Perinatal depression diagnosis and evaluation system based on brain network

Through EEG network analysis technology, a resting brain network is built and combined with a support vector machine classifier, the subjectivity and accuracy of perinatal depression diagnosis is solved, early recognition and severity assessment are achieved, and diagnostic efficiency is improved and accuracy.

CN118942678APending Publication Date: 2024-11-12UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410996951.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The diagnosis of perinatal depression depends on subjective assessment, lacks objectivity and accuracy, and it is difficult for the prior art to identify and quantitatively evaluate the severity of the disease in advance, resulting in misdiagnosis and missed diagnosis.

Method used

Using EEG network analysis technology, through data collection, preprocessing, feature extraction and diagnosis modules, a resting brain network is constructed, network attributes and spatial characteristics are calculated, and the support vector machine classifier is combined to realize the early stage of perinatal depression. Diagnosis and severity assessment.

Benefits of technology

It improves the objectivity and accuracy of diagnosis, realizes early warning, reduces misdiagnosis and missed diagnosis, enhances patient compliance, and provides an objective quantitative assessment of the severity of depression, supporting personalized treatment plans.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118942678A_ABST
    Figure CN118942678A_ABST
Patent Text Reader

Abstract

The invention discloses a perinatal depression diagnosis and evaluation system based on a brain network, and belongs to the crossing field of medical health and brain science and technology. According to the method, more objective physiological indexes are provided through electroencephalogram network analysis, subjectivity in the diagnosis process is reduced, and quantitative evaluation of the severity of the disease is achieved. By analyzing the subtle change of the electroencephalogram activity, the early diagnosis of the perinatal depression is realized. Through a fine electroencephalogram network analysis technology, specific electroencephalogram characteristics related to depression can be captured, and the accuracy and specificity of diagnosis are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the interdisciplinary field of medical health and brain science technology, and specifically relates to a non-invasive electroencephalogram (EEG) detection and evaluation system. Background Art

[0002] Perinatal depression (PD) is a common mood disorder in pregnant and postpartum women, including antenatal depression and postpartum depression. It not only affects the physical and mental health of expectant mothers, but may also have adverse effects on the health of the fetus and newborn, such as premature birth and low birth weight. The incidence of PD is about 15%-25%, and women with a personal or family history of depression and other mental disorders are at greater risk of PD. Therefore, early screening for PD is very important. However, its diagnosis faces many challenges, including low diagnostic accuracy, strong diagnostic subjectivity, and diagnostic delays; objective quantitative assessment of the severity of patients is also an unresolved problem.

[0003] Currently, the diagnosis and assessment of perinatal depression mainly rely on clinical assessment and questionnaire filling, such as the Edinburgh Postnatal Depression Scale. Although these clinical scales are widely used to assess the severity of depression in patients, they are largely affected by the subjective judgment of patients, which reduces the reliability of the assessment, especially for those patients who are shy or have difficulty expressing their emotions. Therefore, the diagnosis of perinatal depression requires the development of more objective, sensitive and early detection technologies to improve the accuracy and timeliness of diagnosis.

[0004] In recent years, EEG, as a non-invasive monitoring method, has been increasingly used in depression research. Changes in brain activity, especially changes in the efficiency of information interaction between specific brain regions, have been found to be associated with cognitive functions destroyed by depression. EEG network analysis quantitatively describes the information interaction and efficiency between brain regions by calculating the strength of functional connections between different brain regions. This method has shown potential in the diagnosis of various mental illnesses and can reveal abnormal brain functional connections in patients with depression. However, the application of EEG and brain networks to the diagnosis of perinatal depression is still in the exploratory stage, and there is a lack of systematic and standard analysis methods. Therefore, the present invention takes EEG network analysis technology as the core, combined with advanced signal processing technology, to develop a new perinatal depression diagnosis and evaluation system, aiming to overcome the shortcomings of existing diagnostic methods, improve diagnostic efficiency and accuracy, and objectively and quantitatively evaluate the severity of depression in individuals.

[0005] Currently, the diagnosis and assessment of perinatal depression mostly rely on self-reports from patients and subjective assessments from clinicians. This diagnostic method is easily affected by the individual expression ability of patients, cultural differences, and the experience of the diagnostician, and lacks sufficient objectivity. Early identification of perinatal depression is crucial for intervention and treatment, but existing diagnostic methods can often only be used to diagnose when symptoms are obvious, missing the best time for intervention. Due to the heterogeneity of perinatal depression, a variety of typical symptoms are often confused with some common emotional reactions during pregnancy, resulting in misdiagnosis and missed diagnosis. Summary of the invention

[0006] The present invention aims to provide more objective physiological indicators through EEG network analysis, reduce subjectivity in the diagnosis process, and achieve quantitative assessment of the severity of the disease. By analyzing subtle changes in EEG activity, early diagnosis of perinatal depression can be achieved. Through sophisticated EEG network analysis technology, specific EEG features related to depression can be captured, improving the accuracy and specificity of diagnosis.

[0007] The present invention proposes a perinatal depression diagnosis and evaluation system based on brain network, which includes: a data acquisition module, a preprocessing module, a feature extraction module, and a diagnosis module;

[0008] The data acquisition module collects EEG signals of women during pregnancy or after childbirth;

[0009] The preprocessing module preprocesses the data collected by the data acquisition module, including: average reference, filtering, and data segmentation;

[0010] The specific processing method of the feature extraction module is:

[0011] Step 1: Construct a resting-state brain network, and then extract the following features based on the brain network:

[0012] Calculate the clustering coefficient C:

[0013]

[0014] Calculate the characteristic path length L:

[0015]

[0016] Calculate the global efficiency Ge:

[0017]

[0018] Calculate the local efficiency Le:

[0019]

[0020] Among them, i, j, h represent the corresponding brain network nodes, dij represents the weighted shortest path length between nodes i and j, d jh (Ψ i ) represents the weighted shortest path length between node j and node h associated with node i, n is the number of nodes, Ψ represents the set of all nodes in a given network, w plv is the connection weight estimated by PLV, where PLV represents the phase lock time value;

[0021] Step 2: Calculate the depression score P based on network attributes.

[0022] P=β0+β1C+β2L+β3Ge+β4Le+ε

[0023] Among them, β 0、1、2、3、4 represents the regression coefficient of the network attribute, and ε represents the error term;

[0024] Step 3: A statistical test was performed on the brain networks of the depressed pregnant women group and the healthy pregnant women group to obtain the differential brain networks. The edges of the two groups of subjects with differences greater than the set threshold on the differential brain networks were preliminarily selected as potential spatial feature edges;

[0025] Step 4: Calculate the correlation between the spatial feature edge obtained in step 3 and the depression score P, set the spatial feature edge with a correlation greater than or equal to the set threshold to "1", and set the other edges to "0" to obtain a logical matrix; perform element-by-element multiplication of the logical matrix and the differential brain network to obtain the modified potential spatial feature edge; calculate the PLV value of the corresponding position of the modified potential spatial feature edge, and the set of all PLV values ​​is defined as the original spatial feature;

[0026] The diagnosis module uses a support vector machine as a classifier, uses a data acquisition module, a preprocessing module, and a feature extraction module to obtain sample data, and uses the obtained sample data to train a classifier. The input of the classifier is the original spatial feature, and the output is divided into two categories: healthy or depressed;

[0027] During actual detection, the trained classifier is used to classify the newly received data.

[0028] Furthermore, the method of first constructing the resting state brain network in the feature extraction module is:

[0029] Phase-locked time-value PLV was used to construct the resting-state brain network of all participants;

[0030]

[0031] In the formula, w plv is the connection weight estimated by PLV, Δt is the sampling period, N is the number of samples, and For the instantaneous phase corresponding to the given time series x(t) and y(t), PLV is calculated for each participant's preprocessed 5s resting-state EEG signal data segment to obtain a 27×27 adjacency matrix. Then, the 27×27 adjacency matrices calculated for all 5s data segments of the participant are averaged to finally obtain the resting-state weighted adjacency matrix of the subject and express it in the form of a brain network.

[0032] Furthermore, the original spatial features are processed as follows;

[0033] Step 1: Data preprocessing: First, perform standardization to ensure that the data in each dimension has zero mean and unit variance, eliminating the impact of different dimensions and magnitudes;

[0034] Step 2: Covariance matrix calculation: Calculate the covariance matrix of the standardized data to quantify the linear relationship between dimensions;

[0035] Step 3: Solving eigenvalues ​​and eigenvectors: Perform eigendecomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​represent the variance of the data in the direction of the corresponding eigenvector, that is, the amount of information.

[0036] Step 4: Principal component selection: According to the size of the eigenvalue, select the eigenvectors corresponding to the first three largest eigenvalues. These eigenvectors are the required principal components.

[0037] Step 5: Data projection: Project the original spatial feature data onto the three principal components to obtain a new 3D data set, which is achieved through linear transformation. The specific formula is:

[0038] New_Data=Original_Data×Eigenvectors (6)

[0039] Among them, New_Data is the data after dimensionality reduction, Original_Data is the original data, and Eigenvectors is the matrix composed of the first three eigenvectors.

[0040] Furthermore, the specific methods of the preprocessing module are: average reference, 1-45Hz bandpass filtering, using a threshold of ±100μV to automatically exclude 5s long EEG data segments with an absolute amplitude exceeding 100μV in any channel, and 5s long data segmentation.

[0041] Furthermore, if the classifier determines that the patient is depressed, the depression score P is used to classify the depression.

[0042] The present invention has the following advantages:

[0043] 1. Improve the objectivity and accuracy of diagnosis: By extracting the spatial characteristics of the brain network of pregnant women, a diagnostic method for perinatal depression based on physiological indicators is provided, which reduces the reliance on subjective reports, improves the objectivity and accuracy of diagnosis, and helps to avoid misdiagnosis or missed diagnosis due to individual differences in patient expression.

[0044] 2. Achieve early warning: Changes in EEG networks may appear before clinical symptoms of depression become obvious. By monitoring these early changes in EEG characteristics, the risk of depression can be identified in early pregnancy, thereby achieving early diagnosis and timely intervention, and reducing adverse effects on pregnant women and their offspring.

[0045] 3. Enhance patient compliance: Compared with diagnostic techniques such as MRI and fMRI, EEG examination is obviously more convenient and cheaper, and is non-invasive and more easily accepted by pregnant women, thereby increasing participation in screening.

[0046] 4. Assist clinical decision support: In addition to diagnosis, the extracted brain network spatial features can also be used to objectively assess the severity of the patient's condition, help clinicians more accurately assess the course of the disease, and provide strong data support for patients to develop personalized treatment plans and prognosis assessments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 To investigate the statistical differences in brain networks between healthy pregnant women and those with perinatal depression;

[0048] Figure 2 Statistical differences in network attributes between healthy pregnant women and pregnant women with perinatal depression. Gray lines and bars represent the PD group, and blue lines and bars represent the HP group.

[0049] Figure 3 HAMD based on EEG index assessment and clinical diagnosis 17 Pearson correlations between scores for (a) evaluation based on network properties and (b) evaluation based on spatial features of brain networks. DETAILED DESCRIPTION

[0050] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0051] 1. Collect EEG data

[0052] Use portable or medical-grade EEG equipment to non-invasively record the subject's EEG signals during a specific period of time during pregnancy or postpartum. The environment during the collection process must be suitable to reduce interference and ensure data quality.

[0053] 2. Pretreatment;

[0054] The raw EEG signals were denoised and several preprocessing procedures were performed, including average reference, 1-45 Hz bandpass filtering, and 5-s data segmentation. A threshold of ±100 μV was used to automatically exclude 5-s EEG data segments with an absolute amplitude exceeding 100 μV in any channel.

[0055] 3. Feature extraction;

[0056] Using brain network analysis methods, phase locking value (PLV) was used to construct the resting-state brain networks of all participants.

[0057]

[0058] In the formula, w plv is the connection weight estimated by PLV, Δt is the sampling period, and N is the number of samples. and is the instantaneous phase corresponding to the given time series x(t) and y(t).

[0059] Specifically, the PLV was calculated for each participant (pregnant women with perinatal depression and healthy pregnant women) in the preprocessed 5s resting-state EEG signal data segment to obtain a 27×27 adjacency matrix, and then the 27×27 adjacency matrix calculated for all 5s data segments of the participant was averaged to finally obtain the resting-state weighted adjacency matrix of the subject and express it in the form of a brain network. Then, the brain networks of pregnant women with perinatal depression and healthy pregnant women were statistically tested to obtain the spatial differences between the two groups of subjects on the brain network.

[0060] In order to measure the efficiency of the brain in processing relevant information as a whole, four traditional network properties, including clustering coefficient (C), characteristic path length (L), global efficiency (Ge), and local efficiency (Le), were used to evaluate the efficiency of the corresponding resting-state brain network.

[0061]

[0062]

[0063] Where d ij represents the weighted shortest path length between nodes i and j, n is the number of nodes, Ψ represents the set of all nodes in a given network, and w plv Estimated connection weight for PLV.

[0064] In addition, considering that traditional network attributes can only measure the overall efficiency of the brain network and cannot capture the specific spatial information contained in the brain network, the spatial features of the brain network were further extracted through the following steps for subsequent diagnostic evaluation. First, the brain networks of the depressed pregnant women group and the healthy pregnant women group were statistically tested to obtain the difference brain network. The edges with significant differences in the difference brain network between the two groups of subjects were preliminarily selected as potential spatial feature edges. Secondly, the potential spatial feature edges were further modified and screened using the correlation between the clinical scale and the resting-state brain network of all subjects. Specifically, after the correlation analysis of the brain network and the psychological scale, the edges significantly correlated with the depression score were retained (set to "1"), and the irrelevant edges were discarded (set to "0"), thereby obtaining a logical matrix; then the logical matrix was element-by-element multiplied with the above-mentioned difference brain network to obtain the Hadamard product (modified potential spatial feature edges). The set of PLV values ​​at the corresponding positions of these modified potential spatial feature edges was regarded as the original spatial feature.

[0065] Since high-dimensional data often contains a lot of redundant information, this not only increases the computational complexity, but may also affect the performance of subsequent classification models. In order to solve this problem, the present invention uses principal component analysis to reduce the original 13-dimensional spatial features to 3 dimensions (brain network spatial features) to simplify the data structure and extract key information. The main steps include:

[0066] a. Data preprocessing: First, the 13-dimensional data set is standardized to ensure that the data in each dimension has zero mean and unit variance to eliminate the impact of different dimensions and magnitudes.

[0067] b. Covariance matrix calculation: Calculate the covariance matrix of the standardized data to quantify the linear relationship between dimensions.

[0068] c. Solving eigenvalues ​​and eigenvectors: Perform eigendecomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The eigenvalue represents the variance of the data in the direction of the corresponding eigenvector, that is, the amount of information.

[0069] d. Principal component selection: According to the size of the eigenvalue, select the eigenvectors corresponding to the first three largest eigenvalues. These eigenvectors are the required principal components.

[0070] e. Data projection: Project the original 13-dimensional data of spatial features onto the three principal components to obtain a new 3-dimensional data set (brain network spatial features). This step is achieved through linear transformation, and the specific formula is:

[0071] New_Data=Original_Data×Eigenvectors (6)

[0072] Among them, New_Data is the data after dimensionality reduction, Original_Data is the original data, and Eigenvectors is a matrix composed of the first three eigenvectors. Using the above principal component analysis method, the dimension of the data is reduced, and the key features in the data are extracted, providing more refined features for subsequent classification and prediction model training.

[0073] 4. Diagnosis and assessment of perinatal depression

[0074] In order to achieve reliable diagnosis, all pregnant women (including depressed and healthy pregnant women) were divided into training and test sets. First, during the training process, the resting-state EEG indicators (network attributes and brain network spatial features) of each participant in the training set were calculated as classification features for training the support vector machine classifier (SVM). Secondly, during the test process, the corresponding resting-state EEG indicators of the participants in the test set were calculated by using the same definition, and then input into the trained SVM classifier, and finally the participants in the test set were distinguished as healthy pregnant women or depressed pregnant women. For the above process, the Leave-One-Out Cross Validation (LOOCV) method was used to cycle all subjects. Specifically, for N samples, N-1 samples were used for training, and the remaining 1 sample was used for testing each LOOCV run. The SVM classifier trained with those N-1 samples was used to classify and identify the remaining 1 sample, and this process was repeated N times until all samples were used as the test set once.

[0075] In the present invention, in order to reflect the advantages of brain network spatial features in diagnosis, different types of resting-state EEG indicators (network attributes and brain network spatial features) are used for classification and identification, and their performance for diagnosis is compared. Taking network attributes as an example, the detailed process of classification and identification is described as follows. In the process of training the SVM classifier, the network attributes (i.e., C, L, Ge, and Le) of each subject are first calculated based on the resting-state brain network constructed by PLV according to the definitions in equations (2)-(5). Then, the SVM classifier is trained accordingly using these network attributes of the training set. After entering the test process, the network attributes of the participants in the test set are first calculated according to equations (2)-(5), and then input into the trained SVM classifier to obtain the final diagnosis result. In addition, considering that the spatial information of the brain network has been widely proven to be helpful in studying the variability between individuals and the cognitive dysfunction of patients with mental illnesses, the relevant brain network spatial features are also applied to classification and identification (the brain network connection edges with significant differences between the healthy pregnant women group and the depressed pregnant women group are defined as the brain network spatial features that can improve the diagnosis of perinatal depression).

[0076] The present invention also considers using these resting-state EEG indicators to further reliably assess the severity of depression in pregnant women. Therefore, network attributes (C, L, Ge and Le) and brain network spatial characteristics are used as variables in a multivariate linear regression model to establish a prediction model to achieve a depression score (HAMD) for each subject. 17 Here, taking four network attributes as an example, the corresponding prediction model formula is as follows:

[0077] P=β0+β1C+β2L+β3Ge+β4Le+ε (7)

[0078] Among them, P represents the HAMD evaluated by the prediction model based on the network attributes 17 Score, β 0…4 Represents the regression coefficient of the network attribute, and ε represents the error term. Similar procedures were also implemented for the evaluation using brain network spatial features to further demonstrate the ability and advantages of resting-state brain network spatial features in assessing the severity of perinatal depression.

[0079] First, 13 pregnant women with clinical diagnosis of PD (aged 22-36 years, average 27.76 years) were recruited from the Department of Obstetrics and Gynecology of a hospital in Chengdu, Sichuan Province and divided into the PD group; 20 healthy pregnant women (aged 23-37 years, average 27.15 years) were divided into the healthy pregnancy group (HP group). Before they participated in the experiment, they were evaluated by using the Hamilton 17-item Depression Rating Scale (HAMD) given by the clinician. 17 ) assessed the severity of their depression. All participants had no history of mental illness and were not currently using any psychiatric drugs. EEG data collection and resting-state EEG analysis (including EEG preprocessing, brain network construction, etc.) were performed on 13 pregnant women clinically diagnosed with PD (PD group) and 20 healthy pregnant women (HP group). After constructing the resting-state brain network of all participants, the brain networks of the PD group and the HP group were statistically compared to extract the corresponding brain network spatial features. At the same time, the corresponding network properties, including C, L, Ge, and Le, were calculated and compared between the two groups of subjects to further quantify the differences between the two groups of subjects in brain networks. The results obtained are shown in the figure below. Figure 1 and Figure 2 shown.

[0080] Different brain networks between HP and PD Figure 1 As shown in Figure 3, the functional connectivity (i.e., blue edges) connecting the left temporal lobe and the right frontal lobe was significantly weaker in PD compared with HP. The corresponding network properties, including C, L, Ge, and Le, were also further quantified. Figure 2 As shown, compared with HP, the C, Ge and Le of the PD group were smaller, and L was longer (p<0.05), which quantitatively reflected the weakened information interaction in the brain of patients with perinatal depression.

[0081] Then, the brain network spatial features and network attributes were used as classification and recognition features for the diagnosis of perinatal depression. As shown in Table 1, we first tried Figure 1 The network attributes in the experiment only achieved an accuracy of 51.5%. Secondly, we tried whether the brain network spatial features could help identify PD from HP, and the results showed that it did improve the accuracy of diagnosis, with an accuracy of 87.9%, a sensitivity of 95.0%, and a specificity of 76.9%. Finally, the fusion of brain network spatial features and traditional network attributes further improved the classification accuracy. These results show that the efficiency of classification and identification (diagnosis) based on brain network spatial features is significantly higher than that of using traditional brain network features (network attributes).

[0082] Table 1 Performance of different EEG indices in the diagnosis of perinatal depression

[0083]

[0084] Finally, two resting-state EEG indices (network attributes and brain network spatial characteristics) were used to assess the severity of perinatal depression in each subject. Figure 3 (a) shows HAMD evaluated using network attributes as predictive features 17 Scoring and actual HAMD 17 There was a significant correlation between the scores (r=0.355, p=0.043). Figure 3 (b) shows HAMD using brain network spatial features as predictive feature evaluation 17 Scoring and actual HAMD 17 The scores were significantly correlated (r=0.543, p=0.001), and its evaluation performance was significantly higher than that using network attributes. The comparison of the two results once again verified the ability and superiority of brain network spatial features in quantitatively assessing the severity of perinatal depression in pregnant women.

Claims

1. A perinatal depression diagnosis and assessment system based on brain network, the system comprising: Data acquisition module, preprocessing module, feature extraction module, diagnosis module; The data acquisition module collects EEG signals of women during pregnancy or after childbirth; The preprocessing module preprocesses the data collected by the data acquisition module, including: average reference, filtering, and data segmentation; The specific processing method of the feature extraction module is: Step 1: Construct a resting-state brain network, and then extract the following features based on the brain network: Calculate the clustering coefficient C: Calculate the characteristic path length L: Calculate the global efficiency Ge: Calculate the local efficiency Le: Among them, i, j, h represent the corresponding brain network nodes, d ij represents the weighted shortest path length between nodes i and j, d jh (Ψ i ) represents the weighted shortest path length between node j and node h associated with node i, n is the number of nodes, Ψ represents the set of all nodes in a given network, w plv is the connection weight estimated by PLV, where PLV represents the phase lock time value; Step 2: Calculate the depression score P based on network attributes. P=β0+β1C+β2L+β3Ge+β4Le+ε Among them, β 0、1、2、3、4 represents the regression coefficient of the network attribute, and ε represents the error term; Step 3: A statistical test was performed on the brain networks of the depressed pregnant women group and the healthy pregnant women group to obtain the differential brain networks. The edges of the two groups of subjects with differences greater than the set threshold on the differential brain networks were preliminarily selected as potential spatial feature edges; Step 4: Calculate the correlation between the spatial feature edge obtained in step 3 and the depression score P, set the spatial feature edge with a correlation greater than or equal to the set threshold to "1", and set the other edges to "0" to obtain a logical matrix; perform element-by-element multiplication of the logical matrix and the differential brain network to obtain the modified potential spatial feature edge; calculate the PLV value of the corresponding position of the modified potential spatial feature edge, and the set of all PLV values ​​is defined as the original spatial feature; The diagnosis module uses a support vector machine as a classifier, uses a data acquisition module, a preprocessing module, and a feature extraction module to obtain sample data, and uses the obtained sample data to train a classifier. The input of the classifier is the original spatial feature, and the output is divided into two categories: healthy or depressed; During actual detection, the trained classifier is used to classify the newly received data.

2. A brain network-based perinatal depression diagnosis and evaluation system as claimed in claim 1, characterized in that: The method of first constructing the resting-state brain network in the feature extraction module is: Phase-locked time-value PLV was used to construct the resting-state brain network of all participants; In the formula, w plv is the connection weight estimated by PLV, Δt is the sampling period, N is the number of samples, and For the instantaneous phase corresponding to the given time series x(t) and y(t), PLV is calculated for each participant's preprocessed 5s resting-state EEG signal data segment to obtain a 27×27 adjacency matrix. Then, the 27×27 adjacency matrices calculated for all 5s data segments of the participant are averaged to finally obtain the resting-state weighted adjacency matrix of the subject and express it in the form of a brain network.

3. A brain network-based perinatal depression diagnosis and evaluation system as claimed in claim 1, characterized in that: The original spatial features are also processed as follows; Step 1: Data preprocessing: First, perform standardization to ensure that the data in each dimension has zero mean and unit variance, eliminating the impact of different dimensions and magnitudes; Step 2: Covariance matrix calculation: Calculate the covariance matrix of the standardized data to quantify the linear relationship between dimensions; Step 3: Solving eigenvalues ​​and eigenvectors: Perform eigendecomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​represent the variance of the data in the direction of the corresponding eigenvector, that is, the amount of information. Step 4: Principal component selection: According to the size of the eigenvalue, select the eigenvectors corresponding to the first three largest eigenvalues. These eigenvectors are the required principal components. Step 5: Data projection: Project the original spatial feature data onto the three principal components to obtain a new 3D data set, which is achieved through linear transformation. The specific formula is: New_Data=Original_Data×Eigenvectors (6) Among them, New_Data is the data after dimensionality reduction, Original_Data is the original data, and Eigenvectors is the matrix composed of the first three eigenvectors.

4. A brain network-based perinatal depression diagnosis and evaluation system as claimed in claim 1, characterized in that: The specific method of the preprocessing module is: average reference, 1-45Hz bandpass filtering, using a threshold of ±100μV to automatically exclude 5s long EEG data segments with an absolute amplitude exceeding 100μV in any channel, and 5s long data segmentation.

5. A brain network-based perinatal depression diagnosis and evaluation system as claimed in claim 1, characterized in that: If the classifier determines that the patient is depressed, the depression score P is used to classify the depression.

Citation Information

Cited By

  • Depression state assessment method based on structure function coupling variability

    CN121196551A