out accurate description.
Brain belongs to chaos system, and EEG signals have nonlinear dissipation, and are interfered by various artifacts, therefore tradition Linearity and non-linearity method EEG signals cannot all be effectively treated.The method of one of processing EEG signals is to calculate Li Ya The classic algorithm of Pu Nuofu index, the classic algorithm for calculating Liapunov exponent assume thatAnd pass through calculating (lnεt-lnε0)/t obtains λ1.Classical theory thinks, as long as λ1Greater than 0, signal is exactly chaos.This hypothesis is asked very much Topic, because it only relies upon ε0Even if real chaos system may also be not suitable for.Can significantly more it see in Fig. 1, εt+δtIt is real It is possible to compare ε on bordertIt is smaller.One bigger difficulty is, for any noise, λ1It may be all easy in this way always greater than 0 Cause confusion noise and real chaos.
Therefore, it is necessary to seek the new signal processing analysis method of one kind to identify epileptiform discharges in EEG signals, and Classify to epileptiform discharges, to make up deficiency of the existing signal processing method in digitlization electroencephalogramsignal signal analyzing.
Summary of the invention
The technical problem to be solved by the present invention is to solve the above shortcomings of the prior art and to provide a kind of epileptiform discharges Identification and classification method, system, device and medium, overcome the defects of traditional Nonlinear Calculation Method, improve to normal The accuracy rate of the identification and classification of EEG signals and epileptiform discharges.
The technical scheme to solve the above technical problems is that
A kind of identification and classification method of epileptiform discharges, comprising the following steps:
Step 1: obtaining multiple original multi-lead eeg datas；
Step 2: each original multi-lead eeg data being pre-processed respectively, is obtained and the original multi-lead Eeg data multiple lead models correspondingly；
Step 3: Liapunov exponent method using power spectral density method and dependent on scale described is led to each respectively Gang mould type carries out feature extraction, obtains and each lead model multiple characteristic index set correspondingly；
Step 4: constructing target random forest grader according to all characteristic index set；
Step 5: EEG signals to be detected being identified and classified according to the target random forest grader, are examined Survey result.
It is pre-processed the beneficial effects of the present invention are: the present invention will acquire the original multi-lead brain electric time, convenient for subsequent According to treated, multi-lead model carries out feature extraction, consequently facilitating obtaining identification and the higher random forest of classification accuracy Classifier；Power spectral density (Power Spetral Density, PSD) and the Liapunov dependent on scale is respectively adopted Index method (Scale-dependent Lyapunov Exponent, SDLE) obtains each one-to-one feature of lead model Index set can overcome the disadvantages that deficiency of traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis, filter out abnormal put The high frequency artifact interference of electricity, reduces influence of the artifact to EEG signal；Traditional Liapunov exponent can also be made up to chaos Obscuring for noise and real chaos, portrays the dynamics of brain electricity；In such a way that linear analysis is in conjunction with nonlinear analysis, respectively Extract EEG signals linear character and nonlinear characteristic, be combined the new feature of composition EEG signals, be input to In machine forest classified device, the classification to normal brain electric signal and epileptiform discharges is realized, the accuracy rate of identification and classification is high, pole It is suitable for electroencephalogramsignal signal analyzings and process field.
Based on the above technical solution, the present invention can also be improved as follows:
Further, the specific steps of the step 2 include:
Step 2.1: each original multi-lead eeg data being exported according to preset format respectively, is obtained and each institute State the one-to-one multiple intermediate multi-lead eeg datas of original multi-lead eeg data；
Step 2.2: function reading corresponding with the preset format in Matlab being called to read each centre respectively Multi-lead eeg data obtains and each intermediate multi-lead eeg data multiple lead time series collection correspondingly It closes；
Step 2.3: respectively according to each lead time series set, building and each intermediate multi-lead brain electricity The one-to-one multiple lead models of data.
The beneficial effect of above-mentioned further scheme is: by exporting original multi-lead eeg data according to preset format Afterwards, the corresponding function reading recalled in Matlab is read out, the lead time series read out convenient for subsequent basis Set carries out feature extraction, consequently facilitating building identification and the higher random forest grader of classification accuracy；Wherein, lattice are preset Formula and corresponding function reading can select according to the actual situation, such as preset format is chosen as edf format, then corresponding reading letter Number is chosen as edfread.m function.
Further, a characteristic index set corresponding with lead model described in any one refers to including fisrt feature Mark, a lead time series set corresponding with intermediate multi-lead eeg data described in any one includes multiple leads Time series；
In the step 3, the specific step of the fisrt feature index in any one described characteristic index set is obtained Suddenly include:
Step 3a.1: each lead time series is obtained in any one described lead time series set when each Between put under potential difference, mean operation is carried out respectively to potential difference of each lead time series under all time points, Obtain each one-to-one potential difference mean value of lead time series in a corresponding lead time series set；
Step 3a.2: the lead time series each in the corresponding one lead time series set is existed respectively Potential difference under each time point carries out difference operation with corresponding potential difference mean value, obtains a corresponding lead time sequence Each one-to-one sequence of differences of lead time series in column set；
Step 3a.3: a pair of to the lead time series one each in the corresponding one lead time series set The sequence of differences answered carries out Fourier transformation respectively, obtains each described in a corresponding lead time series set lead Join the one-to-one transformation results sequence of time series, and squared operation is carried out respectively to each transformation results sequence, obtains Each one-to-one power spectral density of lead time series in a corresponding lead time series set；
The formula of Fourier transformation is carried out the corresponding sequence of differences of any one described lead time series are as follows:
Wherein, x (n) is that any of any one described lead time series set lead time series is corresponding Sequence of differences, n are n-th of time point in the corresponding lead time series, and k is the in the corresponding lead time series K time point, N are the length of the corresponding lead time series, and j is plural number, WNFor twiddle factor, X (k) is corresponding institute State the corresponding transformation results sequence of lead time series；
Step 3a.4: according to multiple predeterminated frequencies, own respectively in the corresponding one lead time series set The corresponding power spectral density of the lead time series is summed, and a corresponding lead time series set is obtained In gross energy one-to-one under each predeterminated frequency；
Step 3a.5: the gross energy under all predeterminated frequencies is integrated into from big to the corresponding one lead time series It successively sorts to small, obtains energy sequence, and successively choose the multiple total of preset quantity since the front end of the energy sequence Energy carries out mean operation, obtains the fisrt feature index of a corresponding lead model.
The beneficial effect of above-mentioned further scheme is: since when monitoring an EEG EEG signals, EEG signals exist each Kind noise jamming, can extract the spectrum information of EEG signal using power Spectral Estimation, by choosing any one described lead time sequence The power spectrum of column predeterminated frequency is summed, and is obtained corresponding to the gross energy that the lead time series presets frequency range, can be filtered Artifact to EEG signal is interfered, and influence of the artifact to EEG signal is reduced；And due in the EEG brain telecommunications containing epileptiform discharges In number, not all lead is all being discharged, and only individual leads are being discharged sometimes, therefore in any one lead model, To the corresponding all energy of any one lead model according to being ranked up from big to small after, obtain energy sequence, then from energy sequence The front end of column starts, and according to preset quantity, chooses multiple energy and carries out mean operation, obtained fisrt feature index can be estimated roughly Count the energy size that lead is discharged in each EEG EEG signals；Wherein, predeterminated frequency and quantity can select according to the actual situation, Such as the adduction operation of energy is carried out to the power spectrum between frequency 0-25Hz, first 10 biggish energy after sequence are carried out Mean operation.
Further, a characteristic index set corresponding with lead model described in any one refers to including second feature Mark, a lead time series set corresponding with intermediate multi-lead eeg data described in any one includes multiple leads Time series；
In the step 3, the specific step of the second feature index in any one described characteristic index set is obtained Suddenly include:
Step 3b.1: according to default spherical shell dimension each of any one lead model lead time series It is reconstructed, obtains each lead time series multiple higher-dimension Spherical Shell Models correspondingly；
In all higher-dimension Spherical Shell Models of the corresponding one lead model, k-th of higher-dimension spherical shell mould Type are as follows:
εk≤||va-vb||≤εk+Δεk,
And
Wherein, εkFor k-th of higher-dimension Spherical Shell Model shell away from Δ εkFor the shell away from variable quantity, vaAnd vbIt is The element of vector where the higher-dimension Spherical Shell Model in phase space, m are the default spherical shell dimension, and L is a corresponding institute State the delay time of lead time series, xa、xa+(w-1)L、xbAnd xb+(w-1)LIt is in a corresponding lead time series Time point；
Step 3b.2: it according to any of the corresponding one lead model higher-dimension Spherical Shell Model, obtains pair The distance between all-pair where one answered the higher-dimension Spherical Shell Model in phase space average magnitude, and calculate corresponding one The Liapunov exponent dependent on scale of a lead model；
Liapunov exponent dependent on scale are as follows:
Wherein, (Va,Vb) it is any of phase space point pair, λ (ε where the higher-dimension Spherical Shell Modelt) it is to rely on In the Liapunov exponent of scale, angle bracket indicate taken in spherical shell between all-pair apart from average magnitude, when t is Between, Δ t is time variation amount, εtFor any of higher-dimension Spherical Shell Model point described in t moment to the distance between, Va、 Va+t、Va+t+Δt、Vb、Vb+tAnd Vb+t+ΔtIt is the vector where the higher-dimension Spherical Shell Model in phase space；
Step 3b.3: the corresponding one higher-dimension Spherical Shell Model is fitted according to the range averaging amount, is obtained The Liapunov exponent curve dependent on scale of a corresponding lead time series calculates described dependent on scale Liapunov exponent curve the slope of curve；
Step 3b.4: step is performed both by higher-dimension Spherical Shell Model described in each of corresponding one described lead model 3b.2 and step 3b.3 is obtained more correspondingly with the lead time series each in the corresponding one lead model The slope of curve of a Liapunov exponent curve dependent on scale；
Step 3b.5: calculating in a corresponding lead model, all Liapunovs for depending on scale The average value of exponential curve slope obtains the second feature index of a corresponding lead model.
The beneficial effect of above-mentioned further scheme is: by the higher-dimension Spherical Shell Model of building, can relatively accurately describe to mix The initial scale of ignorant model, by observe be located at higher-dimension Spherical Shell Model where vector (or point to) in phase space evolution, Higher-dimension Spherical Shell Model is fitted, the obtained Liapunov exponent curve dependent on scale can better describe correspondence Lead time series chaotic model evolution by calculate depend on scale Liapunov exponent, can be obtained any The slope of curve in the Liapunov exponent curve dependent on scale of a lead time series, further to depict EEG The nonlinear characteristic of EEG signals can overcome the disadvantages that traditional Liapunov exponent is to chaotic noise and really mixed through the above steps Ignorant obscures, and the nonlinear characteristic in EEG EEG signals reflected is more accurate；Finally again by all lead time serieses one The local curve of one corresponding all Liapunov exponents is averaged, and obtained second feature index can reflect corresponding lead The linear dynamics of gang mould type (EEG EEG signals) can be more acurrate in such a way that nonlinear analysis and linear analysis combine Ground reflects the feature of EEG EEG signals, realizes the subsequent better identification to normal brain electric signal and epileptiform discharges and divides Class, accuracy rate are higher；Wherein, t and Δ t can select the integral multiple of the frequency of lead acquisition potential difference signal.
Further, the specific steps of the step 4 include:
Step 4.1: multiple lead model samples are extracted from all lead models using Bagging method with putting back to This, and data set is made according to all lead model samples；
Step 4.2: the data set being randomly divided into training set and test set using train_test_split function；
Step 4.3: using all characteristic index set as the categorical attribute of original random forest grader, to described Training set is trained, and obtains the original random forest grader；
Step 4.4: the test set is verified using the original random forest grader, optimize it is described it is original with Machine forest classified device obtains the target random forest grader.
The beneficial effect of above-mentioned further scheme is: through the above steps, the characteristic index collection that will be extracted in abovementioned steps Conjunction is input in random forest grader, and as the categorical attribute of the random forest grader, what can be significantly improved is original The accuracy rate of identification and the classification of random forest grader, then original random forest grader is verified by test set, The validity and accuracy rate of original random forest grader can be further verified, so that the target being further ensured that is gloomy at random The accuracy rate of woods classifier realizes more acurrate efficient identification and classification to normal brain electric signal and epileptiform discharges, pole It is suitable for EEG signals monitoring and analysis process fields.
Another aspect according to the present invention provides the identification and categorizing system of a kind of epileptiform discharges, including data obtain Modulus block, data processing module, characteristic extracting module, classifier building module and identification categorization module；
The data acquisition module, for obtaining multiple original multi-lead eeg datas；
The data processing module is obtained for pre-processing respectively to each original multi-lead eeg data With each original multi-lead eeg data multiple lead models correspondingly；
The characteristic extracting module, for using power spectral density method and dependent on the Liapunov exponent method point of scale It is other that feature extraction is carried out to each lead model, it obtains and each lead model multiple characteristic indexs correspondingly Set；
The classifier constructs module, for according to all characteristic index set building target random forest classification Device；
The identification categorization module, for being known according to the target random forest grader to EEG signals to be detected Not with classification, testing result is obtained.
The beneficial effects of the present invention are: the identification and categorizing system of epileptiform discharges of the invention, are respectively adopted power spectrum Density and Liapunov exponent method dependent on scale obtain each one-to-one characteristic index set of lead model, can be more Deficiency of traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis is mended, the high frequency artifact of paradoxical discharge is filtered out, Reduce influence of the artifact to EEG signal；Traditional Liapunov exponent can also be made up to chaotic noise and real chaos Obscure, portrays the dynamics of brain electricity；In such a way that linear analysis is in conjunction with nonlinear analysis, the line of EEG signals is extracted respectively Property feature and nonlinear characteristic, be combined the new feature of composition EEG signals, be input in random forest grader, it is real Now to the classification of normal brain electric signal and epileptiform discharges, the accuracy rate of identification and classification is high, is extremely applicable to EEG signals Analysis and process field.
Another aspect according to the present invention, provides identification and the sorter of a kind of epileptiform discharges, including processor, Memory is transported with storing in the memory and may operate at the computer program on the processor, the computer program The step in the identification and classification method of a kind of epileptiform discharges of the invention is realized when row.
The beneficial effects of the present invention are: the computer program by storage on a memory, and run on a processor, it is real The identification and classification of existing epileptiform discharges of the invention can overcome the disadvantages that traditional Nonlinear harmonic oscillator method in digitlization EEG signal Deficiency in analysis filters out the high frequency artifact of paradoxical discharge, reduces influence of the artifact to EEG signal；Tradition can also be made up Liapunov exponent chaotic noise and real chaos are obscured, portray the dynamics of brain electricity；Using linear analysis with it is non- The mode that linear analysis combines extracts the linear character and nonlinear characteristic of EEG signals respectively, is combined composition brain The new feature of electric signal, is input in random forest grader, realizes the classification to normal brain electric signal and epileptiform discharges, The accuracy rate of identification and classification is high, is extremely applicable to electroencephalogramsignal signal analyzing and process field.
Another aspect according to the present invention, provides a kind of computer storage medium, and the computer storage medium includes: At least one instruction is performed in the identification and classification method for realizing a kind of epileptiform discharges of the invention in described instruction Step.
The beneficial effects of the present invention are: realizing this hair by executing the computer storage medium comprising at least one instruction The identification and classification of bright epileptiform discharges can overcome the disadvantages that traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis Deficiency, effectively remove interference caused by artifact, compensate for traditional Liapunov exponent to chaotic noise and real chaos Obscure, feature the dynamics of brain electricity；In such a way that linear analysis is in conjunction with nonlinear analysis, EEG signals are extracted respectively Linear character and nonlinear characteristic, be combined the new feature of composition EEG signals, be input to random forest grader In, realize the classification to normal brain electric signal and epileptiform discharges, the accuracy rate of identification and classification is high, is extremely applicable to brain electricity Signal analysis and processing field.
Detailed description of the invention
Fig. 1 is the track schematic diagram in traditional Liapunov exponent calculating process；
Fig. 2 is the identification of epileptiform discharges and the flow diagram of classification method in the embodiment of the present invention one；
Fig. 3 is the flow diagram that multiple lead models are constructed in the embodiment of the present invention one；
Fig. 4 is to obtain the flow diagram of fisrt feature index in the embodiment of the present invention one；
Fig. 5 is to obtain the flow diagram of second feature index in the embodiment of the present invention one；
Fig. 6 is the schematic diagram in the embodiment of the present invention one dependent on the Liapunov exponent curve of scale；
Fig. 7 is the flow diagram that target random forest grader is constructed in the embodiment of the present invention one；
Fig. 8-1 to 8-7 is based on fisrt feature index and second feature index in the embodiment of the present invention one to brain to be detected The result schematic diagram that electric data are classified；
Fig. 9 is the identification of epileptiform discharges and the structural schematic diagram of categorizing system in the embodiment of the present invention two.
Specific embodiment
The principle and features of the present invention will be described below with reference to the accompanying drawings, and the given examples are served only to explain the present invention, and It is non-to be used to limit the scope of the invention.
With reference to the accompanying drawing, the present invention will be described.
Embodiment one, as shown in Fig. 2, a kind of epileptiform discharges identification and classification method, comprising the following steps:
S1: multiple original multi-lead eeg datas are obtained；
S2: pre-processing each original multi-lead eeg data respectively, obtains described original leading with each more Join eeg data multiple lead models correspondingly；
S3: the Liapunov exponent method using power spectral density method and dependent on scale is respectively to each lead mould Type carries out feature extraction, obtains and each lead model multiple characteristic index set correspondingly；
S4: target random forest grader is constructed according to all characteristic index set；
S5: EEG signals to be detected are identified and is classified according to the target random forest grader, are detected As a result.
The present embodiment will acquire the original multi-lead brain electric time and be pre-processed, convenient for subsequent basis treated multi-lead Model carries out feature extraction, consequently facilitating obtaining identification and the higher random forest grader of classification accuracy；Function is respectively adopted Rate spectrum density method and Liapunov exponent method dependent on scale obtain each lead model and correspond characteristic index set, It can overcome the disadvantages that deficiency of traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis, filter out the high frequency of paradoxical discharge Artifact interference reduces influence of the artifact to EEG signals；Can also make up traditional Liapunov exponent to chaotic noise and Obscuring for real chaos, portrays the dynamics of brain electricity；In such a way that linear analysis is in conjunction with nonlinear analysis, brain is extracted respectively The linear character and nonlinear characteristic of electric signal, are combined the new feature of composition EEG signals, are input to random forest In classifier, the classification to normal brain electric signal and epileptiform discharges is realized, the accurate height of identification and classification is extremely applicable to Electroencephalogramsignal signal analyzing and process field.
Specifically, the multiple original multi-lead eeg datas of the present embodiment pick up from No.1 Hospital Attached to Guangxi Medical Univ. and make to face The patient that bed checks, including 59 epileptics, while also picking up from 100 normal subjects.And recording each original lead more When joining eeg data, using 256Hz sample rate, using scalp electrode, electrode is placed to be placed using 10-20 international standard, wherein Electrode corresponding position are as follows: Fz (volume middle line), Cz (central middle line), Pz (top middle line), T3 (left temporo), C3 (left centre), C4 are (right Center), T4 (right side in temporo), Fp1 (left antinion), F7 (left front temporo), T5 (left back temporo), O1 (left pillow), O2 (right pillow), T6 is (after right Temporo), F8 (right before temporo), Fp2 (right antinion), F3 (left volume), F4 (right volume), totally 19 positions P3 (left top) and P4 (right top), with Ears are 6s as reference electrode, 10 Ω of resistive impedance <, record time.It is included in 640 parts of original multi-lead eeg data in total, Wherein, obtain manual sort's result according to artificial figure of readding: slow 174 parts of the complex wave of 69 parts of spike, spine-, 82 parts of sharp wave, point-are slow multiple 72 parts of multiplex, 64 parts of more spikes, more spine-slow 77 parts of complex wave, 2 parts of spine rhythm and pace of moving things wave, 100 parts of normal brain electric signal.
Preferably, as shown in figure 3, the specific steps of S2 include:
S2.1: each original multi-lead eeg data is exported according to preset format respectively, is obtained and each described The one-to-one multiple intermediate multi-lead eeg datas of original multi-lead eeg data；
S2.2: function reading corresponding with the preset format reads each centre respectively and leads more in calling Matlab Join eeg data, obtains and each intermediate multi-lead eeg data multiple lead time series set correspondingly；
S2.3: respectively according to each lead time series set, building and each intermediate multi-lead brain electricity number According to one-to-one corresponding multiple lead models.
After exporting original multi-lead eeg data according to preset format, the corresponding reading in Matlab is recalled Function is read out, and feature extraction is carried out convenient for the lead time series set that subsequent basis is read out, consequently facilitating building Identification and the higher random forest grader of classification accuracy.
Specifically, the present embodiment, subsequently into Matlab computing environment, transfers edfread.m letter with the export of edf format Number, obtains each lead time series set, to construct lead model.
Specifically, the left ear-lobe of the present embodiment and auris dextra, which hang down, is respectively labeled as A1 and A2, ear pole with reference to lead respectively with A1 and A2 is as the end G2, for left hemisphere, A1 (left ear-lobe) be the end G2, then respectively using Fp1, F3 ... as the end G1, together Reason, right hemisphere be also in this way, be formed Fp1-A1, F3-A1, F7-A1, T5-A1, T3-A1, C3-A1, O1-A1, P3-A1, Fp2-A2, F4-A2, F8-A2, T6-A2, T4-A2, C4-A2, O2-A2, P4-A2, Fz-A2, Cz-A2 and Pz-A2 this 19 A lead combines the corresponding 19 lead time serieses to be formed, and constitutes corresponding lead model, wherein Fz-A2, Cz- These three leads of A2 and Pz-A2 can also be Fz-A1, Cz-A1 and Pz-A1 respectively.
Preferably, a characteristic index set corresponding with lead model described in any one refers to including fisrt feature Mark, a lead time series set corresponding with intermediate multi-lead eeg data described in any one includes multiple leads Time series；
As shown in figure 4, obtaining the specific of the fisrt feature index in any one described characteristic index set in S3 Step includes:
S3a.1: each lead time series is obtained in any one described lead time series set in each time Potential difference under point, carries out mean operation to potential difference of each lead time series under all time points respectively, obtains Each one-to-one potential difference mean value of lead time series into the corresponding one lead time series set；
S3a.2: respectively by the lead time series each in the corresponding one lead time series set every Potential difference under a time point carries out difference operation with corresponding potential difference mean value, obtains a corresponding lead time series Each one-to-one sequence of differences of lead time series in set；
S3a.3: the lead time series each in the corresponding one lead time series set is corresponded Sequence of differences carry out Fourier transformation respectively, obtain each lead in a corresponding lead time series set The one-to-one transformation results sequence of time series, and carry out squared operation respectively to each transformation results sequence, obtains pair Each one-to-one power spectral density of lead time series in one answered the lead time series set；
The formula of Fourier transformation is carried out the corresponding sequence of differences of any one described lead time series are as follows:
Wherein, x (n) is that any of any one described lead time series set lead time series is corresponding Sequence of differences, n are n-th of time point in the corresponding lead time series, and k is the in the corresponding lead time series K time point, N are the length of the corresponding lead time series, and j is plural number, WNFor twiddle factor, X (k) is corresponding institute State the corresponding transformation results sequence of lead time series；
S3a.4: according to multiple predeterminated frequencies, respectively to all described in the corresponding one lead time series set The corresponding power spectral density of lead time series is summed, and is obtained a corresponding lead time series and is integrated into One-to-one gross energy under each predeterminated frequency；
S3a.5: to the corresponding one lead time series be integrated into the gross energy under all predeterminated frequencies from greatly to It is small successively to sort, energy sequence is obtained, and multiple total energys of preset quantity are successively chosen since the front end of the energy sequence Amount carries out mean operation, obtains the fisrt feature index of a corresponding lead model.
Since when monitoring an EEG EEG signals, EEG signals, can using power Spectral Estimation there are various noise jammings The spectrum information of EEG signal is extracted, the power spectrum by choosing any one lead time series predeterminated frequency is summed, It obtains corresponding to the gross energy that the lead time series presets frequency range, the artifact interference of EEG signal can be filled into, reduce artifact Influence to EEG signal；And since in the EEG EEG signals containing epileptiform discharges, not all lead is all being put Electricity, only individual leads are being discharged sometimes, therefore in any one lead model, all energy corresponding to any one lead model After amount according to being ranked up from big to small, energy sequence is obtained, then since the front end of energy sequence, according to preset quantity, choosing Multiple energy are taken to carry out mean operation, lead electric discharge in the obtained each EEG EEG signals of fisrt feature index energy rough estimate Energy size；Wherein, predeterminated frequency and quantity can select according to the actual situation, such as to the power spectrum between frequency 0-25Hz The adduction operation for carrying out energy carries out mean operation to first 10 biggish energy after sequence.
The spectrum information that EEG signal is extracted by calculating power spectral density can be filled into the artifact interference of EEG signal, Reduce influence of the artifact to EEG signal；It sums to the power spectrum of predeterminated frequency in each lead time series, what is obtained is total Energy is gross energy when Correspondence lead time series EEG signals discharge, and is pressed to the corresponding all energy of any one lead model According to obtaining energy sequence after being ranked up from big to small, then since the front end of energy sequence according to preset quantity choose energy compared with Big multiple energy carry out mean operation, and lead is put in the obtained each EEG EEG signals of fisrt feature index energy rough estimate The energy size of electricity.
Specifically, the present embodiment leads the potential difference of 19 lead time serieses in any one lead model with corresponding The mean value for joining the potential difference of time series makes the difference, and obtains a sequence of differences, composes operation to sequence of differences power, obtains 19 The corresponding spectrum information of lead time series, and the power spectrum of default frequency range is added, as Correspondence lead EEG signals In the energy of the frequency range, the biggish 10 progress mean operation of energy is chosen, the fisrt feature index of the lead model is obtained.
Preferably, a characteristic index set corresponding with lead model described in any one refers to including second feature Mark, a lead time series set corresponding with intermediate multi-lead eeg data described in any one includes multiple leads Time series；
As shown in figure 5, obtaining the specific of the second feature index in any one described characteristic index set in S3 Step includes:
S3b.1: according to default spherical shell dimension each of any one lead model lead time series into Row reconstruct, obtains each lead time series multiple higher-dimension Spherical Shell Models correspondingly；
In all higher-dimension Spherical Shell Models of the corresponding one lead model, k-th of higher-dimension spherical shell mould Type are as follows:
εk≤||va-vb||≤εk+Δεk,
And
Wherein, εkFor k-th of higher-dimension Spherical Shell Model shell away from Δ εkFor the shell away from variable quantity, vaAnd vbIt is The element of vector where the higher-dimension Spherical Shell Model in phase space, m are the default spherical shell dimension, and L is a corresponding institute State the delay time of lead time series, xa、xa+(w-1)L、xbAnd xb+(w-1)LIt is in a corresponding lead time series Time point；
S3b.2: it according to any of the corresponding one lead model higher-dimension Spherical Shell Model, obtains corresponding The distance between all-pair where one higher-dimension Spherical Shell Model in phase space average magnitude, and calculate a corresponding institute State the Liapunov exponent dependent on scale of lead model；
Liapunov exponent dependent on scale are as follows:
Wherein, (Va,Vb) it is any of phase space point pair, λ (ε where the higher-dimension Spherical Shell Modelt) it is to rely on In the Liapunov exponent of scale, angle bracket indicate taken in spherical shell between all-pair apart from average magnitude, when t is Between, Δ t is time variation amount, εtFor any of higher-dimension Spherical Shell Model point described in t moment to the distance between, Va、 Va+t、Va+t+Δt、Vb、Vb+tAnd Vb+t+ΔtIt is the vector where the higher-dimension Spherical Shell Model in phase space；
S3b.3: the corresponding one higher-dimension Spherical Shell Model is fitted according to the range averaging amount, is corresponded to A lead time series the Liapunov exponent curve dependent on scale, calculate Lee for depending on scale The slope of curve of Ya Punuofu exponential curve；
S3b.4: step is performed both by higher-dimension Spherical Shell Model described in each of corresponding one described lead model S3b.2 and step S3b.3 is obtained one-to-one with the lead time series each in the corresponding one lead model The slope of curve of multiple Liapunov exponent curves dependent on scale；
S3b.5: calculating in a corresponding lead model, all Liapunov exponents for depending on scale The average value of the slope of curve obtains the second feature index of a corresponding lead model.
Above-mentioned steps obtain the Liapunov exponent dependent on scale of any one lead time series, can be further The nonlinear characteristic for depicting EEG EEG signals can overcome the disadvantages that traditional Liapunov exponent to chaotic noise and real chaos Obscure, the nonlinear characteristic in EEG EEG signals reflected is more accurate；Finally again by all lead time serieses one by one Corresponding all Liapunov exponent slopes of curve are averaged, and obtained second feature index can reflect corresponding lead mould The linear dynamics of type (EEG EEG signals) can be more accurately anti-in such a way that nonlinear analysis and linear analysis combine The feature of EEG EEG signals is reflected, realizes subsequent better identification and classification to normal brain electric signal and epileptiform discharges, it is quasi- True rate is higher；Wherein, t and Δ t can select the integral multiple of the frequency of lead acquisition potential difference signal.
Specifically, the Liapunov exponent dependent on scale that the present embodiment is fitted according to above-mentioned steps is bent Line is as shown in Figure 6.
Preferably, as shown in fig. 7, the specific steps of S4 include:
S4.1: extracting multiple lead model samples from all lead models using Bagging method with putting back to, And data set is made according to all lead model samples；
S4.2: the data set is randomly divided into training set and test set using train_test_split function；
S4.3: using all characteristic index set as the categorical attribute of original random forest grader, to the instruction Practice collection to be trained, obtains the original random forest grader；
S4.4: verifying the test set using the original random forest grader, optimizes described original random Forest classified device obtains the target random forest grader.
Through the above steps, the characteristic index set extracted in abovementioned steps is input in random forest grader, is made For the categorical attribute of the random forest grader, the identification of the original random forest grader that can be significantly improved and classification Accuracy rate, then original random forest grader is verified by test set, it can further verify original random forest classification The validity and accuracy rate of device, so that the accuracy rate for the target random forest grader being further ensured that, is realized to normal The more acurrate efficient identification and classification of human brain electric signal and epileptiform discharges, are extremely applicable at EEG signals monitoring and analysis Reason field.
Specifically, the present embodiment extracts N number of lead from multiple lead models using Bagging method with putting back to Then model sample makes data set and is divided into training set and test set, then by two characteristic indexs of each lead model, wraps Include the fisrt feature index obtained based on power spectral density and second feature that the Liapunov exponent dependent on scale obtains Index as the categorical attribute of training set, and randomly selects candidate attribute of one of categorical attribute as split vertexes, and Attribute is selected to be divided according to Gini coefficient when division every time；In the building process of entire random forest, each tree enables it Sufficiently growth generates the post-class processing of a not beta pruning, wherein the post-class processing of single not beta pruning can obtain lower Deviation, to ensure that the preferable classification performance of random forest；Above-mentioned fission process is repeated, until generating N post-class processing； To obtain the original random forest grader of the present embodiment.
Specifically, by thering are 2 classes there was only 1 in EEG EEG signals of 8 classes containing epileptiform discharges that acquire in this present embodiment Part, respectively spine rhythm and pace of moving things wave and spine rhythm and pace of moving things complex wave, it is classified as rhythm and pace of moving things wave by we, therefore the present embodiment is random in building target Before forest classified device, is classified according to fisrt feature index and second feature index to 7 classes eeg data to be detected, obtained The result arrived is to be detected to 7 classes further according to the target random forest grader of the present embodiment building as shown in Fig. 8-1 to Fig. 8-7 EEG signals are detected, and obtain its corresponding classification results, and the classification accuracy exported reaches 99%.
Embodiment two, as shown in figure 9, a kind of epileptiform discharges identification and categorizing system, including data acquisition module, number According to processing module, characteristic extracting module, classifier building module and identification categorization module；
The data acquisition module, for obtaining multiple original multi-lead eeg datas；
The data processing module is obtained for pre-processing respectively to each original multi-lead eeg data With each original multi-lead eeg data multiple lead models correspondingly；
The characteristic extracting module is distinguished for the Liapunov exponent method using power spectral density and dependent on scale Feature extraction is carried out to each lead model, is obtained and each lead model multiple characteristic index collection correspondingly It closes；
The classifier constructs module, for according to all characteristic index set building target random forest classification Device；
The identification categorization module, for being known according to the target random forest grader to EEG signals to be detected Not with classification, testing result is obtained.
The identification and categorizing system of the epileptiform discharges of the present embodiment, are respectively adopted power spectral density and dependent on scale Liapunov exponent method obtains each one-to-one characteristic index set of lead model, can overcome the disadvantages that at traditional nonlinear properties Deficiency of the reason method in digitlization EEG signal analysis, filters out the high frequency artifact interference of paradoxical discharge, reduces artifact to brain electricity The influence of signal；It compensates for traditional Liapunov exponent to obscure chaotic noise and real chaos, portrays the dynamic of brain electricity Mechanics；In such a way that linear analysis is in conjunction with nonlinear analysis, respectively extract EEG signals linear character and non-linear spy Sign, is combined the new feature of composition EEG signals, is input in random forest grader, realizes to normal brain telecommunications The accuracy rate of classification number with epileptiform discharges, identification and classification is high, is extremely applicable to electroencephalogramsignal signal analyzing and process field.
Embodiment three is based on embodiment one and embodiment two, and the present embodiment also discloses a kind of identification of epileptiform discharges With sorter, including processor, memory and stores in the memory and may operate at the calculating on the processor Machine program, the computer program realize the specific steps of S1 to S5 as shown in Figure 3 when running.
It by storing computer program on a memory, and runs on a processor, realizes that epileptic of the invention is put The identification and classification of electricity can overcome the disadvantages that deficiency of traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis, filters out The high frequency artifact of paradoxical discharge reduces influence of the artifact to EEG signal, compensates for traditional Liapunov exponent and make an uproar to chaos Obscuring for sound and real chaos, features the dynamics of EEG signals；In such a way that linear analysis is in conjunction with nonlinear analysis, The linear character and nonlinear characteristic for extracting EEG signals respectively, are combined the new feature of composition EEG signals, input Into random forest grader, the classification to normal brain electric signal and epileptiform discharges, the accuracy rate of identification and classification are realized Height is extremely applicable to electroencephalogramsignal signal analyzing and process field.
The present embodiment also provides a kind of computer storage medium, is stored at least one in the computer storage medium and refers to It enables, described instruction is performed the specific steps for realizing the S1 to S5.
By executing the computer storage medium comprising at least one instruction, the identification of epileptiform discharges of the invention is realized With classification, it can overcome the disadvantages that deficiency of traditional Nonlinear harmonic oscillator method in digitlization EEG signal analysis, filter out paradoxical discharge High frequency artifact, reduce influence of the artifact to EEG signal, compensate for traditional Liapunov exponent to chaotic noise and really Chaos is obscured, and the dynamic characteristic of brain electricity is featured；In such a way that linear analysis is in conjunction with nonlinear analysis, extract respectively The linear character and nonlinear characteristic of EEG signals, are combined the new feature of composition EEG signals, are input to random gloomy In woods classifier, the classification to normal brain electric signal and epileptiform discharges is realized, the accuracy rate of identification and classification is high, extremely suitable For electroencephalogramsignal signal analyzing and process field.
S1 to S5 does not use up details in the present embodiment, and the content of detailed in Example one specifically repeats no more.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.

Claims (8)

1. the identification and classification method of a kind of epileptiform discharges, which comprises the following steps:
Step 1: obtaining multiple original multi-lead eeg datas；
Step 2: each original multi-lead eeg data being pre-processed respectively, is obtained and each original multi-lead Eeg data multiple lead models correspondingly；
Step 3: the Liapunov exponent method using power spectral density method and dependent on scale is respectively to each lead mould Type carries out feature extraction, obtains and each lead model multiple characteristic index set correspondingly；
Step 4: constructing target random forest grader according to all characteristic index set；
Step 5: EEG signals to be detected being identified and classified according to the target random forest grader, obtain detection knot Fruit.
2. the identification and classification method of epileptiform discharges according to claim 1, which is characterized in that the tool of the step 2 Body step includes:
Step 2.1: each original multi-lead eeg data being exported according to preset format respectively, is obtained and each original The one-to-one multiple intermediate multi-lead eeg datas of beginning multi-lead eeg data；
Step 2.2: function reading corresponding with the preset format reads each centre respectively and leads more in calling Matlab Join eeg data, obtains and each intermediate multi-lead eeg data multiple lead time series set correspondingly；
Step 2.3: respectively according to each lead time series set, building and each intermediate multi-lead eeg data One-to-one multiple lead models.
3. the identification and classification method of epileptiform discharges according to claim 2, which is characterized in that led with described in any one The corresponding characteristic index set of gang mould type includes fisrt feature index, with the electricity of intermediate multi-lead brain described in any one The corresponding lead time series set of data includes multiple lead time serieses；
In the step 3, the specific steps packet of the fisrt feature index in any one described characteristic index set is obtained It includes:
Step 3a.1: each lead time series is obtained in any one described lead time series set at every point of time Under potential difference, mean operation is carried out to potential difference of each lead time series under all time points respectively, is obtained Each one-to-one potential difference mean value of lead time series in a corresponding lead time series set；
Step 3a.2: respectively by the lead time series each in the corresponding one lead time series set each Potential difference under time point carries out difference operation with corresponding potential difference mean value, obtains a corresponding lead time series collection Each one-to-one sequence of differences of lead time series in conjunction；
Step 3a.3: one-to-one to each lead time series in the corresponding one lead time series set Sequence of differences carries out Fourier transformation respectively, when obtaining each lead in a corresponding lead time series set Between the one-to-one transformation results sequence of sequence, and squared operation is carried out to each transformation results sequence respectively, is corresponded to A lead time series set in each one-to-one power spectral density of lead time series；
The formula of Fourier transformation is carried out the corresponding sequence of differences of any one described lead time series are as follows:
Wherein, x (n) is the corresponding difference of any of any one described lead time series set lead time series Sequence, n are n-th of time point in the corresponding lead time series, and k is k-th in the corresponding lead time series Time point, N are the length of the corresponding lead time series, and j is plural number, WNFor twiddle factor, X (k) is corresponding described The corresponding transformation results sequence of lead time series；
Step 3a.4: according to multiple predeterminated frequencies, respectively to all described in the corresponding one lead time series set The corresponding power spectral density of lead time series is summed, and is obtained a corresponding lead time series and is integrated into One-to-one gross energy under each predeterminated frequency；
Step 3a.5: the gross energy under all predeterminated frequencies is integrated into from big to small to the corresponding one lead time series It successively sorts, obtains energy sequence, and successively choose multiple gross energies of preset quantity since the front end of the energy sequence Mean operation is carried out, the fisrt feature index of a corresponding lead model is obtained.
4. the identification and classification method of epileptiform discharges according to claim 2, which is characterized in that led with described in any one The corresponding characteristic index set of gang mould type includes second feature index, with the electricity of intermediate multi-lead brain described in any one The corresponding lead time series set of data includes multiple lead time serieses；
In the step 3, the specific steps packet of the second feature index in any one described characteristic index set is obtained It includes:
Step 3b.1: each of any one lead model lead time series is carried out according to default spherical shell dimension Reconstruct, obtains each lead time series multiple higher-dimension Spherical Shell Models correspondingly；
In all higher-dimension Spherical Shell Models of the corresponding one lead model, k-th of higher-dimension Spherical Shell Model are as follows:
εk≤||va-vb||≤εk+Δεk,
And
Wherein, εkFor k-th of higher-dimension Spherical Shell Model shell away from Δ εkFor the shell away from variable quantity, vaAnd vbIt is described The element of vector where higher-dimension Spherical Shell Model in phase space, m are the default spherical shell dimension, and L is to lead described in corresponding one Join the delay time of time series, xa、xa+(w-1)L、xbAnd xb+(w-1)LBe in a corresponding lead time series when Between point；
Step 3b.2: it according to any of the corresponding one lead model higher-dimension Spherical Shell Model, obtains corresponding The distance between all-pair where one higher-dimension Spherical Shell Model in phase space average magnitude, and calculate a corresponding institute State the Liapunov exponent dependent on scale of lead model；
Liapunov exponent dependent on scale are as follows:
Wherein, (Va,Vb) it is any of phase space point pair, λ (ε where the higher-dimension Spherical Shell Modelt) it is dependent on ruler The Liapunov exponent of degree, angle bracket indicate taken in spherical shell between all-pair apart from average magnitude, t is time, Δ t For time variation amount, εtFor any of higher-dimension Spherical Shell Model point described in t moment to the distance between, Va、Va+t、 Va+t+Δt、Vb、Vb+tAnd Vb+t+ΔtIt is the vector where the higher-dimension Spherical Shell Model in phase space；
Step 3b.3: the corresponding one higher-dimension Spherical Shell Model is fitted according to the range averaging amount, is corresponded to A lead time series the Liapunov exponent curve dependent on scale, calculate Lee for depending on scale The slope of curve of Ya Punuofu exponential curve；
Step 3b.4: step 3b.2 is performed both by higher-dimension Spherical Shell Model described in each of corresponding one described lead model With step 3b.3, obtain and the lead time series each in the corresponding one lead model multiple institutes correspondingly State the slope of curve of the Liapunov exponent curve dependent on scale；
Step 3b.5: calculating in a corresponding lead model, all Liapunov exponents for depending on scale The average value of the slope of curve obtains the second feature index of a corresponding lead model.
5. the identification and classification method of epileptiform discharges according to any one of claims 1 to 4, which is characterized in that described The specific steps of step 4 include:
Step 4.1: multiple lead model samples are extracted from all lead models using Bagging method with putting back to, And data set is made according to all lead model samples；
Step 4.2: the data set being randomly divided into training set and test set using train_test_split function；
Step 4.3: using all characteristic index set as the categorical attribute of original random forest grader, to the training Collection is trained, and obtains the original random forest grader；
Step 4.4: the test set being verified using the original random forest grader, is optimized described original random gloomy Woods classifier obtains the target random forest grader.
6. the identification and categorizing system of a kind of epileptiform discharges, which is characterized in that including data acquisition module, data processing mould Block, characteristic extracting module, classifier building module and identification categorization module；
The data acquisition module, for obtaining multiple original multi-lead eeg datas；
The data processing module, for being pre-processed respectively to each original multi-lead eeg data, obtain with often A original multi-lead eeg data multiple lead models correspondingly；
The characteristic extracting module, for right respectively using power spectral density method and the Liapunov exponent method dependent on scale Each lead model carries out feature extraction, obtains and each lead model multiple characteristic index collection correspondingly It closes；
The classifier constructs module, for constructing target random forest grader according to all characteristic index set；
The identification categorization module, for according to the target random forest grader to EEG signals to be detected carry out identification with Classification, obtains testing result.
7. identification and the sorter of a kind of epileptiform discharges, which is characterized in that including processor, memory and be stored in described It in memory and may operate at the computer program on the processor, such as claim realized when the computer program is run Method and step described in any one of 1 to 5 claim.
8. a kind of computer storage medium, which is characterized in that the computer storage medium includes: at least one instruction, in institute It states instruction and is performed realization such as method and step described in any one of claim 1 to 5.
CN201910579663.6A 2019-06-28 2019-06-28 Epileptic discharge identification and classification method, system, device and medium Active CN110338786B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910579663.6A CN110338786B (en) 2019-06-28 2019-06-28 Epileptic discharge identification and classification method, system, device and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910579663.6A CN110338786B (en) 2019-06-28 2019-06-28 Epileptic discharge identification and classification method, system, device and medium

Publications (2)

Publication Number Publication Date
CN110338786A true CN110338786A (en) 2019-10-18
CN110338786B CN110338786B (en) 2020-10-02

Family

ID=68177345

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910579663.6A Active CN110338786B (en) 2019-06-28 2019-06-28 Epileptic discharge identification and classification method, system, device and medium

Country Status (1)

Country Link
CN (1) CN110338786B (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110811609A (en) * 2019-10-28 2020-02-21 杭州电子科技大学 Intelligent epileptic spike detection method based on fusion of adaptive template matching and machine learning algorithm
CN111626153A (en) * 2020-05-13 2020-09-04 电子科技大学 Integrated learning-based partial discharge fault state identification method
CN111631683A (en) * 2020-05-07 2020-09-08 林伟 Method for processing and detecting electrocardiographic data or electroencephalogram data, and storage medium
CN111887811A (en) * 2020-08-07 2020-11-06 浙江环玛信息科技有限公司 Brain abnormal discharge detection method and system based on electroencephalogram signal characteristics
CN111887843A (en) * 2020-08-07 2020-11-06 浙江环玛信息科技有限公司 Fusion epileptic seizure period automatic detection method and system
CN112057068A (en) * 2020-08-27 2020-12-11 武汉工程大学 Epilepsia pathological data classification method and device and storage medium
CN112089414A (en) * 2019-06-17 2020-12-18 阿里健康信息技术有限公司 Method and device for detecting abnormal discharge of brain
CN112244871A (en) * 2020-09-25 2021-01-22 吉林大学 Amplitude integration electroencephalogram classification recognition system based on machine learning
CN112842356A (en) * 2019-11-26 2021-05-28 阿里健康信息技术有限公司 Data processing method, device, equipment and system
CN113425312A (en) * 2021-07-30 2021-09-24 清华大学 Electroencephalogram data processing method and device
CN113662560A (en) * 2020-12-22 2021-11-19 北京航空航天大学 Method for detecting seizure-like discharge between attacks, storage medium and device
CN114115546A (en) * 2022-01-27 2022-03-01 湖南大学 Electroencephalogram data element learning and human-computer intelligent interaction system for multi-robot control

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6442421B1 (en) * 2000-04-27 2002-08-27 Centre National De La Recherche Scientifique Method for the medical monitoring in real time of a patient from the analysis of electroencephalograms to characterize and differentiate between physiological or pathological conditions, and a method for anticipating epileptic seizures
CN2922792Y (en) * 2005-07-21 2007-07-18 高春平 Automatic epilepsia electroencephalo monitoring and diagnosis system
CN102423259A (en) * 2011-09-22 2012-04-25 上海师范大学 Epileptogenic focus positioning device and method
CN103110418A (en) * 2013-01-24 2013-05-22 天津大学 Electroencephalogram signal characteristic extracting method
US20150216436A1 (en) * 2012-09-07 2015-08-06 Children's Medical Center Corporation Detection of epileptogenic brains with non-linear analysis of electromagnetic signals
CN107847138A (en) * 2015-05-10 2018-03-27 阿尔法欧米伽神经科技有限公司 Automatic brain detector guiding system
CN108478216A (en) * 2018-02-06 2018-09-04 杭州电子科技大学 A kind of epileptic seizure intelligent Forecasting early period based on convolutional neural networks
CN109480833A (en) * 2018-08-30 2019-03-19 北京航空航天大学 The pretreatment and recognition methods of epileptic's EEG signals based on artificial intelligence
CN109498000A (en) * 2018-12-25 2019-03-22 杭州电子科技大学 Epileptic attack prediction technique early period based on EMD Yu WPD Fusion Features
US20190142291A1 (en) * 2015-03-23 2019-05-16 Temple University-Of The Commonwealth System Of Higher Education System and Method for Automatic Interpretation of EEG Signals Using a Deep Learning Statistical Model

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6442421B1 (en) * 2000-04-27 2002-08-27 Centre National De La Recherche Scientifique Method for the medical monitoring in real time of a patient from the analysis of electroencephalograms to characterize and differentiate between physiological or pathological conditions, and a method for anticipating epileptic seizures
CN2922792Y (en) * 2005-07-21 2007-07-18 高春平 Automatic epilepsia electroencephalo monitoring and diagnosis system
CN102423259A (en) * 2011-09-22 2012-04-25 上海师范大学 Epileptogenic focus positioning device and method
US20150216436A1 (en) * 2012-09-07 2015-08-06 Children's Medical Center Corporation Detection of epileptogenic brains with non-linear analysis of electromagnetic signals
CN103110418A (en) * 2013-01-24 2013-05-22 天津大学 Electroencephalogram signal characteristic extracting method
US20190142291A1 (en) * 2015-03-23 2019-05-16 Temple University-Of The Commonwealth System Of Higher Education System and Method for Automatic Interpretation of EEG Signals Using a Deep Learning Statistical Model
CN107847138A (en) * 2015-05-10 2018-03-27 阿尔法欧米伽神经科技有限公司 Automatic brain detector guiding system
CN108478216A (en) * 2018-02-06 2018-09-04 杭州电子科技大学 A kind of epileptic seizure intelligent Forecasting early period based on convolutional neural networks
CN109480833A (en) * 2018-08-30 2019-03-19 北京航空航天大学 The pretreatment and recognition methods of epileptic's EEG signals based on artificial intelligence
CN109498000A (en) * 2018-12-25 2019-03-22 杭州电子科技大学 Epileptic attack prediction technique early period based on EMD Yu WPD Fusion Features

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
JIANBO GAO等: "Multiscale Analysis of Biological Data by Scale-Dependent Lyapunov Exponent", 《FRONTIERS IN PHYSIOLOGY》 *
陈爽爽: "基于gradient boosting 算法的癫痫检测", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 *
高玉龙等: "《随机过程与分析原理》", 31 January 2017, 哈尔滨工业大学出版社 *

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112089414A (en) * 2019-06-17 2020-12-18 阿里健康信息技术有限公司 Method and device for detecting abnormal discharge of brain
CN110811609A (en) * 2019-10-28 2020-02-21 杭州电子科技大学 Intelligent epileptic spike detection method based on fusion of adaptive template matching and machine learning algorithm
CN110811609B (en) * 2019-10-28 2022-11-22 杭州电子科技大学 Epileptic spike intelligent detection device based on self-adaptive template matching and machine learning algorithm fusion
CN112842356B (en) * 2019-11-26 2024-03-26 阿里健康信息技术有限公司 Data processing method, device, equipment and system
CN112842356A (en) * 2019-11-26 2021-05-28 阿里健康信息技术有限公司 Data processing method, device, equipment and system
CN111631683A (en) * 2020-05-07 2020-09-08 林伟 Method for processing and detecting electrocardiographic data or electroencephalogram data, and storage medium
CN111631683B (en) * 2020-05-07 2023-10-17 林伟 Electrocardiogram data or electroencephalogram data processing and detecting method and storage medium
CN111626153B (en) * 2020-05-13 2022-10-18 电子科技大学 Integrated learning-based partial discharge fault state identification method
CN111626153A (en) * 2020-05-13 2020-09-04 电子科技大学 Integrated learning-based partial discharge fault state identification method
CN111887843A (en) * 2020-08-07 2020-11-06 浙江环玛信息科技有限公司 Fusion epileptic seizure period automatic detection method and system
CN111887811B (en) * 2020-08-07 2022-08-23 浙江环玛信息科技有限公司 Brain abnormal discharge detection method and system based on electroencephalogram signal characteristics
CN111887811A (en) * 2020-08-07 2020-11-06 浙江环玛信息科技有限公司 Brain abnormal discharge detection method and system based on electroencephalogram signal characteristics
CN112057068A (en) * 2020-08-27 2020-12-11 武汉工程大学 Epilepsia pathological data classification method and device and storage medium
CN112244871A (en) * 2020-09-25 2021-01-22 吉林大学 Amplitude integration electroencephalogram classification recognition system based on machine learning
CN113662560A (en) * 2020-12-22 2021-11-19 北京航空航天大学 Method for detecting seizure-like discharge between attacks, storage medium and device
CN113425312A (en) * 2021-07-30 2021-09-24 清华大学 Electroencephalogram data processing method and device
CN113425312B (en) * 2021-07-30 2023-03-21 清华大学 Electroencephalogram data processing method and device
CN114115546A (en) * 2022-01-27 2022-03-01 湖南大学 Electroencephalogram data element learning and human-computer intelligent interaction system for multi-robot control

Also Published As

Publication number Publication date
CN110338786B (en) 2020-10-02

Similar Documents

Publication Publication Date Title
CN110338786A (en) A kind of identification of epileptiform discharges and classification method, system, device and medium
CN107951485B (en) Ambulatory ECG analysis method and apparatus based on artificial intelligence self study
CN108714026A (en) The fine granularity electrocardiosignal sorting technique merged based on depth convolutional neural networks and on-line decision
Legarreta et al. Continuous wavelet transform modulus maxima analysis of the electrocardiogram: beat characterisation and beat-to-beat measurement
CN100418480C (en) Heart disease automatic classification system based on heart sound analysis and heart sound segmentation method
CN105956624B (en) Mental imagery brain electricity classification method based on empty time-frequency optimization feature rarefaction representation
EP1788937A2 (en) Method for adaptive complex wavelet based filtering of eeg signals
CN102697493A (en) Method for rapidly and automatically identifying and removing ocular artifacts in electroencephalogram signal
CN105997043B (en) A kind of pulse frequency extracting method based on wrist wearable device
CN112603332A (en) Emotion cognition method based on electroencephalogram signal characteristic analysis
CN112932498A (en) T wave morphology classification system with strong generalization capability based on deep learning
CN100538713C (en) A kind of brain electricity fluctuation signal analysis equipment
CN103720471A (en) Factor analysis based ocular artifact removal method
Moser et al. Classification and detection of single evoked brain potentials using time-frequency amplitude features
Elbuni et al. ECG parameter extraction algorithm using (DWTAE) algorithm
Shen et al. Method for extracting time-varying rhythms of electroencephalography via wavelet packet analysis
CN106859641B (en) A method of based on nuclear-magnetism artefact in automatic ICA removal EEG signal
Sanamdikar et al. Extraction of different features of ECG signal for detection of cardiac arrhythmias by using wavelet transformation Db 6
Far et al. Prediction of preterm labor from EHG signals using statistical and non-linear features
Pretorius et al. Feature extraction from ECG for classification by artificial neural networks
CN113491525A (en) Electroencephalogram signal analysis-based method
CN106682630A (en) Feature sampling method based on electrocardiographic signal
Daqrouq et al. Arrhythmia detection using wavelet transform
Akhbardeh et al. Ballistocardiogram classification using a novel transform so-called AliMap and biorthogonal wavelets
Geerts Detection of interictal epileptiform discharge in EEG

Legal Events

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