Method for evaluating mental state based on multi-mode physiological electric signals

Abstract

The invention belongs to the technical field of mental and psychological assessment, and particularly relates to a method for assessing a mental state based on a multi-modal physiological electric signal, which comprises the following steps of: acquiring the significance level of each feature observation value in a second feature matrix by utilizing statistical test analysis; utilizing statistical test to retain feature observation values with significance levels meeting a first threshold value, and obtaining a target feature matrix; judging whether the feature observation value of the target feature matrix meets a preset condition or not; if yes, inputting the target feature matrix into a detection model; if not, performing feature independence processing on the second feature matrix, and inputting the processed second feature matrix into the detection model; and outputting a detection result by using the detection model. And a fitting method under a semi-supervised clustering model is adopted to realize prediction of a continuous gradually-changing mental state.

Description

Technical Field

[0001] This invention belongs to the field of mental and psychological assessment technology, specifically relating to a method for assessing mental state based on multimodal physiological electrical signals. Background Technology

[0002] Itching, pain, fatigue, and attention are all results of the combined effects of multiple physiological electrical signals from the central and peripheral nervous systems. Electroencephalogram (EEG) signals generated by the central nervous system (including the brain and spinal cord) reflect the neural electrical activity of the cerebral cortex and can be used to assess cognitive states (attention, fatigue, pain). Common physiological signals generated by the peripheral nervous system include electrocardiogram (ECG), electromyography (EMG), and electrodermal response (EDS). ECG records cardiac electrical activity, is regulated by the autonomic nervous system, and can be used to detect stress and fatigue. EDS measures changes in skin conductance caused by sweat gland activity, is regulated by the sympathetic nervous system, and can be used to assess emotional stress (fear, anxiety, itching) and cognitive load. To obtain more accurate prediction results, different physiological electrical signals are typically collected under various mental states and scenarios, and multiple physiological signal features are fused. Then, online or offline detection is performed using a detection model. However, due to potential correlations between different scenarios, the multiple physiological electrical signals collected in each scenario may lead to redundancy in the classification results of the detection model, thus affecting the accuracy of the classification results. Summary of the Invention

[0003] This invention provides a method for assessing mental state based on multimodal physiological electrical signals, which integrates multimodal signals from multiple scenarios to assess mental state.

[0004] To address the aforementioned technical problems, this invention provides a method for assessing mental state, comprising: obtaining a first feature matrix for any given scenario. Its elements Indicates the first Physiological electrical signals in the first scenario The first time segment Observation samples corresponding to each feature; obtain the second feature matrix for all scenes. Its elements Indicates the first The first feature matrix corresponding to each scenario Statistical test analysis was performed to obtain the second characteristic matrix. Significance level of each eigenvalue Statistical test, significance level Satisfy the first threshold The second feature matrix is ​​constructed by taking the features of the first feature as the target features and using the observed samples corresponding to the target features as elements. It is then determined whether the target feature matrix meets the preset conditions. If yes, the target feature matrix is ​​input into the detection model; otherwise, the second feature matrix is... The feature is then processed to be independent before being input into the detection model; and the detection model is used to output the detection results.

[0005] Furthermore, the statistical test includes: comparing significance levels. With the first threshold Size; will The features are denoted as target features; if Then determine the relationship between the scenes, i.e., whether they are mutually exclusive or related; if they are mutually exclusive, then... The features are denoted as target features; if they are relevant, then... The features are denoted as target features.

[0006] Furthermore, determining whether the target feature matrix meets the preset conditions includes: if the target features are a non-empty set, then the preset conditions are considered to be met; if the target features are an empty set, then the preset conditions are considered not to be met.

[0007] Furthermore, determining whether the target feature matrix satisfies the preset conditions includes: for the second feature matrix Dimensionality reduction is performed; spatial features with a set contribution rate are retained, and their feature dimension number is recorded as the second threshold. The number of feature dimensions of the observed samples in the target feature matrix is ​​set to . ;when If it is true, then the preset conditions are considered met; when If not, it is considered that the preset conditions are not met.

[0008] Furthermore, the detection model outputs the following: outputs the overall prediction result based on the detection model; extracts the local prediction result with a window length based on the overall prediction result according to the time step; performs data statistics on the local prediction result under the window to achieve smooth fitting; and maps the smooth-fitted prediction result to the mental state level to output the mental state assessment result.

[0009] Furthermore, before outputting the detection results using the detection model, the process includes: training the detection model, i.e., obtaining the major category labels of each element in the target feature matrix; initializing the cluster centers or cluster distribution parameters of the detection model based on the major category labels; calculating the responsivity of each element in the target feature matrix, i.e., the distance of the element to each cluster or the probability of each cluster distribution; and iterating until the parameters of the detection model converge.

[0010] Secondly, the present invention provides a mental state assessment system, comprising: a processor for running the steps of the method described; and a human-computer interaction machine with parameter setting items on its operating interface; wherein the parameter setting items include at least one of scene input items, relationship input items between scenes, a first threshold adjustment item, and a second threshold adjustment item.

[0011] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0012] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method.

[0013] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method.

[0014] The beneficial effect of this invention is that the method for assessing mental state based on multimodal physiological electrical signals utilizes statistical test analysis to obtain the second feature matrix. Significance level of each feature observation in the middle ; retain significance level using statistical tests Satisfy the first threshold The feature observations are used to obtain the target feature matrix; it is then determined whether the feature observations of the target feature matrix meet the preset conditions; if yes, the target feature matrix is ​​input into the detection model; otherwise, the second feature matrix is ​​used. After feature independence processing, the data is input into the detection model; and the detection model outputs the detection results. The relationship between scenes is combined with statistical tests to screen target features. This can reduce the redundancy between features of multimodal physiological electrical signals in different scenes and achieve accurate prediction of continuous and gradual mental states.

[0015] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. To make the foregoing objects, features, and advantages of the invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the methodology for assessing mental state.

[0018] Figure 2 This is a flowchart illustrating the process of filtering the target feature matrix.

[0019] Figure 3 This is a schematic diagram of the clustering results of a Gaussian model under semi-supervised supervision.

[0020] Figure 4 This is a schematic diagram of the clustering results of the Gaussian model under unsupervised conditions.

[0021] Figure 5 This is a histogram showing the data distribution of EEG signals in the beta band of the prefrontal cortex.

[0022] Figure 6 This is a histogram of the data distribution of the parietal lobe EEG signal in the theta band.

[0023] Figure 7 This is a histogram of the EEG signal distribution in the delta band of the occipital lobe.

[0024] Figure 8 This is a schematic diagram showing the results of screening target features using EEG signals from the prefrontal cortex in the δ, θ, α, and β bands.

[0025] Figure 9 This is a schematic diagram showing the results of screening target features using EEG signals from the central-temporal lobe in the δ, θ, α, and β bands.

[0026] Figure 10 This is a schematic diagram showing the results of screening target features using EEG signals from the parietal lobe in the δ, θ, α, and β bands.

[0027] Figure 11 This is a schematic diagram showing the results of screening target features using EEG signals from the occipital lobe in the δ, θ, α, and β bands.

[0028] Figure 12 This is a schematic diagram showing the results of screening target features based on the electrocardiogram signal AVHR index.

[0029] Figure 13 This is a schematic diagram showing the results of filtering target features from electrocardiogram signals using SDNN metrics.

[0030] Figure 14This is a schematic diagram showing the results of screening target features based on the RMSSD index of electrocardiogram signals.

[0031] Figure 15 This is a schematic diagram showing the results of screening target features based on low-frequency energy indicators of electrocardiogram signals.

[0032] Figure 16 This is a schematic diagram showing the results of screening target features based on high-frequency energy indicators of electrocardiogram signals.

[0033] Figure 17 This is a schematic diagram showing the results of screening target features based on the ratio of low-frequency energy to high-frequency energy in electrocardiogram signals.

[0034] Figure 18 This is a schematic diagram of the detection results of the Gaussian model under semi-supervised conditions for itch mental state.

[0035] Figure 19 It is a histogram of the data distribution of electrocardiogram signals on the RMSSD index. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] See Figures 1-4 At least one embodiment provides a method for assessing mental state, comprising: obtaining a first feature matrix for any scenario. Its elements Indicates the first Physiological electrical signals in the first scenario The first time segment Observation samples corresponding to each feature; obtain the second feature matrix for all scenes. Its elements Indicates the first The first feature matrix corresponding to each scenario Statistical test analysis was performed to obtain the second characteristic matrix. Significance level of each eigenvalue Statistical test, significance level Satisfy the first threshold The second feature matrix is ​​constructed by taking the features of the first feature as the target features and using the observed samples corresponding to the target features as elements. It is then determined whether the target feature matrix meets the preset conditions. If yes, the target feature matrix is ​​input into the detection model; otherwise, the second feature matrix is... The feature is then processed to be independent before being input into the detection model; and the detection model is used to output the detection results.

[0038] Optionally, the physiological electrical signals include, but are not limited to, at least two multimodal signals such as electroencephalogram (EEG), electrocardiogram (ECG), electrodermal conductance (EDC), and respiratory signals, representing physiological electrical signals over a period of time under broad category labels or different scenarios, such as multimodal physiological electrical signal data over a period of time under several mental states in a certain scenario. Feature extraction of EEG signals includes, but is not limited to, frequency domain features, spatial domain features, and non-stationary features; time domain features are mainly based on the amplitude characteristics of EEG signals, such as absolute amplitude energy (absEMG), waveform length (WL), and root mean square (RMS) statistical features; frequency domain features are mainly based on the fluctuation characteristics of different frequency bands of EEG signals in different mental states, such as band relative energy, band absolute energy, and band median frequency, with the bands including: theta waves, delta waves, alpha waves, beta waves, low-frequency gamma (γ), and high-frequency gamma (γ); spatial domain features include the correlation between different channels and different brain regions, and spatial template features; non-stationary features include sample entropy and approximate entropy in the channel time domain and frequency domain. Feature extraction from electrocardiogram (ECG) signals includes, but is not limited to, extracting QSR waves and calculating heart rate variability (HRV) features. HRV features include time-domain and frequency-domain features. Time-domain features include: average heart rate (AVHR), standard deviation of normal sinus interval (SDNN), root mean square (RMSSD) of continuous differences between heartbeats, etc. Frequency-domain features include: high-frequency band energy, low-frequency energy, and the ratio of high to low-frequency energy, etc. Feature extraction from electrodermal (ED) signals includes, but is not limited to, time-domain and frequency-domain features. Time-domain features include: statistics based on signal amplitude, such as mean, variance, kurtosis, skewness, number of SCR peaks, and SCR amplitude, etc. Frequency-domain features include: number of SCR peaks and SCR amplitude, etc.

[0039] Optionally, in constructing the first feature matrix Previously, the acquired signals needed to undergo preprocessing and noise reduction. This preprocessing could include time-domain filtering, frequency-domain filtering, spatial-domain filtering, and wavelet decomposition followed by filtering, with different methods employed based on the characteristics of each physiological electrical signal. For example, EEG signals are weak signals at the microvolt level and are easily affected by artifacts of various components, such as ECG interference, eye movement interference, and EMG interference. Preprocessing for these signals includes: finite impulse response (FIR) bandpass filters, 50Hz notch filters, and principal component analysis for noise reduction. ECG and ductal nerve signals are characterized by high signal strength and ease of acquisition, and are low-frequency signals. The filter bank designed for these signals includes: finite impulse response (FIR) bandpass filters, such as 4–20Hz ECG FIR bandpass filters, 0.01–10Hz ductal nerve FIR bandpass filters, and 50Hz notch filters. The noise reduction process removes noise interference from the multimodal signals to obtain a clean signal.

[0040] Optional, see Figure 1 The first characteristic matrix It can be to extract the first The time-domain, frequency-domain, or spatial-domain characteristics of physiological electrical signals in various scenarios are combined to form an initialization. * eigenmatrix Characteristic observations representing different time segments or sliding window slices The feature matrix represents the feature dimension; the feature matrix is ​​standardized or normalized to obtain the first feature matrix. The normalization mentioned is not limited to min-max normalization or Z-score normalization; the normalization process can choose Z-score normalization. Signal feature extraction includes performing sliding window slicing on the preprocessed and denoised clean signal, extracting signal segments of a specific window length according to a certain time step; performing feature extraction on the signal segments, where the window length can be set to 5 seconds and the time step can be set to 0.5 seconds. For example, for a signal segment acquired over a period of 1 minute... In a given scenario, the target EEG signal is processed by a sliding window every 1 second to extract the energy of its four bands, and finally, the feature dimension of the feature matrix of the target signal is obtained. It is 60*4.

[0041] In some embodiments, see Figure 2 By combining statistical tests with scenario-based screening to select target features, the separability of the model can be effectively improved. Generally, the scenario target signal can be a multimodal signal under different targets or scenarios. The collection scenario can be set according to the detection purpose (e.g., multimodal signals under scenarios such as fatigue, moderate fatigue, and wakefulness). The relationship between different scenarios can also be set according to requirements. For example, if two scenarios are mutually exclusive (e.g., fatigue and wakefulness, or itching and no itching), then the signal features of the two scenarios are different and independent; if two scenarios are related (e.g., moderate fatigue and severe fatigue, moderate itching and severe itching), then there is a certain correlation between the signal features of the two scenarios. The second feature matrix of all scenarios is then obtained. Its elements Indicates the first The first feature matrix corresponding to each scenario .

[0042] In some embodiments, the statistical test includes: comparing significance levels. With the first threshold Size; will The features are denoted as target features; if Then determine the relationship between the scenes, i.e., whether they are mutually exclusive or related; if they are mutually exclusive, then... The features are denoted as target features; if they are relevant, then... The features are denoted as target features. Optionally, a first threshold... It can be set to 0.05 or 0.01; the statistical analysis is not limited to t-test, ANOVA, nonparametric tests, etc.; the statistical test is the expectation under the actual application scenario; in practice, the significance level can also be compared first according to the relationship between scenarios. With the first threshold The size. For example, in the first... The signals collected in this scenario are subject to the following assumptions. Assumption 1: If the two scenarios are mutually exclusive, then the signals collected in sequence will be... Perform statistical tests on the eigenvalue variables of dimension , and set the significance level. The features are denoted as target features. Assumption 2: If two scenes are related, then sequentially... Perform statistical tests on the eigenvalue variables of dimension , and set the significance level. The features are denoted as target features. Assumption 3: If there are multiple relationships between two scenes, then observe the relationships sequentially. For eigenvalue variables of dimension 1, a statistical test is performed. The feature that simultaneously satisfies both hypothesis 1 and hypothesis 2 is denoted as the target feature, i.e., the significance level. .

[0043] In some embodiments, preset conditions can be used to determine whether the target feature matrix is ​​suitable as a training set for the detection model, which can improve the parameter optimization effect of the detection model. Specifically, two judgment methods can be selected. The first method is to consider the preset conditions satisfied if the target features are a non-empty set, and not satisfied if the target features are an empty set. The second method is to consider the second feature matrix... Dimensionality reduction is performed; spatial features with a set contribution rate are retained, and their feature dimension number is recorded as the second threshold. The number of feature dimensions of the observed samples in the target feature matrix is ​​set to . ;when If it is true, then the preset conditions are considered met; when If the condition is not met, it is considered that the preset conditions are not met. Preferably, the first method of judgment is performed first: if a feature can be selected as the target feature according to statistical tests, then the target feature matrix is ​​input into the detection model; if no feature is selected as the target feature according to statistical tests, then the second feature matrix is... The feature independence process is then applied to the detection model. This feature independence process includes, but is not limited to, spatial dimensionality reduction or spatial transformation processes such as Principal Component Analysis (PCA), Singular Value Decomposition (SVD), Independent Component Analysis (ICA), and t-distributed random nearest neighbor embedding (t-SNE).

[0044] In some embodiments, see Figure 3Furthermore, the detection model can be trained using class labels before outputting detection results. The parameters of the detection model can be updated under semi-supervised supervision, including obtaining the class labels of each element in the target feature matrix; initializing the cluster center or cluster distribution parameters of the detection model based on the class labels; calculating the responsivity of each element in the target feature matrix, i.e., the distance of the element to each cluster or the probability of each cluster distribution; and iterating until the parameters of the detection model converge. Figure 4 The middle section shows the clustering results of an unsupervised Gaussian mixture model for two features. (Comparison) Figure 3 and Figure 4 It can be seen that training the detection model using class labels results in more convergent parameters under semi-supervised training.

[0045] In some embodiments, outputting detection results using a detection model includes: outputting an overall prediction result based on the detection model; extracting local prediction results of the overall prediction result based on a window length according to a time step; performing data statistics on the local prediction results under the window to achieve smooth fitting; and mapping the smooth-fitted prediction results to a mental state level to output an assessment result of the mental state. The mental state level may include, but is not limited to, continuously varying mental states such as itching, pain, fatigue, and attention.

[0046] Optionally, the detection model is a semi-supervised clustering model, such as a Gaussian mixture model, a hidden Markov tree model, or a hidden Markov random field model. The data statistics can be the average amplitude under a window, the sum of absolute amplitudes, or the amplitude of data distribution at a certain quantile; the smoothing fitting is, for example, but not limited to, mean filter or median filter smoothing fitting; the linear transformation is, for example, but not limited to, max-min normalization or z-score normalization; and the nonlinear transformation is, for example, but not limited to, transformations of the logarithm and its variants over a certain data range, or exponential transformations.

[0047] In some embodiments, a mental state assessment system is also provided, comprising: a processor for running the method; and a human-computer interaction device with parameter setting items on its interface; wherein the parameter setting items include at least one of scene input items, relationship input items between scenes, a first threshold adjustment item, and a second threshold adjustment item.

[0048] In some embodiments, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0049] In some embodiments, the processor may be a central processing unit (CPU), an ASIC, or one or more integrated circuits configured to implement embodiments of the present invention. Specifically, if the memory and processor are implemented independently, they can be interconnected via a bus to communicate with each other. The bus may be an Industry Standard Architecture (ISA) bus, a PCI bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be categorized as an address bus, a data bus, a control bus, etc. If the memory and processor are integrated on a single chip, they can communicate with each other through an internal interface.

[0050] In some embodiments, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method. The storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0051] In some embodiments, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method. The computer program may include program code, which includes computer operation instructions and may be stored in a computer-readable storage medium. Based on this understanding, when the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be implemented in the form of a software product or sold or used as an independent product, the computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method described in the various embodiments of the present invention.

[0052] Experimental example.

[0053] Experimental conditions: Electroencephalogram (EEG) and electrocardiogram (ECG) signals were collected from subjects under itch scenarios, including resting state, histamine-induced itch, and itch relief application. Relative energy characteristics of four wavebands in the prefrontal, centrotemporal, parietal, and occipital lobes were extracted to obtain initial 24-dimensional EEG features. Additionally, six indicators based on heart rate variability (HRV) of the ECG signals were used to obtain initial 6-dimensional ECG features: mean RR interval (AVHR, mean vagal heart rate, reflecting the regulatory effect of the vagus nerve on cardiac activity, in milliseconds), standard deviation of RR intervals (SDNN, in milliseconds), root mean square of the difference between adjacent RR intervals (RMSSD, in milliseconds), and low-frequency energy (frequency range 0.04–0.15 Hz, in milliseconds). 2 High-frequency energy (frequency range 0.15~0.4Hz, unit: ms) 2 ), and the ratio of low-frequency energy to high-frequency energy.

[0054] Experimental results.

[0055] (1) Extract features.

[0056] In the context of itching, a two-sample t-test was used to calculate the significant differences in features between resting-state and histamine 1, resting-state and histamine 2, histamine 1 and histamine 2, and histamine and antipruritic effects. Based on the statistical tests, the extracted feature matrices were derived from the β-band of the prefrontal cortex of the EEG (data distribution as shown in the figure). Figure 5 As shown), the θ band of the apical leaf (data distribution as shown) Figure 6 (as shown) and the delta band of the occipital lobe (data distribution as shown) Figure 7 As shown), and the RMSSD index of the electrocardiogram (data distribution as shown). Figure 19 (As shown).

[0057] Extraction results: Histamine 1 and histamine 2 showed no significant differences in the selected feature distributions for the β band of the prefrontal cortex, the θ band of the parietal cortex, and the δ band of the occipital cortex; the resting state feature distributions were significant under different histamines; and the feature distributions differed between antipruritic and different histamines. This indicates that the EEG and ECG target features selected after statistical testing, combining different physiological signals, have strong interpretability and make a significant contribution to the semi-supervised clustering model.

[0058] (2) Optimize target features.

[0059] See Figures 8-11 Based on the EEG signals of subject 013 in Table 1, the target features were selected after statistical testing of the relative energy characteristics of four bands in the prefrontal, centrotemporal, parietal, and occipital lobes. For the four scenarios mentioned above (resting state, histamine 1-induced itching, histamine 2-induced itching, and the application of anti-itch cream), a total of 16 EEG feature dimensions (i.e., energy of 4 brain regions × 4 bands) were used to determine the correlation between signal features: ① Resting state and histamine 1-induced itching are mutually exclusive; ② Resting state and histamine-2-induced itching are mutually exclusive; ③Itching is a related condition when histamine 1 and histamine 2 are present; ④ Histamine 1 and histamine 2 are mutually exclusive with topical antipruritic agents.

[0060] After performing statistical tests on each EEG feature, the significance level was obtained. The target features are marked as follows: ① Resting state and histamine 1-induced itching, EEG characteristics , denoted as *; ② Resting state and histamine 2-induced itching, EEG characteristics , denoted as *; ③Itching is a related state under histamine 1 and histamine 2, according to EEG characteristics. denoted as ns; ④ Histamine 1 and histamine 2 are mutually exclusive with topical antipruritic agents, as indicated by EEG characteristics. , denoted as *.

[0061] The comparison conditions are met according to the above statistical tests, such as... Figure 8 The relative energy of the beta band in the prefrontal cortex, Figure 10 The relative energy of the θ band of the apical leaf and Figure 11 The relative energy of the delta band in the occipital lobe. Therefore, the target features selected from the EEG features are: the relative energy of the beta band in the prefrontal lobe, the relative energy of the theta band in the parietal lobe, and the relative energy of the delta band in the occipital lobe, and a non-empty set.

[0062] See Figures 12-17 , Figure 19 Based on the heart rate variability of the ECG signal of subject 013 in Table 1, the target features were selected after statistical testing of the features. A total of six ECG feature dimensions were used for the four scenarios mentioned above (resting state, histamine 1-induced itching, histamine 2-induced itching, and the state of applying anti-itch cream). The correlation between signal features was set: ① Resting state and histamine 1-induced itching are mutually exclusive; ② Resting state and histamine-2-induced itching are mutually exclusive; ③Itching is a related condition when histamine 1 and histamine 2 are present; ④ Histamine 1 and histamine 2 are mutually exclusive with topical antipruritic agents.

[0063] After performing statistical tests on each ECG feature, the significance level was determined. The target features are marked as follows: ① Resting state and histamine 1-induced itching, ECG characteristics , denoted as *; ② Resting state and histamine 2-induced itching, ECG characteristics , denoted as *; ③Itching is a related state under histamine 1 and histamine 2, and the ECG characteristics are as follows. denoted as ns; ④ Histamine 1 and histamine 2 are mutually exclusive with topical antipruritic agents, according to ECG characteristics. , denoted as *.

[0064] like Figure 14 Since the RMSSD statistical results meet the comparison conditions, the target feature for ECG feature selection is: RMSSD, non-empty set.

[0065] In summary, based on the 24-dimensional EEG and 6-dimensional ECG characteristics of subject 013 under four scenarios (resting state, histamine 1-induced itching, histamine 2-induced itching, and topical antipruritic application), the optimal features after statistical testing are 4 dimensions: relative energy of the beta band in the prefrontal cortex, relative energy of the theta band in the parietal cortex, and relative energy of the delta band in the occipital cortex, and RMSSD. It is a non-empty set.

[0066] (3) The accuracy of semi-supervised clustering results.

[0067] The results of clustering under resting and histamine conditions for 8 subjects were compared, as shown in Table 1. It can be seen that when the feature dimension after statistical optimization is greater than 1, the clustering accuracy is significantly higher than that after PCA dimensionality reduction. That is, the clustering accuracy based on the target feature matrix is ​​significantly higher than that after PCA dimensionality reduction using the second feature matrix. In Table 1, subjects 040 and 025 did not have their target features optimized after statistical testing; that is, when the optimized target feature was an empty set, semi-supervised clustering could only be performed by inputting the second feature matrix into a Gaussian mixture model after dimensionality reduction.

[0068] Table 1. Comparison of clustering results of resting state and histamine in subjects. Subject number Initial feature dimensions (EEG + ECG) PCA feature dimensionality reduction accuracy % Statistical testing to optimize the dimensions of features (EEG + ECG) Accuracy % after feature optimization 003 30 66.77 2 66.77 034 30 59.86 3 68.05 066 30 60.02 3 72.16 040 30 67.95 0 none 024 30 51.62 2 64.11 025 30 77.74 0 none 013 30 78.81 4 80.9 100 30 51.21 0 none (4) Fit the mental state level.

[0069] Subject number 013 was selected from Table 1. A semi-supervised Gaussian mixture model was used, and the target feature matrix, selected after statistical testing, was fitted to the mental state level to assess the degree of itching. The results are as follows: Figure 18 As shown, it can be seen that in the resting state, the fitted mental state (degree of itching) is low, and the degree of itching is significantly increased after applying histamine; after relieving itching, the degree of itching is effectively reduced and gradually returns to the resting state level.

[0070] Based on the above-described preferred embodiments of the present invention, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention. That is, the technical scope of the present invention is not limited to the contents of the specification.

Claims

1. A method for assessing mental state, characterized in that, include: Obtain the first feature matrix of any scene Its elements Indicates the first Physiological electrical signals in the first scenario The first time segment Observation samples corresponding to each feature; Obtain the second feature matrix of all scenes Its elements Indicates the first The first feature matrix corresponding to each scenario ; Statistical test analysis to obtain the second characteristic matrix Significance level of each eigenvalue ; Statistical test, significance level Satisfy the first threshold The features are used as target features, and the target feature matrix is ​​composed of the observation samples corresponding to the target features; Determine whether the target feature matrix meets the preset conditions; if yes, input the target feature matrix into the detection model; if not, input the second feature matrix. After feature independence processing, the data is input into the detection model; and The detection model is used to output the detection results.

2. The method according to claim 1, characterized in that, The statistical tests include: Comparison of significance levels With the first threshold Size; Will The features are denoted as target features; like Then determine the relationship between the scenes, i.e., whether they are mutually exclusive or related; if they are mutually exclusive, then... The features are denoted as target features; if they are relevant, then... The features are denoted as target features.

3. The method according to claim 1, characterized in that, Determining whether the target feature matrix meets the preset conditions includes: If the target feature is a non-empty set, then the preset condition is considered to be met; If the target feature is an empty set, it is considered that the preset conditions are not met.

4. The method according to claim 1, characterized in that, Determining whether the target feature matrix meets the preset conditions includes: For the second characteristic matrix Perform dimensionality reduction processing; Retain the spatial features with a set contribution rate and denote their feature dimension number as the second threshold. ; Set the feature dimension of the observed samples in the target feature matrix to be ;when If it is true, then the preset conditions are considered met; when If not, it is considered that the preset conditions are not met.

5. The method according to claim 1, characterized in that, The detection results output by the detection model include: The overall prediction results are output based on the detection model; The local prediction results are extracted from the overall prediction results based on the time step and the window length. Statistical analysis is performed on the local prediction results within the window to achieve a smooth fit; The predictions from the smooth fit are mapped to mental state levels to output an assessment of mental state.

6. The method according to claim 1, characterized in that, Before outputting detection results using the detection model, the process also includes: training the detection model, i.e. Obtain the major category label of each element in the target feature matrix; Initialize the cluster center or cluster distribution parameters of the detection model based on the major category label; Calculate the responsivity of each element in the target feature matrix, which is the distance of the element to each cluster or the probability of the distribution of each cluster; Iterate until the parameters of the detection model converge.

7. A mental state assessment system, characterized in that, include: The processor performs the steps of the method as described in any one of claims 1-6; Human-computer interaction machine, with parameter setting items on its operation interface; in The parameter settings include at least one of the following: scene input, relationship input between scenes, first threshold adjustment, and second threshold adjustment.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method as claimed in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images