A method and system for detecting a neurofeedback non-responder

Abstract

The application provides a method and system for detecting non-responders of neurofeedback, which can extract features from electroencephalogram (EEG) signals after obtaining the EEG signals and behavior data, extract EEG features, calculate an intervention effect value according to the EEG features and the behavior data, evaluate the intervention result of neurofeedback training according to the intervention effect value, and then label the EEG category of each EEG feature. Then, a neural network model is trained according to the EEG features labeled with EEG category labels to obtain a neurofeedback detection classification model. The intervention effect value is used to represent the intervention effect of neurofeedback training. The method combines EEG data and behavior data at different stages to more comprehensively and accurately measure the intervention effect of neurofeedback training, thereby improving the detection accuracy of the neurofeedback result.

Description

Technical Field

[0001] This application relates to the field of neurofeedback detection technology, and in particular to a method and system for detecting neurofeedback non-responders. Background Technology

[0002] Neurofeedback, also known as EEG biofeedback or brainwave feedback, is an analytical method based on brain activity characteristics. Neurofeedback involves collecting a user's electroencephalogram (EEG) signals and extracting specific EEG activity characteristics as feedback indicators during neurofeedback training. These indicators are then used to predict psychological, brain structural, and neurophysiological indicators. Neurofeedback can assist users in learning self-regulation of neurophysiological parameters. For example, during neurofeedback training, neurofeedback indicators are mapped to feedback methods, providing real-time feedback through auditory, visual, or tactile stimuli. This feedback loop can induce changes in the user's brain function and behavior, enabling them to better control their neurophysiological parameters.

[0003] There are various neurofeedback training programs, but due to individual differences among users, neurofeedback training will produce different intervention effects for different users. Therefore, in order to assess the changing trends of users during neurofeedback training, the intervention effect of neurofeedback results can be evaluated to distinguish between responders and non-responders. Responders are users whose feedback parameters are higher than the evaluation index value during training. This involves detecting whether neurofeedback can change the user's EEG pattern and evaluating the effectiveness of neurofeedback training. When assisting users in self-regulating neurophysiological parameters, users with poor feedback results can be predicted and screened based on neurofeedback results, and their neurofeedback training programs can be adjusted accordingly.

[0004] However, the detection of neurofeedback results focuses on group differences and cannot meet the need for precise analysis of individuals. Indicators for assessing the magnitude of user intervention effects are easily affected by users' own baseline outliers, which may lead to errors in the assessment of the user intervention effect. Furthermore, since improvements in the EEG baseline do not necessarily lead to behavioral changes, assessment indicators may classify users without behavioral improvement as respondents, failing to meet the actual needs of users. Summary of the Invention

[0005] This application provides a method and system for detecting non-responders in neurofeedback, in order to solve the problem of errors in the detection of neurofeedback results.

[0006] Firstly, this application provides a method for detecting non-responders in neurofeedback, comprising:

[0007] Acquire a set of electroencephalogram (EEG) signals and behavioral data. The set of EEG signals includes multiple EEG signals, which include a baseline EEG signal and an interventional EEG signal. The baseline EEG signal is the EEG signal before neurofeedback training, and the interventional EEG signal is the EEG signal after neurofeedback training. The behavioral data consists of behavioral characteristics before and after neurofeedback training.

[0008] EEG features are extracted from the EEG signals, and an intervention effect value is calculated based on the EEG features and the behavioral data. The intervention effect value includes an EEG intervention effect value and a behavioral intervention effect value. The EEG intervention effect value is used to characterize the intervention effect of the EEG signals before and after neurofeedback training, and the behavioral intervention effect value is used to characterize the intervention effect of the target behavior before and after neurofeedback training.

[0009] The EEG category of the EEG features is labeled according to the intervention effect value;

[0010] A neural network model is constructed, and the neural network model is trained based on EEG features labeled with EEG categories to obtain a neural feedback detection classification model;

[0011] The EEG characteristics of the EEG signal to be tested are input into the neural feedback detection and classification model to obtain the classification result output by the neural feedback detection and classification model.

[0012] Secondly, this application provides a detection system for neurofeedback non-responders, including an electroencephalogram (EEG) signal acquisition device and a controller connected to the EEG signal acquisition device. The EEG signal acquisition device is used to acquire EEG signals, and the controller is configured to execute the following program steps:

[0013] Acquire a set of electroencephalogram (EEG) signals and behavioral data. The set of EEG signals includes multiple EEG signals, which include a baseline EEG signal and an interventional EEG signal. The baseline EEG signal is the EEG signal before neurofeedback training, and the interventional EEG signal is the EEG signal after neurofeedback training. The behavioral data consists of behavioral characteristics before and after neurofeedback training.

[0014] EEG features are extracted from the EEG signals, and intervention effect values ​​are calculated based on the EEG features and behavioral data. The intervention effect values ​​include EEG intervention effect values ​​and behavioral intervention effect values. The EEG intervention effect values ​​are used to characterize the intervention effect of EEG signals before and after neurofeedback training, and the behavioral intervention effect values ​​are used to characterize the intervention effect of target behavior before and after neurofeedback training.

[0015] The EEG category of the EEG features is labeled according to the intervention effect value;

[0016] A neural network model is constructed, and the neural network model is trained based on EEG features labeled with EEG categories to obtain a neural feedback detection classification model;

[0017] The EEG characteristics of the EEG signal to be tested are input into the neural feedback detection and classification model to obtain the classification result output by the neural feedback detection and classification model.

[0018] As can be seen from the above technical solutions, this application provides a method and system for detecting non-responders in neurofeedback. The method, after acquiring EEG signals and behavioral data, performs feature extraction on the EEG signals, extracts EEG features, calculates the intervention effect value based on the EEG features and behavioral data, and evaluates the intervention results of neurofeedback training based on the intervention effect value, thereby labeling the EEG category of each EEG feature. Then, a neural network model is trained based on the EEG features labeled with EEG categories to obtain a neurofeedback detection classification model. The EEG signals include baseline EEG signals and intervention EEG signals, which are the EEG signals before and after neurofeedback training, respectively. The behavioral data are the target behavioral data before and after neurofeedback training, and the intervention effect value is used to characterize the intervention effect of neurofeedback training. By combining EEG data and behavioral data from different stages, this method more comprehensively and accurately measures the intervention effect of neurofeedback training, thereby improving the detection accuracy of neurofeedback results. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram showing the placement of the standard electrodes in an embodiment of this application;

[0021] Figure 2 This is a flowchart illustrating the process of labeling EEG signals in an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of the process of training the neurofeedback training model in the embodiments of this application;

[0023] Figure 4 This is a schematic diagram of a single-case experiment of type AB in the embodiments of this application;

[0024] Figure 5 This is a schematic diagram of the calculation matrix table in the embodiments of this application;

[0025] Figure 6This is a schematic diagram of the process for generating weighted EEG features in an embodiment of this application. Detailed Implementation

[0026] To make the objectives and implementation methods of this application clearer, the exemplary implementation methods of this application will be clearly and completely described below with reference to the accompanying drawings of the exemplary embodiments of this application. Obviously, the exemplary embodiments described are only some embodiments of this application, and not all embodiments.

[0027] Neurofeedback, also known as EEG biofeedback or brainwave feedback, is an intervention method based on brain activity characteristics. Neurofeedback involves collecting the user's electroencephalogram (EEG) signals and extracting specific EEG activity characteristics as feedback indicators during neurofeedback training. These indicators are then used to predict psychological, brain structural, and neurophysiological indicators. Neurofeedback can assist users in learning self-regulation of neurophysiological parameters. For example, during neurofeedback training, neurofeedback indicators are mapped to feedback methods, providing real-time feedback through auditory, visual, or tactile stimuli. This process can teach users how to enhance or inhibit specific neurophysiological parameters through operant conditioning, enabling them to autonomously regulate their brain activity. Neurofeedback can also be used to intervene in a user's psychological state and behavior for medical assistance.

[0028] To assess the changing trends of users during neurofeedback training, the intervention effect of neurofeedback results can be evaluated, thus distinguishing between responders and non-responders. Responders are those whose feedback parameters exceed the evaluation index value during training. This involves detecting whether neurofeedback can change the user's EEG pattern and evaluating the effectiveness of neurofeedback training. When assisting users in self-regulating neurophysiological parameters, users with different feedback results can be predicted and screened based on neurofeedback results, and their neurofeedback training programs can be adjusted accordingly.

[0029] When evaluating the effect size of neurofeedback outcomes, the evaluation results differ depending on the definition of effect size. In some embodiments, the evaluation indicators for the neurofeedback intervention effect include three: the difference between the feedback parameters of the first and last neurofeedback training sessions; the average difference between the feedback parameters of the remaining neurofeedback training sessions and the feedback parameters of the first neurofeedback training session; and the trend of feedback parameter changes throughout the entire neurofeedback training period using logistic regression analysis. If any one of the evaluation indicators is greater than 0, the neurofeedback outcome is considered to be of the responder type.

[0030] In some embodiments, because neurophysiological indicators are not only easily acquired and quantified, neurofeedback outcomes can be predicted based on these indicators. This is achieved by collecting the user's resting-state EEG activity before neurofeedback training and predicting the outcome based on resting-state neurophysiological indicators, thereby distinguishing between responders and non-responders. For example, Beta / Theta neurofeedback can predict the user's neurofeedback outcome using the resting-state Beta amplitude before training; Alpha neurofeedback can be predicted using the resting-state Alpha amplitude / power before training combined with other frequency bands; and SMR neurofeedback can be predicted using the resting-state SMR power before training.

[0031] In some embodiments, neurofeedback training can employ a single-case experiment design. Single-case experiments are used to analyze the relationship between one or more protocols and changes in individual physiological or behavioral outcomes. They can be used to evaluate the effectiveness of neurofeedback training. In a single-case experiment, individual participants undergo repeated evaluations at different stages based on one or more indicators, analyzing individual changes and allowing for better control over data variability. Single-case experiments include various types, such as AB design (interrupted time series design), reversal design, and alternation design.

[0032] In some embodiments, an AB-type single-case experiment can be employed, comprising a baseline phase (Phase A) and an intervention phase (Phase B). In the AB-type single-case experiment, characteristics prior to neurofeedback training are obtained in the baseline phase (Phase A), meaning no neurofeedback training is performed on the user in Phase A. Characteristics post-neurofeedback training are obtained in the intervention phase (Phase B), meaning neurofeedback training is performed on the user in Phase B. The AB-type single-case experiment allows for the assessment of the impact of intervention training on individual behavior with fewer data points and is simple and easy to implement.

[0033] However, the detection of neurofeedback results typically focuses on group differences, neglecting the analysis and quantification of individual differences within groups, thus failing to meet the need for precise individual analysis. Due to significant individual variability among users, indicators for assessing the intervention effect are easily influenced by users' own baseline outliers, potentially leading to errors in the assessment. Furthermore, since improvements in baseline EEG do not necessarily lead to behavioral changes, assessment indicators may classify users without behavioral improvement as respondents, failing to meet the actual needs of users.

[0034] To improve the accuracy of neurofeedback detection, some embodiments of this application provide a method for detecting neurofeedback non-responders, applied to a neurofeedback non-responder detection system. The detection system includes an electroencephalogram (EEG) signal acquisition device and a controller connected to the EEG signal acquisition device. The EEG signal acquisition device is used to acquire EEG signals. The EEG signal acquisition device can be an EEG signal sensor, which acquires EEG signals by placing electrode sensors outside the user's cerebral cortex. To improve signal accuracy, the EEG signal sensor can use semi-wet electrodes as the conductive medium to enhance conductivity.

[0035] Since the calculation of neurofeedback indices only involves 2-4 leads, in order to improve EEG signal processing performance and achieve real-time feedback, EEG signal acquisition equipment can use a three-lead EEG signal sensor, such as... Figure 1 The diagram shows a standard electrode placement. When acquiring EEG signals, two leads are placed in the Fp1 and Fp2 regions of the frontal brain to serve as the EEG signal acquisition potential, with region A1 or A2 used as the reference potential. Compared to EEG sensors with a larger number of leads, the three-lead EEG sensor meets the EEG signal acquisition requirements for neural feedback while improving EEG signal processing performance, effectively reducing the user's operational difficulty and minimizing device wear time.

[0036] In some embodiments, the EEG signal acquisition device can be connected to the controller wirelessly or via a wired connection. For example, the EEG signal acquisition device can wirelessly connect to the controller via Bluetooth to transmit EEG signals. The EEG data acquisition device may include digital circuits and analog circuits. The digital circuits include an A / D converter, a DSP (Digital Signal Processing), a USB (Universal Serial Bus) chip, a DC correction circuit, and an AC impedance detection circuit. The analog circuits include a preamplifier circuit and a filter circuit. The filter circuit consists of a notch filter circuit and a low-pass filter circuit, with the notch filter circuit handling power frequency interference during EEG signal acquisition.

[0037] The controller is then configured to execute the aforementioned detection method, such as... Figure 2 , Figure 3 As shown, Figure 2 This is a flowchart illustrating the process of labeling EEG signals in an embodiment of this application. Figure 3 This is a flowchart illustrating the training process of the neurofeedback training model in this application embodiment, specifically including the following:

[0038] S100, acquires the set of EEG signals.

[0039] The EEG signal set includes multiple EEG signals. The controller connects to the EEG signal acquisition device, which collects the EEG signals. During the acquisition process, the EEG signal acquisition device may be affected by noise signals, such as baseline drift caused by sensor heating, and power frequency noise from electronic devices and wires. Additionally, there are artifacts from the user's own EMG, EOG, and ECG signals. Therefore, after acquiring the EEG signals, preprocessing is required to filter out artifacts such as baseline drift, power frequency interference, EMG noise, and EOG noise to obtain clean EEG signal data.

[0040] Baseline drift is caused by the physical characteristics of the acquisition equipment and the contact characteristics of the electrodes, resulting in baseline interference, i.e., low-frequency noise, in the EEG signals transmitted by the EEG signal acquisition equipment. Therefore, median filtering can be used to remove baseline drift. Power frequency interference is caused by conducted interference from the equipment's power supply, resulting in some overlap with the frequency range of the EEG signals. Therefore, notch filters can be used to remove power frequency interference. For example, a 50Hz notch filter can be used to remove power frequency interference. Electromyographic (EMG) noise can be removed using an infinite-long unit impulse response (IIR) bandpass filter with a Blackman window function. Electrooculographic (EOG) noise is caused by eye movement or blinking. The frequency range of EOG noise overlaps to some extent with the frequency range of the EEG signals, and EOG noise significantly interferes with the EEG signals. Therefore, a model combining wavelet transform and Kalman filtering can be used to remove EOG noise. Wavelet transform is used to locate the EOG region in the EEG signal, constructing EOG artifacts, and then Kalman filtering is used to extract a clean EEG signal.

[0041] To reduce the impact of outliers and user-specific trend changes on intervention effect assessment, neurofeedback training employs an AB single-case experimental design, such as... Figure 4 As shown, it includes a baseline phase (Phase A) and an intervention phase (Phase B). The user's EEG signals are collected in the two phases respectively. That is, the EEG signals include baseline EEG signals and intervention EEG signals. The baseline EEG signals are the EEG signals before neurofeedback training, and the intervention EEG signals are the EEG signals after neurofeedback training.

[0042] The number of data collections refers to the number of times EEG signals are collected at each stage, with the dependent variable being the changes in the user's EEG signals. The number of EEG signal collections at different stages can be set according to needs. For example, to facilitate the analysis of EEG data trends, at least three observation points are required for analysis. Therefore, three EEG signal collections can be set at the baseline stage (with the time interval consistent with the training interval in the intervention stage) to reduce the impact of outliers and thus evaluate the intervention effect of neurofeedback training.

[0043] S200, acquire behavioral data.

[0044] To improve the accuracy of the detection, user behavioral data can be recorded during both the baseline and intervention phases. This behavioral data includes target behavioral data before and after neurofeedback training; specifically, user behavioral data is recorded during the baseline phase (Phase A) and the intervention phase (Phase B). By acquiring this behavioral data, the intervention effect of neurofeedback training can be evaluated based on both EEG and behavioral data. The behavioral data can be statistically analyzed in both phases using a scale-based approach.

[0045] For example, during the baseline phase, 5 minutes of resting-state EEG data and one scale data (behavioral data) are collected from the user weekly for 3 weeks, resulting in three sets of resting-state EEG data (EEG signals) and scale data (behavioral data) during the baseline phase. The semi-wet electrode type three-lead EEG signal sensor can be attached to the Fp1 and Fp2 electrode positions on the user's forehead while the user is relaxed with their eyes closed and remains quiet. The three-lead EEG signal sensor collects EEG signals and transmits the data to the controller in real time. The controller performs preprocessing such as noise reduction and artifact removal on the EEG signals before storing them.

[0046] During the intervention phase, users received 40 minutes of mindfulness-based neurofeedback training weekly, with 5 minutes of resting-state EEG data collected before each training session. A 2-minute interval was followed by 30 minutes of mindfulness-based neurofeedback training. After each training session, 5 minutes of resting-state EEG data and questionnaire data were collected after a 2-minute interval. This process continued for 8 weeks. In other words, during the intervention phase, specific neurofeedback training was conducted, and EEG and questionnaire data were collected.

[0047] Understandably, when neurofeedback is used to assist users in self-regulating neurophysiological parameters, the user's electroencephalogram (EEG) signals can be collected during neurofeedback training (e.g., mindfulness meditation training). After preprocessing and feature extraction of the EEG signals, the user's neural feedback index (FAA) is calculated and normalized. The normalized neural feedback index is then mapped; for example, different combinations of natural sounds are mapped based on the user's neural feedback index, where the volume of the natural sounds changes with the neural feedback training index, thus providing the user with auditory neural feedback stimulation. The user adjusts their self-state based on the auditory stimulation, thereby changing their own EEG signals.

[0048] S300 extracts EEG features from EEG signals.

[0049] After acquiring the EEG signal, feature extraction can be performed to extract EEG features. These features include linear and nonlinear characteristics. To more comprehensively analyze and predict neural feedback results, EEG bands such as Theta, Alpha, Beta, and Gamma waves are extracted from the EEG signal. Linear features of these EEG bands are calculated using frequency domain analysis, and nonlinear features are calculated using nonlinear dynamics methods. As shown in the table below, the EEG features include correlation features of the four EEG bands (Theta, Alpha, Beta, and Gamma waves), correlation features across all bands, and feedback features recorded based on the feedback paradigm, totaling 63 EEG features.

[0050]

[0051] S400 calculates the intervention effect value based on EEG characteristics and behavioral data.

[0052] After extracting EEG features, the intervention effect value can be calculated based on the EEG features and behavioral data to assess the intervention effect of neurofeedback training. Then, the EEG features can be classified to assess whether the user has benefited from neurofeedback training, thus distinguishing between responders and non-responders.

[0053] The intervention effect value is used to characterize the intervention effect of neurofeedback training. In order to measure the intervention effect more comprehensively and accurately, the intervention effect can be evaluated based on the EEG signals and behavioral data of the baseline and intervention phases. That is, the intervention effect value includes the EEG intervention effect value and the behavioral intervention effect value. The EEG intervention effect value is used to characterize the intervention effect of EEG signals before and after neurofeedback training, and the behavioral intervention effect value is used to characterize the intervention effect of target behavior before and after neurofeedback training.

[0054] In some embodiments, the intervention effect value can be calculated based on nonparametric methods, such as the Tau-U index calculation method (Tau for nonoverlap with baseline trend control). By measuring the nonoverlap between the baseline and the intervention phase and the phase trend, adverse baseline trends can be estimated and corrected, thereby measuring the size of the intervention effect more comprehensively and accurately.

[0055] When calculating the intervention effect value using nonparametric methods, calculation matrix tables can be established based on EEG characteristics and behavioral data (scale data), and the intervention effect value can be calculated based on these calculation matrix tables. The calculation matrix tables include a comparison matrix, a baseline matrix, and an intervention matrix. The comparison matrix is ​​used to characterize the changes in EEG signals and behavioral data before and after neurofeedback training; the baseline matrix is ​​used to characterize the EEG signals and behavioral data before neurofeedback training; and the intervention matrix is ​​used to characterize the EEG signals and behavioral data after neurofeedback training.

[0056] For example, based on an AB-type single-case experiment, the EEG data collected in the baseline phase (Phase A) were 2, 3, 5, 3, and the EEG data collected in the intervention phase (Phase B) were 4, 5, 5, 7, 6, as shown. Figure 5 As shown, a calculation matrix table is established, in which the data of each stage are arranged in a preset order to construct a comparison matrix, a baseline matrix, and an intervention matrix.

[0057] Comparing the values ​​in the rows and columns of the calculation matrix involves comparing each pair of data in the matrix and marking the comparison result at the intersection of each pair. For example, ... Figure 5 As shown, the symbol "+" indicates that the row value is less than the column value, representing improved data; the symbol "-" indicates that the row value is greater than the column value, representing worsened data; and the symbol "T" indicates that the row value is equal to the column value, representing stable data. The algorithm retrieves the number of positive and negative comparison logarithms. The positive comparison logarithm is calculated by counting the number of rows where the value is less than the column value in the matrix table, and the negative comparison logarithm is calculated by counting the number of rows where the value is greater than the column value in the matrix table.

[0058] The intervention effect value is calculated based on the positive and negative comparison logarithms, such as... Figure 5 As shown, the matrix in the upper left corner is the comparison matrix for comparing EEG data from phase A and phase B. The intervention effect value for the uncontrolled baseline trend in EEG data can be calculated based on this matrix. :

[0059]

[0060] in, This is the positive comparison logarithm within the matrix. This is the negative comparison logarithm within the matrix.

[0061]

[0062] in, To compare the difference between the positive and negative comparison logarithms in the comparison matrix, This represents the total number of comparison pairs in the comparison matrix.

[0063] The matrix in the upper right corner is the baseline matrix of the trend within stage A. The baseline trend value of the EEG data can be calculated based on this matrix. :

[0064]

[0065] in, This represents the difference between the positive and negative comparison logarithms in the baseline matrix. This represents the total number of comparison logarithms in the baseline matrix.

[0066] The matrix in the lower left corner is the intervention matrix for the trend within stage B. The intervention trend value of the EEG data can be calculated based on this matrix. :

[0067]

[0068] in, This represents the difference between the positive and negative logarithms of the comparisons in the intervention matrix. This represents the total number of comparison logs in the intervention matrix.

[0069] The intervention effect value of EEG data controlling the baseline trend :

[0070]

[0071] In this embodiment, the baseline trend can be controlled based on the Tau-U index calculation method. However, the absence of a baseline trend may lead to a deviation in the intervention effect value. Therefore, a baseline threshold can be set to control the baseline trend when it is greater than or equal to the threshold, i.e., controlling the trend when it appears and is relatively significant. In some embodiments, the baseline trend value can be calculated based on a baseline matrix, where the baseline trend value includes the baseline trend value of EEG signals and the baseline trend value of behavioral data. The baseline trend value is used to characterize the changing trend of EEG signals and behavioral data before neurofeedback training. By comparing the baseline trend value with the trend threshold, if the baseline trend value is greater than or equal to the trend threshold, it is determined that there is a significant baseline trend in the EEG data and behavioral data, which needs to be corrected. Then, the intervention effect value is calculated based on the comparison matrix. If the baseline trend value is less than the trend threshold, it is determined that there is no significant baseline trend in the EEG data and behavioral data, which does not need to be corrected. Then, the intervention effect value is calculated based on the comparison matrix and the baseline matrix.

[0072] For example, if the trend threshold is set to 0.33, then the effect value of the EEG intervention is:

[0073]

[0074] Similarly, for the effect value of behavioral intervention, the effect value of intervention without controlling for baseline trends in behavioral data is... The baseline trend value of the behavioral data is The intervention effect value of behavioral data controlling baseline trends is ', then the effect value of the behavioral intervention is:

[0075]

[0076] in, and These represent the percentage improvement in EEG and behavior compared to baseline after neurofeedback training. The intervention effect of neurofeedback training can be measured based on the intervention effect value, and the user's EEG characteristics can be classified accordingly.

[0077] In this embodiment, the intervention effect value is calculated by obtaining the changing trends of EEG data and behavioral data at the baseline and intervention stages. The EEG data is the feedback feature (neurofelive index) extracted from the EEG features. For example, for mindfulness neurofeedback training, the EEG signals of the user are collected, and the EEG signals are preprocessed and feature extracted to obtain the neurofeedback index (FAA). The EEG intervention effect value is calculated based on the changing trend of the neurofeedback index (FAA).

[0078] S500, which categorizes EEG features based on intervention effect values.

[0079] After obtaining the intervention effect value, EEG characteristics can be classified based on the intervention effect value to determine whether EEG and behavioral data have improved, thereby distinguishing between responders and non-responders. For example, as shown in the table below, EEG categories can be divided into four types:

[0080]

[0081] When classifying EEG features, an intervention effect threshold can be preset. The EEG intervention effect value and the intervention effect threshold are compared, as are the behavioral intervention effect value and the intervention effect threshold. If both the EEG intervention effect value and the behavioral intervention effect value are greater than the intervention effect threshold, the EEG feature is labeled as the first category. If either the EEG intervention effect value or the behavioral intervention effect value is less than or equal to the intervention effect threshold, the EEG feature is labeled as the second category.

[0082] For example, if the intervention effect threshold is 0, the EEG intervention effect value is... The effect value of behavioral intervention is .according to and Calculate the classification result EBEI:

[0083]

[0084] In this context, square brackets represent Iverson brackets. If the expression within the brackets is true, the EBEI is 1; otherwise, the EBEI is 0. If the EBEI is 1, the EEG feature is labeled as category 1, indicating that the user corresponding to that EEG feature is a responder. If the EBEI is 0, the EEG feature is labeled as category 2, indicating that the user corresponding to that EEG feature is a non-responder.

[0085] S600 constructs a neural network model and trains the neural network model based on EEG features labeled with EEG categories to obtain a neural feedback detection classification model.

[0086] After labeling EEG features with category tags, a neural network model can be trained based on the EEG features labeled with EEG category tags to obtain a neural feedback detection classification model.

[0087] In some embodiments, since the median data point can accurately reflect the state of the outcome variable before intervention, the median of each collected EEG feature can be selected as the EEG feature. For example, three EEG features are collected at the baseline stage, and the median of the three data points for each EEG feature is taken as the final EEG feature for training the neurofeedback detection classification model.

[0088] In some embodiments, when the number of EEG features is large, to improve the performance of the neural feedback detection and classification model, features with high detection and classification accuracy can be extracted from the EEG features and used as data for training the neural feedback detection and classification model. A predetermined number of sub-features can be selected from the EEG features based on a feature selection algorithm to construct multiple feature subsets. The accuracy of each feature subset is calculated based on a classification algorithm, and a target feature subset is selected based on the accuracy. The target feature subset can then be used as data for training the neural feedback detection and classification model.

[0089] For example, four feature selection algorithms are used: chi-square test-based feature selection algorithm, mutual information-based feature selection algorithm, recursive feature elimination algorithm, and L1 paradigm-based feature selection algorithm. For each algorithm, from the original feature matrix of EEG features, every possible feature subset is selected, starting with the fewest features and increasing in number. For each selected feature subset, a classification experiment is performed using a classification algorithm, and the accuracy is used as the evaluation criterion to assess the feature subset, thus selecting the target feature subset. The target feature subset can be a feature subset with an accuracy greater than or equal to a preset threshold, or it can be the feature subset with the highest accuracy among the selected feature subsets.

[0090] In some embodiments, since EEG features include multiple features and each EEG feature has different importance, in order to improve the performance and classification effect of the neural feedback detection classification model, a precision-based feature weighting strategy (PFWS) can be used. For each feature in the EEG features, an evaluation function is used to calculate the weight of each feature in the EEG features, and a weight is assigned to each feature in the EEG features, thereby generating weighted EEG features for training the neural feedback detection classification model. This allows the better performing features to play their own advantages and improve the classification effect.

[0091] The feature weighting strategy uses classification accuracy as the basis for weighting. By traversing the EEG features, it calculates the classification accuracy of the remaining EEG features after removing one EEG feature, and calculates the weight of each EEG feature based on the classification accuracy. The weight is then assigned to the EEG feature to generate a weighted EEG feature.

[0092] The weighted EEG characteristics are calculated using the following formula:

[0093]

[0094] Where A is the weighted EEG feature, w is the weight matrix, and a is the EEG feature matrix.

[0095] like Figure 6 As shown, the classification precision (P) of the EEG feature matrix excluding itself is calculated for each feature. A higher classification precision indicates a lower impact on classification precision after excluding the feature, and a lower weight for that feature in the EEG feature matrix. Therefore, (1-P) is used as the weight of that feature. Repeating the above process yields the weight matrix, and multiplying the weight matrix by the EEG feature matrix gives the weighted EEG features.

[0096] For example, the EEG feature matrix is =( traverse each feature To calculate the removal of features The set of classification accuracy of the post-EEG feature matrix According to the classification accuracy set Calculate the weight matrix The weighted EEG features can be obtained by multiplying the weight matrix by the EEG feature matrix:

[0097]

[0098] In this embodiment, responders are defined as positive class samples. Classification accuracy is the proportion of all samples predicted as positive and actually being positive out of all samples predicted as positive. Higher accuracy means fewer non-responders are identified as responders. Therefore, the feature weighting strategy based on classification accuracy uses classification accuracy as the evaluation function of feature weights. When applied to datasets where the number of positive class samples is greater than the number of negative class samples, it can better leverage the characteristics of each feature to achieve better recognition results.

[0099] After obtaining the weighted EEG features and their category labels, machine learning techniques can be used to build a classification model for training. For example, five classification algorithms can be used: K-Nearest Neighbor (KNN), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), and Support-Vector Machine (SVM). Accuracy and F1 score are selected as model evaluation metrics, and the best classification model is chosen as the detection classification model for neural feedback.

[0100] S700 inputs the EEG characteristics of the EEG signal to be tested into the neural feedback detection and classification model to obtain the classification result output by the neural feedback detection and classification model.

[0101] The classification results include responders and non-responders. EEG and behavioral data were obtained before and after neurofeedback training through single-case experiments. The intervention effect on the data was evaluated, and the data were labeled accordingly. A neurofeedback detection classification model was then trained based on this labeled data. Therefore, the predictive metric of the neurofeedback detection classification model is the EEG features extracted in the baseline phase, and the prediction target is the category of the EEG features, i.e., the responder and non-responder labels for the user.

[0102] By inputting the EEG characteristics of the EEG signal to be tested into a neurofeedback detection and classification model, the EEG category of the EEG data can be detected through the neurofeedback detection and classification model. For example, according to the above method steps, a neurofeedback detection and classification model for mindfulness neurofeedback training can be trained. By inputting the EEG characteristics of the user's baseline EEG signal into the neurofeedback detection and classification model, the classification results output by the neurofeedback detection and classification model can be obtained. Based on the classification results, responders and non-responders can be distinguished to determine whether mindfulness neurofeedback training is beneficial to the user.

[0103] Based on the above detection method, some embodiments of this application also provide a detection system for neurofeedback non-responders. The system includes an electroencephalogram (EEG) signal acquisition device and a controller connected to the EEG signal acquisition device. The EEG signal acquisition device is used to acquire EEG signals, and the controller is configured to execute the following program steps:

[0104] The system acquires a set of electroencephalogram (EEG) signals and behavioral data. The EEG signal set includes multiple EEG signals, which are a baseline EEG signal and an interventional EEG signal. The baseline EEG signal is the EEG signal before neurofeedback training, and the interventional EEG signal is the EEG signal after neurofeedback training. The behavioral data is the target behavioral data before and after neurofeedback training.

[0105] EEG features are extracted from EEG signals, and intervention effect values ​​are calculated based on EEG features and behavioral data. The intervention effect values ​​include EEG intervention effect values ​​and behavioral intervention effect values. The EEG intervention effect values ​​are used to characterize the intervention effect of EEG signals before and after neurofeedback training, and the behavioral intervention effect values ​​are used to characterize the intervention effect of target behavior before and after neurofeedback training.

[0106] EEG characteristics are categorized based on the intervention effect value.

[0107] A neural network model is constructed, and the neural network model is trained based on EEG features labeled with EEG categories to obtain a neural feedback detection classification model.

[0108] The EEG characteristics of the EEG signal to be tested are input into the neural feedback detection and classification model to obtain the classification results output by the neural feedback detection and classification model. The classification results include responders and non-responders.

[0109] As can be seen from the above technical solutions, this application provides a method and system for detecting non-responders in neurofeedback. The method, after acquiring EEG signals and behavioral data, performs feature extraction on the EEG signals, extracts EEG features, calculates the intervention effect value based on the EEG features and behavioral data, and evaluates the intervention results of neurofeedback training based on the intervention effect value, thereby labeling the EEG category of each EEG feature. Then, a neural network model is trained based on the EEG features labeled with EEG categories to obtain a neurofeedback detection classification model. The EEG signals include baseline EEG signals and intervention EEG signals, which are the EEG signals before and after neurofeedback training, respectively. The behavioral data are the target behavioral data before and after neurofeedback training, and the intervention effect value is used to characterize the intervention effect of neurofeedback training. By combining EEG data and behavioral data from different stages, this method more comprehensively and accurately measures the intervention effect of neurofeedback training, thereby improving the detection accuracy of neurofeedback results.

[0110] For ease of explanation, the above description has been provided in conjunction with specific embodiments. However, the exemplary discussion above is not intended to be exhaustive or to limit the embodiments to the specific forms disclosed above. Various modifications and variations can be obtained based on the above teachings. The selection and description of the above embodiments are for the purpose of better explaining the principles and practical applications, thereby enabling those skilled in the art to better utilize the embodiments and various different variations of embodiments suitable for specific application considerations.

Claims

1. A method for detecting non-responders in neural feedback, characterized in that, include: Acquire a set of electroencephalogram (EEG) signals and behavioral data. The set of EEG signals includes multiple EEG signals, which include a baseline EEG signal and an interventional EEG signal. The baseline EEG signal is the EEG signal before neurofeedback training, and the interventional EEG signal is the EEG signal after neurofeedback training. The behavioral data is the target behavioral data before and after neurofeedback training. EEG features are extracted from the EEG signals, and an intervention effect value is calculated based on the EEG features and the behavioral data. The intervention effect value includes an EEG intervention effect value and a behavioral intervention effect value. The EEG intervention effect value is used to characterize the intervention effect of the EEG signals before and after neurofeedback training, and the behavioral intervention effect value is used to characterize the intervention effect of the target behavior before and after neurofeedback training. The EEG category of the EEG features is labeled according to the intervention effect value; A neural network model is constructed, and the neural network model is trained based on EEG features labeled with EEG categories to obtain a neural feedback detection classification model; The EEG characteristics of the EEG signal to be tested are input into the neural feedback detection and classification model to obtain the classification result output by the neural feedback detection and classification model, which includes responders and non-responders.

2. The detection method according to claim 1, characterized in that, Prior to the step of extracting EEG features from the EEG signals, the method further includes: The EEG signals are preprocessed to filter out baseline drift, power line interference, electromyography noise, and electrooculography noise.

3. The detection method according to claim 1, characterized in that, The steps for calculating the intervention effect value based on the EEG characteristics and the behavioral data include: A calculation matrix table is established based on the EEG characteristics and the behavioral data. The calculation matrix table includes a comparison matrix, a baseline matrix, and an intervention matrix. The comparison matrix is ​​used to characterize the changes in EEG signals and behavioral data before and after neurofeedback training. The baseline matrix is ​​used to characterize the EEG signals and behavioral data before neurofeedback training. The intervention matrix is ​​used to characterize the EEG signals and behavioral data after neurofeedback training. The intervention effect value is calculated based on the calculation matrix table.

4. The detection method according to claim 3, characterized in that, The steps for calculating the intervention effect value based on the calculation matrix table include: Compare the values ​​of rows and columns in the computation matrix table to obtain positive and negative comparison logarithms. The positive comparison logarithm is the number of rows in the computation matrix table where the value is less than the column value, and the negative comparison logarithm is the number of rows in the computation matrix table where the value is greater than the column value. The intervention effect value is calculated based on the positive and negative comparison logarithms.

5. The detection method according to claim 3, characterized in that, The steps for calculating the intervention effect value based on the calculation matrix table include: Baseline trend values ​​are calculated based on the baseline matrix, and these baseline trend values ​​are used to characterize the changing trends of EEG signals and behavioral data before and after neurofeedback training. If the baseline trend value is greater than or equal to the trend threshold, the intervention effect value is calculated based on the comparison matrix; If the baseline trend value is less than the trend threshold, the intervention effect value is calculated based on the comparison matrix and the baseline matrix.

6. The detection method according to claim 1, characterized in that, The step of labeling the EEG features according to the intervention effect value includes: Obtain the threshold of intervention effect; Compare the EEG intervention effect value with the intervention effect threshold, and compare the behavioral intervention effect value with the intervention effect threshold; If both the EEG intervention effect value and the behavioral intervention effect value are greater than the intervention effect threshold, the EEG feature is labeled as the first category. If the EEG intervention effect value and the behavioral intervention effect value are both less than or equal to the intervention effect threshold, the EEG feature is labeled as a second category.

7. The detection method according to claim 1, characterized in that, Before the step of training the neural network model based on EEG features labeled with EEG categories, the method further includes: A predetermined number of sub-features are selected from the EEG features to construct multiple feature subsets; The accuracy of each feature subset is calculated based on the classification algorithm; A target feature subset is selected based on the accuracy rate, wherein the target feature subset is a feature subset with an accuracy rate greater than or equal to a preset threshold.

8. The detection method according to claim 1, characterized in that, Before the step of training the neural network model based on EEG features labeled with EEG categories, the method further includes: Obtain the weight of each feature in the EEG features; Each of the EEG features is assigned a weight to generate a weighted EEG feature.

9. The detection method according to claim 8, characterized in that, The step of obtaining the weight of each feature in the EEG features further includes: Traverse the described EEG features; After removing one EEG feature, the classification accuracy of the remaining EEG features is calculated sequentially, and the weight of each EEG feature is calculated based on the classification accuracy.

10. A detection system for non-responders in neurofeedback, characterized in that, The system includes an electroencephalogram (EEG) signal acquisition device and a controller connected to the EEG signal acquisition device. The EEG signal acquisition device is used to acquire EEG signals, and the controller is configured to execute the following program steps: Acquire a set of electroencephalogram (EEG) signals and behavioral data. The set of EEG signals includes multiple EEG signals, which include a baseline EEG signal and an interventional EEG signal. The baseline EEG signal is the EEG signal before neurofeedback training, and the interventional EEG signal is the EEG signal after neurofeedback training. The behavioral data is the target behavioral data before and after neurofeedback training. EEG features are extracted from the EEG signals, and intervention effect values ​​are calculated based on the EEG features and behavioral data. The intervention effect values ​​include EEG intervention effect values ​​and behavioral intervention effect values. The EEG intervention effect values ​​are used to characterize the intervention effect of EEG signals before and after neurofeedback training, and the behavioral intervention effect values ​​are used to characterize the intervention effect of target behavior before and after neurofeedback training. The EEG category of the EEG features is labeled according to the intervention effect value; A neural network model is constructed, and the neural network model is trained based on EEG features labeled with EEG categories to obtain a neural feedback detection classification model; The EEG characteristics of the EEG signal to be tested are input into the neural feedback detection and classification model to obtain the classification result output by the neural feedback detection and classification model, which includes responders and non-responders.

Images