阅读说明：本技术 一种区别精神***症幻听症状和其他疾病幻听症状的方法 (Method for distinguishing auditory hallucination symptoms of schizophrenia from other diseases ) 是由 林小东 陈策 禚传君 于 2020-06-22 设计创作，主要内容包括：本发明属于医学信号处理领域,具体公开了一种区别精神分裂症幻听症状和其他疾病幻听症状的方法,基于多频段小波包熵的不对称率提取特征为主要客观依据,根据患者幻听症状量表评分为辅助客观依据,以汉密尔顿抑郁量表和杨氏躁狂量表为辅助客观依据,进行融合计算,并参照正常状态下,健康个体的脑电信号构建多分类器系统,所建立的分类算法,能较为准确的区分出精神分裂症幻听症状与抑郁症幻听和双相障碍幻听等2种不同疾病种类幻听的特异性特征,最终可以输出待检测脑电信号的精神分裂症特有的幻听症状所属概率值,得到支持分裂症幻听早期确诊决策的建议。(The invention belongs to the field of medical signal processing, and particularly discloses a method for distinguishing schizophrenia auditory hallucinations from other diseases auditory hallucinations, which takes asymmetry extraction characteristics based on multi-band wavelet packet entropy as a main objective basis, performing fusion calculation by taking the auditory symptom scale score of the patient as an auxiliary objective basis and taking the Hamilton depression scale and the Young mania scale as auxiliary objective basis, and a multi-classifier system is constructed by referring to the electroencephalograms of healthy individuals under a normal state, the established classification algorithm can accurately distinguish the specific characteristics of the schizophrenia auditory hallucination symptoms and 2 different disease types of auditory hallucinations such as depression auditory hallucinations and bipolar disorder auditory hallucinations, and finally the probability value of the specific auditory hallucinations of the schizophrenia of the electroencephalograms to be detected can be output, so that the suggestion supporting the early diagnosis decision of the schizophrenia auditory hallucinations is obtained.)

1. A method of distinguishing auditory hallucinations of schizophrenia from other diseases, comprising the steps of:

acquiring electroencephalogram data;

preprocessing the acquired data;

acquiring multi-scale wavelet energy spectrums of segmented electroencephalogram signals of all leads of an individual;

solving the asymmetry ratio among all leads to obtain the characteristic mode vector of the individual segmented electroencephalogram signals;

evaluating the scale score of the individual to be tested to obtain an individual clinical characteristic vector;

constructing a classifier of a multi-core fusion support vector machine according to the feature mode vector of the segmented electroencephalogram signal and the feature vector of the clinical scale;

and analyzing the electroencephalogram signals of the patient with the auditory hallucinations symptom by using a multi-core fusion classifier, and predicting whether the auditory hallucinations symptom is the specific auditory hallucinations of the schizophrenia disease.

2. A method of distinguishing between schizophrenia phantom symptoms and other diseases phantom symptoms according to claim 1, wherein said pre-treatment comprises:

positioning an electrode; re-referencing; correcting the artifact; filtering; segmenting; deleting errors, tried, deleted and rebuilding bad leads; removing the artifacts; removing the bad section; baseline correction and normalization.

3. The method for distinguishing the auditory hallucinations of schizophrenia from other diseases according to claim 1, wherein the obtaining the multi-scale wavelet energy spectrum of the segmented brain electrical signals of all leads of the individual comprises:

performing wavelet packet decomposition on the individual segmented electroencephalogram signals one by one to obtain a multilayer decomposition tree;

and (4) calculating the wavelet packet entropy of the nodes in the tree to obtain the multi-scale wavelet energy spectrum.

4. The method for distinguishing the auditory hallucinations of schizophrenia from other diseases according to claim 3, wherein wavelet packet decomposition is performed on electroencephalogram data to obtain information of different frequency bands.

5. The method of distinguishing auditory hallucinations of schizophrenia from other diseases according to claim 1, wherein said deriving the asymmetry ratio between leads comprises:

the individual segmented electroencephalogram data comprises multi-lead signals, wherein half of the leads are from the left half brain, and half of the leads are from the right lead;

and (3) solving the asymmetric rate as a characteristic value by utilizing the combination of the leads of the left half brain and the right half brain to form a characteristic mode vector of the electroencephalogram signal.

6. The method for differentiating the auditory hallucinations of schizophrenia from other diseases according to claim 5, wherein the characteristic pattern vector of the brain electrical signals is a vector consisting of asymmetric rates of wavelet packet entropy of left and right brain leads in segmented and specific frequency band components.

7. The method of distinguishing between schizophrenia phantom auditory symptoms and other maladies phantom auditory symptoms according to claim 1, wherein said scoring an individual's tested scale comprises:

selecting related scores of a auditory symptom scale, a Hamilton depression scale and a Young mania scale to form a feature vector;

and standardizing the characteristic vectors and giving weights to the characteristic vectors to obtain the characteristic vectors which finally represent clinical behaviours.

8. The method of distinguishing between schizophrenia and auditory hallucinations of other diseases according to claim 1, wherein the multi-classifier model is constructed based on a classification accuracy maximization criterion.

9. The method of claim 8, wherein the similarity or difference between the EEG feature pattern vectors and clinical scale feature vectors of different individuals can be measured by any Euclidean distance.

10. The method of distinguishing auditory hallucinations for schizophrenia from other diseases according to claim 8, wherein the fusion of multinuclear functions is constructed based on two types of features of electroencephalogram signals and clinical scales.

11. A method of distinguishing between schizophrenia auditory hallucinations and other disease auditory hallucinations according to claim 8, wherein the multiple classifiers are trained and adapted based on schizophrenia auditory hallucinations, depression auditory hallucinations, bipolar disorder auditory hallucinations, and normal control data.

12. The method of claim 1, wherein the electroencephalogram signal and clinical scale information of the test individual are analyzed to obtain auditory hallucinations characteristics and classification information.

Technical Field

The invention belongs to the field of medical signal processing, and particularly relates to a method for distinguishing the auditory hallucinations symptoms of schizophrenia from other diseases.

Background

At present, electroencephalogram plays more and more important roles in basic research and clinical application of brain activity, the current research methods for electroencephalogram are mainly divided into linear methods and nonlinear analysis, the linear methods comprise time domain analysis, frequency domain analysis and frequency domain analysis, the time domain analysis mainly takes waveform characteristics as the leading factor and can only reflect the resolution of the electroencephalogram in the time domain, the frequency domain analysis mainly converts the electroencephalogram into electroencephalogram power information through time-frequency transformation and explores the change condition and rhythm distribution condition of the electroencephalogram along with frequency, the typical representative of the time-frequency domain analysis method is wavelet analysis which can simultaneously process the electroencephalogram in the time domain and the frequency domain, and the nonlinear analysis methods for the electroencephalogram have correlation dimension, Lyapunov index, entropy-based analysis method and the like.

The electroencephalogram signal is a time-varying and non-stationary nonlinear dynamic signal, the linear analysis method cannot effectively extract electroencephalogram data characteristics and has certain limitation, in the electroencephalogram nonlinear analysis method based on the information theory, wavelet packet entropy is an effective characteristic extraction mode, the wavelet packet entropy is based on wavelet packet transformation, rhythm signals can be extracted through the wavelet packet transformation, frequency resolution better than that of wavelet transformation is achieved, the stability of the signals is almost not required, the electroencephalogram signal is decomposed through wavelet packets to obtain different rhythm signals, the spectrum entropy of the electroencephalogram signal is calculated through an energy spectrum, the concentration or dispersion degree of a power spectrum of the electroencephalogram signal can be reflected, and each rhythm of the electroencephalogram signal can be effectively extracted and the local complexity of the electroencephalogram signal can be estimated.

The early diagnosis of the schizophrenia patient with the auditory symptom has important significance for the individual diagnosis and treatment of the schizophrenia, a large amount of researches neglect the specific auditory hallucinations in the schizophrenia to carry out electroencephalogram analysis, and the practical application limitation is realized, so that the multi-classification prediction model for the schizophrenia with the auditory symptom is built, and the deep research significance and the wide application value are realized.

Disclosure of Invention

The present invention is directed to a method for distinguishing the auditory hallucinations of schizophrenia from other diseases, which solves the problems set forth in the background art.

In order to achieve the purpose, the invention provides the following technical scheme:

a method of distinguishing auditory hallucinations of schizophrenia from other illnesses, comprising the steps of:

acquiring electroencephalogram data;

preprocessing the acquired data;

acquiring multi-scale wavelet energy spectrums of segmented electroencephalogram signals of all leads of an individual;

solving the asymmetry ratio among all leads to obtain the characteristic mode vector of the individual segmented electroencephalogram signals;

evaluating the scale score of the individual to be tested to obtain an individual clinical characteristic vector;

constructing a classifier of a multi-core fusion support vector machine according to the feature mode vector of the segmented electroencephalogram signal and the feature vector of the clinical scale;

and analyzing the electroencephalogram signals of the patient with the auditory hallucinations symptom by using a multi-core fusion classifier, and predicting whether the auditory hallucinations symptom is the specific auditory hallucinations of the schizophrenia disease.

As a still further scheme of the invention: the pretreatment comprises the following steps: positioning an electrode; re-referencing; correcting the artifact; filtering; segmenting; deleting errors, tried, deleted and rebuilding bad leads; removing the artifacts; removing the bad section; baseline correction and normalization.

As a still further scheme of the invention: the acquiring of the multi-scale wavelet energy spectrum of the segmented electroencephalogram signals of all leads of an individual comprises the following steps:

performing wavelet packet decomposition on the individual segmented electroencephalogram signals one by one to obtain a multilayer decomposition tree;

and (4) calculating the wavelet packet entropy of the nodes in the tree to obtain the multi-scale wavelet energy spectrum.

As a still further scheme of the invention: and carrying out wavelet packet decomposition on the electroencephalogram signal data to acquire information of different frequency bands.

As a still further scheme of the invention: the step of calculating the asymmetry ratio among the leads comprises the following steps:

the individual segmented electroencephalogram data comprises multi-lead signals, wherein half of the leads are from the left half brain, and half of the leads are from the right lead;

and (3) solving the asymmetric rate as a characteristic value by utilizing the combination of the leads of the left half brain and the right half brain to form a characteristic mode vector of the electroencephalogram signal.

As a still further scheme of the invention: the characteristic mode vector of the electroencephalogram signal is a vector formed by asymmetric rates of wavelet packet entropies of components of segmented and specific frequency bands of left and right brain leads.

As a still further scheme of the invention: said scoring the subject's scale comprises:

selecting related scores of a auditory symptom scale, a Hamilton depression scale and a Young mania scale to form a feature vector;

and standardizing the characteristic vectors and giving weights to the characteristic vectors to obtain the characteristic vectors which finally represent clinical behaviours.

As a still further scheme of the invention: the multi-classifier model is constructed based on the classification accuracy maximization criterion.

As a still further scheme of the invention: the similarity or difference between the EEG feature pattern vectors and the clinical scale feature vectors of different individuals can be measured by any Euclidean distance.

As a still further scheme of the invention: constructing multi-core function fusion based on the two characteristics of the electroencephalogram signal and the clinical scale.

As a still further scheme of the invention: the multi-classifier is trained and debugged based on schizophrenia auditory hallucinations, depression auditory hallucinations, bipolar disorder auditory hallucinations and normal control data.

As a still further scheme of the invention: and carrying out predictive analysis on the electroencephalogram signals of the tested individuals and the clinical scale information to obtain auditory hallucinations characteristics and classification information.

Compared with the prior art, the invention has the beneficial effects that: the invention is based on the asymmetry extraction characteristic of the entropy of the multi-band wavelet packet as the main objective basis, the invention is divided into auxiliary objective basis according to the rating of the auditory symptom scale of the patient, the Hamilton depression scale and the Young mania scale are taken as the auxiliary objective basis, the fusion calculation is carried out, and the multi-classifier system is constructed by referring to the electroencephalogram of the healthy individual under the normal state, the established classification algorithm can accurately distinguish the auditory symptom of schizophrenia and the specific characteristics of the auditory symptoms of 2 different diseases such as depressive auditory hallucinations, bipolar disorder auditory hallucinations and the like, finally the probability value of the auditory symptom of the specific auditory hallucinations of the electroencephalogram to be detected can be output, and the suggestion supporting the early diagnosis decision of the auditory hallucinations is obtained, the system comprises a general preprocessing flow, a data extraction flow, model building and classification prediction output, and an interactive interface can realize the self-defined preprocessing, the classification and, The invention can predict whether the auditory hallucination symptom is the specific auditory hallucination of the schizophrenia disease by utilizing objective basis of individual electroencephalogram signals, scale evaluation and the like, and can provide objective basis for early diagnosis of the schizophrenia disease with the auditory hallucination symptom.

Detailed Description

The following describes in detail various problems involved in the technical solutions of the present invention. It should be noted that the described embodiments are only intended to facilitate the understanding of the invention and do not serve any limiting purpose.

The method collects multi-lead electroencephalogram signal data, and each individual electroencephalogram data is a 2-dimensional signal comprising lead information (1-dimensional) and time information (1-dimensional);

the collection of the EEG signals is completed on an EEG signal collecting instrument with a multi-lead electrode cap;

the specific parameters of the acquisition have no special requirements, the sampling time is preferably 2-4 minutes, and the head is kept still as much as possible in the data acquisition process.

Preprocessing electroencephalogram data, which mainly comprises the following steps: positioning an electrode; re-referencing; carrying out artifact correction on the continuous data; filtering; segmenting; deleting errors, tried, deleted and rebuilding bad leads; removing the artifacts; removing the bad section; baseline correction and normalization.

The preprocessed electroencephalogram signals are segmented signals, namely electroencephalogram data of the same individual are divided into a plurality of multi-lead signals with equal length, the data volume is amplified, and the electroencephalogram signals have dynamic characteristics;

performing wavelet packet decomposition on each lead data of the segmented signals;

wavelet packet transformation can decompose non-stationary signals into weighted sums of different-scale base wavelets, and can guarantee high frequency resolution of a high frequency band.

The wavelet packet function is expressed as:

where ψ (t) mother wavelet is used, and h (k), g (k) are weight coefficients.

The recursive relationships at levels j and j +1 are:

wavelet coefficientComprises the following steps:

thus, the electroencephalogram signal can be expressed as a plurality of wavelet packet sets, which correspond to different frequency ranges and can be changed as needed.

Defining electroencephalogram signal components using wavelet packet nodes

The energy in (1) is:

the total energy of the electroencephalogram signals is as follows:

defining the energy of a particular frequency band (signal subband) as E_sWhich is the sum of the energies of the contained signal components, normalized to be expressed as:

finally defining the wavelet packet entropy as:

S_wp＝-∑p_sln[p_s]

therefore, assuming that the electroencephalogram signal of an individual is represented by P (number of leads) × M (number of signals in a segmented short time window) × N (number of segments), the wavelet packet entropy characteristic thereof is P (number of leads) × N (number of segments), and the leads in each piece of data are from the left and right half brain signals, and assuming that 3 of them are from the left half brain (a1, a2, A3), and 3 of them are from the right half brain (B1, B2, B3), the combinations of the arrangements of the left and right half brains (a1, B1), (a1, B2), (a1, B3), (a2, B1), (a2, B2), (a2, B3), (A3, B1), (A3, B2), (A3, B3), the asymmetry ratio e of the wavelet packet entropy values is found as:

wherein R represents leads from the right semi-brain and L represents leads from the left semi-brain;

the data used for modeling is 16-lead electroencephalogram signals, and the left half brain and the right half brain are respectively 8-lead, so that 64 characteristic values can be obtained to form a characteristic vector of a signal in a certain frequency band corresponding to a data segment, and wavelet packet entropy and asymmetry rate are repeatedly calculated in different frequency bands to obtain a corresponding characteristic vector;

the final segmented electroencephalogram of the independent individual can use the F (frequency segment number) × 64 vector V_signalTo indicate.

The invention combines the related scores of the clinical behavior scale (auditory symptom scale, Hamilton depression scale and Young mania scale) to form a characteristic vector, the score weight in the scale is set as 1, and the normalized score is a vector V_behavior。

The classifier based on pattern recognition builds an optimal multi-classification model, and the feature vector V of multiple frequency bands is used in the training process_signalAnd clinical scale feature vector V_behaviorStraightening into a one-dimensional characteristic vector, and inputting the one-dimensional characteristic vector into a classifier;

in view of V_signalThe influence of the feature dimension on the classifier, the dimension reduction is realized by adopting a principal component analysis method in the implementation process of the invention, and the contribution rate of the selected components is ensured to reach 99 percent; then, carrying out multi-class training on the segmented data of the training individual by using a Support Vector Machine (SVM);

in particular, the invention will characterize V of different characteristics_signal、V_behaviorThe multi-core function combination method is used for feature fusion, so that the multi-core function has more accurate and stronger mapping capability and classification performance.

kernal＝αk₁(V_signal)+(1-αk₂(V_behavior))

The SVM kernel function can select a radial basis kernel function, a linear kernel function and the like to define distance measurement, and the distance measurement is output as a posterior probability value and a classification label;

the classification result of the classifier is determined by selecting corresponding parameters, particularly optimizing a multi-kernel function fusion parameter alpha, and performing ten-fold cross validation on a data set to generate a plurality of classifiers, wherein the classification labels are schizophrenia auditory hallucination data, depression auditory hallucination data and bipolar disorder auditory hallucination data and normal control data.

For new test data, firstly, preprocessing to obtain segment data and clinical scale data of the electroencephalogram signal to be predicted, and then sequentially predicting all segment data feature vectors through a built multi-classifier model system;

because the test data comprises a plurality of segmented results, the classifier also generates a plurality of segmented prediction conditions, and all prediction results are subjected to statistical analysis;

finally, the system outputs probability values and decision suggestions of all categories to which the segmented signals belong, different segmented prediction results can be comprehensively considered in practical application, and schizophrenia patients with auditory symptoms can be predicted.

The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention are equivalent to or changed within the technical scope of the present invention.