Brain activeness index extraction method based on electroencephalogram signals
By preprocessing and modeling resting-state EEG signals, brain activity indicators are extracted, solving the problem of large individual differences in rTMS treatment and enabling stable prediction of rTMS treatment effects and provision of individualized treatment plans.
Patent Information
- Application Number
- CN202510808705.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-17
- Publication Date
- 2025-11-14
AI Technical Summary
When using repetitive transcranial magnetic stimulation (rTMS) to treat depression and cognitive impairment, there are significant individual differences and a lack of effective objective biomarkers to predict treatment response, resulting in low standardized response rates for treatment regimens.
By preprocessing resting-state EEG signals, brain activity indicators are extracted using the pairwise maximum entropy model and the sorted vector entropy method. Combined with the improvement rate of N-back task indicators, the cognitive improvement trend after rTMS treatment is predicted.
It enables stable and objective prediction of the effects of rTMS treatment, provides a basis for individualized treatment plans, and improves the accuracy and effectiveness of treatment.
Smart Images
Figure CN120938338A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of brain science technology, and in particular to a method for extracting brain activity indicators based on electroencephalogram (EEG) signals. Background Technology
[0002] Repetitive transcranial magnetic stimulation (rTMS) is widely used in the treatment of mental and neurological disorders, particularly demonstrating significant efficacy in the intervention of depression and cognitive impairment. Current research indicates that depression, as a severe mental illness, is often accompanied by impairments in attention, executive function, and working memory. High-frequency rTMS stimulation of the left dorsolateral prefrontal cortex (Left DLPFC) at 10 Hz can significantly improve patients' working memory performance, and when applied in the treatment of depression, it can further improve patients' depressive symptoms. However, the efficacy of rTMS treatment varies considerably among individuals. This heterogeneity limits the response rate of standardized rTMS treatment regimens, and some patients may experience ineffective long-term treatment. Therefore, identifying objective biomarkers that can predict treatment response remains a crucial issue. Summary of the Invention
[0003] Purpose of the invention: The purpose of this invention is to provide a method for extracting brain activity indicators based on electroencephalogram (EEG) signals, which can be used to predict the level of cognitive improvement in individuals after rTMS treatment, provide a scientific basis for the treatment of depression and other related neurological diseases, and help doctors develop more precise treatment plans.
[0004] Technical Solution: A method for extracting brain activity indicators based on electroencephalogram (EEG) signals, comprising the following steps: S1, preprocesses the raw resting-state EEG signal and selects multiple channels of interest; S2 converts the preprocessed EEG signal into a binary EEG sequence, and uses a pairwise maximum entropy model to construct an energy topography map representing brain state transitions; based on the model probabilities of all instantaneous brain states obtained from the maximum entropy model, the energy values of all instantaneous states are calculated. S3. Based on the instantaneous state transition sequence of the brain in a resting state and the energy values of all instantaneous states of the brain, a time series of energy changes in the brain in a resting state is generated; the entropy value of the time series is calculated by applying the sorted vector entropy method to obtain a quantitative brain activity index. S4. Correlation analysis was performed on the baseline brain activity of the subjects and the improvement rate of N-back indicators after one course of rTMS treatment. Based on the correlation analysis, the cognitive improvement trend of clinical depression patients in rTMS intervention was predicted.
[0005] Furthermore, the preprocessing steps for the raw resting-state EEG signals include: S11 locates the original EEG signal channel and removes electrooculography (EOG) and electromyography (EMG); S12, select EEG signals from 1 to 100 Hz and use a notch filter from 49.5 to 50.5 Hz to remove 50 Hz power frequency interference; S13, identify channels with obvious drift or severe noise as bad conductors, and use the built-in Spherical algorithm of EEGLAB to perform interpolation on bad conductors; S14, the selected EEG signal is divided into non-overlapping 1-second segments, and segments containing extreme values with amplitudes exceeding 100 microvolts and segments with a standard deviation greater than 15 are removed. S15, the EEG signal is truncated into signals of uniform length, with the length selected as 50s; S16, perform rereference in the form of whole-brain average reference; S17 uses independent principal component analysis to perform inverse operations on the EEG signal and decompose it into several components, and removes artifact components to obtain a relatively clean EEG signal.
[0006] Furthermore, step S2, the step of obtaining the binarized EEG sequence, includes: S211, For each selected EEG channel, perform a Hilbert transform to obtain an analytical signal, and use the absolute value of the analytical signal as the amplitude envelope. ; S212, The system divides the time frame into multiple non-overlapping time windows. For each EEG channel within each time window, the average EEG signal value for each channel is calculated. This average value is then used as the binarization threshold for that EEG channel within that time window. Values greater than the threshold are assigned +1, and values less than the threshold are assigned -1. Thus, each EEG channel is transformed into a binary sequence, with the i-th EEG channel represented as... The binarized sequence of the i-th EEG channel is represented as: : , in, T represents the length of the EEG signal. This indicates that the i-th brainwave channel is active at time point t. This indicates that the i-th brainwave channel is inactive at time point t. Therefore, at each time point t, we obtain an N-dimensional vector: , in, This represents the brain state at time point t, and there are a total of There are several states, where N represents the number of EEG channels selected.
[0007] Further, in step S2, after binarizing the EEG signal, the original resting-state EEG signal is transformed into a sequence of brain activity state transitions. The observed brain activity state transition sequence is then fitted to a Boltzmann distribution using a pairwise maximum entropy model to obtain the model probability of each brain state, and the energy value of each state is calculated. The specific steps are as follows: In the maximum entropy model, the average value for each EEG channel of interest is calculated as follows: , The average value of the product of each pair of EEG channels of interest is: , Will , As a constraint, the system entropy is maximized. The expression is: , Brain states derived from Boltzmann distribution probability : , in, Represents brain state Energy: , in, Represents brainwave channels The initial activation level, Represents brainwave channels and The strength of their interaction; Represents brainwave channels In brain state Activation status; ; Parameter optimization using gradient ascent algorithm and Iteratively update parameters and The expression is as follows: , , , , Among them, scaling parameters The initial value is set to 0.1, and the parameters are optimized. and Until the gradient value is less than 10 -5 .
[0008] Furthermore, the specific steps for calculating brain activity indicators are as follows: Let the sampling frequency and low-pass frequency of the time series X be respectively... and Select a sub-window containing W sample points from X, where the sample points are integers. The interval is defined as: , If the child window starts from the first Starting with one sample, we get a vector: , vector The data is sorted according to the magnitude of the signal amplitude corresponding to each sample point and converted into a sorting vector. : , For a sub-window of length W, we have Possible sorting vectors; mapping sorting vectors to state symbols ,Depend on A sorting vector mapping table is formed by sorting vectors, and each sorting vector... Corresponding to an index value Thus, any sorting vector is converted into a state symbol; When the sub-window slides across the complete time series, multiple sorting vectors are generated, thus an energy conversion sequence of length T yields a sequence of length T. A sequence of state symbols; based on the frequency of each state symbol in the entire state symbol transition sequence. Calculate its entropy value RVE: , Entropy value (RVE) was used as an indicator of brain activity in the subjects.
[0009] Furthermore, the N-back indicators include reaction time RT0 and accuracy ACC0 for the 0-back task, and reaction time RT2 and accuracy ACC2 for the 2-back task. The improvement value of the N-back indicators refers to the improvement value of the four N-back indicators compared with the baseline immediately after stimulation. Pearson correlation tests were performed on the baseline brain activity of the subjects and the improvement values of the four N-back indicators after receiving one course of rTMS treatment. If a significant statistical relationship was found between the two, the activity indicators were considered to have predictive potential for cognitive ability. Based on significant correlations, a prediction function was fitted using brain activity indicators as independent variables and behavioral indicators as dependent variables.
[0010] Compared with the prior art, the significant advantages of this invention are as follows: This invention utilizes a pairwise maximum entropy model to analyze electroencephalogram (EEG) signals, calculating the energy values of all instantaneous states. Based on the instantaneous state transition sequence of the brain in a resting state and the energy values of all instantaneous states, a time series of energy changes in the brain in a resting state is generated. The entropy value of the time series is calculated using the sorted vector entropy method, resulting in a quantified brain activity index. This allows for the acquisition of stable and objective individual neural state indicators from resting-state EEG, providing auxiliary decision support for neuromodulation methods such as rTMS. Experimental results show that this invention has good predictive efficacy, enabling early assessment of intervention benefits and providing a basis for personalized treatment. Attached Figure Description
[0011] Figure 1 This is a flowchart of the present invention; Figure 2 This is a schematic diagram illustrating the construction process of the pairwise maximum entropy model; Figure 3 This is a schematic diagram illustrating the process of extracting the entropy of the sorted vector. Figure 4 This is a flowchart of the experiment with healthy subjects in Example 1; Figure 5 This is a schematic diagram showing the correlation analysis results between the baseline brain activity of the subjects and the immediate increase in RT0 during the 0-back response in Example 1. Figure 6 This is a flowchart of the experiment with depressed subjects in Example 2; Figure 7 This is a schematic diagram showing the correlation analysis results between the baseline brain activity of the subjects and the immediate increase in 2-back accuracy (ACC2) after stimulation in Example 2. Detailed Implementation
[0012] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0013] like Figure 1 As shown, this invention provides a method for extracting brain activity indicators based on electroencephalogram (EEG) signals, comprising the following steps: Step 1: Preprocess the EEG signals; The raw resting-state EEG signal underwent routine preprocessing such as bandpass filtering and artifact removal, and 10 channels representing the whole brain distribution were selected. Specifically, the process included the following steps: Step 11: Import the raw EEG signal, locate the channels of the raw EEG signal, and remove useless channels such as electrooculography (EOG) and electromyography (EMG).
[0014] Step 12, Filtering: Bandpass filtering, select EEG signals of 1-100Hz, and use a notch filter of 49.5-50.5Hz to remove 50Hz power frequency interference.
[0015] Step 13, Interpolate bad leads: Identify channels with significant drift or severe noise as bad leads, and use EEGLAB's built-in Spherical algorithm to interpolate the bad leads.
[0016] Step 14, Remove bad segments: Divide the selected EEG signal into non-overlapping 1-second segments, remove segments containing extreme values with amplitudes exceeding 100 microvolts, and segments with a standard deviation greater than 15.
[0017] Step 15, Truncate to a uniform length: Truncate the EEG signal into signals of uniform length to ensure data consistency in subsequent analysis. The length is selected as 50s (50,000 sampling points). Step 16, Rereference: Rereference the EEG signals of uniform length using a whole-brain average reference format.
[0018] Step 17, Independent Principal Component Analysis (ICA): Using EEGLAB, perform independent principal component analysis on the rereferenced EEG signal to decompose it into several components. Remove artifact components such as blinking components and muscle movement components to obtain a relatively clean EEG signal.
[0019] Step 18, Selecting Channels: In this embodiment, 10 channels of interest distributed throughout the brain are selected as research objects. The 10 selected EEG channels are: FP1, FP2, F3, F4, C3, C4, P3, P4, O1, and O2.
[0020] Step 2, Binarization of EEG state and maximum entropy modeling; Using a pairwise maximum entropy model to analyze EEG (electroencephalogram) data, an energy topography map representing brain state transitions is constructed; based on the model probabilities of all instantaneous brain states obtained from the maximum entropy model, the energy values of all instantaneous states are calculated. Figure 2 As shown, the specific steps include: Step 21: Construct a binary EEG sequence; The processed, clean EEG signal contains N (in this embodiment, N=10) EEG channels of interest: The EEG signal of each EEG channel is represented as , T represents the length of the EEG signal. First, the hilbert function in Matlab is used to analyze the EEG signal of each EEG channel. Perform a Hilbert transform to obtain the analytic signal. The amplitude envelope of the EEG signal is obtained by taking the absolute value of the analytical signal using the abs function in Matlab. .
[0021] After that, The system is divided into multiple non-overlapping time windows, each 1 second long, corresponding to 1000 sampling points. For each EEG channel within each time window, the average value of the EEG signal for each channel is calculated. This average value is then used as the binarization threshold for that EEG channel within that time window. Values greater than the threshold are assigned +1, and values less than the threshold are assigned -1. Thus, each EEG channel is transformed into a binary sequence, with the i-th EEG channel represented as... The binarized sequence of the i-th EEG channel is represented as: : (1) Where T is the length of the EEG signal. This indicates that the i-th brainwave channel is active at time point t. This indicates that the i-th brainwave channel is inactive at time point t.
[0022] Therefore, at each time point t, we obtain an N-dimensional vector: (2) in, This represents the activation state of the i-th EEG channel at time t; This represents the brain state at time point t, and there are a total of A type of brain state, represented as , N represents the number of EEG channels selected.
[0023] Therefore, the original brainwave signals are represented as this A sequence of state transitions.
[0024] Step 22: Fit a pairwise maximum entropy model; After binarization, the original resting-state EEG signal is transformed into a sequence of brain activity state transitions. The observed brain activity state transition sequence is fitted to the Boltzmann distribution using a pairwise maximum entropy model to obtain the model probability of each brain state, and the energy value of each state is calculated accordingly. The specific steps are as follows: In the maximum entropy model, the average value for each EEG channel of interest is calculated as follows: (3) The average value of the product of each pair of EEG channels of interest is: (4) , These two quantities, as fixed characteristics representing the system's properties, serve as constraints to maximize the system's entropy. : (5) Brain states derived from the Boltzmann distribution probability : (6) in, Represents brain state Energy: (7) in, Represents brainwave channels The initial activation level, Represents brainwave channels and The strength of their interaction Represents brainwave channels In brain state The activation status.
[0025] Parameter optimization using gradient ascent algorithm and The two parameters are updated iteratively using formulas (8) and (9): (8) (9) in, and The calculation formula is as follows: (10) (11) Among them, scaling parameters The initial value is set to 0.1, and the parameters are optimized. and Until the gradient value is less than 10 -5 .
[0026] The parameters are obtained iteratively according to formulas (8)-(11). and Substituting this into formula (7), we obtain all states. energy value By substituting the energy values of all states into the time series of the binarized EEG signal, a state transition sequence with energy labels is obtained. : (12) in, Indicates the brain state at time T. The energy value.
[0027] Step 3, extraction of brain activity indicators; Based on the instantaneous state transition sequence of the brain in a resting state and the energy values of all instantaneous brain states, a time series of energy changes in the brain in a resting state is generated; the entropy value of the time series is calculated using the sorted vector entropy method, resulting in a quantified index of brain activity. Figure 3 As shown, the specific method for calculating brain activity indicators is as follows: Let X be a time series, and let X have a sampling frequency and a low-pass frequency of 1 / 2. and Select a sub-window from X containing W sample points, where the sample points are integers. For intervals, Defined as: (13) If the child window starts from the first Starting with one sample, we can obtain a vector: (14) vector The data is sorted according to the magnitude of the signal amplitude corresponding to each sample point and converted into a RankVector. : (15) For a sub-window of length W, we have There are several possible sort vectors. Assume we have a table (as shown in Table 1) that maps the sort vectors to state symbols. Table 1 is derived from this. It consists of 120 sort vectors. If W=5, there will be 120 possible sort vectors, each of which... The corresponding index value can be found through Table 1. This converts any sorting vector into a state symbol. For example, a sample within a sub-window is... Sort these five sample points in ascending order, and the corresponding position index values form a sorting vector. This indicates that the minimum value is the third value and the maximum value is the fifth value. Then, in Table 1, the value [2,3,1,4,5] corresponds to number 37, which is the status symbol. .
[0028] Table 1 Sorting Vector Mapping Table
[0029] As the sub-window slides across the complete time series, multiple sorting vectors are generated, which are then converted into multiple state symbols. Ultimately, a state transition sequence of length T with energy labels can be used to obtain a sequence of length T. The state symbol transition sequence; the frequency of each state symbol in the entire state symbol transition sequence is counted. And calculate its entropy value RVE: (16) Entropy value (RVE) was used as an indicator of brain activity in the subjects.
[0030] Step 4: Correlation analysis and prediction model construction were performed on the baseline brain activity of the subjects and the improvement values of the four N-back indicators after one course of rTMS treatment. Based on the correlation analysis, the cognitive improvement trend of clinical depression patients in rTMS intervention was predicted. The specific method for predicting the effect is as follows: Working memory of the subjects is measured using four indicators in the N-back task: 0-back task reaction time RT0 and accuracy ACC0, and 2-back task reaction time RT2 and accuracy ACC2. The improvement of these four N-back indicators from baseline immediately after stimulation is calculated to measure the improvement of working memory in the immediate period after one course of rTMS treatment. These improvement values are represented as: RT0_diff, ACC0_diff, RT2_diff, and ACC2_diff. Pearson correlation tests are performed on the baseline brain activity of the subjects and the improvement values of the four N-back indicators after one course of rTMS treatment. If a significant statistical relationship is found (p < 0.05), the activity indicator is considered to have predictive potential for cognitive ability. Based on the significant correlation, a linear regression model is further constructed, with brain activity indicators as independent variables and behavioral indicators as dependent variables, to fit the prediction function: (17) Where x is the brain activity index, y is the predicted improvement in cognitive task performance, and a and b are the parameters of the prediction function.
[0031] The constructed regression model is used to generate a prediction curve, enabling numerical prediction of the improvement in working memory indicators. This prediction curve can be used to estimate the improvement in an individual's cognitive state after rTMS intervention.
[0032] Example 1 In this embodiment, 23 healthy subjects (HE) without mental or neurological disorders were selected, and their resting-state EEG data were collected during the baseline period.
[0033] like Figure 4 As shown, the EEG acquisition frequency was 1000Hz, using a 64-channel acquisition system, and the acquisition time was 5 minutes. Subsequently, the signal underwent standard preprocessing (including bandpass filtering, artifact removal, rereference, etc.), and 10 representative channels distributed in the frontal lobe, parietal lobe, occipital lobe, and central region were selected.
[0034] The preprocessed EEG signals were subjected to Hilbert transform, and the absolute value of the analytic signal was taken as the amplitude envelope. Subsequently, the amplitude envelopes of the 10-channel analytic signals within each time window (1 second in length, 1000 sample points) were binarized. The state sequence was modeled based on the pairwise maximum entropy model (pMEM), and the energy of each brain state was calculated to construct a brain state energy topography map. Within each sliding time window, the transition trajectory of the brain state on the energy topography map was tracked, and its ranking vector entropy was calculated over the entire time series, serving as an indicator of the subject's brain activity.
[0035] Subsequently, high-frequency rTMS stimulation (10Hz, left dorsolateral prefrontal cortex DLPFC, lasting 40 minutes) was applied to each participant, and reaction times RT2, RT0 and accuracy ACC2, ACC0 for the 2-back and 0-back tasks in the n-back task were recorded before and after stimulation. The working memory improvement rate after rTMS (the improvement of n-back indicators compared to baseline: RT2_diff, RT0_diff, ACC2_diff, ACC0_diff) was calculated, and Pearson correlation analysis was performed with the participant's baseline brain activity. Figure 5 As shown, the results indicate that there is a significant positive correlation between baseline activity and the increase in 0-back reaction time RT0_diff (r = 0.4225, p = 0.0446), indicating that the index of this invention can effectively predict the degree of cognitive improvement after rTMS in healthy individuals.
[0036] Example 2 In this embodiment, 18 subjects clinically diagnosed with major depressive disorder (MDD) were selected. Their resting-state EEG signals were collected before rTMS treatment, and the acquisition settings were the same as in Example 1. After acquisition, the same preprocessing and 10-channel selection operations were performed. The flowchart of the depressive subject experiment is shown below. Figure 6As shown.
[0037] The brain activity index of the subject was extracted using the method of this invention, and the improvement of the 2-back accuracy (ACC2) in the n-back task compared to baseline was evaluated immediately after the subject received one treatment. The improvement of ACC2 before and after rTMS was used as the dependent variable (ACC2_diff) to analyze its relationship with the baseline brain activity index.
[0038] like Figure 7 As shown, the results indicate that there is a significant positive correlation between this indicator and the improvement in 2-back accuracy (r = 0.5497, p = 0.0181), indicating that the activity index proposed in this invention is not only applicable to healthy individuals, but can also effectively predict the trend of cognitive improvement in clinical patients with depression during rTMS intervention.
[0039] In summary, the brain activity index proposed in this invention can effectively predict an individual's response level to rTMS intervention. In representative healthy individuals and clinically depressed individuals, the baseline brain activity calculated using this method showed a significant positive correlation with the improvement rate of working memory task scores (n-back task reaction time and accuracy) after rTMS, preliminarily verifying its feasibility and practical value as a predictive neuromarker.
Claims
1. A method for extracting brain activity indicators based on electroencephalogram (EEG) signals, characterized in that, The steps include the following: S1, preprocesses the raw resting-state EEG signal and selects multiple channels of interest; S2 converts the preprocessed EEG signals into binary EEG sequences and uses a pairwise maximum entropy model to construct an energy topography map representing brain state transitions. Based on the model probabilities of all instantaneous states of the brain obtained from the maximum entropy model, calculate the energy value of all instantaneous states; S3. Based on the instantaneous state transition sequence of the brain in a resting state and the energy values of all instantaneous states of the brain, a time series of energy changes in the brain in a resting state is generated; the entropy value of the time series is calculated by applying the sorted vector entropy method to obtain a quantitative brain activity index. S4. Correlation analysis was performed on the baseline brain activity of the subjects and the improvement rate of N-back indicators after one course of rTMS treatment. Based on the correlation analysis, the cognitive improvement trend of clinical depression patients in rTMS intervention was predicted.
2. The method for extracting brain activity indicators based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The steps for preprocessing raw resting-state EEG signals include: S11 locates the original EEG signal channel and removes the electrooculogram (EOG) and electromyogram (EMG) channels. S12, select EEG signals from 1 to 100 Hz and use a notch filter from 49.5 to 50.5 Hz to remove 50 Hz power frequency interference; S13, identify channels with obvious drift or severe noise as bad conductors, and use the built-in Spherical algorithm of EEGLAB to perform interpolation on bad conductors; S14, the selected EEG signal is divided into non-overlapping 1-second segments, and segments containing extreme values with amplitudes exceeding 100 microvolts and segments with a standard deviation greater than 15 are removed. S15, the EEG signal is truncated into signals of uniform length, with the length selected as 50s; S16, perform rereference in the form of whole-brain average reference; S17 uses independent principal component analysis to perform inverse operations on the EEG signal and decompose it into several components, and removes artifact components to obtain a relatively clean EEG signal.
3. The method for extracting brain activity indicators based on electroencephalogram (EEG) signals according to claim 1, characterized in that, Step S2, the steps for obtaining the binarized EEG sequence, include: S211, For each selected EEG channel, perform a Hilbert transform to obtain an analytical signal, and use the absolute value of the analytical signal as the amplitude envelope. ; S212, The system divides the time frame into multiple non-overlapping time windows. For each EEG channel within each time window, the average EEG signal value for each channel is calculated. This average value is then used as the binarization threshold for that EEG channel within that time window. Values greater than the threshold are assigned +1, and values less than the threshold are assigned -1. Thus, each EEG channel is transformed into a binary sequence, with the i-th EEG channel represented as... The binarized sequence of the i-th EEG channel is represented as: : ， in, T represents the length of the EEG signal. This indicates that the i-th brainwave channel is active at time point t. This indicates that the i-th brainwave channel is inactive at time point t. Therefore, at each time point t, we obtain an N-dimensional vector: ， in, This represents the brain state at time point t, and there are a total of There are several states, where N represents the number of EEG channels selected.
4. The method for extracting brain activity indicators based on electroencephalogram (EEG) signals according to claim 3, characterized in that, In step S2, after binarizing the EEG signal, the original resting-state EEG signal is transformed into a sequence of brain activity state transitions. The observed brain activity state transition sequence is then fitted to a Boltzmann distribution using a pairwise maximum entropy model to obtain the model probability of each brain state, and the energy value of each state is calculated. The specific steps are as follows: In the maximum entropy model, the average value for each EEG channel of interest is calculated as follows: ， The average value of the product of each pair of EEG channels of interest is: ， Will , As a constraint, the system entropy is maximized. The expression is: ， Brain states derived from Boltzmann distribution probability : ， in, Represents brain state Energy: ， in, Represents brainwave channels The initial activation level, Represents brainwave channels and The strength of their interaction; Represents brainwave channels In brain state Activation status; ; Parameter optimization using gradient ascent algorithm and Iteratively update parameters and The expression is as follows: ， ， ， ， Among them, scaling parameters The initial value is set to 0.1, and the parameters are optimized. and Until the gradient value is less than 10 -5 .
5. The method for extracting brain activity indicators based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The specific steps for calculating brain activity indicators are as follows: Let the sampling frequency and low-pass frequency of the time series X be respectively... and Select a sub-window containing W sample points from X, where the sample points are integers. The interval is defined as: ， If the child window starts from the first Starting with one sample, we get a vector: ， vector The data is sorted according to the magnitude of the signal amplitude corresponding to each sample point and converted into a sorting vector. : ， For a sub-window of length W, we have Possible sorting vectors; mapping sorting vectors to state symbols ,Depend on A sorting vector mapping table is formed by sorting vectors, and each sorting vector... Corresponding to an index value Thus, any sorting vector is converted into a state symbol; When the sub-window slides across the complete time series, multiple sorting vectors are generated, thus an energy conversion sequence of length T yields a sequence of length T. A sequence of state symbols; based on the frequency of each state symbol in the entire state symbol transition sequence. Calculate its entropy value RVE: ， Entropy value (RVE) was used as an indicator of brain activity in the subjects.
6. The method for extracting brain activity indicators based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The N-back indicators include reaction time RT0 and accuracy ACC0 for the 0-back task, and reaction time RT2 and accuracy ACC2 for the 2-back task. The improvement value of the N-back indicators refers to the improvement value of the four N-back indicators compared with the baseline immediately after stimulation. The baseline brain activity of the subjects and the improvement values of the four N-back indicators after receiving one course of rTMS treatment were respectively subjected to Pearson correlation test. If a significant statistical relationship was found between the two, the activity indicators were considered to have predictive potential for cognitive ability. Based on significant correlations, a prediction function was fitted using brain activity indicators as independent variables and behavioral indicators as dependent variables.