Resting eeg mental stress prediction method fusing bidirectional decoupling and potential feature mining

Abstract

The application discloses a resting EEG psychological stress prediction method fusing bidirectional decoupling and potential feature mining, which is applied to a resting EEG psychological stress prediction system fusing bidirectional decoupling and potential feature mining. The prediction system is obtained by training according to historical electroencephalogram (EEG) data and corresponding psychological stress labels. The prediction method comprises the following steps: preprocessing collected original EEG data, and dividing the preprocessed continuous EEG data into a plurality of slices according to a sliding time window, wherein each slice comprises a task-state EEG signal and a resting-state EEG signal; adding random noise to the task-state EEG signal, so that the noisy task-state EEG signal is randomly distributed; introducing a guiding mechanism of the task-state electroencephalogram to the resting-state potential feature, so as to improve the feature separability in a weak feature scenario; meanwhile, effectively separating individual differences and depression-related features by using a bidirectional decoupling structure, extracting the time-space features of the EEG by using a wavelet-Riemann technology, and improving the psychological stress prediction efficiency and objectivity.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining. Background Technology

[0002] Electroencephalography (EEG) has become one of the most commonly used methods for measuring psychological stress due to its non-invasiveness, low cost, high mobility, and high temporal resolution. EEG can be obtained through resting-state or task-related experiments, namely resting-state EEG and task-based EEG. In comparison, resting-state EEG reveals more profound psychological activity and is easier to obtain; while task-based EEG, although more time-consuming, provides key task information related to psychological function.

[0003] However, the relevant technologies still have the following limitations in practical applications: First, the intrinsic mapping relationship between resting-state EEG signals and task-state EEG signals is unclear, which makes it very difficult to extract task-related information from resting-state EEG signals; Second, there is a lack of effective guidance for task-state feature extraction, which leads to unstable features and a lack of specificity, thereby limiting prediction performance; Third, multi-stage feature extraction and classification processes are usually adopted, resulting in complex model structures, low efficiency, and difficulty in achieving end-to-end automated detection. Summary of the Invention

[0004] This application provides a resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining.

[0005] In a first aspect, embodiments of this application provide a resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining, which is applied to a resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining. The resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining is trained based on historical EEG data and its corresponding psychological stress labels. The method includes: The raw EEG data collected is preprocessed, and the preprocessed continuous EEG data is divided into several slices according to a sliding time window. Each slice includes task-state EEG signals and resting-state EEG signals. Random noise is added to the task-state EEG signal to make the noisy task-state EEG signal randomly distributed; The resting-state EEG signal and the noisy task-state EEG signal are mapped to a shared latent space by a task-resting-state encoder, and the corresponding resting-state EEG features and task-state EEG features are extracted. By using a bidirectional decoupled attention decoder with task-state EEG features as query vectors and resting-state EEG features as keys and values, and calibrating task-state EEG features through a cross-attention mechanism, self-attention modeling of task-state EEG features in the time and channel dimensions is achieved, thereby obtaining the reconstructed task-state EEG signal. Extract the temporal and spatial features of the reconstructed task-state EEG signal; The temporal and spatial features are concatenated and input into a pre-built fully connected classifier to obtain psychological stress prediction results, which include psychological stress classification labels and their corresponding prediction probabilities.

[0006] Secondly, embodiments of this application provide a resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining, including: The preprocessing module is configured to preprocess the acquired raw EEG data and divide the preprocessed continuous EEG data into several slices according to a sliding time window, each slice including task-state EEG signal and resting-state EEG signal. The bidirectional decoupled conditional diffusion module is configured to add random noise to the task-state EEG signal, making the noisy task-state EEG signal randomly distributed. The task-resting state encoder maps the resting state EEG signal and the noisy task-state EEG signal to a shared latent space, extracting the corresponding resting state EEG features and task-state EEG features. The bidirectional decoupled attention decoder uses the task-state EEG features as the query vector and the resting state EEG features as the key and value, and calibrates the task-state EEG features through a cross-attention mechanism, realizing self-attention modeling of the task-state EEG features in the time and channel dimensions, thereby obtaining the reconstructed task-state EEG signal. The wavelet-Riemann feature extraction module is configured to extract the temporal and spatial features of the reconstructed task-state EEG signal; the temporal and spatial features are concatenated and input into a pre-built fully connected classifier to obtain the psychological stress prediction result, which includes a psychological stress classification label and its corresponding prediction probability.

[0007] Thirdly, embodiments of this application provide an electronic device comprising: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the steps of the resting EEG psychological stress prediction method integrating bidirectional decoupling and latent feature mining as described in any embodiment of the first aspect.

[0008] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps of the resting EEG psychological stress prediction method integrating bidirectional decoupling and latent feature mining as described in any embodiment of the first aspect.

[0009] Fifthly, embodiments of this application provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining as provided in the first aspect of embodiments of this application.

[0010] The resting EEG psychological stress prediction method of this application, which integrates bidirectional decoupling and latent feature mining, improves feature separability in weak feature scenarios by introducing a task-oriented EEG guidance mechanism for resting latent features; at the same time, it effectively separates individual differences and depression-related features using a bidirectional decoupling structure, extracts spatiotemporal features of EEG using wavelet-Riemann technology to improve the generalization performance of the model, and constructs a complete end-to-end detection framework to improve the efficiency and objectivity of psychological stress prediction. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining, as provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining, provided in an embodiment of this application. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0013] Figure label: Electronic device 300, processor 301, memory 302, communication interface 303, bus 310. Detailed Implementation

[0014] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0015] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0016] It should be noted that the acquisition, storage, use, and processing of data in this application embodiment all comply with the relevant provisions of national laws and regulations.

[0017] In recent years, depression has become a major concern in global public health. Subclinical depression (SD), as an early and reversible stage of depression, can effectively prevent the condition from worsening if detected early. Currently, SD is usually assessed using questionnaires such as the Hamilton Depression Rating Scale and the Perceived Stress Scale. However, these methods are susceptible to response bias and subjective results, and are time-consuming, leading to poor early detection and intervention outcomes. Although current real-time monitoring systems based on electrocardiogram (ECG) and pulse signals (such as wristbands and bracelets) can effectively track psychological stress, they are not suitable for detecting SD.

[0018] Electroencephalography (EEG) has become one of the most commonly used methods for measuring psychological stress due to its non-invasiveness, low cost, high mobility, and high temporal resolution. EEG can be obtained through resting-state or task-related experiments, namely resting-state EEG and task-based EEG. In comparison, resting-state EEG reveals more profound psychological activity and is easier to obtain; while task-based EEG, although more time-consuming, provides key task information related to psychological function.

[0019] In related technologies, the spectral and functional connectivity features of resting-state EEG signals have been used to improve the detection capability of mental stress. However, these technologies only use resting-state EEG signals for classification rather than directly predicting different levels of mental stress, and fail to extract sufficient stress-related features from resting-state EEG signals.

[0020] However, the relevant technologies still have the following limitations in practical applications: First, the intrinsic mapping relationship between resting-state EEG signals and task-state EEG signals is unclear, which makes it very difficult to extract task-related information from resting-state EEG signals; second, there is a lack of effective guidance for task-state feature extraction, which leads to unstable features and a lack of specificity, thus limiting prediction performance; third, most existing systems adopt multi-stage feature extraction and classification processes, with complex model structures and low efficiency, making it difficult to achieve end-to-end automated detection.

[0021] To address the problems in related technologies, this application provides a resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining.

[0022] The following description, in conjunction with the accompanying drawings, details the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining provided in this application, through specific embodiments and application scenarios.

[0023] It should be noted that the embodiments of this application are applied to a resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining. The resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining is trained based on historical EEG data and their corresponding psychological stress labels.

[0024] Figure 1 This diagram illustrates a flowchart of a resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining, according to an embodiment of this application. Figure 1 As shown, the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining can specifically include the following steps: S101. The raw EEG data collected is preprocessed, and the preprocessed continuous EEG data is divided into several slices according to a sliding time window. Each slice includes task-state EEG signal and resting-state EEG signal. S102. Add random noise to the task-state EEG signal so that the noisy task-state EEG signal is randomly distributed; S103. Map the resting-state EEG signal and the noisy task-state EEG signal to the shared latent space using the task-resting-state encoder, and extract the corresponding resting-state EEG features and task-state EEG features. S104. Using a bidirectional decoupled attention decoder, the task-state EEG features are used as the query vector, and the resting-state EEG features are used as the key and value. The task-state EEG features are calibrated through a cross-attention mechanism to achieve self-attention modeling of the task-state EEG features in the time and channel dimensions, thereby obtaining the reconstructed task-state EEG signal. S105. Extract the temporal and spatial features of the reconstructed task-state EEG signal; S106. The time features and spatial features are concatenated and input into a pre-built fully connected classifier to obtain the psychological stress prediction result, which includes the psychological stress classification label and its corresponding prediction probability.

[0025] Therefore, this study addresses the gap in the field of subthreshold depression by implementing automated EEG detection applicable to this area, filling the technical gap that existing severe depression identification schemes cannot be applied to individuals with mild mental disorders. A task-state-resting-state latent guidance mechanism is introduced, using emotion-regulation-related features of task-state EEG to guide the latent space learning of resting-state EEG, thus enhancing the previously weakly salient resting-state EEG features. A bidirectional decoupling technique is employed to separate individual differences from subthreshold depression features, addressing the problem of weak generalization ability caused by excessive individual differences in EEG data across subjects. Wavelet-Riemann techniques are used to extract spatiotemporal features from EEG, enhancing the comprehensiveness of feature extraction. A complete end-to-end psychological stress prediction system is achieved, with the entire process—feature extraction, decoupling, latent guidance, and classification prediction—automatically completed by a deep model, improving detection efficiency, reducing human intervention, and enhancing objectivity.

[0026] The specific implementation methods for each of the above steps are described below.

[0027] In some embodiments, in S101, the raw EEG data first undergoes a systematic preprocessing procedure, including: filtering and downsampling the acquired raw EEG data to make the raw EEG data have a preset dimension; and using independent component analysis to separate and remove noise components from the raw EEG data, wherein the noise components include at least electrooculography (EOG) and electromyography (EMG) artifact data.

[0028] In practice, to remove power frequency interference and high-frequency noise, the signal is initially processed using a 50 Hz low-pass filter; then the sampling rate is uniformly reduced to 250 Hz, thereby reducing the data dimensionality and improving computational efficiency while preserving the main neural rhythms; in order to further eliminate artifacts such as electrooculography and electromyography, independent component analysis is used in the preprocessing stage to separate and remove these noise components.

[0029] Furthermore, it should be understood that the preprocessed EEG data is a continuous EEG sequence, which can be used... Let U represent the number of subjects and L represent the signal length. Further, the preprocessed continuous EEG data is divided into several slices according to a sliding time window, and each of the resulting slices can be represented as... ,in, This indicates the task-state EEG signal. This represents the resting-state EEG signal, where d represents the number of EEG channels, and the length of each slice is 1 second.

[0030] In some embodiments, in S102, random noise is added to the task-state EEG signal according to the following formula (1): (1) in, This indicates a noisy task-mode EEG signal; , , Indicates the first i The retention coefficients of the signal at each time step. i Indicates the index of the time step. It is a series of scalars that gradually increase with time step t and satisfy the corresponding constraints; This indicates that Gaussian noise is added gradually; T represents the total number of time steps, which can be set empirically, for example, T=1000.

[0031] Furthermore, in some optional embodiments, a noisy task-state EEG signal is received via a task-resting encoder. The original resting-state EEG signal and diffusion time step encoding Therefore, time step t is converted into an eigenvector that can be used by the synthesis process according to the following formula (2): (2) Where D represents the embedding dimension, and i is the index of the embedding dimension.

[0032] Then, through adaptive normalization and global attention mechanisms, the resting-state EEG signal and the noisy task-state EEG signal are mapped to the shared latent space, and latent features with cross-subject consistency are extracted to obtain the corresponding resting-state EEG features and task-state EEG features, i.e., S103 is executed.

[0033] Next, in S104, the task-state EEG features are input into the bidirectional decoupled attention decoder to reconstruct the task-state EEG signal, including a task-state self-attention branch and a resting-state-guided cross-attention branch. The task-state self-attention branch models the task-state EEG features with self-attention in both the temporal and channel dimensions to capture the intrinsic relationships between different time slices and channels, thereby enhancing the expressive power of task-related features. The resting-state-guided cross-attention branch uses the task-state EEG features as the query vector and the resting-state EEG features as the key and value. It calibrates the task-state EEG features through a cross-attention mechanism, i.e., it performs a weighted summation of the features to model the inherent neural activity patterns of an individual in a resting state, separating out task-irrelevant individual noise. Finally, a clean task-state EEG slice is reconstructed through one-dimensional convolution, and the reconstructed task-state EEG signal can be defined as: .

[0034] In this way, task-oriented EEG is used to guide the mining of relevant latent features in resting-state EEG signals. To address the problem that task-related neural information is difficult to extract explicitly from resting-state EEG signals, task-oriented EEG signals are introduced as guiding information. Through a bidirectional decoupling modeling mechanism, latent feature representations that are highly correlated with task-oriented neural activity are mined from resting-state EEG signals.

[0035] Furthermore, in some embodiments, in S105, based on Gabor wave convolution, the temporal characteristics of the reconstructed task-state EEG signal in different frequency bands are adaptively obtained through a learnable center frequency and time width; the covariance matrix between the EEG channels corresponding to the reconstructed task-state EEG signal is calculated, and the covariance matrix is ​​used to characterize the amplitude variation relationship of the EEG signal of each channel in the time dimension; the covariance matrix is ​​processed by shrinkage regularization so that the covariance matrix is ​​located on a symmetric positive definite matrix manifold; the covariance matrix is ​​mapped to the tangent space to obtain the tangent space matrix; the elements of a preset region in the tangent space matrix are extracted, and the elements are vectorized to obtain the corresponding spatial features.

[0036] In other words, a dual-branch structure is adopted: the temporal feature branch is based on Gabor wave convolution, which adaptively captures the multi-band time-frequency structure of EEG through learnable center frequency and time width to obtain the temporal features of different frequency bands of EEG; the spatial feature branch is based on Riemannian manifold learning, which first calculates the covariance matrix between EEG channels, then uses shrinkage regularization to ensure that the matrix is ​​located on a symmetric positive definite matrix manifold, and maps it to the tangent space to maintain the topological structure between channels. Finally, the upper triangular part of the tangent space matrix is ​​extracted and vectorized to obtain the corresponding spatial feature representation.

[0037] In practice, temporal feature extraction is performed based on Gabor wavelet convolution. Unlike traditional convolution kernels with fixed shapes, Gabor wavelet convolution can adaptively adjust the filter shape in the time domain. This adaptability makes extracting multi-band features from the generated task-state EEG signals more effective. Gabor wavelets were chosen as the convolution kernel because of their recognized effectiveness in capturing multi-band EEG features. This convolution method is used to obtain the temporal features of the EEG signal. .

[0038] In practice, to preserve the inherent spatial topology of the EEG channel (which is often destroyed during Euclidean space processing), the reconstructed task-state EEG signal will be... Mapped to a Riemannian manifold. First, computed through a covariance pooling layer. The covariance matrix of each sample in the matrix, which is denoted as The linear correlation between EEG channels (i=1,…,M) was captured. Although the theoretical covariance matrix of EEG signals is symmetric positive definite (SPD), empirical estimates are usually only semi-positive definite due to noise. To ensure that this matrix lies on the SPD manifold, shrinkage regularization was used. Finally, the following was obtained: Spatial characteristics in Riemann .

[0039] In this way, to address the problem of complex time-frequency characteristics and strong channel spatial correlation of EEG signals, a learnable wavelet convolution and Riemannian geometric modeling method are combined to jointly model the time-frequency features and spatial covariance structure of EEG signals, thereby realizing the extraction of spatiotemporal features of EEG.

[0040] Furthermore, in some embodiments, in S106, the time feature is... Spatial features After concatenation, the input is fed into a fully connected classifier, and the psychological stress classification label P and its corresponding predicted probability are output through a Softmax layer, thus obtaining the final stress recognition result.

[0041] Therefore, the resting-state EEG psychological stress prediction method integrating bidirectional decoupling and latent feature mining in this application embodiment can effectively improve the discriminativeness of resting-state EEG features without increasing the cost of task-state data acquisition. This allows resting-state EEG signals to possess diagnostic information content similar to task-state EEG, thereby significantly improving the accuracy and robustness of subthreshold depression detection. It can simultaneously characterize the dynamic changes of EEG signals at different frequency bands and the spatial topological relationships between multiple channels, avoiding the destruction of the intrinsic geometric structure of EEG caused by traditional Euclidean space modeling. This effectively improves the stability and discriminative ability of feature expression, further enhancing the overall performance of the subthreshold depression detection system. Compared with existing detection methods that rely on task-induced experiments, this application does not require subjects to complete specific tasks, significantly reducing the difficulty of data acquisition and experimental burden, improving detection efficiency and clinical scalability, while enhancing stability and generalization ability across subject scenarios.

[0042] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0043] Based on the same technical concept, corresponding to any of the above embodiments, this application also provides a resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining.

[0044] like Figure 2 As shown, this resting EEG psychological stress prediction system, which integrates bidirectional decoupling and latent feature mining, may include: The preprocessing module is configured to preprocess the acquired raw EEG data and divide the preprocessed continuous EEG data into several slices according to a sliding time window, each slice including task-state EEG signal and resting-state EEG signal. The bidirectional decoupled conditional diffusion module is configured to add random noise to the task-state EEG signal, making the noisy task-state EEG signal randomly distributed. The task-resting state encoder maps the resting state EEG signal and the noisy task-state EEG signal to a shared latent space, extracting the corresponding resting state EEG features and task-state EEG features. The bidirectional decoupled attention decoder uses the task-state EEG features as the query vector and the resting state EEG features as the key and value, and calibrates the task-state EEG features through a cross-attention mechanism, realizing self-attention modeling of the task-state EEG features in the time and channel dimensions, thereby obtaining the reconstructed task-state EEG signal. The wavelet-Riemann feature extraction module is configured to extract the temporal and spatial features of the reconstructed task-state EEG signal; the temporal and spatial features are concatenated and input into a pre-built fully connected classifier to obtain the psychological stress prediction result, which includes a psychological stress classification label and its corresponding prediction probability.

[0045] As an optional embodiment, the resting EEG psychological stress prediction system that integrates bidirectional decoupling and latent feature mining also includes a training module, which is configured to collaboratively learn the bidirectional decoupling conditional diffusion module and the wavelet-Riemann feature extraction module under the condition of shared gradient, thereby achieving joint training of the loss function corresponding to the bidirectional decoupling conditional diffusion module and the loss function corresponding to the wavelet-Riemann feature extraction module. The total loss function is: ; (3) in, This represents the weighting parameter, used to balance the contributions of the BDCD and WRFE modules, both of which play equally important roles in the framework. 0.5 can be taken; This represents the loss function corresponding to the bidirectional decoupling conditional diffusion module. The loss function of the BDCD module consists of reconstruction loss and multidimensional covariance loss. This represents the loss function corresponding to the wavelet-Riemann feature extraction module. The WRFE module is optimized using the cross-entropy loss function during training.

[0046] Specifically, it includes: (4) (5) (6) (7) in, This represents the mean square error between the reconstructed task-state EEG signal and the task-state EEG signal. This indicates that Gaussian noise is added gradually. This indicates noise reduction processing, where t represents the time step. This indicates the task-state EEG signal. This indicates the resting-state EEG signal; Let M represent the multidimensional covariance loss, M represent the number of samples in the reconstructed task-state EEG signal, and d represent the number of EEG channels. m The index represents the number of samples in the reconstructed task-state EEG signal, and j represents the index of the number of channels; This represents the covariance matrix of the task-state EEG signal. This represents the covariance matrix of the reconstructed task-state EEG signal. Denotes the Frobenius norm. Indicates the first m Labels of a resting-state EEG sample The label represents the m-th reconstructed task-state EEG signal.

[0047] In other words, the resting EEG psychological stress prediction system integrating bidirectional decoupling and latent feature mining in this application embodiment comprises four sequential stages: signal preprocessing, a bidirectional decoupled conditional diffusion (BDCD) module, a wavelet-Riemannian feature extraction (WRFE) module, and an end-to-end training stage. Specifically, the signal preprocessing stage provides clean and reliable EEG signals, and then the BDCD module extracts latent features from these signals to generate high-fidelity, task-state-related EEG signals; subsequently, the WRFE module extracts discriminative temporal and spatial features from the generated signals, ultimately achieving stress level classification; to enable system optimization, the framework employs joint training, allowing the BDCD and WRFE modules to mutually optimize each other.

[0048] Optionally, the BDCD module aims to generate more discriminative task-state EEG based on resting-state EEG using a diffusion model structure, explicitly decoupling subthreshold depressive features from individual variability factors in the process. This module consists of three parts: a forward diffusion process, a task-resting-state encoder, and a bidirectional decoupled attention decoder. The forward diffusion process continuously adds random noise to the task-state slices, gradually making the signal randomly distributed, thus establishing a conditional modeling foundation for subsequent reverse generation. Given the weak abnormal features and large individual variability of resting-state EEG signals in subthreshold depressive states, directly identifying based on resting-state EEG often leads to feature instability and insufficient cross-subject generalization ability. By introducing task-state EEG as potential guiding information, an intrinsic mapping relationship between resting-state and task-state EEG is established without increasing the task burden in the actual detection stage. By employing a bidirectional decoupling conditional diffusion mechanism, individual differences and depression-related neural features are explicitly separated during the generation process. This enhances depression-related information that is difficult to detect in resting-state EEG, thereby effectively improving feature separability and cross-subject consistency in weak feature scenarios and solving the problem that existing methods cannot stably identify subthreshold depression.

[0049] Optionally, the WRFE module focuses on extracting spatiotemporal features from the reconstructed task-state EEG signals for subthreshold depression classification. Based on the aforementioned potential guidance and decoupled modeling, a learnable wavelet convolution and Riemannian geometric modeling method are combined to perform spatiotemporal joint feature extraction on the generated EEG signals. This effectively characterizes the spatial covariance structure between channels while preserving multi-band dynamic features, avoiding the disruption of the intrinsic geometric relationships of EEG by traditional Euclidean space modeling. By constructing an end-to-end framework for joint optimization of generation and discrimination, feature extraction, decoupled modeling, and classification prediction learn collaboratively in a unified network. This not only simplifies the traditional multi-stage processing flow but also significantly improves the accuracy, stability, and application efficiency of subthreshold depression detection, enhancing the system's scalability and practical value in real-world scenarios.

[0050] Optionally, to achieve an overall improvement in the generation of task-oriented EEGs and the ability to identify subthreshold depression, an end-to-end joint optimization framework is constructed on top of the BDCD and WRFE modules. This framework integrates generative decoupling and discriminative recognition tasks into a unified network that can be updated simultaneously, enabling the two modules to learn collaboratively under shared gradient conditions, jointly improving the generalization ability and classification performance of cross-subject EEGs. Specifically, the end-to-end framework is driven by two complementary loss terms: one is the reconstruction loss and covariance consistency loss of the BDCD module, used to improve the structural integrity and cross-subject consistency of the generated signal; the other is the classification loss of the WRFE module, used to promote the discriminative expression of temporal and spatial features. During joint optimization, these two types of losses are weighted and summed according to a set weight, so that the network simultaneously focuses on generation quality and classification performance, achieving integrated generation-recognition learning. That is, this model will... Loss and Joint training was conducted to mitigate the losses.

[0051] Therefore, this resting EEG psychological stress prediction system, which integrates bidirectional decoupling and latent feature mining, fills the gap in the end-to-end detection model of subthreshold depression. By jointly modeling the generation and discrimination process of EEG signals, it realizes the automated analysis from the original resting EEG signal to the output of the subthreshold depression level, thereby significantly improving the accuracy, stability and application efficiency of subthreshold depression identification.

[0052] It should be noted that, for ease of description, the above system is described by dividing it into various modules based on their functions. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware components.

[0053] The system described in the above embodiments is used to implement the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0054] Based on the same technical concept, corresponding to any of the above embodiments, this application also provides an electronic device.

[0055] Figure 3 A schematic diagram of a more specific electronic device hardware structure provided in this embodiment is shown.

[0056] The electronic device 300 may include a processor 301 and a memory 302 storing computer program instructions.

[0057] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0058] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory.

[0059] In certain embodiments, the memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this application.

[0060] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the resting EEG psychological stress prediction methods that integrate bidirectional decoupling and latent feature mining in the above embodiments.

[0061] In some examples, the electronic device 300 may also include a communication interface 303 and a bus 310. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 310 and complete communication with each other.

[0062] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0063] Bus 310 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not as a limitation, bus 310 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 310 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0064] For example, the electronic device 300 can be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc.

[0065] Based on the same technical concept, corresponding to any of the methods in the above embodiments, this application also provides a non-transitory computer-readable storage medium. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the resting EEG psychological stress prediction methods in the above embodiments that integrate bidirectional decoupling and latent feature mining. Examples of computer-readable storage media include non-transitory computer-readable storage media such as portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, etc.

[0066] Based on the same technical concept, corresponding to any of the above embodiments, this application also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer to cause the computer and / or the processor to execute the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining. Corresponding to the execution entity for each step in each embodiment of the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining, the processor executing the corresponding step can belong to the corresponding execution entity.

[0067] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0068] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0069] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0070] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0071] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for predicting resting EEG-induced psychological stress that integrates bidirectional decoupling and latent feature mining, characterized in that, A resting EEG psychological stress prediction system integrating bidirectional decoupling and latent feature mining is applied. The resting EEG psychological stress prediction system integrating bidirectional decoupling and latent feature mining is trained based on historical EEG data and their corresponding psychological stress labels. The method includes: The raw EEG data collected is preprocessed, and the preprocessed continuous EEG data is divided into several slices according to a sliding time window. Each slice includes task-state EEG signals and resting-state EEG signals. Random noise is added to the task-state EEG signal to make the noisy task-state EEG signal randomly distributed; The resting-state EEG signal and the noisy task-state EEG signal are mapped to a shared latent space by a task-resting-state encoder, and the corresponding resting-state EEG features and task-state EEG features are extracted. By using a bidirectional decoupled attention decoder with task-state EEG features as query vectors and resting-state EEG features as keys and values, and calibrating task-state EEG features through a cross-attention mechanism, self-attention modeling of task-state EEG features in the time and channel dimensions is achieved, thereby obtaining the reconstructed task-state EEG signal. Extract the temporal and spatial features of the reconstructed task-state EEG signal; The temporal and spatial features are concatenated and input into a pre-built fully connected classifier to obtain psychological stress prediction results, which include psychological stress classification labels and their corresponding prediction probabilities.

2. The method according to claim 1, characterized in that, The extraction of the temporal and spatial features of the reconstructed task-state EEG signal includes: Based on Gabor wave convolution, the temporal characteristics of the reconstructed task-state EEG signal in different frequency bands are adaptively obtained through a learnable center frequency and time width.

3. The method according to claim 1, characterized in that, The extraction of the temporal and spatial features of the reconstructed task-state EEG signal includes: Calculate the covariance matrix between the EEG channels corresponding to the reconstructed task-state EEG signal. The covariance matrix is ​​used to characterize the amplitude variation relationship of the EEG signal of each channel in the time dimension. The covariance matrix is ​​processed using shrinkage regularization so that it lies on a symmetric positive definite matrix manifold; The covariance matrix is ​​mapped to the tangent space to obtain the tangent space matrix; Extract the elements of the preset region in the tangent space matrix and vectorize the elements to obtain the corresponding spatial features.

4. The method according to claim 1, characterized in that, The slices obtained by dividing the preprocessed continuous EEG data according to a sliding time window are represented as follows: ,in, This indicates the task-state EEG signal. This represents the resting-state EEG signal, where d represents the number of EEG channels, and the length of each slice is 1 second. The step of adding random noise to the task-state EEG signal to make the noisy task-state EEG signal randomly distributed includes: Random noise is added to the task-state EEG signal according to the following formula: in, This indicates a noisy task-mode EEG signal; , , Indicates the first i The retention coefficients of the signal at each time step. i Indicates the index of the time step. It is a series of scalars that gradually increase with time step t; This represents the Gaussian noise added gradually; T represents the total number of time steps.

5. The method according to claim 1, characterized in that, The preprocessing of the collected raw EEG data includes: The collected raw EEG data was filtered and downsampled to ensure it conformed to a preset dimension; and, Independent component analysis was used to separate and remove noise components from the raw EEG data, and the noise components included at least electrooculography and electromyography artifact data.

6. A resting EEG psychological stress prediction system integrating bidirectional decoupling and latent feature mining, characterized in that, include: The preprocessing module is configured to preprocess the acquired raw EEG data and divide the preprocessed continuous EEG data into several slices according to a sliding time window, each slice including task-state EEG signal and resting-state EEG signal. The bidirectional decoupling conditional diffusion module is configured to add random noise to the task-state EEG signal, so that the noisy task-state EEG signal is randomly distributed; the resting-state EEG signal and the noisy task-state EEG signal are mapped to the shared latent space by the task-resting-state encoder, and the corresponding resting-state EEG features and task-state EEG features are extracted. By using a bidirectional decoupled attention decoder with task-state EEG features as query vectors and resting-state EEG features as keys and values, and calibrating task-state EEG features through a cross-attention mechanism, self-attention modeling of task-state EEG features in the time and channel dimensions is achieved, thereby obtaining the reconstructed task-state EEG signal. The wavelet-Riemann feature extraction module is configured to extract the temporal and spatial features of the reconstructed task-state EEG signal; The temporal and spatial features are concatenated and input into a pre-built fully connected classifier to obtain psychological stress prediction results, which include psychological stress classification labels and their corresponding prediction probabilities.

7. The system according to claim 6, characterized in that, The system also includes a training module configured to collaboratively learn the bidirectional decoupled conditional diffusion module and the wavelet-Riemann feature extraction module under the condition of shared gradient, thereby achieving joint training of the loss function corresponding to the bidirectional decoupled conditional diffusion module and the loss function corresponding to the wavelet-Riemann feature extraction module. The total loss function is: ； in, Represents the weight parameters. This represents the loss function corresponding to the bidirectional decoupling conditional diffusion module. This represents the loss function corresponding to the wavelet-Riemann feature extraction module; in, This represents the mean square error between the reconstructed task-state EEG signal and the task-state EEG signal. This indicates that Gaussian noise is added gradually. This indicates noise reduction processing, where t represents the time step. This indicates the task-state EEG signal. This indicates the resting-state EEG signal; Let M represent the multidimensional covariance loss, M represent the number of samples in the reconstructed task-state EEG signal, and d represent the number of EEG channels. m The index represents the number of samples in the reconstructed task-state EEG signal, and j represents the index of the number of channels; This represents the covariance matrix of the task-state EEG signal. , represents the covariance matrix of the reconstructed task-state EEG signal. Denotes the Frobenius norm. Indicates the first m Labels of a resting-state EEG sample The label represents the m-th reconstructed task-state EEG signal.

8. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; when the processor invokes the computer program instructions, it implements the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining as described in any one of claims 1-5.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when invoked by a processor, implement the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining as described in any one of claims 1-5.

10. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the resting EEG psychological stress prediction method that integrates bidirectional decoupling and latent feature mining as described in any one of claims 1-5.

Images