Teenager schizophrenia classification method
By combining discrete wavelet transform (DWT) and integrated empirical modal decomposition (EMD) to extract and fusion the EEG signals in adolescents, the problem in the prior art is solved that it is difficult to effectively distinguish adolescent schizophrenia patients from healthy control groups, and a higher classification accuracy is achieved.
Patent Information
- Application Number
- CN202411593061.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-02-19
- Filing Date
- 2024-11-08
- Publication Date
- 2025-04-25
AI Technical Summary
Prior Art In the auxiliary diagnosis of schizophrenia in adolescent schizophrenia, it is difficult to effectively distinguish adolescent schizophrenia patients from healthy control groups when using discrete wavelet transform (DWT) or integrated empirical modal decomposition (EMD).
The combination of DWT and EEMD is used to decompose the signal of the EEG signal in adolescents, extract the standard deviation, kurtitude and center of gravity frequency characteristics, and improve the classification accuracy through feature fusion.
The accuracy of schizophrenia classification in adolescents was improved, with higher classification accuracy in all models than when using DWT or EEMD alone, and with higher classification accuracy compared to other studies on the same data set.
Smart Images
Figure CN119867751A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of machine learning, and in particular to a method for classifying adolescent schizophrenia. Background Art
[0002] Schizophrenia is one of the most common mental illnesses currently. Patients will have abnormal behaviors in perception, thinking, and other aspects, as well as incoordination of mental activities. Studies have shown that compared with adult schizophrenia, the clinical symptoms of adolescent schizophrenia are more severe. At the same time, adolescent schizophrenia patients are often accompanied by some dangerous behaviors when they become ill, which in turn increases the burden on their families and society. With the continuous development of brain-computer interfaces and artificial intelligence technologies in recent years, machine learning technology has gradually begun to be used in the auxiliary diagnosis of mental illnesses such as depression, anxiety, and schizophrenia. It is one of the current research hotspots for schizophrenia identification methods.
[0003] Discrete wavelet transform (DWT) and ensemble empirical mode decomposition (EMD) have been widely used in the research of brain-computer interface such as motor imagery and epilepsy diagnosis in recent years due to their excellent nonlinear signal decomposition capabilities. Among them, DWT can decompose EEG signals into multiple frequency bands of different scales, which can better describe the detailed characteristics of EEG signals, and EEMD is an improved method of EMD, which can decompose signals according to the time scale characteristics of signals and effectively improve the mode aliasing phenomenon caused by EMD. Many previous studies have proved the effectiveness of DWT and EEMD in analyzing EEG signals. However, most of the current studies on schizophrenia still use DWT or EEMD alone to study EEG signals. Summary of the invention
[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a method for classifying adolescent schizophrenia to solve the problems raised in the above background technology.
[0005] A method for classifying adolescent schizophrenia comprises the following steps:
[0006] S1. Data acquisition, adolescent EEG acquisition target schizophrenia EEG dataset;
[0007] S2. Data processing: Data preprocessing: Preprocess the target schizophrenia EEG data to obtain the target frequency band schizophrenia EEG data, perform signal decomposition on the target frequency band schizophrenia EEG data to obtain decomposed EEG signal data, extract features from the decomposed EEG signal data based on the decomposed EEG signal data extracted by different methods, and perform feature fusion on the extracted EEG signal features;
[0008] S3. Data output, EEG signal characteristics.
[0009] Preferably, in S2, discrete wavelet decomposition is performed on the target schizophrenia EEG data to obtain decomposed EEG signal data.
[0010] Preferably, in S2, integrated empirical mode decomposition of superimposed Gaussian white noise is performed on the target schizophrenia EEG data to obtain decomposed EEG signal data.
[0011] Preferably, the feature extraction and feature fusion steps include:
[0012] SS1, extract features from the two decomposed EEG signal data to obtain standard deviation, kurtosis and centroid frequency;
[0013] SS2: The standard deviation, kurtosis and centroid frequency extracted by the two methods are normalized and serially connected to obtain the fused EEG signal features.
[0014] Preferably, the EEG signal features in the data output are fused EEG signal features.
[0015] Preferably, the original EEG signal is filtered at 0.5-50 Hz to obtain the target EEG signal data. For the input EEG signal S(t), its continuous wavelet transform can be defined as:
[0016]
[0017] Where f is the mother wavelet, a represents the scale factor, and b represents the time translation factor. The EEG signal is processed using discrete wavelet transform. The a in the wavelet basis function is discretized as m, and the b in the wavelet basis function is discretized as n. The discrete wavelet transform of the EEG signal S(t) is defined as
[0018]
[0019] The original EEG signal is decomposed four times using Daubechies4's db4 as the mother wavelet, and the result of discrete wavelet transform is used for feature extraction.
[0020] Preferably, the original EEG signal is filtered at 0.5-50 Hz to obtain the target EEG signal data, a white Gaussian noise signal g(t) with an amplitude of a is added to the original EEG signal S(t), and then the noise-added S(t) is subjected to EMD decomposition to obtain K IMF components. and a residual component Repeat the above process N times to obtain N IMF components of the noisy signal. and the residual component Respectively and Calculate the average value, and finally get the Kth IMF component of the original EEG signal S(t):
[0021]
[0022] The residual component is
[0023]
[0024] Preferably, the decomposition signals of DWT and EEMD are overlapped and sampled using a 3-second time window and a 1-second step size, and the data segments of the DWT and EEMD decomposition signals are divided. The standard deviation, kurtosis and centroid frequency extracted are used as the features of each data segment, and each data segment is recorded as
[0025] X={x(i)|i=1,2,…,N}, N=384.
[0026] The formula for calculating the standard deviation is
[0027]
[0028] The calculation formula of kurtosis is
[0029]
[0030] The formula for calculating the center of gravity frequency is:
[0031]
[0032] Wherein, ω1 represents the upper limit of the frequency band of the current data segment, ω2 represents the lower limit of the frequency band of the current data segment, and P(ω) represents the power spectrum density of the current data segment.
[0033] Preferably, the features extracted based on DWT are recorded as The features extracted based on EEMD are recorded as Respectively and Perform normalization and then use the normalized and Perform serial connection to obtain DWT-EEMD fusion features Right now
[0034] F Fusion =[F DWT ,F EEMD ]
[0035] Beneficial effects of the present invention: When the adolescent schizophrenia classification method uses the DWT-EEMD fusion feature for classification, the classification accuracy of all models is higher than the accuracy when only the DWT or EEMD features are used for classification. Compared with other research results on the same data set in recent years, it has a higher classification accuracy, which proves the effectiveness of the method proposed in the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 Schematic diagram of discrete wavelet decomposition of the present invention.
[0037] Figure 2 The discrete wavelet transform results of the EEG signals of the healthy control group (a) and the adolescent schizophrenia patients (b) of the present invention are shown.
[0038] Figure 3 The EEMD decomposition result diagram of the EEG signals of the healthy control group (a1) and the adolescent schizophrenia patient (b1) of the present invention.
[0039] Figure 4 Schematic diagram of the classification accuracy of each model of the present invention.
[0040] Figure 5 It is a schematic diagram of the classification of adolescent schizophrenia according to the present invention. DETAILED DESCRIPTION
[0041] Embodiment 1, a method for classifying adolescent schizophrenia, comprising the following steps:
[0042] S1. Data acquisition, adolescent EEG acquisition target schizophrenia EEG dataset (targeting the public schizophrenia EEG dataset of Lomonosov Moscow State University);
[0043] S2. Data processing, data preprocessing: preprocess the target schizophrenia EEG data to obtain the target frequency band schizophrenia EEG data (bandpass filtering), perform signal decomposition (using discrete wavelet transform decomposition) on the target schizophrenia EEG data, i.e., discrete wavelet decomposition, obtain decomposed EEG signals, and perform feature extraction on the decomposed EEG signals;
[0044] S3. Data output, EEG signal characteristics.
[0045] The original EEG signal is filtered at 0.5-50 Hz to obtain the target EEG signal data.
[0046] For the input EEG signal S(t), its continuous wavelet transform can be defined as
[0047]
[0048] Where f is the mother wavelet, a and b represent the scale factor and time translation factor respectively. In most studies, discrete wavelet transform is often used to process EEG signals. The a and b in the wavelet basis function are discretized and recorded as m and n respectively. Then the discrete wavelet transform of the EEG signal S(t) is defined as
[0049]
[0050] Use Daubechies4's db4 as the mother wavelet to decompose the original EEG signal four times. The decomposition diagram is shown in the figure. Figure 1 As shown in the figure, A1, A2, A3, and A4 represent the approximate coefficients obtained by decomposition, and D1, D2, D3, and D4 represent the detail coefficients obtained by decomposition. Since D4 has more high-frequency noise, D2, D3, D4, and A4 are selected as the results of discrete wavelet transform for feature extraction. The discrete wavelet transform results of the EEG signals of the healthy control group (a) and the adolescent schizophrenia patients (b) are shown in the figure. Figure 2 shown.
[0051] Embodiment 2, a method for classifying adolescent schizophrenia, comprising the following steps:
[0052] S1. Data acquisition, adolescent EEG data acquisition target schizophrenia EEG dataset (targeting the public schizophrenia EEG dataset of Lomonosov Moscow State University);
[0053] S2. Data processing: Data preprocessing: preprocess the target schizophrenia EEG data to obtain the target frequency band schizophrenia EEG data (bandpass filtering), perform signal decomposition on the target schizophrenia EEG data (using integrated empirical mode decomposition of superimposed Gaussian white noise), and perform feature extraction on the decomposed EEG signal;
[0054] S3. Data output, EEG signal characteristics.
[0055] The original EEG signal is filtered at 0.5-50 Hz to obtain the target EEG signal data.
[0056] A white Gaussian noise signal g(t) with an amplitude of a is added to the original EEG signal S(t), and then the noise-added S(t) is decomposed by EMD to obtain K IMF components c(t) and a residual component r(t). The above process is repeated N times to obtain N IMF components c(t) of the noise-added signal. K，X (t) and the residual component r X (t), respectively for c K，X (t) and r X (t) calculates the average value, and finally obtains the Kth IMF component of the original EEG signal S(t):
[0057]
[0058] The residual component is
[0059]
[0060] The EEMD white noise amplitude a is set to 0.2, the number of integrations N is set to 100, and IMF1, IMF2, IMF3 and IMF4 with larger correlation coefficients are selected as the EEMD results for feature extraction. The EEMD decomposition results of the EEG signals of the healthy control group (a1) and the adolescent schizophrenia patients (b1) are shown in Figure 3 shown.
[0061] Embodiment 3, a method for classifying adolescent schizophrenia, comprising the following steps:
[0062] S1. Data acquisition, adolescent EEG acquisition target schizophrenia EEG dataset (targeting the public schizophrenia EEG dataset of Lomonosov Moscow State University);
[0063] S2. Data processing, data preprocessing: preprocess the target schizophrenia EEG data to obtain the target frequency band schizophrenia EEG data (bandpass filtering), perform signal decomposition on the target schizophrenia EEG data (using discrete wavelet transform decomposition), that is, discrete wavelet decomposition, perform signal decomposition on the target schizophrenia EEG data (using integrated empirical mode decomposition of superimposed Gaussian white noise, based on the decomposed EEG signal data extracted by the two different methods, obtain the decomposed EEG signal features of feature fusion;
[0064] S3. Data output, fused EEG signal features.
[0065] The decomposition signals of DWT and EEMD are overlapped and sampled using a 3-second time window and a 1-second step size. The data segments of the DWT and EEMD decomposition signals are divided, and the standard deviation, kurtosis, and centroid frequency extracted are used as the features of each data segment. Each data segment is recorded as X = {x(i)|i = 1, 2, ..., N}, N = 384. The calculation formula of the standard deviation is
[0066]
[0067] The calculation formula of kurtosis is
[0068]
[0069] The formula for calculating the center of gravity frequency is:
[0070]
[0071] Among them, ω1 and ω2 represent the upper and lower limits of the frequency band of the current data segment, respectively, and P(ω) represents the power spectrum density of the current data segment. The features extracted based on DWT are recorded as The feature extracted based on EEMD is denoted as F EEMD , respectively for F DWT and F EEMD Normalize it, and then use the normalized F DWT and F EEMD Perform serial connection to obtain the DWT-EEMD fusion feature F Fushion Right now
[0072] F Fusion =[F DWT ,F EEMD ]
[0073] Combining Example 1, Example 2 and Example 3, the final classification is into a classification model, that is, classification into health and schizophrenia.
[0074] The classification results of Example 1 and Example 2 are compared. The classification results of the three methods are as follows: Figure 4 shown.
[0075] Based on the evaluation indicators such as accuracy, recall, precision, F1 score, and Kappa coefficient, the comprehensive performance of each machine learning model in classifying adolescent schizophrenia when using the DWT-EEMD fusion feature is analyzed. The results are shown in Table 1. The results show that the KNN model has the best classification performance in all evaluation indicators, with an accuracy of 97.48%, a recall of 95.71%, a precision of 99.56%, an F1 score of 97.59%, and a Kappa coefficient of 0.9494.
[0076] Table 1 Classification performance indicators of various machine learning models
[0077]
[0078] In summary, when the DWT-EEMD fusion feature is used for classification, the classification accuracy of all models of the adolescent schizophrenia classification method is higher than that when only DWT or EEMD features are used for classification. Compared with other research results on the same data set in recent years, it has a higher classification accuracy, which proves the effectiveness of the method proposed in the present invention.
Claims
1. A method for classifying adolescent schizophrenia, characterized in that: The following steps are involved: S1. Data acquisition, adolescent EEG acquisition target schizophrenia EEG dataset; S2. Data processing: Data preprocessing: Preprocess the target schizophrenia EEG data to obtain the target frequency band schizophrenia EEG data, perform signal decomposition on the target frequency band schizophrenia EEG data to obtain decomposed EEG signal data, extract features from the decomposed EEG signal data based on the decomposed EEG signal data extracted by different methods, and perform feature fusion on the extracted EEG signal features; S3. Data output, EEG signal characteristics.
2. A method for classifying adolescent schizophrenia according to claim 1, characterized in that: In the S2, discrete wavelet decomposition is performed on the target schizophrenia EEG data to obtain decomposed EEG signal data.
3. A method for classifying adolescent schizophrenia according to claim 2, characterized in that: In the S2, integrated empirical mode decomposition of superimposed Gaussian white noise is performed on the target schizophrenia EEG data to obtain decomposed EEG signal data.
4. A method for classifying adolescent schizophrenia according to claim 3, characterized in that: The feature extraction and feature fusion steps include: SS1, respectively extracting features from the two decomposed EEG signal data to obtain standard deviation, kurtosis and center of gravity frequency; SS2. The standard deviation, kurtosis and center of gravity frequency extracted by the two methods are normalized and serially connected to obtain the fused EEG signal features.
5. A method for classifying adolescent schizophrenia according to claim 4, characterized in that: The EEG signal features in the data output are fused EEG signal features.
6. A method for classifying adolescent schizophrenia according to claim 5, characterized in that: The original EEG signal is filtered at 0.5-50 Hz to obtain the target EEG signal data. For the input EEG signal S(t), its continuous wavelet transform can be defined as: Where f is the mother wavelet, a represents the scale factor, and b represents the time translation factor. The EEG signal is processed using discrete wavelet transform. The a in the wavelet basis function is discretized as m, and the b in the wavelet basis function is discretized as n. The discrete wavelet transform of the EEG signal S(t) is defined as The original EEG signal is decomposed four times using Daubechies4's db4 as the mother wavelet, and the result of discrete wavelet transform is used for feature extraction.
7. A method for classifying adolescent schizophrenia according to claim 6, characterized in that: The original EEG signal is filtered at 0.5-50Hz to obtain the target EEG signal data. The white Gaussian noise signal g(t) with amplitude a is added to the original EEG signal S(t). Then, the noise-added S(t) is decomposed by EMD to obtain K IMF components c(t) and a residual component r(t). The above process is repeated N times to obtain N IMF components c(t) of the noise-added signal. K，N (t) and the residual component r N (t), respectively for c K，N (t) and r N (t) calculates the average value, and finally obtains the Kth IMF component of the original EEG signal S(t): The residual component is 8. A method for classifying adolescent schizophrenia according to claim 7, characterized in that: The decomposition signals of DWT and EEMD are overlapped and sampled using a 3-second time window and a 1-second step size. The data segments of the DWT and EEMD decomposition signals are divided, and the extracted standard deviation, kurtosis and centroid frequency are used as the features of each data segment. Each data segment is denoted as X = {x(i)|i = 1, 2, …, N}, N = 384. The formula for calculating the standard deviation is The calculation formula of kurtosis is The formula for calculating the center of gravity frequency is: Wherein, ω1 represents the upper limit of the frequency band of the current data segment, ω2 represents the lower limit of the frequency band of the current data segment, and P(ω) represents the power spectrum density of the current data segment.
9. A method for classifying adolescent schizophrenia according to claim 8, characterized in that: The feature extracted based on DWT is denoted as F DWT , the feature extracted based on EEMD is recorded as F EEMD , respectively for F DWT and F EEMD Normalize it, and then use the normalized F DWT and F EEWD Perform serial connection to obtain the DWT-EEMD fusion feature F Fusion ,Right now F Fusion =[F DWT ,F EEMD ]