Electroencephalogram signal monitoring device and method for depression and anxiety

Abstract

The application provides an electroencephalogram monitoring device and method for depression and anxiety, and relates to the technical field of artificial intelligence, and comprises the following steps: collecting electroencephalogram signals of a subject under resting state and task state conditions to obtain original electroencephalogram time series data; performing band-pass filtering on the original electroencephalogram time series data to obtain filtered electrical signals; removing electrooculogram and electromyogram artifacts from the filtered electrical signals to obtain preprocessed electroencephalogram signals; extracting electroencephalogram characteristic frequency band parameters and brain network connection topological features from the preprocessed electroencephalogram signals; constructing a feature parameter set according to the electroencephalogram characteristic frequency band parameters and the brain network connection topological features; and positioning three recording electrodes of a frontal lobe, a temporal lobe and a central region as feature sensing nodes. The application improves the accuracy and stability of electroencephalogram monitoring of depression and anxiety.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a device and method for monitoring electroencephalogram (EEG) signals for depression and anxiety. Background Technology

[0002] In practical applications such as rapid mood screening in psychiatric outpatient clinics and follow-up visits to community residents' mental health, EEG signal monitoring technology for depression and anxiety is gradually being introduced into the auxiliary assessment process. However, existing conventional technical solutions have technical shortcomings in actual implementation.

[0003] These technologies typically rely on standard multi-channel EEG acquisition equipment to simultaneously acquire whole-brain EEG signals of subjects under resting eye-open, resting eye-closed, and emotional cognitive task states. After preprocessing such as basic bandpass filtering and artifact removal, feature parameters of the whole brain dimension and the overall brain network connectivity topology features are directly extracted. The feature set is then input into a classification model to complete the discrimination of depressive and anxious emotional states.

[0004] The technical defects of the existing technology in actual implementation are as follows: it does not establish a dedicated feature sensing node localization and feature focusing extraction mechanism for the three core brain regions directly related to emotion regulation: the prefrontal cortex, temporal lobe and central region. Instead, it uses the average features of the whole brain channels as the basis for analysis, which makes it difficult to accurately capture the unique EEG signal distribution features of key brain regions. This results in the extraction of emotion-related features with missing information from the core brain regions. Summary of the Invention

[0005] This invention provides an EEG signal monitoring device and method for depression and anxiety, which improves the accuracy and stability of EEG monitoring for depression and anxiety.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for monitoring electroencephalogram (EEG) signals targeting depressive and anxious moods, the method comprising: The acquisition module is used to acquire EEG signals of subjects under resting and task conditions to obtain raw EEG time-series data; bandpass filtering is performed on the raw EEG time-series data to obtain filtered electrical signals; and ocular and electromyographic artifacts are removed from the filtered electrical signals to obtain preprocessed EEG signals. The extraction module is used to extract EEG characteristic frequency band parameters and brain network connection topology features from the preprocessed EEG signals; construct a feature parameter set based on the EEG characteristic frequency band parameters and brain network connection topology features; and locate three recording electrodes in the prefrontal lobe, temporal lobe and central region as feature sensing nodes. The module is used to construct a virtual irregular spatial feature surface based on the EEG amplitude sequence and phase synchronization measurement collected by three feature sensing nodes; to perform grid-based discretization and regional curvature integral operation on the virtual irregular spatial feature surface to obtain the initial spatial distribution compensation field; to perform minimum coverage ellipse fitting on the projection point set of the three feature sensing nodes in the virtual space to obtain geometric feature parameters; and to perform weighted adjustment of the initial spatial distribution compensation field through the eccentricity and major axis direction vector in the geometric feature parameters to obtain the final spatial distribution compensation field. The correction module is used to perform weight compensation correction on the brain network connection topology features in the feature parameter set through the final spatial distribution compensation field quantity, so as to obtain the corrected feature parameter set. The recognition module is used to input the corrected feature parameter set into the pre-trained EEG feature classification model, perform pattern recognition on EEG signal features, and obtain analysis result data to complete EEG signal feature monitoring.

[0007] Secondly, methods for monitoring EEG signals related to depression and anxiety include: EEG signals were collected from subjects under resting and task conditions to obtain raw EEG time-series data; bandpass filtering was performed on the raw EEG time-series data to obtain filtered electrical signals; ocular and electromyographic artifacts were removed from the filtered electrical signals to obtain preprocessed EEG signals; EEG characteristic frequency band parameters and brain network connection topology features are extracted from the preprocessed EEG signals; a set of characteristic parameters is constructed based on the EEG characteristic frequency band parameters and brain network connection topology features; three recording electrodes in the prefrontal lobe, temporal lobe and central region are located as characteristic sensing nodes. Based on the EEG amplitude sequence and phase synchronization measurement collected by three feature sensing nodes, a virtual irregular spatial feature surface is constructed. The virtual irregular spatial feature surface is then subjected to grid-based discretization and regional curvature integral calculation to obtain the initial spatial distribution compensation field. The projection point set of the three feature sensing nodes in the virtual space is fitted with a minimum coverage ellipse to obtain geometric feature parameters. The initial spatial distribution compensation field is then weighted and adjusted using the eccentricity and major axis direction vector in the geometric feature parameters to obtain the final spatial distribution compensation field. The brain network connectivity topology features in the feature parameter set are corrected by weight compensation correction of the final spatial distribution compensation field quantity to obtain the corrected feature parameter set. The corrected feature parameter set is input into the pre-trained EEG feature classification model to perform pattern recognition on the EEG signal features and obtain the analysis results data to complete the monitoring of EEG signal features.

[0008] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0009] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0010] The above-described solution of the present invention has at least the following beneficial effects: By employing three key brain regions—the prefrontal cortex, temporal lobe, and central region—as feature sensing nodes, a virtual irregular spatial feature surface is constructed. This surface is then combined with gridded subdivision, curvature integral, and minimum coverage ellipse fitting to generate a spatial distribution compensation field. This process corrects the weighted topological features of the brain network connectivity, and the corrected features are input into a pre-trained model to complete pattern recognition. This technique overcomes the technical problems of existing EEG monitoring technologies, such as the inability to focus on the core brain regions for emotion regulation, insufficient utilization of EEG spatial distribution features, unreasonable weighting of brain network features leading to low accuracy in emotion recognition, and poor robustness of monitoring results. As a result, it achieves the accurate capture of core EEG features related to depression and anxiety. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the electroencephalogram (EEG) signal monitoring method for depression and anxiety provided in an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of an electroencephalogram (EEG) signal monitoring device for depression and anxiety provided in an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram comparing the time domain of EEG signals before and after preprocessing.

[0014] Figure 4 This is a schematic diagram of the simulation of virtual irregular spatial feature surfaces.

[0015] Figure 5 This is a schematic diagram of the minimum coverage ellipse fitting of feature nodes.

[0016] Figure 6 This is a schematic diagram comparing the brain network topological features before and after correction.

[0017] Figure 7 This is a diagram illustrating the trend of emotion recognition accuracy and loss.

[0018] Figure 8 This is a schematic diagram of the thermal distribution of the spatially distributed compensation field. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0020] like Figure 2 As shown, embodiments of the present invention propose an electroencephalogram (EEG) signal monitoring device for depression and anxiety, comprising: The acquisition module is used to acquire EEG signals of subjects under resting and task conditions to obtain raw EEG time-series data; bandpass filtering is performed on the raw EEG time-series data to obtain filtered electrical signals; and ocular and electromyographic artifacts are removed from the filtered electrical signals to obtain preprocessed EEG signals. The extraction module is used to extract EEG characteristic frequency band parameters and brain network connection topology features from the preprocessed EEG signals; construct a feature parameter set based on the EEG characteristic frequency band parameters and brain network connection topology features; and locate three recording electrodes in the prefrontal lobe, temporal lobe and central region as feature sensing nodes. The module is used to construct a virtual irregular spatial feature surface based on the EEG amplitude sequence and phase synchronization measurement collected by three feature sensing nodes; to perform grid-based discretization and regional curvature integral operation on the virtual irregular spatial feature surface to obtain the initial spatial distribution compensation field; to perform minimum coverage ellipse fitting on the projection point set of the three feature sensing nodes in the virtual space to obtain geometric feature parameters; and to perform weighted adjustment of the initial spatial distribution compensation field through the eccentricity and major axis direction vector in the geometric feature parameters to obtain the final spatial distribution compensation field. The correction module is used to perform weight compensation correction on the brain network connection topology features in the feature parameter set through the final spatial distribution compensation field quantity, so as to obtain the corrected feature parameter set. The recognition module is used to input the corrected feature parameter set into the pre-trained EEG feature classification model, perform pattern recognition on EEG signal features, and obtain analysis result data to complete EEG signal feature monitoring.

[0021] like Figure 1 As shown, embodiments of the present invention also provide a method for monitoring electroencephalogram (EEG) signals for depressive and anxious moods, including: Step 1: Collect EEG signals from subjects under resting and task conditions to obtain raw EEG time-series data; perform bandpass filtering on the raw EEG time-series data to obtain filtered electrical signals; remove ocular and electromyographic artifacts from the filtered electrical signals to obtain preprocessed EEG signals. Step 2: Extract EEG characteristic frequency band parameters and brain network connection topology features from the preprocessed EEG signals; construct a feature parameter set based on the EEG characteristic frequency band parameters and brain network connection topology features; locate three recording electrodes in the prefrontal lobe, temporal lobe, and central region as feature sensing nodes. Step 3: Based on the EEG amplitude sequence and phase synchronization measurement collected by the three feature sensor nodes, a virtual irregular spatial feature surface is constructed; the virtual irregular spatial feature surface is subjected to grid-based discretization and regional curvature integral operation to obtain the initial spatial distribution compensation field; the projection point set of the three feature sensor nodes in the virtual space is fitted with a minimum coverage ellipse to obtain geometric feature parameters; the initial spatial distribution compensation field is weighted and adjusted by the eccentricity and major axis direction vector in the geometric feature parameters to obtain the final spatial distribution compensation field. Step 4: The brain network connectivity topology features in the feature parameter set are weighted and corrected by the final spatial distribution compensation field quantity to obtain the corrected feature parameter set. Step 5: Input the corrected feature parameter set into the pre-trained EEG feature classification model, perform pattern recognition on the EEG signal features, and obtain the analysis result data to complete the EEG signal feature monitoring.

[0022] In this embodiment of the invention, by employing techniques such as acquiring resting-state and task-state EEG signals and performing filtering and artifact removal preprocessing, extracting EEG feature frequency band parameters and brain network connection topology features, and locating electrodes in the prefrontal, temporal, and central regions as feature sensing nodes, constructing virtual irregular spatial feature surfaces based on these sensing nodes, obtaining spatial distribution compensation field quantities through gridded subdivision, curvature integral, and minimum coverage ellipse fitting, and performing weight correction on the brain network topology features, and inputting the corrected feature parameter set into a pre-trained EEG feature classification model to complete pattern recognition, this invention overcomes the technical problems of existing EEG monitoring methods, such as the inability to focus on core brain regions for emotion regulation, insufficient utilization of EEG spatial distribution features, and unreasonable brain network feature weight allocation leading to low accuracy in identifying depression and anxiety, and poor stability of monitoring results. This achieves accurate extraction of core EEG features related to depression and anxiety, optimization of feature distribution weights, and improved emotional feature discrimination and monitoring robustness.

[0023] In a preferred embodiment of the present invention, step 1 above may include: Step 1.1: Simultaneously acquire multi-channel raw electrical signals of the subject under resting and task-oriented conditions using an EEG acquisition device, and integrate and splice them into raw EEG time-series data. Specifically, this includes: Before signal acquisition, completing the electrode layout of the EEG acquisition device according to internationally accepted EEG electrode distribution standards, firmly attaching the reference electrode, ground electrode, and acquisition electrodes of each channel to the corresponding points on the subject's head, ensuring that the contact impedance between the electrodes and the scalp meets the acquisition requirements. During the acquisition phase, the subject is controlled to sequentially enter a resting state and an emotional task state. The resting state includes two sub-stages: resting with eyes open and resting with eyes closed. The emotional task state is the stage in which the subject completes a standardized emotional cognitive task. The EEG acquisition device continuously acquires brain electrical signals in the above different states at a constant sampling frequency. Each acquisition channel independently records the value of the potential change of the corresponding brain region over time to obtain discrete time-series electrical signals of multiple channels under different behavioral states.

[0024] After data acquisition, using the uniformly generated sampling timestamp as a benchmark, point-by-point timing matching is performed on the timing signals of all channels to ensure that the signal sampling points of all channels correspond at the same sampling time and that there is no timing misalignment. According to the preset electrode channel numbering and sorting rules, the timing signals of each channel are arranged sequentially by channel number. The timing signals of each single channel are aligned, superimposed, and integrated in the time dimension, so that the multiple dispersed single-channel signals are integrated into a set of continuous timing data containing information from all channels, ultimately obtaining the raw EEG timing data. Let the total number of acquisition channels be... N , No. The potential signal of each acquisition channel at sampling time t is The original EEG time-series data is , The discrete sampling time, representing the time point when the EEG acquisition device records signals at a constant sampling frequency, is represented by the formula for calculating the integration and splicing of multiple signals: .

[0025] Step 1.2: Apply a preset passband filter to the raw EEG time-series data to remove low-frequency baseline drift and high-frequency environmental noise, obtaining the filtered electrical signal. Specifically, this includes applying a preset passband filter to the raw EEG time-series data. The calculation formula for the time-domain bandpass filter is: The original EEG time-series data is ( The filter kernel coefficient sequence is h(k), and the sliding window length for the convolution operation is... The baseline correction value is , It is the position index variable within the sliding window. This represents the original signal sample value corresponding to the k-th position within the sliding window, and the filtered electrical signal is... The process involves filtering out low-frequency baseline drift and high-frequency environmental noise to obtain a filtered electrical signal. Specifically, this includes integrating the original EEG time-series data. ( The input filtering module is pre-defined with a passband frequency range adapted to the EEG signal. This range retains valid EEG signal components while blocking low-frequency baseline drift below the lower passband limit and high-frequency environmental noise above the upper passband limit. The filtering process is implemented using temporal convolution operations. During the operation, the filtering kernel operates at a fixed step size... ( Slide the slider point by point, and at each calculation time t, extract the data. The endpoint is and the length is A continuous segment of the original signal, which contains from arrive of One original signal sample value .

[0026] Perform the calculation according to the weighted product rules in the formula, and then calculate the value of the first product in the window. each position Coefficients with the same position as the filter kernel Multiply each element one by one to obtain X(t-k+1). Calculation of the weighted product of h(k); filter kernel coefficients To pre-determine the weighting, higher weights are assigned to effective EEG signals, while lower weights are assigned to noise and baseline drift. Through the complete filtering operation of the above-mentioned formula, low-frequency baseline drift caused by human physiological fluctuations in the original EEG time series data, as well as high-frequency noise introduced by environmental electromagnetic fields and equipment, can be eliminated, while retaining the effective signal components that reflect the real neural electrical activity of the brain.

[0027] Step 1.3 involves waveform component identification and interference separation processing of the filtered electrical signal to remove artifacts related to electrooculography (EOG) and electromyography (EMG), retaining the valid EEG waveform and realigning the time series to obtain the preprocessed EEG signal. Specifically, this includes: identifying waveform components of the filtered electrical signal point by point. The identification process is achieved through preset amplitude and waveform slope thresholds: signal components whose amplitude fluctuations exceed the normal EEG range and whose waveform rise and fall slopes are significantly different from the EEG waveform are identified as EOG artifact components and EMG artifact components, respectively. Signal components whose amplitude and slope are both within the normal range are identified as valid EEG waveforms. After artifact component identification, interference separation is performed on the filtered electrical signal to remove EOG and EMG artifact components by subtracting them point by point, retaining the pure valid EEG waveform.

[0028] Because artifact separation operations can cause timing shifts in local signals, after artifact removal, the effective EEG waveform is re-aligned point-by-point using the original sampling time axis as a reference. This corrects the timing misalignment caused by the separation process, ensuring that the timing of each channel signal is consistent with the original acquisition timing. The result is a pre-processed EEG signal with regular timing and free from noise and artifact interference. Let the filtered electrical signal be... The component of electrooculography artifacts is The electromyographic artifact component is The intermediate signal after artifact removal is The time-aligned preprocessed EEG signals are The formula for artifact separation and removal is: Timing alignment is achieved by... Matching the original sampling time point by point The process is completed, and finally standardized preprocessed EEG signals are obtained.

[0029] In this embodiment of the invention, because the technical means of synchronously acquiring multi-channel EEG signals in resting and task states and integrating and splicing them according to the time axis to generate original time-series data, filtering out low-frequency baseline drift and high-frequency environmental noise in the original signal through time-domain convolution filtering, separating artifacts through waveform recognition and removing EEG and EMG interference by subtraction operation, and then re-aligning the time sequence, the technical problems of inconsistent multi-channel time sequence of the original EEG signal, high and low frequency noise interference, and distortion of effective signal caused by the mixing of EEG and EMG artifacts are overcome, thereby achieving the acquisition of pure pre-processed EEG signals with regular time sequence and thorough removal of noise and artifacts.

[0030] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1 involves performing frequency domain analysis and inter-channel correlation calculations on the preprocessed EEG signal to extract the energy distribution attributes within the target frequency band and the connection strength attributes between nodes in each brain region, thereby obtaining the EEG characteristic frequency band parameters and brain network connection topology features. Specifically, this includes selecting continuous valid data segments from the preprocessed EEG signal, performing discrete Fourier transform operations to convert the time-fluctuating potential signal in the time domain into energy spectrum data distributed with frequency in the frequency domain. The specific calculation process is as follows: for a single channel at the sampling time... signal sequence Perform the transformation, where Discrete sampling time represents each time point in the EEG acquisition device recording the signal; For a single channel at the sampling time The preprocessed EEG potential signal is a pure EEG value after noise reduction and artifact removal. The total number of sampling points participating in the transformation refers to the total number of signal sampling points contained in the selected continuous valid data segment. It is a frequency variable in the frequency domain, used to distinguish different frequency components of EEG signals; The imaginary unit in mathematical operations is used to perform Fourier transform calculations in the complex field. The corresponding discrete Fourier transform formula is expressed as: In the formula For frequency The corresponding frequency domain amplitude represents the magnitude of the vibration amplitude of the EEG signal at that frequency. Through this transformation, the time-domain EEG signal can be completely converted into frequency domain data, and the energy distribution density data of the signal in the entire frequency range can be obtained.

[0031] Based on a preset target emotion-related frequency range, multiple analysis frequency bands are defined. The frequency domain energy data within each band is integrated and accumulated to calculate the total energy value within that band. The energy percentage is then normalized. The specific calculation method is as follows: In the formula The total energy value of a single target frequency band is a quantitative result of the intensity of brain electrical activity in that frequency band; The lower limit frequency of the target frequency band. The upper limit of the target frequency band, together with the upper limit of the frequency band, defines the analytical range of the frequency band related to emotion. The square of the amplitude in the frequency domain represents the frequency. The energy density of the EEG signal is calculated by integration, which involves summing the energy densities of all frequencies within the band. Through energy statistics and feature extraction on a frequency band basis, EEG characteristic frequency band parameters reflecting the intensity changes of brain neural electrical activity under different frequency components are obtained.

[0032] By combining phase synchronization analysis and correlation analysis, pairwise correlation calculations are performed on the signals from all acquisition channels to calculate the covariance value of any two channel signals. This reflects the degree of linear correlation between the two channel signals: in, Representing the Preprocessed EEG signals from each acquisition channel Representing the Preprocessed EEG signals from each acquisition channel and These are the average values ​​of the two channel signals, respectively. This represents the expectation operation, calculating the standard deviation of two channel signals. and The standardized correlation coefficient was obtained. The formula is expressed as: By traversing all channel combinations, a complete channel correlation coefficient matrix is ​​constructed. Combined with physiological prior knowledge related to depression and anxiety, connection paths between key brain regions such as the prefrontal cortex, temporal lobe, and central region are identified. The connection strength values ​​and topological distribution patterns of connection paths between each node are extracted to form the topological features of brain network connections.

[0033] Step 2.2 involves associating and fusing the EEG feature frequency band parameters with the brain network connection topology features according to preset data dimensions to obtain a feature parameter set. Specifically, this includes: establishing the spatial and temporal correspondence between the two types of features; unifying the sampling time axis of the EEG feature frequency band parameters with the time series of the brain network connection topology features to ensure complete synchronization in the time dimension; simultaneously, establishing a mapping index between channel numbers and the frequency dimension, so that each node channel in the brain network topology features can accurately correspond to the relevant frequency energy data in the frequency domain analysis, achieving seamless alignment of multi-dimensional features. This fusion and integration is performed using a combination of weighted concatenation and feature vector expansion, expressed by the following formula: In the actual fusion process, to balance the weight contributions of the two types of features, a preset weight coefficient can be introduced. and Scaling the features to satisfy ,Right now: ,in, This represents the final set of feature parameters formed by fusion. Represents the frequency band parameter vector of EEG features. This represents the topological feature vector of brain network connectivity. Through this operation, the energy distribution features in the frequency domain and the network connectivity features in the spatial domain are correlated, mapped, fused, and encapsulated to finally obtain a comprehensive feature parameter set with multi-dimensional information such as frequency energy, connectivity strength, and topological structure.

[0034] Step 2.3: Based on the physiological region mapping relationship corresponding to the feature parameter set, locate three specific recording electrodes covering the prefrontal cortex, temporal lobe, and central region in the standard EEG electrode distribution. Mark these three specific recording electrodes as feature sensing nodes. Specifically, this includes: calling the internationally recognized standard EEG electrode distribution map to clarify the specific spatial location of each acquisition electrode channel number on the subject's scalp surface, as well as the corresponding cerebral cortex physiological region. Focus on identifying the distribution range of the three core brain regions—the prefrontal cortex, temporal lobe, and central region—which are highly related to the regulation of depression and anxiety and the execution of cognitive functions. Based on the feature peak values, select target nodes and analyze the energy distribution data and topological connectivity strength in the feature parameter set. According to the method, the channel locations corresponding to the peak points of energy distribution and peak points of connectivity strength are found. The peak points usually correspond to the areas with the most active neural electrical activity and the strongest functional connectivity, which can effectively reflect the core changes in emotional state. Based on the mapping relationship of physiological regions, three specific recording electrodes covering the prefrontal lobe, temporal lobe and central region are selected from the standard atlas. It is ensured that the selected electrodes are located at the key sampling points of each core brain region to capture the changes in electrical activity characteristics of the corresponding brain regions. After the feature sensing nodes are marked and recorded and the localization is completed, the three specific recording electrodes are officially marked as feature sensing nodes. The spatial coordinate information, channel number and corresponding physiological brain region name of each node are recorded to establish a node spatial location database.

[0035] In this embodiment of the invention, by employing technical means such as frequency domain analysis and inter-channel correlation calculation to extract feature frequency band parameters and topological features from preprocessed EEG signals, as well as associating, mapping, fusing, and encapsulating them into a feature parameter set according to preset dimensions, and locating electrodes in core brain regions such as the prefrontal cortex based on physiological mapping relationships as feature sensing nodes, the invention overcomes the technical problems of traditional feature extraction that only focuses on a single dimension, does not integrate EEG frequency and network topological features, and lacks targeted sensing nodes, resulting in incomplete capture of emotional features. Thus, it achieves the technical effect of comprehensively integrating emotion-related EEG information, accurately locating core emotional feature acquisition points, and constructing a highly effective feature parameter set, providing feature support for spatial compensation and emotion pattern recognition.

[0036] In a preferred embodiment of the present invention, step 2 above may include: Step 2.4 involves mapping the real-time EEG amplitude sequences and phase synchronization metrics acquired by the three feature sensing nodes to a three-dimensional virtual coordinate system, thus constructing a virtual irregular spatial feature surface. Specifically, this includes: based on the three feature sensing nodes with localization markers, initiating a real-time data synchronization acquisition process; continuously acquiring EEG potential amplitude data from the three feature sensing nodes at each sampling time according to the fixed sampling frequency of the EEG acquisition device, obtaining three independent real-time EEG amplitude sequences; ensuring that the sampling timestamps of the three sequences are completely aligned; retrieving three sets of phase synchronization metrics, which correspond to the degree of synchronization of brain functional connections between each pair of the three feature sensing nodes, quantifying the coordinated state of neural electrical activity between core brain regions; and constructing a three-dimensional virtual coordinate system, using the standard three-dimensional spatial coordinates of the three feature sensing nodes on the scalp surface as a reference, and defining the axes of the coordinate system: with the horizontal direction of the scalp as the reference. The axis and the anterior-posterior direction of the scalp are The values ​​of the axis and EEG dynamic characteristics are z The axes ensure that the coordinate system can accurately map the spatial position of the scalp surface and the dynamic changes of EEG features. Real-time EEG amplitude sequences and phase synchronization measurements are mapped to the constructed three-dimensional virtual coordinate system: the real-time EEG amplitude sequences of the three feature sensor nodes are mapped to the corresponding coordinate system. z The dynamic height component of the axis reflects the intensity changes of EEG activity at each node; the three sets of phase synchronization measures are mapped to the spatial connectivity weight components between the three nodes.

[0037] To achieve continuous spatial representation of EEG features, a cubic spline spatial interpolation algorithm is employed. This results in a large number of continuous surface sampling points between the reference spatial points corresponding to the three feature sensing nodes. The phase connection weights of the z-axis heights of these continuous surface sampling points are obtained through interpolation based on the values ​​of adjacent reference points. The calculation formula is as follows: In the formula, This represents the z-axis height value of any sampling point after interpolation. , For this sampling point in the coordinate system - Coordinates of a plane to The interpolation coefficients are obtained by solving the coordinates of three reference nodes and the corresponding EEG amplitude data. Through the above interpolation operation, the discrete node features are transformed into a continuous spatial surface, and finally a virtual irregular spatial feature surface that dynamically changes with the EEG signal is constructed.

[0038] Step 2.5 involves performing a meshing discretization process on the virtual irregular spatial feature surface, dividing it into discrete surface elements covering the entire surface, and integrating them into a set of discrete surface elements. Specifically, this includes: using the Delaunay trigonometric meshing algorithm to perform meshing discretization on the constructed virtual irregular spatial feature surface. The core purpose is to decompose the continuous surface into discrete, computable surface elements, laying the foundation for local curvature extraction and integration. The specific implementation process is as follows: setting the meshing accuracy parameters, determining the size of the discrete surface elements after meshing based on the complexity of the virtual irregular spatial feature surface, ensuring that the elements accurately fit the surface shape, and simultaneously... To avoid computational redundancy caused by an excessive number of micro-elements, all sampling points of the virtual irregular spatial feature surface are traversed. Triangular discrete surface micro-elements are constructed by grouping three adjacent sampling points together. During the construction process, the core rules of Delaunay triangulation are strictly followed to ensure that each micro-element is a non-degenerate triangle and that all micro-elements are non-overlapping and complete, fully covering the entire virtual irregular spatial feature surface. After the triangulation is completed, all discrete surface micro-elements are numbered and counted, and the coordinates of the three vertices, the side length, and the spatial position of each micro-element are recorded. All discrete surface micro-elements are integrated into a discrete surface micro-element set, and the correspondence between micro-elements and surface sampling points is established.

[0039] Step 2.6: Extract the local curvature distribution data of each discrete surface element from the set of discrete surface elements, perform regional curvature integral operations on the local curvature distribution data, and spatially accumulate and summarize the integral results of each element to obtain the initial spatial distribution compensation field. Specifically, this includes: extracting the local curvature distribution data of each discrete surface element; for each element in the set of discrete surface elements, calculating the unit normal vector of the plane containing the element using the spatial coordinates of its three vertices; the calculation formula is as follows: In the formula, Let be the unit normal vector of the plane containing the infinitesimal element. , Let be two adjacent edge vectors of a infinitesimal element. This is the cross product of two edge vectors. is the modulus of the cross product result.

[0040] Calculate the two principal curvatures of each discrete surface element based on the unit normal vector. and Principal curvature is the core parameter characterizing the degree of curvature of a micro-element surface, among which Let be the curvature of the direction of maximum curvature on the infinitesimal surface. The curvature in the direction of minimum bending is calculated using the surface curvature formula: In the formula, , The coefficients are the second fundamental form coefficients of the surface. , The first fundamental form coefficients of the surface are all calculated using the coordinates of the vertex of the infinitesimal element and the unit normal vector. The arithmetic mean of the two principal curvatures is taken as the local mean curvature of the discrete surface element, and the calculation formula is as follows: In the formula, To find the local mean curvature of a discrete surface element, a region curvature integral operation is performed on the local curvature distribution data. For each discrete surface element, its local mean curvature is multiplied by the area of ​​that element to obtain the local curvature integral value. The calculation formula is as follows: In the formula, Let be the local curvature integral value of a single discrete surface element. Let the area of ​​this discrete surface element be the coordinates of its three vertices, calculated using Heron's formula. In the formula, , , Let be the lengths of the three sides of the infinitesimal element. The integral of local curvature of all infinitesimal elements in the discrete surface element set is calculated by summing the integral values ​​of their local curvature in space, which is half the perimeter of the infinitesimal element. The formula is as follows: In the formula, The initial spatial distribution compensation field quantity, This represents the total number of infinitesimal elements in the discrete surface element set. For the first The local curvature integral values ​​of each discrete surface element are accumulated and integrated into the global curvature characteristics of the entire virtual surface to obtain the initial spatial distribution compensation field.

[0041] In this embodiment of the invention, by employing techniques such as mapping the real-time EEG amplitude sequence and phase synchronization metric of three feature sensing nodes to a three-dimensional virtual coordinate system, constructing a virtual irregular spatial feature surface through cubic spline interpolation, using the Delaunay triangular meshing algorithm to discretize the surface into a set of discrete surface micro-elements, extracting the local curvature of each micro-element, and obtaining the initial spatial distribution compensation field quantity through regional curvature integral and summation, the technical problems in the prior art—namely, the inability to integrate the EEG amplitude and functional connectivity synchronization of the core brain region into spatial features, the lack of refined discretization methods in surface processing, and the inability to quantify the surface morphology through curvature, thus making it difficult to obtain accurate spatial compensation field quantities—are overcome. This achieves the technical effect of realizing continuous spatial representation of EEG features, accurately capturing spatial distribution differences of EEG features in the core brain region, accurately quantifying surface morphology changes through curvature integral, and obtaining an initial spatial distribution compensation field quantity that conforms to the spatial distribution law of EEG, providing spatial feature support for field quantity optimization and the recognition of depressive and anxious emotions.

[0042] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1: By performing boundary constraint iterative fitting processing, the minimum covering elliptical boundary contour covering all projection points is obtained. The center point coordinates, major axis length, minor axis length, and spatial rotation angle of the minimum covering elliptical boundary contour are extracted to obtain the basic geometric fitting parameters. Specifically, based on the projection point set of the three feature sensor nodes obtained by projection, the boundary constraint iterative fitting process is started to obtain an elliptical boundary contour that completely covers all projection points and meets the minimum coverage requirement. The specific implementation steps are: Select the horizontal and vertical coordinate data of all projection points in the projection point set to construct the basic input dataset for ellipse fitting. Use the least squares ellipse fitting algorithm with boundary constraints to construct the general equation of the ellipse. The equation expression is: In the formula, , These are the x and y coordinates of a single projection point in the projection point set, respectively. , , , , , These are the undetermined coefficients of the ellipse, and they must satisfy the constraints of ellipse fitting, i.e. This ensures that the fitted result is a valid ellipse rather than a hyperbola or parabola.

[0043] In the initial fitting stage, a subset of projection points are randomly selected as the initial fitting benchmark, and the initial undetermined coefficients of the ellipse are obtained. Boundary constraints are introduced, and the coefficients are iteratively optimized step by step: with the coverage of all projection points as the core constraint, the coordinates of all projection points are substituted one by one to verify whether the ellipse equation is satisfied. If there are any projection points that are not covered, the undetermined coefficients are adjusted until all projection points fall inside or on the ellipse boundary. At the same time, with the minimum ellipse area as the optimization objective, the ellipse boundary range is continuously reduced through iterative calculations, and finally the minimum covering ellipse boundary contour covering all projection points is obtained. After completing the fitting of the minimum covering ellipse boundary contour, its basic geometric fitting parameters are extracted. The specific calculation process is as follows: the x and y coordinates of the ellipse center point are solved by the mean of the coordinates of all projection points, and the formula is expressed as: In the formula, The x-coordinate of the center point The ordinate of the center point is... The total number of projection points in the projection point set. , The first The horizontal and vertical coordinates of each projection point.

[0044] Based on the undetermined coefficients of the ellipse, the general equation of the ellipse is transformed to its standard form through coordinate transformation. The lengths of the major and minor semi-axes are then obtained by solving for them. Multiplying each by 2 yields the lengths of the major and minor axes. The major axis is the distance between the two farthest points on the ellipse boundary, and the minor axis is the maximum width of the ellipse boundary perpendicular to the major axis. The specific calculations are performed using the parameters of the standard ellipse equation, expressed by the following formula: In the formula, The length of the major semi-axis The length of the minor axis is 2, and the length of the major axis is 2. The minor axis length is 2 The angle between the major axis and the x-axis of the virtual coordinate system, i.e., the spatial rotation angle, is determined by using the undetermined coefficients of the ellipse. The formula is expressed as: In the formula, The spatial rotation angle represents the spatial orientation of the minimum covering ellipse. The coordinates of the center point, the length of the major axis, the length of the minor axis, and the spatial rotation angle are integrated into the basic geometric fitting parameters.

[0045] Step 3.2: Calculate the numerical ratio of the minor axis length to the major axis length in the basic geometric fitting parameters to quantify the spatial dispersion of the projection point set and obtain the eccentricity; analyze the spatial orientation of the spatial rotation angle in the three-dimensional coordinate system to obtain the major axis direction vector; encapsulate and combine the eccentricity and the major axis direction vector to obtain the geometric feature parameters. Specifically, this includes: based on the basic geometric fitting parameters, using an elliptical geometric analysis algorithm based on centroid constraints to analyze and encapsulate the geometric feature parameters. This algorithm uses the centroid coordinates of the projection point set as a rigid geometric constraint, locks the ellipse center reference through point set mean calculation, and completes the normalization analysis of the major and minor axis ratios and spatial orientations by combining the standard geometric definition of the ellipse. This effectively suppresses the calculation offset caused by the discrete distribution of projection points and ensures the consistency of shape quantization and direction extraction. The specific steps are: the shape quantization rules of the elliptical geometric analysis algorithm based on centroid constraints. The algorithm uses the centroid coordinates of the projection point set as a reference to perform coaxial calibration of the major and minor axis parameters, where the centroid coordinates are calculated using the point set mean formula. In the formula, Let x be the x-coordinate of the center point of the ellipse. The ordinate of the center point of the ellipse is y. The total number of projection points. , The first After calibration, the horizontal and vertical coordinates of each projection point are selected from the basic geometric fitting parameters, specifically the minor axis and major axis lengths. The algorithm's built-in geometric ratio calculation logic is used to calculate their numerical ratio, which quantitatively characterizes the dispersion of the projection point set in virtual space. This ratio is the core basis for calculating eccentricity, expressed by the formula: In the formula, For eccentricity, For the length of the shorter half-axis, The length of the semi-major axis is denoted by 0. The value of eccentricity ranges from 0 to 1. When the eccentricity approaches 0, the distribution of the projection point set is close to a circle, and the spatial dispersion is low. When the eccentricity approaches 1, the distribution of the projection point set is significantly stretched, and the spatial dispersion is high. This value can be used to accurately quantify the spatial morphological characteristics of the projection point set.

[0046] The spatial orientation analysis rule of the elliptic geometric analytical algorithm based on centroid constraints normalizes the spatial rotation angle with the centroid as the coordinate origin to eliminate the angle error caused by center offset. It analyzes the spatial orientation of the spatial rotation angle in the three-dimensional virtual coordinate system from the basic geometric fitting parameters, obtaining the major axis direction vector. Using the constructed three-dimensional virtual coordinate system as a reference, the spatial rotation angle is converted into its direction component in three-dimensional space. The specific calculation process is as follows: Based on the rotation angle, the major axis is determined... xy The projection direction of the plane, combined with the spatial mapping rules of the three-dimensional coordinate system and the vector generation logic of the algorithm, constructs the major axis direction vector, whose vector expression is: In the formula, The major axis direction vector, For spatial rotation angle, For the major axis and z The included angle of the axis, in this embodiment, is due to the projection point set being located at... x- y In a plane, Φ takes the value of 0, so the major axis direction vector simplifies to ( , ,0), representing the dominant extension direction of the projection point set.

[0047] Step 3.3 involves spatially aligning the geometric feature parameters with the initial spatial distribution compensation field, and then performing morphological stretching or compression adjustment on the radial amplitude distribution of the initial spatial distribution compensation field to obtain a morphologically adaptive field. This morphologically adaptive field includes adjusted amplitude components. Specifically, this includes: using the center point coordinates of the minimum covering ellipse as a reference, unifying the spatial coordinate origin of the initial spatial distribution compensation field so that the spatial distribution center of the initial field completely coincides with the center of the projection point set; aligning the radial dimension of the initial spatial distribution compensation field with the major and minor axis dimensions in the geometric feature parameters to ensure precise matching between the radial distribution range of the field and the spatial coverage range of the projection point set, eliminating spatial misalignment and dimensional deviation; and, based on the eccentricity in the geometric feature parameters, performing morphological stretching or compression adjustment on the radial amplitude distribution of the initial spatial distribution compensation field to obtain the morphologically adaptive field. The specific calculation process is as follows: determining the stretching or compression adjustment coefficient, which is positively correlated with the eccentricity, expressed by the formula: In the formula, For adjustment coefficients, The eccentricity is the coefficient of adjustment. When the eccentricity is greater than 0.5, the adjustment coefficient is greater than 1, which stretches the radial amplitude distribution of the initial field quantity and strengthens the field quantity amplitude in the spatial stretching region. When the eccentricity is less than 0.5, the adjustment coefficient is less than 1, which compresses the radial amplitude distribution of the initial field quantity and weakens the field quantity amplitude in the region with low spatial dispersion. When the eccentricity is equal to 0.5, the adjustment coefficient is equal to 1, which keeps the radial amplitude distribution of the initial field quantity unchanged.

[0048] The radial amplitude components of the initial spatial distribution compensation field are weighted and adjusted, as expressed by the following formula: In the formula, The radial amplitude component of the morphological adaptive field is adjusted. The radial amplitude component of the initial spatial distribution compensation field. The radial distance from the spatial point to the center point is used for weighting operations to adaptively match the spatial dispersion pattern of the projection point set with the radial amplitude distribution of the initial field quantity.

[0049] Step 3.4: Using the major axis direction vector as a spatial guiding reference, the gradient propagation direction of the morphological adaptive field is oriented and rotated for correction to obtain the corrected direction component. The corrected direction component is then vector-synthesized with the adjusted amplitude component to obtain the final spatial distribution compensation field. Specifically, this includes: analyzing the initial gradient propagation direction of the morphological adaptive field, which is determined by the amplitude change gradient of the field and represents the natural extension trend of the spatial distribution of EEG features; using the major axis direction vector as a spatial guiding reference, the initial gradient propagation direction is rotated for correction to ensure that the gradient propagation direction of the field is consistent with the dominant direction of functional connectivity in the core brain regions. The specific rotation correction formula is as follows: In the formula, This is the corrected gradient propagation direction vector. For rotation matrix, This represents the rotation angle corresponding to the major axis direction vector. The initial gradient propagation direction vector of the shape-adaptive field is: The rotation matrix expression is: This rotation operation eliminates the deviation between the initial gradient propagation direction and the direction of functional connectivity in brain regions.

[0050] The corrected direction component and amplitude component are vector-synthesized to generate the final spatial distribution compensation field, expressed by the formula: In the formula, To compensate for the final spatial distribution of the field quantity, For amplitude components, To obtain the unit vector of the corrected direction vector, the spatial distribution compensation field quantity that accurately corresponds to the functional connection direction of the brain region is obtained by precisely coupling the amplitude and direction to the shape of the adaptive matching projection point set.

[0051] In this embodiment of the invention, because it employs the technical means of obtaining the minimum covering elliptical boundary contour covering all projection points through boundary constraint iterative fitting and extracting basic geometric parameters, calculating eccentricity based on the ratio of minor axis to major axis, analyzing the major axis rotation angle to obtain the major axis direction vector and encapsulating geometric feature parameters, aligning the geometric parameters with the initial field quantity space and adjusting the amplitude distribution based on eccentricity, correcting the gradient propagation direction based on the major axis direction vector and vector synthesizing the final field quantity, it overcomes the technical problems in the prior art of mismatch between the spatial distribution compensation field quantity shape and the dispersed features of the projection point set, deviation between the field quantity propagation direction and the brain region functional connection direction, and insufficient accuracy of emotional feature compensation due to the inability to adaptively adjust the field quantity distribution. This achieves the technical effects of accurately matching the spatial shape of the projection point set to adaptively adjust the field quantity amplitude, accurately correcting the field quantity propagation direction to be consistent with the brain region functional connection direction, improving the effectiveness and robustness of the spatial distribution compensation field quantity, providing accurate and suitable spatial compensation basis for the identification of depression and anxiety, and enhancing the accuracy of monitoring.

[0052] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1: Analyze the field intensity gradient values ​​of the final spatial distribution compensation field at the corresponding spatial coordinates of each feature sensing node. Map the field intensity gradient values ​​spatially with the brain network connection topology features in the feature parameter set to obtain the initial spatial weight coefficient sequence. Specifically, this includes: analyzing the field intensity gradient values ​​of the final spatial distribution compensation field at the corresponding spatial coordinates of each feature sensing node. Using the spatial coordinates of the three feature sensing nodes as reference points, determine the field intensity distribution region corresponding to each node in the virtual space. Calculate the gradient components of the field at the spatial coordinates of that node using numerical differentiation, specifically including the gradient components along the x-axis and y-axis of the virtual coordinate system. Obtain the field intensity gradient amplitude by performing a square root operation on the sum of the squares of the two gradient components. This amplitude directly reflects the intensity of the EEG feature distribution at the spatial location of that node. The calculation formula is: In the formula, The magnitude of the field strength gradient. for axial gradient components, for The gradient component along the axial direction is used to perform spatial location mapping and matching between the field strength gradient value and the topological features of the brain network connection. All topological connection paths in the topological features of the brain network connection are traversed, and the position coordinates of the node pair corresponding to each path in virtual space are determined. These position coordinates are then mapped to the spatial coordinates of the feature sensing nodes to establish a spatial correspondence between the topological connection path and the field strength gradient value. Based on the positive correlation logic between spatial distance and gradient magnitude, the initial spatial weight coefficient corresponding to each topological connection path is calculated. That is, the weight of the topological connection path is directly proportional to the field strength gradient magnitude at the corresponding location and inversely proportional to the spatial distance from the path to the gradient peak region. The calculation formula is as follows: In the formula, These are the initial spatial weighting coefficients. The distance between the topological connection path and the peak region of the field intensity gradient is represented by 1. Adding 1 is to avoid the calculation error of the denominator being zero when the distance is zero. Through the above mapping matching and calculation, the initial spatial weight coefficient sequence corresponding to each topological connection path is finally obtained.

[0053] Step 4.2: Based on the initial spatial weight coefficient sequence, the original node connection strength values ​​in the brain network connection topology features are scaled node-by-node to obtain a local connection strength matrix pre-weighted by the field quantity. The extreme values ​​of numerical fluctuations in the local connection strength matrix are extracted, and neighborhood smoothing is performed on the matrix using these extreme values ​​as constraints to obtain an optimized spatially weighted topology feature set. Specifically, this includes: scaling the original node connection strength values ​​in the brain network connection topology features node-by-node; traversing each original node connection strength value in the brain network connection topology feature set; finding the matching weight of its corresponding topology connection path in the initial spatial weight coefficient sequence; and performing a weighted operation to complete the node-by-node scaling, resulting in a local connection strength matrix pre-weighted by the field quantity. The calculation formula is as follows: In the formula, This represents the local connectivity strength value after preliminary weighting. This represents the original node connection strength value. The initial spatial weight coefficients are used for matching. This operation integrates the spatial distribution features of EEG into the weight allocation of topological features, enabling precise matching between connection strength and the spatial distribution patterns of EEG features. It extracts the extreme values ​​of numerical fluctuations in the local connection strength matrix, traverses all values ​​in the local connection strength matrix, compares and analyzes the magnitude of each value, and finds the maximum and minimum values ​​in the matrix to determine the possible abnormal weight intervals in the matrix.

[0054] Neighborhood smoothing is performed on the matrix with numerical fluctuation extrema as constraints to eliminate interference from local outlier weights. Taking each element in the local connectivity strength matrix as the center element, all elements within a predetermined neighborhood range are selected to form a neighborhood dataset. The arithmetic mean of all elements within the neighborhood is calculated as the smoothing value of the center element. The calculation formula is as follows: In the formula, The optimized connection strength value after smoothing. This represents the total number of elements in the neighborhood. For the neighboring region During the smoothing process, the numerical values ​​of each element are constrained by the extreme values ​​of numerical fluctuations to limit the range of values ​​for neighborhood smoothing, ultimately resulting in an optimized spatially weighted topological feature set.

[0055] Step 4.3 involves comparing the data dimensions and replacing the structure of the optimized spatial weighted topological feature set and the feature parameter set to obtain the replaced weighted features. The replaced weighted features are then isomorphically concatenated with the retained EEG feature frequency band parameters to obtain the intermediate feature parameter set to be verified. Specifically, this includes: performing data dimension comparison and structure replacement between the optimized spatial weighted topological feature set and the feature parameter set; analyzing the dimensional structure of the feature parameter set, the feature types of each dimension, and the data range; and comparing the number of dimensions, dimension types, and data distribution characteristics of the optimized spatial weighted topological feature set. If the number of dimensions is inconsistent, dimension alignment is achieved through dimension completion or dimensionality reduction. If the dimension types differ, structural adaptation is completed through type conversion. After dimension alignment, the original brain network connection topological feature parts corresponding to the feature parameter set are replaced with the optimized spatial weighted topological feature set, while retaining other non-topological feature parts in the feature parameter set, ensuring the integrity and dimensionality matching of the replaced data structure.

[0056] The replaced weighted features are isomorphically concatenated with the retained EEG feature frequency band parameters. The core of isomorphic concatenation is to unify the structural form and data organization logic of the two types of features. This integration is achieved using vector concatenation. The replaced weighted feature vector and the EEG feature frequency band parameter vector are concatenated end-to-end in a preset order to obtain a new unified feature vector. The calculation formula is as follows: In the formula, This is the set of intermediate feature parameters to be verified. The weighted feature vector after replacement. To preserve the frequency band parameter vectors of EEG features, isomorphic splicing is used to achieve deep fusion of frequency domain EEG features and spatial domain weighted topological features, thereby constructing an intermediate feature parameter set containing multi-dimensional emotion-related information.

[0057] Step 4.4: Perform multidimensional data consistency verification on the intermediate feature parameter set to be verified. Standardize and encapsulate the verified feature data to obtain the corrected feature parameter set. Specifically, this includes: checking whether the number and order of each sub-feature dimension in the intermediate feature parameter set are completely consistent with the preset standard dimensions for emotion recognition features. If there are missing dimensions, redundant dimensions, or disordered order, the dimension verification is deemed to have failed. Then, verify whether each feature value in the intermediate feature parameter set falls within the corresponding range of the preset reasonable numerical range for each sub-feature. If there are values ​​exceeding the range or abnormal values, the numerical verification is deemed to have failed. Finally, combine the physiological priors of depression and anxiety emotion regulation. The knowledge is used to verify whether the numerical correlation between the frequency band parameters of EEG features and the spatially weighted topological features conforms to physiological laws, such as the positive / negative correlation between EEG energy and topological connection strength during emotional fluctuations. If the correlation is abnormal, it is judged as a logical verification failure. The feature data that has passed the above three verifications are standardized and encapsulated. In order to eliminate the difference in dimensionality and numerical scale deviation between different feature dimensions, the minimum-maximum normalization method is used to complete the standardization process, and each feature value is uniformly mapped to the interval between 0 and 1. After standardization, the feature data is structured and encapsulated to clarify the meaning, data type and value range of each feature dimension, and finally obtain the corrected feature parameter set.

[0058] In this embodiment of the invention, by employing the following technical means—namely, analyzing the field strength gradient value of the final spatial distribution compensation field quantity and matching it with the spatial mapping of the brain network connection topology feature space to generate an initial weight sequence, scaling the original connection strength node by node based on the initial weight sequence and smoothing the neighborhood through extreme value constraints to obtain an optimized topology feature set, comparing the dimensions of the optimized topology feature set with the feature parameter set and replacing the structure, and isomorphically splicing it with the EEG feature frequency band parameters to obtain an intermediate feature parameter set, and performing multi-dimensional consistency verification and standardized encapsulation on the intermediate feature parameter set to obtain a corrected feature parameter set—this overcomes the technical problems in the prior art, such as the mismatch between the weights of the brain network connection topology feature and the spatial distribution law of EEG features, the existence of local abnormal weight interference in the topology feature leading to poor robustness, the heterogeneous structure of the feature parameter set preventing deep fusion, and the lack of data verification leading to insufficient feature effectiveness, thus affecting the accuracy of emotion recognition. This achieves the technical effects of accurately matching the spatial distribution of the field quantity with the connection strength of the topology feature, improving the anti-interference ability and effectiveness of the topology feature, unifying the structure and dimensions of multi-dimensional features, ensuring the consistency and standardization of feature data, providing a high-quality and highly reliable feature parameter set for the recognition of depression and anxiety, and enhancing the accuracy and robustness of the emotion recognition model.

[0059] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1 involves performing data dimension alignment and numerical range standardization on the corrected feature parameter set to eliminate dimensional differences between feature parameters and obtain a standardized feature vector sequence. Specifically, this includes: accurately normalizing the corrected feature parameter set according to the fixed input dimension length of the pre-trained EEG feature classification model. The specific implementation process is as follows: retrieve the input interface parameters of the pre-trained model, clarify the required feature dimension length, and statistically compare the current dimension of the corrected feature parameter set. If the current dimension of the feature parameter set is less than the required model length, constant zero-value padding is used to complete the dimension, ensuring the zero values ​​are consistent with the type of the original feature data. If the current dimension of the feature parameter set exceeds the required model length, mutual information is used to calculate the contribution of each feature dimension to the recognition of depression and anxiety. The calculation formula is: In the formula, For feature dimensions With emotion tags Mutual information value; Eigenvalues With emotion tags The joint probability, Eigenvalues The marginal probability, Emotional tags The marginal probabilities are used to sort all feature dimensions from high to low based on mutual information values, and redundant dimensions with the lowest contribution and the lowest ranking are removed until the feature dimensions match the requirements of the model input interface.

[0060] Numerical range standardization is performed, mapping all feature values ​​to a standardized range of 0 to 1 to eliminate dimensional differences between different feature parameters. Each feature value is then individually verified to ensure it falls within the 0-1 range. If outliers are found, they are corrected. The Clip function applies a range constraint to feature values. The full name of the Clip function is numerical clipping function. It is the core function used in data preprocessing to eliminate outlier values. Its function is to forcibly constrain values ​​that exceed the specified legal range to the range boundary. Function operation rules: ,in, For standardized feature values, when the standardized value is less than 0, the function outputs 0; when the standardized value is greater than 1, the function outputs 1; when the value is in the range of 0 to 1, the function directly outputs the original value. Through this function constraint, it is ensured that all standardized feature values ​​meet the model input requirements. Through the complete process of the above dimension regularization and numerical standardization, a standardized feature vector sequence that conforms to the model input interface specification is finally obtained.

[0061] Step 5.2 involves loading the standardized feature vector sequence into the pre-trained EEG feature classification model, driving the model to perform multi-level nonlinear feature extraction and pattern matching operations to obtain probability distribution mapping values. Specifically, this includes: loading the standardized feature vector sequence in batches into the EEG feature classification model, which has been pre-trained and validated using a large number of clinical EEG samples. This model employs a CNN and multilayer perceptron fusion structure and has been pre-trained with thousands of clinical EEG samples representing normal emotions, mild depression and anxiety, moderate depression and anxiety, and severe depression and anxiety. It possesses stable multi-level feature extraction and emotion pattern matching capabilities, driving the EEG feature classification model to progressively perform multi-level nonlinear feature extraction and pattern matching operations. The specific implementation process is as follows: the EEG feature classification model uses a CNN structure to perform spatial feature enhancement on the standardized feature vector sequence, and uses a multilayer perceptron structure to perform layer-by-layer nonlinear feature transformation on the enhanced features. Each layer of neurons strictly performs a weighted summation operation, calculated using the following formula: In the formula, For the first The linear computation results of layer neurons, For the first Layer neurons and the previous layer Connection weights between features For the next level One standardized feature value, For the first Bias terms of layer neurons.

[0062] After the weighted summation operation at each layer is completed, the ReLU nonlinear activation function is introduced to activate the linear output result, enhancing the discriminative power of the features and suppressing the interference of invalid features. The calculation formula of the activation function is as follows: In the formula, For the first The activation output of layer neurons, when When ≥0, the activation output is Retain valid features; when When the value is less than 0, the activation output is 0, and invalid features are eliminated. After 3-5 layers of multi-level nonlinear feature extraction, the EEG feature classification model inputs the finally extracted high-order features into the fully connected layer to perform emotion pattern matching. The output of the fully connected layer is converted into a probability distribution mapping value through normalization exponential operation, ensuring that the sum of all probability values ​​is 1. The calculation formula is as follows: In the formula, For the first The probability distribution mapping value of various emotional states. The first output of the fully connected layer The raw scores of various emotional states. The total number of emotional states. is a natural constant, and the resulting probability distribution mapping values ​​correspond to the confidence levels of four different emotional states.

[0063] Step 5.3 involves comparing the probability distribution mapping values ​​with the preset clinical emotional state judgment thresholds one by one, selecting the target state label with the highest confidence level, and extracting the corresponding risk intensity index to obtain preliminary emotion recognition results. Specifically, this includes comparing the probability distribution mapping values ​​with the preset clinical emotional state judgment thresholds item by item and category by category. The judgment rules and specific implementation process are as follows: retrieve the preset clinical emotional state judgment thresholds, which are determined based on the statistical results of EEG characteristics of a large number of clinical depression and anxiety cases. Set corresponding judgment thresholds for the four emotional states. During the comparison process, compare the probability distribution mapping value of each emotional state with the corresponding judgment threshold one by one. The judgment rule is: when the probability distribution mapping value of a certain emotional state is greater than or equal to its corresponding judgment threshold, the emotional state is judged as valid and reliable; if the probability distribution mapping values ​​of all emotional states are less than the corresponding judgment thresholds, the recognition result is determined to be uncertain, and the EEG characteristic data at the current moment needs to be retrieved again, and the calculation process is repeated until a valid and reliable judgment result is obtained.

[0064] Among all emotional states that meet the threshold conditions, the probability values ​​are compared one by one, and the one with the highest probability value and the highest confidence is selected as the target emotional state label. At the same time, based on the target emotional state label, the depression and anxiety risk intensity level index pre-bound to the label is retrieved. The risk intensity strictly corresponds to the four emotional states and is divided into four levels: no risk, mild risk, moderate risk, and severe risk. Each risk level corresponds to a clear numerical range. The target emotional state label and the corresponding risk intensity index are combined to obtain the preliminary emotion recognition results.

[0065] Step 5.4 involves performing temporal continuity verification and structured data encapsulation on the preliminary emotion recognition results. The verified emotion classification labels, risk intensity indicators, and corresponding monitoring timestamps are then fused and recombined to obtain the final analysis results data. This completes the EEG signal monitoring for depression and anxiety. Specifically, this includes: performing temporal continuity verification and structured data encapsulation on the preliminary emotion recognition results to ensure the stability, reliability, and standardization of the recognition results, ultimately completing the EEG signal monitoring process for depression and anxiety. The specific implementation process is as follows: Temporal continuity verification is performed, the core purpose of which is to eliminate fluctuations in the recognition results caused by transient abnormal EEG signals, ensuring that the emotion recognition results conform to the natural changing patterns of depression and anxiety. The verification method is: retrieving the preliminary emotion recognition results at the current moment, and simultaneously retrieving the recognition results from the previous 5 consecutive monitoring moments. The risk intensity indicator for each moment is extracted, and the difference between the risk intensity indicators of adjacent moments is calculated one by one. The calculation formula is: In the formula, This represents the difference in risk intensity between adjacent time points. This is an indicator of the current level of risk. This represents the risk intensity index from the previous moment.

[0066] A preset reasonable fluctuation range for risk intensity is established. The calculated difference in risk intensity between adjacent moments is compared with this preset range. If the differences between all adjacent moments are within the preset reasonable fluctuation range, the preliminary emotion recognition result is considered to be sequentially continuous and valid, requiring no correction. If the difference between any adjacent moments exceeds the preset fluctuation range, the recognition result at that moment is marked as abnormal, and a smoothing correction is performed using a moving average method. The correction formula is as follows: In the formula, To correct the risk intensity index at the current moment, For the risk intensity indicators at the first two time points, the average of three adjacent time points is used to smoothly correct abnormal results. After time series verification, the valid results that pass the verification are encapsulated in structured data. The encapsulation format strictly follows the clinical monitoring data specifications. The verified emotion classification labels, risk intensity indicators, and corresponding monitoring timestamps are merged and recombined according to a fixed structure: the emotion classification labels are clearly marked as one of the following: normal emotion, mild depression, anxiety, moderate depression, anxiety, or severe depression, anxiety; the risk intensity indicators are marked with both the level and the specific value. After encapsulation, complete and standardized analysis results data are obtained. This data can be directly used for clinical diagnostic reference and tracking of emotion change trends, ultimately completing the entire process of EEG signal monitoring for depression and anxiety.

[0067] In this embodiment of the invention, because the mutual information method is used to calculate the feature contribution to complete dimension alignment, the standardized vector is loaded into the pre-trained model that integrates CNN and MLP, and nonlinear feature extraction and probability mapping are completed through multi-level weighted summation and ReLU activation operation, the probability value is compared with the clinical statistical threshold and combined with mutual information sorting to ensure the reliability of the results, and the results are stable and standardized through temporal continuity verification and structured encapsulation, the technical means of overcoming the technical problems of poor feature dimension adaptability, model instability due to dimensional differences, low emotion classification accuracy due to insufficient feature extraction, abnormal fluctuations in results due to lack of temporal verification, and output data that cannot be directly used in clinical practice due to lack of standardized format in traditional EEG emotion recognition are achieved. Thus, the invention achieves the following: standardizing model input to improve operational stability, accurately extracting high-order emotion-related features, improving the accuracy of emotion classification and risk quantification, and ensuring the temporal continuity and reliability of recognition results, thereby realizing accurate and stable EEG monitoring.

[0068] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0069] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0070] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0071] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A brainwave signal monitoring device for depression and anxiety, characterized in that, include: The acquisition module is used to acquire the EEG signals of the subjects under resting and task conditions to obtain raw EEG time-series data; The raw EEG time-series data were bandpass filtered to obtain the filtered electrical signal; The filtered electrical signal was processed to remove artifacts related to electrooculography and electromyography, resulting in a preprocessed electroencephalogram (EEG) signal. The extraction module is used to extract EEG characteristic frequency band parameters and brain network connection topology features from the preprocessed EEG signals; construct a feature parameter set based on the EEG characteristic frequency band parameters and brain network connection topology features; and locate three recording electrodes in the prefrontal lobe, temporal lobe and central region as feature sensing nodes. The module is used to construct a virtual irregular spatial feature surface based on the EEG amplitude sequence and phase synchronization measurement collected by three feature sensing nodes; to perform grid-based discretization and regional curvature integral operation on the virtual irregular spatial feature surface to obtain the initial spatial distribution compensation field; to perform minimum coverage ellipse fitting on the projection point set of the three feature sensing nodes in the virtual space to obtain geometric feature parameters; and to perform weighted adjustment of the initial spatial distribution compensation field through the eccentricity and major axis direction vector in the geometric feature parameters to obtain the final spatial distribution compensation field. The correction module is used to perform weight compensation correction on the brain network connection topology features in the feature parameter set through the final spatial distribution compensation field quantity, so as to obtain the corrected feature parameter set. The recognition module is used to input the corrected feature parameter set into the pre-trained EEG feature classification model, perform pattern recognition on EEG signal features, and obtain analysis result data to complete EEG signal feature monitoring.

2. The EEG signal monitoring device for depression and anxiety according to claim 1, characterized in that, EEG signals were collected from subjects under resting and task conditions to obtain raw EEG time-series data; The raw EEG time-series data were bandpass filtered to obtain the filtered electrical signal; The filtered electrical signal is then processed to remove artifacts related to electrooculography (EOG) and electromyography (EMG), resulting in a preprocessed electroencephalogram (EEG) signal, including: The EEG acquisition device simultaneously acquires multi-channel raw electrical signals of the subject under resting and task conditions, and integrates and splices them into raw EEG time-series data. The raw EEG time series data is filtered by a preset passband range to remove low-frequency baseline drift and high-frequency environmental noise, resulting in a filtered electrical signal. The filtered electrical signal is subjected to waveform component identification and interference separation processing to remove the artifacts of electrooculography and electromyography, retain the effective EEG waveform, and realign the time series to obtain the preprocessed EEG signal.

3. The EEG signal monitoring device for depression and anxiety according to claim 2, characterized in that, Extract EEG characteristic frequency band parameters and brain network connection topology features from the preprocessed EEG signals; construct a feature parameter set based on the EEG characteristic frequency band parameters and brain network connection topology features; Three recording electrodes, located in the prefrontal, temporal, and central regions, were used as feature sensing nodes, including: Frequency domain analysis and interchannel correlation calculation were performed on the preprocessed EEG signals to extract the energy distribution attributes and connection strength attributes between brain region nodes in the target frequency band, thereby obtaining the EEG characteristic frequency band parameters and brain network connection topology features. The frequency band parameters of EEG features are associated, mapped, fused, and encapsulated with the topological features of brain network connections according to a preset data dimension to obtain a set of feature parameters; Based on the physiological region mapping relationship corresponding to the feature parameter set, three specific recording electrodes covering the prefrontal lobe, temporal lobe and central region are located in the standard EEG electrode distribution, and the three specific recording electrodes are marked as feature sensing nodes.

4. The EEG signal monitoring device for depression and anxiety according to claim 3, characterized in that, Based on the acquisition of EEG amplitude sequences and phase synchronization measurements from three feature sensing nodes, a virtual irregular spatial feature surface is constructed. The virtual irregular spatial feature surface is then subjected to gridded discretization and regional curvature integral calculations to obtain the initial spatial distribution compensation field, including: The real-time EEG amplitude sequence and phase synchronization metric collected by three feature sensing nodes are mapped to a three-dimensional virtual coordinate system to construct a virtual irregular spatial feature surface. The virtual irregular spatial feature surface is subjected to grid-based discrete subdivision processing to obtain discrete surface micro-elements covering the entire surface, which are then integrated into a set of discrete surface micro-elements. The local curvature distribution data of each discrete surface element is extracted from the discrete surface element set. Regional curvature integral operation is performed on the local curvature distribution data, and the integral results of each element are spatially accumulated and summarized to obtain the initial spatial distribution compensation field.

5. The EEG signal monitoring device for depression and anxiety according to claim 4, characterized in that, Minimum coverage ellipse fitting is performed on the projection point set of the three feature sensing nodes in the virtual space to obtain the geometric feature parameters; The initial spatial distribution compensation field quantity is weighted and adjusted using the eccentricity and major axis direction vector in the geometric characteristic parameters to obtain the final spatial distribution compensation field quantity, including: By performing boundary constraint iterative fitting, the minimum covering elliptical boundary contour covering all projection points is obtained. The center point coordinates, major axis length, minor axis length and spatial rotation angle of the minimum covering elliptical boundary contour are extracted to obtain the basic geometric fitting parameters. Based on the numerical ratio of the minor axis length to the major axis length in the basic geometric fitting parameters, the spatial dispersion pattern of the projection point set is quantitatively characterized to obtain the eccentricity; the spatial orientation of the spatial rotation angle in the three-dimensional coordinate system in the basic geometric fitting parameters is analyzed to obtain the major axis direction vector; the eccentricity and the major axis direction vector are encapsulated and combined to obtain the geometric feature parameters. The geometric feature parameters are spatially aligned and matched with the initial spatial distribution compensation field quantity. The radial amplitude distribution of the initial spatial distribution compensation field quantity is morphologically stretched or compressed to obtain the morphologically adaptive field quantity, which includes the adjusted amplitude component. Using the major axis direction vector as a spatial guiding reference, the gradient propagation direction of the shape adaptive field quantity is oriented and rotated for correction to obtain the corrected direction component. The corrected direction component and the adjusted amplitude component are then vector-synthesized to obtain the final spatial distribution compensation field quantity.

6. The EEG signal monitoring device for depression and anxiety according to claim 5, characterized in that, By applying weight compensation correction to the brain network connectivity topology features in the feature parameter set using the final spatial distribution compensation field, the corrected feature parameter set is obtained, including: The field strength gradient values ​​of the final spatial distribution compensation field at the corresponding spatial coordinates of each feature sensing node are analyzed, and the field strength gradient values ​​are spatially mapped and matched with the brain network connection topology features in the feature parameter set to obtain the initial spatial weight coefficient sequence. Based on the initial spatial weight coefficient sequence, the original node connection strength values ​​in the brain network connection topology features are calculated by scaling the original node proportionally to obtain a local connection strength matrix with preliminary field weighting; the numerical fluctuation extrema in the local connection strength matrix are extracted, and neighborhood smoothing is performed on the matrix with the numerical fluctuation extrema as a constraint to obtain the optimized spatial weighted topology feature set; The optimized spatial weighted topological feature set and feature parameter set are compared in data dimension and the structure is replaced to obtain the replaced weighted features; the replaced weighted features are isomorphically spliced ​​with the retained EEG feature frequency band parameters to obtain the intermediate feature parameter set to be verified. Perform multidimensional data consistency verification on the intermediate feature parameter set to be verified, and standardize and encapsulate the verified feature data to obtain the corrected feature parameter set.

7. The EEG signal monitoring device for depression and anxiety according to claim 6, characterized in that, The corrected feature parameter set is input into the pre-trained EEG feature classification model to perform pattern recognition on the EEG signal features, obtaining analysis results data to complete EEG signal feature monitoring, including: The corrected feature parameter set is subjected to data dimension alignment and numerical range standardization to eliminate the dimensional differences between feature parameters and obtain a standardized feature vector sequence. The standardized feature vector sequence is loaded into the pre-trained EEG feature classification model, which drives the model to perform multi-level nonlinear feature extraction and pattern matching operations to obtain probability distribution mapping values. The probability distribution mapping value is compared with the preset clinical emotion state judgment threshold one by one. The target state label with the highest confidence is selected and the corresponding risk intensity index is extracted to obtain the preliminary emotion recognition result. The preliminary emotion recognition results are verified for temporal continuity and encapsulated into structured data. The verified emotion classification labels, risk intensity indicators and corresponding monitoring timestamps are then fused and recombined to obtain the final analysis results data, so as to complete the monitoring of EEG signals for depressive and anxious emotions.

8. A method for monitoring electroencephalogram (EEG) signals for depression and anxiety, the method being used to execute the apparatus as described in any one of claims 1 to 7, characterized in that, include: EEG signals were collected from subjects under resting and task conditions to obtain raw EEG time-series data; The raw EEG time-series data were bandpass filtered to obtain the filtered electrical signal; The filtered electrical signal was processed to remove artifacts related to electrooculography and electromyography, resulting in a preprocessed electroencephalogram (EEG) signal. EEG characteristic frequency band parameters and brain network connection topology features are extracted from the preprocessed EEG signals; a set of characteristic parameters is constructed based on the EEG characteristic frequency band parameters and brain network connection topology features; three recording electrodes in the prefrontal lobe, temporal lobe and central region are located as characteristic sensing nodes. Based on the EEG amplitude sequence and phase synchronization measurement collected by three feature sensing nodes, a virtual irregular spatial feature surface is constructed. The virtual irregular spatial feature surface is then subjected to grid-based discretization and regional curvature integral calculation to obtain the initial spatial distribution compensation field. The projection point set of the three feature sensing nodes in the virtual space is fitted with a minimum coverage ellipse to obtain geometric feature parameters. The initial spatial distribution compensation field is then weighted and adjusted using the eccentricity and major axis direction vector in the geometric feature parameters to obtain the final spatial distribution compensation field. The brain network connectivity topology features in the feature parameter set are corrected by weight compensation correction of the final spatial distribution compensation field quantity to obtain the corrected feature parameter set. The corrected feature parameter set is input into the pre-trained EEG feature classification model to perform pattern recognition on the EEG signal features and obtain the analysis results data to complete the monitoring of EEG signal features.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to perform the method as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in claim 8.

Images