阅读说明：本技术 基于脑电信号特征的大脑异常放电检测方法及系统 (Brain abnormal discharge detection method and system based on electroencephalogram signal characteristics ) 是由 董芳 刘俊飙 吴端坡 孙乐 于 2020-08-07 设计创作，主要内容包括：本发明提供一种基于脑电信号特征的大脑异常放电检测方法及系统,该方法包括：采集原始脑电信号并进行数据处理；提取预处理后的脑电信号中的多域特征和棘波特征,多域特征包括时频特征和非线性特征；基于提取时频特征和非线性特征采用已训练完成的随机森林模型来得到脑电信号异常放电的第一检测结果；基于提取的棘波特征中棘波比和设定阈值的关系来获得脑电信号异常放电的第二检测结果；融合第一检测结果和第二检测结果以得到最终的大脑异常放电检测结果。(The invention provides a brain abnormal discharge detection method and system based on electroencephalogram signal characteristics, wherein the method comprises the following steps: collecting original electroencephalogram signals and performing data processing; extracting multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features; obtaining a first detection result of abnormal discharge of the electroencephalogram signal by adopting a trained random forest model based on the extracted time-frequency characteristics and nonlinear characteristics; obtaining a second detection result of abnormal discharge of the electroencephalogram signal based on the relation between the spike ratio in the extracted spike characteristics and a set threshold; and fusing the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result.)

1. A brain abnormal discharge detection method based on EEG signal characteristics is characterized by comprising the following steps:

collecting original electroencephalogram signals and performing data processing;

extracting multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features;

obtaining a first detection result of abnormal discharge of the electroencephalogram signal by adopting a trained random forest model based on the extracted time-frequency characteristics and nonlinear characteristics;

obtaining a second detection result of abnormal discharge of the electroencephalogram signal based on the relation between the spike ratio in the extracted spike characteristics and a set threshold;

and fusing the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result.

2. The brain abnormal discharge detection method based on electroencephalogram signal characteristics according to claim 1, wherein the extracting the multi-domain characteristics in the preprocessed electroencephalogram signal comprises:

performing statistical analysis on the electroencephalogram signals to obtain time domain characteristics, wherein the time domain characteristics comprise line length, kurtosis, skewness, maximum value, minimum value, mean value and variance characteristics of a waveform;

extracting different rhythm waves from the electroencephalogram signal by wavelet packet transformation, decomposing the electroencephalogram signal into five frequency bands, and mutually corresponding the different rhythm waves and the five frequency bands to establish a time-frequency relationship, wherein the five frequency bands are respectively as follows: 0.5 Hz-4 Hz wave, 4 Hz-8 Hz theta wave, 8H-12 Hz alpha wave, 12 Hz-23 Hz beta 1 wave and 23 Hz-30 Hz beta 2 wave;

and calculating the nonlinear characteristics of abnormal discharge waveforms of the brain by adopting a Hurst index.

3. The brain abnormal discharge detection method based on EEG signal characteristics as claimed in claim 2, wherein the step of establishing time-frequency relationship between different rhythm waves and five frequency bands comprises:

first, db6 is selected as the wavelet basis function and each epoch is defined to be 5 seconds;

secondly, extracting different rhythm waves from each epoch through wavelet packet transformation, wherein the formula is as follows:

wherein Φ (t) is a scale function;is a wavelet basis function; i (0,1, 2.., 2.)^j-1) number of nodes for wavelet packet transformation; j is a decomposition layer; h (n) and g (n) ═ 1)^1-nh (1-n) is a pair of orthogonal image filters;and

let the wavelet packet coefficient of the electroencephalogram signal x (t) of the kth node of the jth layer be as follows:

whereinAnd

then, the wavelet packet coefficients of the five frequency bands are normalized, and the formula is as follows:

wherein K_l(r) wavelet packet coefficient, p, for each frequency band_l(r) is as followsNormalizing the wavelet packet coefficients; n is the length of the wavelet packet coefficient; l is the number of frequency segments;

finally, performing L-layer wavelet packet decomposition will generate 2^LA node; when the j layer is decomposed, the signal is decomposed into a high-frequency part and a low-frequency part, and the decomposed high-frequency part and the decomposed low-frequency part are continuously decomposed to obtain a new high-frequency part and a new low-frequency part; the decomposition is repeated until the rhythm waves corresponding to the five frequency bands are matched.

4. The brain abnormal discharge detection method based on EEG signal characteristics according to claim 2, wherein the step of calculating the nonlinear characteristics of the brain abnormal discharge waveform by using the Hurst index comprises:

estimation was performed using R/S analysis with X ═ X1; x 2; the following steps are carried out; xC represents a channel in the epoch of the electroencephalogram signal, and the R/S analysis method has the following calculation formula:

wherein H is the Herster index; r is the maximum distance, i.e. the difference between the maximum and minimum values of X; s is the standard deviation of X; c is the length of X.

5. The method for detecting brain abnormal discharge based on electroencephalogram signal characteristics according to claim 1, wherein the first detection result of the brain abnormal discharge of the electroencephalogram signal is obtained by adopting a trained random forest model after the dimensionality of the characteristic parameters is reduced by adopting linear discriminant analysis after the time-frequency characteristics and the nonlinear characteristics are extracted.

6. The brain abnormal discharge detection method based on EEG signal characteristics according to claim 1, wherein a morphological filter is used to extract spike characteristics, and the steps thereof comprise:

first, since the shape of the spike is similar to a triangle, the triangle structural element g (k) is selected, and the formula is as follows:

wherein A is the center height of the structural element, D is half of the width of the structural element, and k is the number of sampling points in a time sequence;

then, each channel of the electroencephalogram signal is divided into electroencephalogram segments with a certain length, and the geometric characteristics of each electroencephalogram segment are matched with preset structural elements by using a morphological filter.

7. The brain abnormal discharge detection method based on electroencephalogram signal characteristics according to claim 6, characterized in that the process of matching the geometric characteristics in each electroencephalogram segment with preset structural elements by using a morphological filter is as follows:

firstly, using the selected structural elements to carry out open-close and close-open operations on each electroencephalogram segment signal, wherein the formula is as follows:

wherein, Y (k) represents a background signal, and X (k) represents each electroencephalogram segment signal; k is the number of sampling points in one electroencephalogram segment;

then, calculating the difference between the background signal and each electroencephalogram segment signal, wherein the calculated result is a spike signal, and the formula is as follows:

z (k) ═ x (k) -y (k) formula eight

Wherein, Z (k) is the difference between the background signal and each electroencephalogram segment signal;

wherein SR is the spike ratio, expressed as the mean of the spike signal Z (k); num_j(i) Represents the spike wave quantity in the ith epoch of the jth channel, t represents the time length of each electroencephalogram signal segment, and k is an electroencephalogram sheetNumber of samples in a segment.

8. The method for detecting abnormal brain discharge based on electroencephalogram characteristics according to claim 1, wherein when the first detection result and the second detection result both represent that the detected electroencephalogram is in abnormal discharge, the final result after the first detection result and the second detection result are fused is the abnormal discharge.

9. The method for detecting abnormal brain discharge based on EEG signal features of claim 1, wherein a 5 th order Butterworth band-pass filter with frequency range of 0.5 Hz-32 Hz is used to remove noise and artifacts from EEG signal during data preprocessing.

10. A brain abnormal discharge detection system based on EEG signal characteristics is characterized by comprising:

the data acquisition and processing module is used for acquiring original electroencephalogram signals and processing data;

the feature extraction module is used for extracting multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features;

the first detection module is used for obtaining a first detection result of abnormal discharge of the electroencephalogram signal by adopting a trained random forest model based on the extracted time-frequency characteristics and nonlinear characteristics;

the second detection module is used for obtaining a second detection result of abnormal discharge of the electroencephalogram signal based on the relation between the spike ratio in the extracted spike characteristics and the set threshold value;

and the fusion module is used for fusing the first detection result and the second detection result to obtain a final abnormal brain discharge detection result.

Technical Field

The invention relates to the field of computers, in particular to a brain abnormal discharge detection method and system based on electroencephalogram signal characteristics.

Background

The abnormal cerebral discharge disease is a chronic disease of transient cerebral dysfunction caused by sudden abnormal discharge of cerebral neurons, and epilepsy is a typical disease caused by abnormal cerebral discharge. Over sixty-five million people worldwide suffer from epilepsy, with about nine million people in epilepsy in china, and this figure is increasing at a rate of 40 million per year. Seizures are paroxysmal, repetitive and unpredictable and may occur at any age.

Electroencephalograms (EEG) are patterns obtained by recording spontaneous biopotentials of the brain from the scalp by amplifying them with a precision electronic instrument, and are spontaneous and rhythmic electrical activities of brain cell groups recorded by electrodes. In clinical medicine, EEG remains an important tool for diagnosing certain brain diseases, and can also provide an effective treatment for certain brain diseases and play an important role in the detection of diseases caused by abnormal brain discharges.

Spike waves are typical abnormal brain discharge signatures, usually recorded in electroencephalograms, are sharp with respect to background waveforms, and have high amplitude and transient characteristics. At present, clinical examination of epilepsia electroencephalogram is mainly to identify spike waves in electroencephalogram signals through manual detection, but the efficiency is low, the subjectivity is strong, the accuracy of results cannot be guaranteed, and therefore the spike wave automatic detection technology receives more and more attention in recent years. Through researches on spike detection in recent years, a plurality of spike automatic detection methods based on threshold values have appeared, but the threshold value selection still has difficulty; if the threshold value is too low, the false detection rate is higher; too high a threshold value will miss many spikes.

Disclosure of Invention

The invention provides a brain abnormal discharge detection method and system based on electroencephalogram signal characteristics, which have high detection efficiency and high detection precision and are used for overcoming the problems of strong subjectivity and low accuracy of the existing brain abnormal discharge detection method.

In order to achieve the above object, the present invention provides a method for detecting abnormal brain discharge based on electroencephalogram characteristics, comprising:

collecting original electroencephalogram signals and performing data processing;

extracting multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features;

obtaining a first detection result of abnormal discharge of the electroencephalogram signal by adopting a trained random forest model based on the extracted time-frequency characteristics and nonlinear characteristics;

obtaining a second detection result of abnormal discharge of the electroencephalogram signal based on the relation between the spike ratio in the extracted spike characteristics and a set threshold;

and fusing the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result.

According to an embodiment of the present invention, extracting multi-domain features in the preprocessed electroencephalogram signal includes:

carrying out statistical analysis on the electroencephalogram signals to obtain time domain characteristics, wherein the time domain characteristics comprise line length, kurtosis, skewness, maximum value, minimum value, mean value and variance characteristics of a waveform;

extracting different rhythm waves from the electroencephalogram signal by wavelet packet transformation, decomposing the electroencephalogram signal into five frequency bands, and mutually corresponding the different rhythm waves and the five frequency bands to establish a time-frequency relationship, wherein the five frequency bands are respectively as follows: 0.5 Hz-4 Hz wave, 4 Hz-8 Hz theta wave, 8H-12 Hz alpha wave, 12 Hz-23 Hz beta 1 wave and 23 Hz-30 Hz beta 2 wave;

and calculating the nonlinear characteristics of abnormal discharge waveforms of the brain by adopting a Hurst index.

According to an embodiment of the present invention, the step of establishing a time-frequency relationship between the different rhythmic waves and the five frequency bands includes:

first, db6 is selected as the wavelet basis function and each epoch is defined to be 5 seconds;

secondly, extracting different rhythm waves from each epoch through wavelet packet transformation, wherein the formula is as follows:

wherein Φ (t) is a scale function;is a wavelet basis function; i (0,1, 2.., 2.)^j-1) number of nodes for wavelet packet transformation; j is a decomposition layer; h (n) and g (n) ═ 1)^1-nh (1-n) is a pair of orthogonal image filters;andis a wavelet basis function based on dual scales; n is the length of the filter;

let the wavelet packet coefficient of the electroencephalogram signal x (t) of the kth node of the jth layer be as follows:

whereinAnddetail coefficients and approximation coefficients of wavelet packet decomposition of the kth node of the jth layer, respectively;

then, the wavelet packet coefficients of the five frequency bands are normalized, and the formula is as follows:

wherein, K_l(r) wavelet packet coefficient, p, for each frequency band_l(r) is the normalized wavelet packet coefficient; n is the length of the wavelet packet coefficient; l is the number of frequency segments;

finally, performing L-layer wavelet packet decomposition will generate 2^LA node; in the j-th layer decomposition, the signal is decomposed into a high frequency part and a low frequency part, and the decomposed high frequency part and the decomposed low frequency part are continued to be subjected toDecomposing the components to obtain a new high-frequency part and a new low-frequency part; the decomposition is repeated until the rhythm waves corresponding to the five frequency bands are matched.

According to an embodiment of the present invention, the step of calculating the nonlinear characteristics of the abnormal brain discharge waveform using the hestert index comprises:

estimation was performed using R/S analysis with X ═ X1; x 2; the following steps are carried out; xC represents a channel in the epoch of the electroencephalogram signal, and the R/S analysis method has the following calculation formula:

wherein H is the Herster index; r is the maximum distance, i.e. the difference between the maximum and minimum values of X; s is the standard deviation of X; c is the length of X.

According to an embodiment of the invention, after the time-frequency feature and the nonlinear feature are extracted, linear discriminant analysis is adopted to reduce the dimension of the feature parameter, and then a trained random forest model is adopted to obtain a first detection result of the abnormal discharge of the electroencephalogram signal.

According to an embodiment of the present invention, a morphological filter is used to extract spike features, which includes the steps of:

first, since the shape of the spike is similar to a triangle, the triangle structural element g (k) is selected, and the formula is as follows:

wherein A is the center height of the structural element and D is half the width of the structural element; k is the number of sampling points in a time series;

then, each channel of the electroencephalogram signal is divided into electroencephalogram segments with a certain length, and the geometric characteristics of each electroencephalogram segment are matched with preset structural elements by using a morphological filter.

According to an embodiment of the present invention, the process of matching the geometric features in each electroencephalogram segment with the preset structural elements by using the morphological filter is as follows:

firstly, using the selected structural elements to carry out open-close and close-open operations on each electroencephalogram segment signal, wherein the formula is as follows:

wherein, Y (k) represents a background signal, and X (k) represents each electroencephalogram segment signal; k is the number of sampling points in one electroencephalogram segment;

then, calculating the difference between the background signal and each electroencephalogram segment signal, wherein the calculated result is a spike signal, and the formula is as follows:

z (k) ═ x (k) -y (k) formula eight

Wherein, Z (k) is the difference between the background signal and each electroencephalogram segment signal;

wherein SR is the spike ratio, expressed as the mean of the spike signal Z (k); num_j(i) The number of spikes in the ith epoch of the jth channel is represented, t represents the duration of each electroencephalogram signal segment, and k is the number of sampling points in one electroencephalogram segment.

According to an embodiment of the invention, when the first detection result and the second detection result both represent that the detected electroencephalogram signal is in abnormal discharge, the final result after the first detection result and the second detection result are fused is the abnormal discharge.

In accordance with an embodiment of the invention, a 5 th order Butterworth bandpass filter with a frequency range of 0.5Hz to 32Hz is used to remove noise and artifacts from EEG signals during data pre-processing.

Correspondingly, the invention also provides a brain abnormal discharge detection system based on the electroencephalogram signal characteristics, which comprises a data acquisition and processing module, a characteristic extraction module, a first detection module, a second detection module and a fusion module. The data acquisition and processing module acquires original electroencephalogram signals and performs data processing. The feature extraction module extracts multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features. The first detection module obtains a first detection result of abnormal discharge of the electroencephalogram signal by adopting a trained random forest model based on the extracted time-frequency characteristics and nonlinear characteristics. The second detection module obtains a second detection result of abnormal discharge of the electroencephalogram signal based on the relation between the spike ratio in the extracted spike characteristics and the set threshold value. And the fusion module fuses the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result.

In summary, the method and system for detecting abnormal brain discharge based on electroencephalogram features provided by the invention are based on detection of multi-domain features and spike features, the method fuses the first detection result based on the multi-domain features on the basis of the second detection result based on the spike features, and the two detection results are fused to form the final detection result of abnormal brain discharge. The fusion of the first detection result based on the multi-domain features weakens the relevance degree of threshold selection in the existing threshold-based spike wave automatic detection method, so that the detection precision is greatly improved.

In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

Fig. 1 is a flowchart of a method for detecting abnormal brain discharge based on electroencephalogram characteristics according to an embodiment of the present invention.

Fig. 2 is a flowchart illustrating the extraction of multi-domain features from the preprocessed electroencephalogram signal in fig. 1.

Fig. 3 is a flow chart showing the process of using wavelet packet transformation to establish the time-frequency relationship between different rhythmic waves and five frequency bands.

FIG. 4 is a flow chart illustrating random forest model generation.

FIG. 5 is a flow chart illustrating spike feature extraction using a morphological filter.

Fig. 6 is a schematic block diagram of a brain abnormal discharge detection system based on electroencephalogram characteristics according to an embodiment of the present invention.

Detailed Description

As shown in fig. 1, the method for detecting abnormal brain discharge based on electroencephalogram characteristics according to the present embodiment includes: the original brain electrical signal is acquired and data processing is performed (step S1). Extracting multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features (step S2). And obtaining a first detection result of the abnormal discharge of the electroencephalogram signal by adopting the trained random forest model based on the extraction time-frequency characteristics and the nonlinear characteristics (step S3). A second detection result of abnormal discharge of the electroencephalogram signal is obtained based on the relationship between the spike ratio in the extracted spike characteristic and the set threshold (step S4). And fusing the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result (step S5). The working principle of the brain abnormal discharge detection method based on electroencephalogram signal characteristics provided by the present embodiment will be described in detail below with reference to fig. 1 to 5.

The method for detecting abnormal brain discharge based on electroencephalogram characteristics provided by the embodiment starts in step S1, and in the step, a multi-lead electroencephalograph is used for collecting long-range monitoring electroencephalograms of a patient, the sampling frequency is 256Hz, the electrode distribution adopts the international 10-20 electroencephalogram collection standard, and 8-channel electroencephalogram data are collected in total. During data processing, a 5-order Butterworth band-pass filter is adopted to filter frequency components above 32Hz and below 0.5Hz, and the interference of noise and artifacts is reduced.

Thereafter, step S2 is performed, which includes extraction of multi-domain features and extraction of spike features. For the extraction of multi-domain features, the specific process is as follows:

and step S21, performing statistical analysis on the electroencephalogram signals to obtain time domain characteristics, wherein the time domain characteristics comprise line length, kurtosis, skewness, maximum value, minimum value, mean value and variance characteristics of the waveform.

Step S22, extracting different rhythm waves from the EEG signal by wavelet packet transformation, decomposing the EEG signal into five frequency bands and corresponding the different rhythm waves and the five frequency bands to establish a time-frequency relationship, wherein the five frequency bands are respectively as follows: 0.5 Hz-4 Hz wave, 4 Hz-8 Hz theta wave, 8H-12 Hz alpha wave, 12 Hz-23 Hz beta 1 wave and 23 Hz-30 Hz beta 2 wave.

In this step, the step of establishing a time-frequency relationship between the different rhythmic waves and the five frequency bands includes:

step S221, a scale function and a wavelet basis function are selected. Specifically, db6 is chosen as the wavelet basis function and each epoch is defined to be 5 seconds;

step S222, extracting different rhythm waves from each epoch through wavelet packet transformation, wherein the formula is as follows:

wherein Φ (t) is a scale function;is a wavelet basis function; i (0,1, 2.., 2.)^j-1) number of nodes for wavelet packet transformation; j is a decomposition layer; h (n) and g (n) ═ 1)^1-nh (1-n) is a pair of orthogonal image filters;andis a wavelet basis function based on dual scales; n is the length of the filter;

let the wavelet packet coefficient of the electroencephalogram signal x (t) of the kth node of the jth layer be as follows:

wherein

Anddetail coefficients and approximation coefficients of wavelet packet decomposition of the kth node of the jth layer, respectively;

step S223, decomposing the electroencephalogram signal into five frequency bands, which are: 0.5 Hz-4 Hz wave, 4 Hz-8 Hz theta wave, 8H-12 Hz alpha wave, 12 Hz-23 Hz beta 1 wave and 23 Hz-30 Hz beta 2 wave.

Step S224, normalizing the wavelet packet coefficients of the five frequency bands, wherein the formula is as follows:

wherein K_l(r) wavelet packet coefficient, p, for each frequency band_l(r) is the normalized wavelet packet coefficient; n is the length of the wavelet packet coefficient; l is the number of frequency segments;

and step S225, decomposing the rhythm waves extracted from the wavelet packet of each layer and finding out the corresponding frequency band. Specifically, wavelet packet decomposition is performed on the L-th layer electroencephalogram signal, and 2 is generated^LA node; when the j layer is decomposed, the signal is decomposed into a high-frequency part and a low-frequency part, and the decomposed high-frequency part and the decomposed low-frequency part are continuously decomposed to obtain a new high-frequency part and a new low-frequency part; the decomposition is repeated until the rhythm waves corresponding to the five frequency bands are matched.

And step S23, calculating the nonlinear characteristics of the abnormal discharge waveform of the brain by adopting a Hurst index. The hurst index is a non-linear measure of the brain electrical signal, and its calculation is estimated using R/S analysis with X ═ X1; x 2; the following steps are carried out; xC represents a channel in the epoch of the electroencephalogram signal, and the R/S analysis method has the following calculation formula:

wherein H is the Herster index; r is the maximum distance, i.e. the difference between the maximum and minimum values of X; s is the standard deviation of X; c is the length of X.

And executing step S24 after the time-frequency characteristics and the nonlinear characteristics are extracted, and reducing the dimension of the characteristic parameters by adopting linear discriminant analysis. Since the final detection results are both seizure and non-seizure, the dimension is reduced to 2-dimension. However, the present invention is not limited thereto.

In step S2, spike features are extracted using a morphological filter, which includes the steps of:

in step S24, a structural element is selected. Since the shape of the spike is approximately triangular, the triangular structure element g (k) is chosen, whose formula is expressed as follows:

where A is the center height of the structuring element, D is half the width of the structuring element, and k is the number of sampling points in a time series.

Typically, the spike is 20 to 70 milliseconds in width; when the sampling frequency is 256Hz, 2L has a value of 2 to 9. Therefore, when the value of D is 5, the features in the electroencephalogram signal can be effectively extracted.

And step S25, dividing each channel of the electroencephalogram signal into electroencephalogram segments with certain lengths. In this embodiment, each electroencephalogram segment is 1 second in length. However, the present invention is not limited thereto.

And step S26, matching the geometric characteristics of each electroencephalogram segment with preset structural elements by using a morphological filter. Specifically, as shown in fig. 5, the process of matching the geometric features in each electroencephalogram segment with the preset structural elements by using the morphological filter is as follows:

step S261, performing on-off and on-off operations on each electroencephalogram segment signal using the structural element selected in step S24, and the formula is as follows:

wherein, Y (k) represents a background signal, and X (k) represents each electroencephalogram segment signal; k is the number of sampling points in one electroencephalogram segment;

step S262, calculating the difference between the background signal and each electroencephalogram segment signal, wherein the calculation result is a spike signal, and the formula is as follows:

z (k) ═ x (k) -y (k) formula eight

Wherein, Z (k) is the difference between the background signal and each electroencephalogram segment signal;

and finally calculating the spike ratio SR according to the following formula nine:

wherein SR is the spike ratio, expressed as the mean of the spike signal Z (k); num_j(i) The number of spikes in the ith epoch of the jth channel is represented, t represents the duration of each electroencephalogram signal segment, and k is the number of sampling points in one electroencephalogram segment.

Although the present embodiment numbers each step in the form of a serial number, the multi-domain feature extraction of steps 22 to 23 and the spike feature extraction of steps 24 to 26 are parallel, and the execution order of both is not limited at all.

After the feature extraction of step S2 is completed, steps S3 and S4 are performed; likewise, the present invention does not limit the execution sequence of steps S3 and S4. For step S3, a trained random forest model is used to obtain a first detection result of abnormal discharge of the electroencephalogram signal based on the extracted time-frequency features and nonlinear features. The training of the random forest model in this embodiment will be described in detail below with reference to fig. 4. In order to train the random forest model, a sample database needs to be established first. Firstly, the method in step S1 is adopted to collect a large number of electroencephalograms of experimental subjects of different sexes and different ages, thereby forming a training sample set. And then, marking a plurality of electroencephalogram signal samples by a professional doctor, and marking the abnormal discharge waveform in each channel of the electroencephalogram signals. Then, the electroencephalogram signals are filtered, and step S2 is executed to extract multi-domain features in each electroencephalogram signal sample, so as to form feature vectors. Training a random forest model by taking multi-domain characteristics in abnormal discharge waveforms in each electroencephalogram signal as input, wherein the training steps are as follows:

first, a new training set having the same number of samples as the training set is extracted in a training sample set so as to be replaced.

Then, the feature vector of each sample is randomly and unreplaceably sampled from the new training set to form a candidate feature vector set.

And then, selecting features from the feature vectors to be selected by using the uncertainty of the kini to generate a node, calculating the optimal splitting mode of each node, splitting until the node cannot be split, and marking as a leaf node.

And then, repeating the splitting step and adding the recorded leaf nodes into the decision tree.

And repeating the steps, continuously adding the leaf nodes into the decision tree and storing the generated decision tree until the data of the decision tree meets the requirement of generating the random forest.

In step S3, the multi-domain features obtained in step S2 are input into the trained random forest model for prediction, so as to obtain a first detection result of abnormal discharge of the electroencephalogram signal.

In step S4, a spike ratio is calculated and a threshold is set to form a second detection result. Specifically, when the spike ratio is greater than a threshold value, abnormal discharge of the brain is judged; otherwise, the discharge is not abnormal.

And finally, executing a step S5, fusing the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result. In this embodiment, the electroencephalogram signal is considered to be in abnormal discharge when both the first detection result and the second detection result indicate that the electroencephalogram signal is in abnormal discharge. However, the present invention is not limited in any way to the manner of fusion. In other embodiments, each detection result may be added with a weight for fusion.

Correspondingly, as shown in fig. 6, the present embodiment further provides a brain abnormal discharge detection system based on electroencephalogram signal characteristics, which includes a data acquisition and processing module 1, a characteristic extraction module 2, a first detection module 3, a second detection module 4, and a fusion module 5. The data acquisition and processing module 1 acquires original electroencephalogram signals and performs data processing. The feature extraction module 2 extracts multi-domain features and spike wave features in the preprocessed electroencephalogram signals, wherein the multi-domain features comprise time-frequency features and nonlinear features. The first detection module 3 obtains a first detection result of the abnormal discharge of the electroencephalogram signal by adopting a trained random forest model based on the extraction time-frequency characteristics and the nonlinear characteristics. The second detection module 4 obtains a second detection result of abnormal discharge of the electroencephalogram signal based on the relationship between the spike ratio in the extracted spike characteristic and the set threshold value. And the fusion module fuses 5 the first detection result and the second detection result to obtain a final abnormal cerebral discharge detection result.

The working principle of the brain abnormal discharge detection system based on electroencephalogram signal characteristics provided by this embodiment is as described in steps S1 to S5, which is not described herein again.

In summary, the method and system for detecting abnormal brain discharge based on electroencephalogram features provided by the invention are based on detection of multi-domain features and spike features, the method fuses the first detection result based on the multi-domain features on the basis of the second detection result based on the spike features, and the two detection results are fused to form the final detection result of abnormal brain discharge. The fusion of the first detection result based on the multi-domain features weakens the relevance degree of threshold selection in the existing threshold-based spike wave automatic detection method, so that the detection precision is greatly improved.

Although the present invention has been described with reference to the preferred embodiments, it should be understood that various changes and modifications can be made therein by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.