CN110338786B - Epileptic discharge identification and classification method, system, device and medium - Google Patents
Epileptic discharge identification and classification method, system, device and medium Download PDFInfo
- Publication number
- CN110338786B CN110338786B CN201910579663.6A CN201910579663A CN110338786B CN 110338786 B CN110338786 B CN 110338786B CN 201910579663 A CN201910579663 A CN 201910579663A CN 110338786 B CN110338786 B CN 110338786B
- Authority
- CN
- China
- Prior art keywords
- lead
- time sequence
- lead time
- models
- spherical shell
- 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.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—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
Abstract
The invention relates to a method, a system, a device and a medium for recognizing and classifying epileptic discharge, wherein the method comprises the steps of acquiring a plurality of original multi-lead electroencephalogram data; preprocessing each original multi-lead electroencephalogram data to obtain a plurality of lead models corresponding to each original multi-lead electroencephalogram data one by one; respectively extracting the characteristics of each lead model by adopting a power spectral density method and a scale-dependent Lyapunov exponent method to obtain a plurality of characteristic index sets corresponding to each lead model one by one; constructing a target random forest classifier according to all the characteristic index sets; and identifying and classifying the electroencephalogram signals to be detected according to the target random forest classifier to obtain a detection result. The invention can make up the defects of the traditional nonlinear signal processing method in the digital EEG signal analysis, realizes the classification of the normal EEG signal and the epileptic discharge, and has high identification and classification accuracy.
Description
Technical Field
The invention relates to the technical field of digital electroencephalogram signal processing and analysis, in particular to a method, a system, a device and a medium for recognizing and classifying epilepsy-like discharge.
Background
The brain, as the most important organ of human body, has very complex structure and function, and with the continuous development of neuroelectrophysiological technology, the research on human brain function has progressed from comprehensive brain electrical activity recorded by scalp electrode to recording single-channel potential of cell membrane by using patch clamp, and combined with molecular biology technology, further elucidating the mysterious process of human brain activity. In clinical application, Electroencephalogram (EEG) is the spontaneous and rhythmic electrical activity of brain cell populations recorded by electrodes, is the most sensitive method for detecting brain function, is an important means for auxiliary diagnosis and treatment of neurological diseases, and has irreplaceable effects particularly on solving the qualitative and positioning problems of paroxysmal brain dysfunction such as epilepsy.
Epilepsy is a common chronic syndrome, and is clinically characterized by seizures. In patients with clinically typical seizures, epileptiform discharges can be found in EEG examinations for about 80% of the patients. The waveforms of epileptiform discharges are mainly as follows:
1. spike wave
Spikes are one of the most typical epileptic signatures. The phase of the phase is mostly negative, and some are positive. The amplitude of the potential difference of the spike wave is generally above 100 μ V, a few are less than 50 μ V, and the period is within 80 ms. Usually, the spike is generated when there is a stimulating lesion. By observing whether spikes occur in a normal background or in a slow wave, the location of the focus can be inferred, and if spikes suddenly increase, an impending seizure is generally predicted. Spike waves can be seen in various types of epilepsy.
2. Spike wave
A sharp wave is also a typical characteristic waveform of epilepsy. The ascending branch of the sharp wave has a steep slope, while the descending branch appears gentle. The period of the spike is generally greater than 80ms and less than 200ms, the amplitude of the potential difference is between 100-200 μ V, and the phase is generally negative. Its focus is generally large and the discharge process is relatively slow. Sharp waves are seen in various epileptics.
3. Spike-slow wave complex
The spine-slow complex wave consists of two waves, namely a spine wave and a slow wave with the period of 200-500 ms, and is basically negative in phase. The amplitude of the potential difference is much larger than that of the former two waveforms, and can reach more than 500 muV at most. The spike-slow wave belongs to the complex wave, the main component is slow wave, the regularity is particularly strong, and the amplitude of the spike wave is lower than that of the slow wave. And different from the former two waveforms, it is not possessed by various types of epilepsy and has small range.
4. Point-slow wave complex
The spike-slow complex consists of two waves, a spike and a slow wave. The occurrence forms are various, and epileptic seizures are irregular and asynchronous. The foci of the tip-slow complex are typically located deep in the brain tissue.
5. Multiple spikes
Multiple spikes generally occur when multiple spikes are connected together, with one to a few slow waves being added occasionally. When spikes occur continuously, it is generally predicted that the epilepsy will be about to occur, and multiple spikes will occur at the beginning of the seizure.
6. Multiple spike-slow wave complex
The multiple spike-slow complex consists of multiple spikes plus a slow wave, and the spikes appear irregularly and have non-uniform amplitudes. It often occurs during the onset of myoclonic epilepsy.
7. Paroxysmal or explosive activity
Also known as paroxysmal rhythm waves. Paroxysmal or explosive activity is characterized by a sudden appearance without any sign of disappearance. The focus of paroxysmal or explosive activity is generally located in the central brain system and originates from the central brain system, and the wave amplitude is particularly high and the destructive power is particularly strong during the attack.
In clinical EEG examinations it is important to identify whether epileptiform discharges occur in the EEG. At present, an irregularly occurring transient signature associated with epilepsy is identified from EEG signals of patients suspected of having epilepsy or epilepsy, typically by a specialist visually analyzing a large number of EEG signals. Because of the complexity and uncertainty of the EEG signal, the EEG signal is difficult to automatically identify and classify by means of instruments, so that the mass data of the EEG signal generated by long-time continuous monitoring can be only manually observed by a professional, the workload is extremely high, and the EEG signal is difficult to judge in real time and feed back to clinic for timely intervention. Accordingly, methods for digitally analyzing, identifying and classifying EEG signals are continually being sought.
At present, the methods applied to EEG signal analysis are based on time domain analysis or frequency domain analysis theory. The time domain analysis reflects the dynamic change of voltage on a time scale, and the EEG signal is observed and analyzed mainly by utilizing the properties of EEG signal waveforms, such as amplitude, mean value, variance, skewness, kurtosis and the like; however, such methods are prone to false negative results because part of the abnormal electrical activity is dominated by changes in frequency only, making it difficult for the associated automated analysis software to identify. The frequency domain analysis is based on Fourier transform on the premise that the waveform of the EEG signal is assumed to have stationarity, and the waveform of the EEG signal is identified and detected in a frequency domain; however, because the EEG signals have randomness and non-stationarity, information in the EEG is easily lost by simply adopting a time domain or frequency domain analysis method, and the change condition of the EEG cannot be accurately described.
The brain belongs to a chaotic system, electroencephalogram signals have nonlinearity and complexity and are interfered by various artifacts, so that the traditional linear and nonlinear methods cannot effectively process the electroencephalogram signals. One of the methods for processing the electroencephalogram signal is a classical algorithm for calculating the Lyapunov exponent, and the classical algorithm for calculating the Lyapunov exponent assumes that
And by calculating (ln)t-ln0) T to obtain lambda1. The classical theory holds that only lambda1Above 0, the signal is chaotic. This assumption is problematic since it relies only on0Even true chaotic systems may not be suitable. As can be seen more clearly in figure 1,t+tin practice it is possible to comparetAnd is smaller. One more difficulty is that for any noise, λ1May always be greater than 0, which easily results in aliasing noise and true chaos.
Therefore, a new signal processing and analyzing method is needed to identify epileptic discharges in electroencephalogram signals and classify the epileptic discharges so as to make up for the defects of the existing signal processing method in digital electroencephalogram signal analysis.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a method, a system, a device and a medium for recognizing and classifying epileptic-like discharges, which overcome the defects in the conventional nonlinear calculation method, and improve the accuracy of recognizing and classifying normal electroencephalograms and epileptic-like discharges.
The technical scheme for solving the technical problems is as follows:
a method for recognizing and classifying epileptiform discharges comprises the following steps:
step 1: acquiring a plurality of original multi-lead electroencephalogram data;
step 2: preprocessing each original multi-lead electroencephalogram data to obtain a plurality of lead models corresponding to the original multi-lead electroencephalogram data one by one;
and step 3: respectively extracting the characteristics of each lead model by adopting a power spectral density method and a scale-dependent Lyapunov exponent method to obtain a plurality of characteristic index sets corresponding to each lead model one by one;
and 4, step 4: constructing a target random forest classifier according to all the characteristic index sets;
and 5: and identifying and classifying the electroencephalogram signals to be detected according to the target random forest classifier to obtain a detection result.
The invention has the beneficial effects that: according to the invention, the acquired original multi-lead electroencephalogram time is preprocessed, so that the feature extraction is conveniently carried out subsequently according to the processed multi-lead model, and therefore, a random forest classifier with higher identification and classification accuracy is conveniently obtained; the method comprises the steps that a Power Spectral Density (PSD) and a Scale-dependent Lyapunov Exponent method (SDLE) are respectively adopted to obtain a feature index set corresponding to each lead model one by one, so that the defects of a traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, high-frequency pseudo-error interference of abnormal discharge is filtered, and the influence of pseudo-error on EEG signals is reduced; the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos can be compensated, and the dynamics of electroencephalogram is described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
On the basis of the technical scheme, the invention can be further improved as follows:
further: the specific steps of the step 2 comprise:
step 2.1: respectively exporting each original multi-lead electroencephalogram data according to a preset format to obtain a plurality of intermediate multi-lead electroencephalogram data corresponding to each original multi-lead electroencephalogram data one to one;
step 2.2: calling a reading function corresponding to the preset format in Matlab to respectively read each intermediate multi-lead electroencephalogram data to obtain a plurality of lead time sequence sets corresponding to each intermediate multi-lead electroencephalogram data one by one;
step 2.3: and constructing a plurality of lead models which are in one-to-one correspondence with the middle multi-lead electroencephalogram data according to each lead time sequence set.
The beneficial effects of the further scheme are as follows: the original multi-lead electroencephalogram data are derived according to a preset format, and then the corresponding reading function in Matlab is called to read, so that feature extraction is conveniently carried out subsequently according to the read lead time sequence set, and a random forest classifier with higher identification and classification accuracy is conveniently constructed; the preset format and the corresponding read function may be selected according to actual conditions, for example, the preset format may be selected as an edf format, and the corresponding read function may be selected as an edfread.
Further: one characteristic index set corresponding to any one lead model comprises a first characteristic index, and one lead time sequence set corresponding to any one intermediate multi-lead electroencephalogram data comprises a plurality of lead time sequences;
in step 3, the specific step of obtaining the first feature index in any one of the feature index sets includes:
the formula for performing fourier transform on the difference sequence corresponding to any one of the lead time sequences is as follows:
wherein x (N) is a difference sequence corresponding to any one of the lead time sequences in any one of the lead time sequence sets, N is the nth time point in the corresponding lead time sequence, k is the kth time point in the corresponding lead time sequence, N is the length of the corresponding lead time sequence, j is a complex number, W is a length of the lead time sequence, N is a length of the lead time sequenceNFor the rotation factor, x (k) is the corresponding transformation result sequence of the lead time sequence;
The beneficial effects of the further scheme are as follows: when an EEG signal is monitored, various noise interferences exist in the EEG signal, the power spectrum estimation can be used for extracting the frequency spectrum information of the EEG signal, the total energy of the preset frequency band corresponding to any lead time sequence is obtained by selecting the power spectrum of the preset frequency of the lead time sequence for summation, the artifact interference of the EEG signal can be filtered, and the influence of artifacts on the EEG signal is reduced; in the EEG signals containing epileptiform discharge, not all leads are discharged, but sometimes only individual leads are discharged, so that in any lead model, all energy corresponding to any lead model is sequenced from large to small to obtain an energy sequence, then a plurality of energies are selected from the front end of the energy sequence according to a preset number to perform mean value operation, and the obtained first characteristic index can roughly estimate the energy size of lead discharge in each EEG signal; the preset frequency and number can be selected according to actual conditions, for example, the sum of the energy of the power spectrum between 0 and 25Hz and the average of the sorted top 10 larger energies are calculated.
Further: one characteristic index set corresponding to any one lead model comprises a second characteristic index, and one lead time sequence set corresponding to any one intermediate multi-lead electroencephalogram data comprises a plurality of lead time sequences;
in the step 3, the specific step of obtaining the second feature index in any one of the feature index sets includes:
among all the high-dimensional spherical shell models of a corresponding one of the lead models, the kth high-dimensional spherical shell model is:
k≤||va-vb||≤k+Δk，
and is
Wherein the content of the first and second substances,kfor the hull pitch, Δ, of the kth said high-dimensional hull modelkIs the amount of change of the shell pitch, vaAnd vbAll elements of the vector in the phase space of the high-dimensional spherical shell model, m is the preset spherical shell dimension, L is the delay time of the corresponding lead time sequence, and xa、xa+(w-1)L、xbAnd xb+(w-1)LAre time points in a corresponding one of the lead time sequences;
the scale-dependent lyapunov exponent is:
wherein (V)a,Vb) For any one of the point pairs in the phase space of the high-dimensional spherical shell model, λ: (t) For scale-dependent lyapunov exponent, the closed angle brackets denote the average amount of distance between all pairs of points taken within the spherical shell, t is time, Δ t is the amount of change in time,tis the distance between any one point pair in the high-dimensional spherical shell model at the time t, Va、Va+t、Va+t+Δt、Vb、Vb+tAnd Vb+t+ΔtAll the vectors are vectors in the phase space of the high-dimensional spherical shell model;
The beneficial effects of the further scheme are as follows: the initial scale of the chaotic model can be accurately described through the constructed high-dimensional spherical shell model, the scale-dependent Lyapunov exponent curve obtained by observing the evolution of a vector (or a point pair) in a phase space of the high-dimensional spherical shell model can better describe the evolution of the chaotic model of the corresponding lead time sequence by calculating the scale-dependent Lyapunov exponent, so that the curve slope in the scale-dependent Lyapunov exponent curve of any lead time sequence can be obtained to further describe the nonlinear characteristics of EEG (electroencephalogram) signals, the traditional Lyapunov exponents can be compensated for chaotic noise and true chaos, and the nonlinear characteristics in the EEG signals are more accurate; finally, averaging partial curves of all Lyapunov indexes corresponding to all lead time sequences one by one to obtain a second characteristic index which can reflect the linear dynamic characteristic of a corresponding lead model (EEG (electroencephalogram), and can more accurately reflect the characteristics of the EEG through a mode of combining nonlinear analysis and linear analysis, so that the subsequent better identification and classification of the electroencephalogram signals of normal people and epileptic discharge are realized, and the accuracy is higher; wherein t and Δ t can be selected as integer multiples of the frequency of the lead acquisition potential difference signal.
Further: the specific steps of the step 4 comprise:
step 4.1: extracting a plurality of lead model samples from all the lead models in a replacing manner by adopting a Bagging method, and manufacturing a data set according to all the lead model samples;
step 4.2: randomly dividing the data set into a training set and a testing set by using a train _ test _ split function;
step 4.3: taking all the characteristic index sets as classification attributes of an original random forest classifier, and training the training set to obtain the original random forest classifier;
step 4.4: and verifying the test set by adopting the original random forest classifier, and optimizing the original random forest classifier to obtain the target random forest classifier.
The beneficial effects of the further scheme are as follows: through the steps, the feature index set extracted in the steps is input into the random forest classifier to serve as the classification attribute of the random forest classifier, the identification and classification accuracy of the obtained original random forest classifier can be obviously improved, the original random forest classifier is verified through the test set, the effectiveness and accuracy of the original random forest classifier can be further verified, the accuracy of the obtained target random forest classifier is further guaranteed, accurate and efficient identification and classification of electroencephalogram signals and epileptic discharge of normal people are achieved, and the method is extremely suitable for the field of electroencephalogram signal monitoring and analysis and processing.
According to another aspect of the invention, an epileptic discharge identification and classification system is provided, which comprises a data acquisition module, a data processing module, a feature extraction module, a classifier construction module and an identification and classification module;
the data acquisition module is used for acquiring a plurality of original multi-lead electroencephalogram data;
the data processing module is used for respectively preprocessing each original multi-lead electroencephalogram data to obtain a plurality of lead models which are in one-to-one correspondence with each original multi-lead electroencephalogram data;
the characteristic extraction module is used for respectively extracting the characteristics of each lead model by adopting a power spectral density method and a scale-dependent Lyapunov exponent method to obtain a plurality of characteristic index sets corresponding to each lead model one by one;
the classifier building module is used for building a target random forest classifier according to all the characteristic index sets;
and the identification and classification module is used for identifying and classifying the electroencephalogram signals to be detected according to the target random forest classifier to obtain detection results.
The invention has the beneficial effects that: according to the system for recognizing and classifying epileptic sample discharge, the power spectral density and the scale-dependent Lyapunov exponent method are respectively adopted to obtain the feature index sets corresponding to each lead model one by one, so that the defects of the traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, high-frequency pseudo-errors of abnormal discharge are filtered, and the influence of the pseudo-errors on EEG signals is reduced; the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos can be compensated, and the dynamics of electroencephalogram is described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
According to another aspect of the present invention, there is provided an apparatus for recognizing and classifying epileptiform discharges, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program realizes the steps of a method for recognizing and classifying epileptiform discharges when running.
The invention has the beneficial effects that: through the computer program stored in the memory and running on the processor, the epileptic sample discharge identification and classification are realized, the defects of the traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, the high-frequency artifact of abnormal discharge is filtered, and the influence of the artifact on the EEG signal is reduced; the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos can be compensated, and the dynamics of electroencephalogram is described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
In accordance with another aspect of the present invention, there is provided a computer storage medium comprising: at least one instruction which, when executed, performs a step in a method of epileptiform discharge identification and classification of the present invention.
The invention has the beneficial effects that: the epileptic sample discharge identification and classification are realized by executing the computer storage medium containing at least one instruction, the defects of the traditional nonlinear signal processing method in the digital EEG signal analysis can be overcome, the interference caused by the artifact is effectively removed, the confusion of the traditional Lyapunov exponent on the chaotic noise and the true chaos is overcome, and the electroencephalo dynamics is characterized; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
Drawings
FIG. 1 is a schematic diagram of a trajectory in a conventional Lyapunov exponent calculation process;
FIG. 2 is a schematic flow chart illustrating a method for recognizing and classifying epileptiform discharges according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a process for constructing multiple lead models according to a first embodiment of the present invention;
FIG. 4 is a schematic flow chart illustrating a process of obtaining a first characteristic index according to a first embodiment of the present invention;
FIG. 5 is a schematic flow chart illustrating a process of obtaining a second characteristic index according to a first embodiment of the present invention;
FIG. 6 is a diagram illustrating a scale-dependent Lyapunov exponent curve according to an embodiment of the present invention;
FIG. 7 is a schematic flow chart of constructing a target random forest classifier according to a first embodiment of the present invention;
8-1 to 8-7 are schematic diagrams illustrating results of classifying electroencephalogram data to be detected based on a first characteristic index and a second characteristic index in the first embodiment of the present invention;
fig. 9 is a schematic structural diagram of a system for recognizing and classifying epileptiform discharges in accordance with a second embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
The present invention will be described with reference to the accompanying drawings.
In a first embodiment, as shown in fig. 2, a method for recognizing and classifying epileptiform discharges includes the following steps:
s1: acquiring a plurality of original multi-lead electroencephalogram data;
s2: preprocessing each original multi-lead electroencephalogram data to obtain a plurality of lead models corresponding to each original multi-lead electroencephalogram data one by one;
s3: respectively extracting the characteristics of each lead model by adopting a power spectral density method and a scale-dependent Lyapunov exponent method to obtain a plurality of characteristic index sets corresponding to each lead model one by one;
s4: constructing a target random forest classifier according to all the characteristic index sets;
s5: and identifying and classifying the electroencephalogram signals to be detected according to the target random forest classifier to obtain a detection result.
In the embodiment, the acquired original multi-lead electroencephalogram time is preprocessed, so that the feature extraction is conveniently carried out according to the processed multi-lead model, and a random forest classifier with higher identification and classification accuracy is conveniently obtained; the power spectral density method and the scale-dependent Lyapunov exponent method are respectively adopted to obtain the one-to-one corresponding characteristic index set of each lead model, so that the defects of the traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, high-frequency artifact interference of abnormal discharge can be filtered, and the influence of artifact on EEG signals can be reduced; the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos can be compensated, and the dynamics of electroencephalogram is described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
Specifically, in the present embodiment, the multiple pieces of raw multi-lead electroencephalogram data are collected from the patients who are clinically examined at the first subsidiary hospital of the medical university in Guangxi, including 59 patients with epilepsy, and are also collected from 100 normal subjects. And when recording each original multi-lead electroencephalogram data, using 256Hz sampling rate, adopting scalp electrodes, and adopting 10-20 international standard placement for electrode placement, wherein the corresponding positions of the electrodes are as follows: fz (midline of the forehead), Cz (midline of the center), Pz (midline of the vertex), T3 (left medial temporal), C3 (left central), C4 (right central), T4 (right medial temporal), Fp1 (left frontal pole), F7 (left anterior temporal), T5 (left posterior temporal), O1 (left occipital), O2 (right occipital), T6 (right posterior temporal), F8 (right anterior temporal), Fp2 (right frontal pole), F3 (left frontal), F4 (right frontal), P3 (left vertex) and P4 (right vertex) in 19 positions, with ears as reference electrodes, resistance impedance < 10 Ω, recording time 6 s. The total number of the original multi-lead electroencephalogram data is 640, wherein the artificial classification result is obtained according to the artificial reading: 69 parts of spike, 174 parts of spike-slow composite wave, 82 parts of spike, 72 parts of spike-slow composite wave, 64 parts of multi-spike wave, 77 parts of multi-spike-slow composite wave, 2 parts of spike rhythm wave and 100 parts of normal human brain electrical signals.
Preferably, as shown in fig. 3, the specific step of S2 includes:
s2.1: respectively exporting each original multi-lead electroencephalogram data according to a preset format to obtain a plurality of intermediate multi-lead electroencephalogram data corresponding to each original multi-lead electroencephalogram data one to one;
s2.2: calling a reading function corresponding to the preset format in Matlab to respectively read each intermediate multi-lead electroencephalogram data to obtain a plurality of lead time sequence sets corresponding to each intermediate multi-lead electroencephalogram data one by one;
s2.3: and constructing a plurality of lead models which correspond to the middle multi-lead electroencephalogram data one by one according to each lead time sequence set.
The original multi-lead electroencephalogram data are derived according to a preset format, and then the corresponding reading function in Matlab is called to read, so that feature extraction is conveniently carried out subsequently according to the read lead time sequence set, and a random forest classifier with higher identification and classification accuracy is conveniently constructed.
Specifically, the embodiment is derived in an edf format, and then enters a Matlab computing environment to call an edfread.m function to obtain each lead time sequence set, thereby constructing the lead model.
Specifically, in the present embodiment, the left and right earlobes are labeled a1 and a2, respectively, the leading pole reference lead uses a2 and a2 as the G2 end, and a2 (left earlobe) is the G2 end for the left cerebral hemisphere, respectively, Fp2, F2, a.... as the G2 end, similarly, the right cerebral hemisphere is also the same, so that Fp2-a2, F2-a 2, T2-a 2, C2-a 2, O2-a2, P2-a2, F2-a 2, T2-a 2, a 3619, a2, Cz-A1 and Pz-A1.
Preferably, one set of the characteristic indicators corresponding to any one of the lead models comprises a first characteristic indicator, and one set of the lead time sequences corresponding to any one of the intermediate multi-lead electroencephalogram data comprises a plurality of lead time sequences;
as shown in fig. 4, in S3, the specific step of obtaining the first feature index in any one of the feature index sets includes:
s3a.1: acquiring the potential difference of each lead time sequence in any one lead time sequence set at each time point, and respectively performing mean value operation on the potential difference of each lead time sequence at all time points to obtain the mean value of the potential difference corresponding to each lead time sequence in one corresponding lead time sequence set;
s3a.2: performing difference operation on the potential difference of each lead time sequence in the corresponding lead time sequence set at each time point and the corresponding potential difference mean value to obtain a difference value sequence corresponding to each lead time sequence in the corresponding lead time sequence set one by one;
s3a.3: performing Fourier transform on a difference sequence corresponding to each lead time sequence in one corresponding lead time sequence set respectively to obtain a transform result sequence corresponding to each lead time sequence in one corresponding lead time sequence set, and performing squaring operation on each transform result sequence respectively to obtain a power spectral density corresponding to each lead time sequence in one corresponding lead time sequence set;
the formula for performing fourier transform on the difference sequence corresponding to any one of the lead time sequences is as follows:
wherein x (N) is a difference sequence corresponding to any one of the lead time sequences in any one of the lead time sequence sets, N is the nth time point in the corresponding lead time sequence, k is the kth time point in the corresponding lead time sequence, N is the length of the corresponding lead time sequence, j is a complex number, W is a length of the lead time sequence, N is a length of the lead time sequenceNFor twiddle factor, X (k) is the corresponding transformation of the lead time sequenceA sequence of results;
s3a.4: according to a plurality of preset frequencies, summing power spectral densities corresponding to all the lead time sequences in one corresponding lead time sequence set respectively to obtain total energy of one corresponding lead time sequence set under each preset frequency;
s3a.5: and sequentially sequencing the total energy of one corresponding lead time sequence set under all preset frequencies from large to small to obtain an energy sequence, and sequentially selecting a plurality of total energies of a preset number from the front end of the energy sequence to perform mean value operation to obtain the first characteristic index of one corresponding lead model.
When an EEG signal is monitored, various noise interferences exist in the EEG signal, the power spectrum estimation can be used for extracting the frequency spectrum information of the EEG signal, the total energy of the preset frequency band corresponding to any lead time sequence is obtained by selecting the power spectrum of the preset frequency of the lead time sequence for summation, the artifact interference of the EEG signal can be filtered, and the influence of artifacts on the EEG signal is reduced; in the EEG signals containing epileptiform discharge, not all leads are discharged, but sometimes only individual leads are discharged, so that in any lead model, all energy corresponding to any lead model is sequenced from large to small to obtain an energy sequence, then a plurality of energies are selected from the front end of the energy sequence according to a preset number to perform mean value operation, and the obtained first characteristic index can roughly estimate the energy size of lead discharge in each EEG signal; the preset frequency and number can be selected according to actual conditions, for example, the sum of the energy of the power spectrum between 0 and 25Hz and the average of the sorted top 10 larger energies are calculated.
By calculating the power spectral density and extracting the frequency spectrum information of the EEG signal, the artifact interference of the EEG signal can be filtered, and the influence of artifact on the EEG signal is reduced; summing power spectrums with preset frequencies in each lead time sequence to obtain total energy which is the total energy when the EEG signals of the corresponding lead time sequences are discharged, sequencing all energy corresponding to any lead model from large to small to obtain an energy sequence, selecting a plurality of energies with larger energy from the front end of the energy sequence according to the preset quantity to perform mean value operation, and obtaining a first characteristic index which can roughly estimate the energy size of lead discharge in each EEG signal.
Specifically, in this embodiment, the difference is made between the potential difference of 19 lead time sequences in any lead model and the mean value of the potential difference of the corresponding lead time sequences to obtain a difference sequence, power spectrum operation is performed on the difference sequence to obtain frequency spectrum information corresponding to the 19 lead time sequences, the power spectrums in a preset frequency band are added to obtain the energy of the corresponding lead electroencephalogram signal in the frequency band, and 10 lead time sequences with larger energy are selected to perform mean value operation to obtain the first characteristic index of the lead model.
Preferably, one set of the characteristic indicators corresponding to any one of the lead models comprises a second characteristic indicator, and one set of the lead time sequences corresponding to any one of the intermediate multi-lead electroencephalogram data comprises a plurality of lead time sequences;
as shown in fig. 5, in S3, the specific step of obtaining the second feature index in any one of the feature index sets includes:
s3b.1: reconstructing each lead time sequence in any one lead model according to a preset spherical shell dimension to obtain a plurality of high-dimensional spherical shell models corresponding to each lead time sequence one by one;
among all the high-dimensional spherical shell models of a corresponding one of the lead models, the kth high-dimensional spherical shell model is:
k≤||va-vb||≤k+Δk，
and is
Wherein the content of the first and second substances,kfor the hull pitch, Δ, of the kth said high-dimensional hull modelkIs the amount of change of the shell pitch, vaAnd vbAll elements of the vector in the phase space of the high-dimensional spherical shell model, m is the preset spherical shell dimension, L is the delay time of the corresponding lead time sequence, and xa、xa+(w-1)L、xbAnd xb+(w-1)LAre time points in a corresponding one of the lead time sequences;
s3b.2: obtaining the average distance between all point pairs in a phase space where a corresponding one of the high-dimensional spherical shell models is located according to any one of the high-dimensional spherical shell models in the corresponding one of the lead models, and calculating a scale-dependent Lyapunov exponent of the corresponding one of the lead models;
the scale-dependent lyapunov exponent is:
wherein (V)a,Vb) For any one of the point pairs in the phase space of the high-dimensional spherical shell model, λ: (t) For scale-dependent lyapunov exponent, the closed angle brackets denote the average amount of distance between all pairs of points taken within the spherical shell, t is time, Δ t is the amount of change in time,tis the distance between any one point pair in the high-dimensional spherical shell model at the time t, Va、Va+t、Va+t+Δt、Vb、Vb+tAnd Vb+t+ΔtAll the vectors are vectors in the phase space of the high-dimensional spherical shell model;
s3b.3: fitting the corresponding high-dimensional spherical shell model according to the distance average quantity to obtain a scale-dependent Lyapunov exponent curve of the corresponding lead time sequence, and calculating the curve slope of the scale-dependent Lyapunov exponent curve;
s3b.4: executing step S3b.2 and step S3b.3 on each high-dimensional spherical shell model in one corresponding lead model to obtain the curve slopes of a plurality of scale-dependent Lyapunov exponent curves in one-to-one correspondence with each lead time sequence in one corresponding lead model;
s3b.5: and calculating the average value of all the scale-dependent Lyapunov exponent curve slopes in the corresponding one of the lead models to obtain the second characteristic index of the corresponding one of the lead models.
The Lyapunov exponent depending on the scale of any lead time sequence is obtained in the steps, the nonlinear characteristics of the EEG can be further drawn, the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos can be made up, and the nonlinear characteristics in the EEG are reflected more accurately; finally, averaging all the slope values of the Lyapunov exponent curves corresponding to all the lead time sequences one by one, wherein the obtained second characteristic index can reflect the linear dynamic characteristic of a corresponding lead model (EEG (electroencephalogram), and can more accurately reflect the characteristics of the EEG through the mode of combining nonlinear analysis and linear analysis, so that the subsequent better identification and classification of the electroencephalogram and epileptic discharge of normal people are realized, and the accuracy is higher; wherein t and Δ t can be selected as integer multiples of the frequency of the lead acquisition potential difference signal.
Specifically, the scale-dependent lyapunov exponential curve obtained by fitting according to the above steps in the present embodiment is shown in fig. 6.
Preferably, as shown in fig. 7, the specific step of S4 includes:
s4.1: extracting a plurality of lead model samples from all the lead models in a replacing manner by adopting a Bagging method, and manufacturing a data set according to all the lead model samples;
s4.2: randomly dividing the data set into a training set and a testing set by using a train _ test _ split function;
s4.3: taking all the characteristic index sets as classification attributes of an original random forest classifier, and training the training set to obtain the original random forest classifier;
s4.4: and verifying the test set by adopting the original random forest classifier, and optimizing the original random forest classifier to obtain the target random forest classifier.
Through the steps, the feature index set extracted in the steps is input into the random forest classifier to serve as the classification attribute of the random forest classifier, the identification and classification accuracy of the obtained original random forest classifier can be obviously improved, the original random forest classifier is verified through the test set, the effectiveness and accuracy of the original random forest classifier can be further verified, the accuracy of the obtained target random forest classifier is further guaranteed, accurate and efficient identification and classification of electroencephalogram signals and epileptic discharge of normal people are achieved, and the method is extremely suitable for the field of electroencephalogram signal monitoring and analysis and processing.
Specifically, in the embodiment, a Bagging method is adopted to extract N lead model samples from a plurality of lead models in a put-back manner, then a data set is made and divided into a training set and a test set, then two characteristic indexes of each lead model, including a first characteristic index obtained based on power spectral density and a second characteristic index obtained based on a scale-dependent lyapunov index, are used as classification attributes of the training set, one of the classification attributes is randomly extracted as a candidate attribute of a splitting node, and the splitting is performed according to a kini coefficient selection attribute during each splitting; in the whole construction process of the random forest, each tree is fully grown to generate a classification regression tree without pruning, wherein a single classification regression tree without pruning can obtain lower deviation, so that the better classification performance of the random forest is ensured; repeating the splitting process until N classification regression trees are generated; thus, the original random forest classifier of the embodiment is obtained.
Specifically, since only 1 part of 2 types of 8 types of EEG electroencephalograms containing epileptiform discharges are collected in this embodiment, which are respectively a spine rhythm wave and a spine rhythm complex wave, and we classify the signals into rhythm waves, in this embodiment, before a target random forest classifier is constructed, 7 types of electroencephalograms to be detected are classified according to a first characteristic index and a second characteristic index, and the obtained result is shown in fig. 8-1 to 8-7, and then the target random forest classifier constructed according to this embodiment detects the 7 types of electroencephalograms to be detected to obtain a corresponding classification result, and the output classification accuracy reaches 99%.
In a second embodiment, as shown in fig. 9, a system for recognizing and classifying epileptiform discharges includes a data acquisition module, a data processing module, a feature extraction module, a classifier construction module, and a recognition and classification module;
the data acquisition module is used for acquiring a plurality of original multi-lead electroencephalogram data;
the data processing module is used for respectively preprocessing each original multi-lead electroencephalogram data to obtain a plurality of lead models which are in one-to-one correspondence with each original multi-lead electroencephalogram data;
the characteristic extraction module is used for respectively extracting the characteristics of each lead model by adopting a power spectral density and a scale-dependent Lyapunov exponent method to obtain a plurality of characteristic index sets corresponding to each lead model one by one;
the classifier building module is used for building a target random forest classifier according to all the characteristic index sets;
and the identification and classification module is used for identifying and classifying the electroencephalogram signals to be detected according to the target random forest classifier to obtain detection results.
According to the system for recognizing and classifying epileptic-like discharge, the power spectral density and the scale-dependent Lyapunov exponent method are respectively adopted to obtain the feature index sets corresponding to each lead model one by one, so that the defects of the traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, high-frequency artifact interference of abnormal discharge is filtered, and the influence of artifact on EEG signals is reduced; the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos is made up, and the dynamics of electroencephalogram is described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
In a third embodiment, based on the first embodiment and the second embodiment, the present embodiment further discloses an epileptiform discharge identification and classification device, which includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, and when the computer program runs, the specific steps of S1 to S5 shown in fig. 3 are implemented.
Through the computer program stored in the memory and running on the processor, the epileptic sample discharge recognition and classification are realized, the defects of the traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, the high-frequency artifact of abnormal discharge is filtered, the influence of the artifact on the EEG signal is reduced, the confusion of the traditional Lyapunov exponent on chaotic noise and true chaos is overcome, and the dynamics of the EEG signal is described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
The present embodiment also provides a computer storage medium having at least one instruction stored thereon, where the instruction when executed implements the specific steps of S1-S5.
The epileptic sample discharge identification and classification method disclosed by the invention is realized by executing the computer storage medium containing at least one instruction, so that the defects of a traditional nonlinear signal processing method in digital EEG signal analysis can be overcome, high-frequency pseudo-errors of abnormal discharge can be filtered, the influence of the pseudo-errors on the EEG signal can be reduced, the confusion of traditional Lyapunov indexes on chaotic noise and true chaos can be compensated, and the dynamic characteristics of the EEG can be described; the linear characteristic and the nonlinear characteristic of the electroencephalogram signal are respectively extracted by adopting a mode of combining linear analysis and nonlinear analysis, the linear characteristic and the nonlinear characteristic are combined to form a new characteristic of the electroencephalogram signal, the new characteristic is input into a random forest classifier, the classification of the electroencephalogram signal of a normal person and epileptic discharge is realized, the accuracy of identification and classification is high, and the method is extremely suitable for the field of electroencephalogram signal analysis and processing.
Details of S1 to S5 in this embodiment are not described in detail in the first embodiment, and are not described again in detail.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (5)
1. A system for recognizing and classifying epileptic discharge is characterized by comprising a data acquisition module, a data processing module, a feature extraction module, a classifier construction module and a recognition and classification module;
the data acquisition module is used for acquiring a plurality of original multi-lead electroencephalogram data;
the data processing module is used for respectively preprocessing each original multi-lead electroencephalogram data to obtain a plurality of lead models which are in one-to-one correspondence with each original multi-lead electroencephalogram data;
the characteristic extraction module is used for respectively extracting the characteristics of each lead model by adopting a power spectral density method and a scale-dependent Lyapunov exponent method to obtain a plurality of characteristic index sets corresponding to each lead model one by one;
the classifier building module is used for building a target random forest classifier according to all the characteristic index sets;
the recognition and classification module is used for recognizing and classifying the electroencephalogram signals to be detected according to the target random forest classifier to obtain detection results;
wherein the data processing module is specifically configured to:
respectively exporting each original multi-lead electroencephalogram data according to a preset format to obtain a plurality of intermediate multi-lead electroencephalogram data corresponding to each original multi-lead electroencephalogram data one to one;
calling a reading function corresponding to the preset format in Matlab to respectively read each intermediate multi-lead electroencephalogram data to obtain a plurality of lead time sequence sets corresponding to each intermediate multi-lead electroencephalogram data one by one;
constructing a plurality of lead models which are in one-to-one correspondence with the middle multi-lead electroencephalogram data according to each lead time sequence set;
wherein one of the feature index sets corresponding to any one of the lead models comprises a first feature index and a second feature index, and one of the lead time sequence sets corresponding to any one of the intermediate multi-lead electroencephalogram data comprises a plurality of lead time sequences;
the feature extraction module is specifically configured to:
acquiring the potential difference of each lead time sequence in any one lead time sequence set at each time point, and respectively performing mean value operation on the potential difference of each lead time sequence at all time points to obtain the mean value of the potential difference corresponding to each lead time sequence in one corresponding lead time sequence set;
performing difference operation on the potential difference of each lead time sequence in the corresponding lead time sequence set at each time point and the corresponding potential difference mean value to obtain a difference value sequence corresponding to each lead time sequence in the corresponding lead time sequence set one by one;
performing Fourier transform on a difference sequence corresponding to each lead time sequence in one corresponding lead time sequence set respectively to obtain a transform result sequence corresponding to each lead time sequence in one corresponding lead time sequence set, and performing squaring operation on each transform result sequence respectively to obtain a power spectral density corresponding to each lead time sequence in one corresponding lead time sequence set;
the formula for performing fourier transform on the difference sequence corresponding to any one of the lead time sequences is as follows:
wherein x (N) is a difference sequence corresponding to any one of the lead time sequences in any one of the lead time sequence sets, N is the nth time point in the corresponding lead time sequence, k is the kth time point in the corresponding lead time sequence, N is the length of the corresponding lead time sequence, j is a complex number, W is a length of the lead time sequence, N is a length of the lead time sequenceNFor the rotation factor, x (k) is the corresponding transformation result sequence of the lead time sequence;
according to a plurality of preset frequencies, summing power spectral densities corresponding to all the lead time sequences in one corresponding lead time sequence set respectively to obtain total energy of one corresponding lead time sequence set under each preset frequency;
and sequentially sequencing the total energy of one corresponding lead time sequence set under all preset frequencies from large to small to obtain an energy sequence, and sequentially selecting a plurality of total energies of a preset number from the front end of the energy sequence to perform mean value operation to obtain the first characteristic index of one corresponding lead model.
2. The system for identification and classification of epileptiform discharges as recited in claim 1, wherein the feature extraction module is further specifically configured to:
reconstructing each lead time sequence in any one lead model according to a preset spherical shell dimension to obtain a plurality of high-dimensional spherical shell models corresponding to each lead time sequence one by one;
among all the high-dimensional spherical shell models of a corresponding one of the lead models, the kth high-dimensional spherical shell model is:
k≤||va-vb||≤k+Δk，
and is
Wherein the content of the first and second substances,kis the kth oneHull pitch, delta, of high dimensional spherical hull modelskIs the amount of change of the shell pitch, vaAnd vbAll elements of the vector in the phase space of the high-dimensional spherical shell model, m is the preset spherical shell dimension, L is the delay time of the corresponding lead time sequence, and xa、xa+(w-1)L、xbAnd xb+(w-1)LAre time points in a corresponding one of the lead time sequences;
obtaining the average distance between all point pairs in a phase space where a corresponding one of the high-dimensional spherical shell models is located according to any one of the high-dimensional spherical shell models in the corresponding one of the lead models, and calculating a scale-dependent Lyapunov exponent of the corresponding one of the lead models;
the scale-dependent lyapunov exponent is:
wherein (V)a,Vb) For any one of the point pairs in the phase space of the high-dimensional spherical shell model, λ: (t) For scale-dependent lyapunov exponent, the closed angle brackets denote the average amount of distance between all pairs of points taken within the spherical shell, t is time, Δ t is the amount of change in time,tis the distance between any one point pair in the high-dimensional spherical shell model at the time t, Va、Va+t、Va+t+Δt、Vb、Vb+tAnd Vb+t+ΔtAll the vectors are vectors in the phase space of the high-dimensional spherical shell model;
fitting the corresponding high-dimensional spherical shell model according to the distance average quantity to obtain a scale-dependent Lyapunov exponent curve of the corresponding lead time sequence, and calculating the curve slope of the scale-dependent Lyapunov exponent curve;
obtaining a plurality of curve slopes of the scale-dependent Lyapunov exponent curves corresponding to each lead time sequence in the corresponding one of the lead models one to one according to each of the high-dimensional spherical shell models in the corresponding one of the lead models;
and calculating the average value of all the scale-dependent Lyapunov exponent curve slopes in the corresponding one of the lead models to obtain the second characteristic index of the corresponding one of the lead models.
3. The system for the identification and classification of epileptiform discharges as claimed in claim 1 or 2, wherein the classifier building module is specifically configured to:
extracting a plurality of lead model samples from all the lead models in a replacing manner by adopting a Bagging method, and manufacturing a data set according to all the lead model samples;
randomly dividing the data set into a training set and a testing set by using a train _ test _ split function;
taking all the characteristic index sets as classification attributes of an original random forest classifier, and training the training set to obtain the original random forest classifier;
and verifying the test set by adopting the original random forest classifier, and optimizing the original random forest classifier to obtain the target random forest classifier.
4. An apparatus for the identification and classification of epileptiform discharges, comprising a processor, a memory and a computer program stored in the memory and executable on the processor, the computer program when executed implementing the functions of the system for the identification and classification of epileptiform discharges as claimed in any one of claims 1 to 3.
5. A computer storage medium, the computer storage medium comprising: at least one instruction which, when executed, implements the functionality of the epileptiform discharge identification and classification system of any of claims 1 to 3.
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 CN110338786A (en) | 2019-10-18 |
| CN110338786B true 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) |
Families Citing this family (12)
| 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 |
| 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 |
| 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 |
| CN111887843B (en) * | 2020-08-07 | 2022-08-19 | 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 |
| 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 |
| CN112674782B (en) * | 2020-12-22 | 2021-11-26 | 北京航空航天大学 | Device and method for detecting epileptic-like electrical activity of epileptic during inter-seizure period |
| CN113425312B (en) * | 2021-07-30 | 2023-03-21 | 清华大学 | Electroencephalogram data processing method and device |
| CN114115546B (en) * | 2022-01-27 | 2022-06-17 | 湖南大学 | Electroencephalogram data element learning and human-computer intelligent interaction system for multi-robot control |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
Family Cites Families (8)
| 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 |
| US10278608B2 (en) * | 2012-09-07 | 2019-05-07 | Children's Medical Center Corporation | Detection of epileptogenic brains with non-linear analysis of electromagnetic signals |
| CN103110418B (en) * | 2013-01-24 | 2015-04-08 | 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 |
| 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 |
| CN109498000B (en) * | 2018-12-25 | 2021-08-17 | 杭州电子科技大学 | Epileptic seizure early-stage prediction method based on EMD and WPD feature fusion |
-
2019
- 2019-06-28 CN CN201910579663.6A patent/CN110338786B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
Non-Patent Citations (2)
| Title |
|---|
| Multiscale Analysis of Biological Data by Scale-Dependent Lyapunov Exponent;Jianbo Gao等;《frontiers in PHYSIOLOGY》;20120124;第2卷;第1-13页 * |
| 基于gradient boosting 算法的癫痫检测;陈爽爽;《中国优秀硕士学位论文全文数据库 医药卫生科技辑》;20131215(第S2期);第30-48页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110338786A (en) | 2019-10-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110338786B (en) | Epileptic discharge identification and classification method, system, device and medium | |
| Sangaiah et al. | An intelligent learning approach for improving ECG signal classification and arrhythmia analysis | |
| CN109171712B (en) | Atrial fibrillation identification method, atrial fibrillation identification device, atrial fibrillation identification equipment and computer readable storage medium | |
| CN110811609B (en) | Epileptic spike intelligent detection device based on self-adaptive template matching and machine learning algorithm fusion | |
| Subha et al. | EEG signal analysis: a survey | |
| Beyramienanlou et al. | Shannon’s energy based algorithm in ECG signal processing | |
| Halford | Computerized epileptiform transient detection in the scalp electroencephalogram: Obstacles to progress and the example of computerized ECG interpretation | |
| JP5271718B2 (en) | How to identify fetal and maternal ECGs across multiple time segments | |
| CN110598676B (en) | Deep learning gesture electromyographic signal identification method based on confidence score model | |
| CN106236080B (en) | The removing method of myoelectricity noise in EEG signals based on multichannel | |
| CN108742697B (en) | Heart sound signal classification method and terminal equipment | |
| CN112603332A (en) | Emotion cognition method based on electroencephalogram signal characteristic analysis | |
| CN112971795B (en) | Electrocardiosignal quality evaluation method | |
| CN108937916A (en) | A kind of electrocardiograph signal detection method, device and storage medium | |
| Vimala | Stress causing Arrhythmia detection from ECG Signal using HMM | |
| JP2017042562A (en) | Music hearing experience presence/absence estimation method, music hearing experience presence/absence estimation device and music hearing experience presence/absence estimation program | |
| Dilber et al. | EEG based detection of epilepsy by a mixed design approach | |
| CN111887811A (en) | Brain abnormal discharge detection method and system based on electroencephalogram signal characteristics | |
| CN107320097A (en) | The method and apparatus that muscular fatigue feature is extracted using electromyographic signal marginal spectrum entropy | |
| Sanamdikar et al. | Extraction of different features of ECG signal for detection of cardiac arrhythmias by using wavelet transformation Db 6 | |
| Pretorius et al. | Feature extraction from ECG for classification by artificial neural networks | |
| Biran et al. | Automatic qrs detection and segmentation using short time fourier transform and feature fusion | |
| Anishchenko et al. | Comparative analysis of methods for classifying the cardiovascular system's states under stress | |
| Plesnik et al. | ECG signal acquisition and analysis for telemonitoring | |
| Talatov et al. | Methodology for processing and analysis of diagnostic indicators electrocardiogram based on Labview |
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 |