Intelligent identification system for mental diseases

Abstract

The invention provides an intelligent identification system for mental diseases, and the system comprises a data collection device which is used for carrying out data collection; the data analysis device is used for cleaning the collected data and removing noise; the model training device is used for training the analyzed data by using a residual neural network and graph convolutional network combined recognition model, and establishing a mental disease intelligent recognition model; the diagnosis result output device is used for inputting new data into the trained intelligent recognition model, and the model gives a diagnosis result; the follow-up visit and monitoring device is used for regularly monitoring the diagnosed patient and collecting symptom change data of the patient; according to the method, the analyzed data is trained by using a residual neural network and graph convolutional network combined recognition model, an intelligent mental disease recognition model is established, the adaptability and generalization ability of the model on different tasks and data sets are further improved, a more robust and robust recognition model is constructed, and the recognition ability of the model is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recognition technology, and in particular to an intelligent recognition system for mental illness. Background Art

[0002] Mental illness is a complex health problem, the root causes of which involve the intertwined influence of biological, psychological and social environmental factors. These diseases lead to disorders in brain function, and then cause varying degrees of impairments in mental aspects such as cognition, emotion, will and behavior. There are many types of mental illness, covering many types such as autism, depression and paranoia, and the patient population is huge.

[0003] Currently, there are two main diagnostic devices in the field of depression diagnosis technology, but they each have limitations. The first one relies on the collection and analysis of physiological signal data. Although this method can reflect the patient's physiological state from a biological perspective, it often ignores the impact of the psychological and social environment on the patient's psychological state. Therefore, the comprehensiveness and accuracy of the diagnostic results are limited; the other device diagnoses by presenting different types of emotional stimulation videos and collecting eye and facial images of the test subject.

[0004] Based on the above-mentioned existing technical solutions, although the above method takes psychological factors into consideration to a certain extent, it is obviously not enough to judge depression only by facial expressions and eye changes. Therefore, the present invention proposes an intelligent recognition system for mental illness to solve the above problem. Summary of the invention

[0005] The present invention provides an intelligent recognition system for mental illness, which is used to solve the defect of incomplete consideration of the physiological and psychological aspects of patients in the prior art.

[0006] In one aspect, the present invention provides an intelligent identification system for mental illness, comprising:

[0007] Data collection device, used to collect symptom information, behavioral data, medical data of patients, and perform data integration;

[0008] Data analysis device, used to clean and organize various collected data, remove noise and outliers, standardize data to a specific range, and extract data features;

[0009] A model building training device is used to build a model and train a recognition model combining a residual neural network and a graph convolutional network using the analyzed data;

[0010] The diagnostic result output device is used to input new patient data into the trained intelligent recognition model, and the model gives the corresponding mental illness diagnosis result according to the input data characteristics;

[0011] Follow-up and monitoring devices are used to conduct regular follow-up and monitoring of patients after diagnosis, and to collect data on patients' symptom changes and treatment effects.

[0012] The graph convolutional neural network module is used to represent objects and scenes in the image as graph structures, capture the relationship between them, and improve recognition accuracy. The graph convolutional neural network module includes: a graph convolution operator definition unit, a spectral graph convolution filter unit, and a graph convolutional neural network definition unit; the graph convolution operator definition unit is used to preliminarily define the graph convolutional neural network through graph inverse Fourier transform and graph Laplace matrix Z; the spectral graph convolution filter unit is used to further define the graph convolutional neural network by referring to Chebyshev polynomials; the graph convolutional neural network definition unit is used to fully define the propagation rules of the graph convolutional neural network by introducing Chebyshev's first-order approximation and normalization.

[0013] The residual neural network module is used to extract features through multiple convolutional layers, batch normalization layers, and nonlinear activation functions. In the residual neural network, the main path contains two 3x3 convolutional layers, each followed by a batch normalization layer and a ReLU activation function. The stride of the second convolutional layer is 1 to ensure that the size of the output feature map is consistent with the input feature map. Figure 1 Add the input patient data directly to the output of the main path.

[0014] According to an intelligent identification system for mental illness provided by the present invention, the data collection device includes: the patient's symptom information including emotional state, sleep conditions, and behavioral changes; detecting the patient's symptom information through wearable devices; recording the patient's behavioral data in a specific environment through video monitoring equipment; and extracting the patient's medical history, diagnosis results, and treatment plan medical data from the hospital's electronic medical record system.

[0015] On the other hand, the present invention also provides an intelligent identification system for mental illness, wherein the data analysis device comprises:

[0016] Data preprocessing module, used to clean and organize various collected data and remove noise and outliers;

[0017] Data standardization module: used to standardize data from different sources and scales;

[0018] The feature extraction module is used to extract features related to mental illness from the preprocessed data;

[0019] The physiological index processing module is used to calculate the statistical characteristics of the physiological index data, such as mean, variance and peak value.

[0020] On the other hand, the present invention also provides an intelligent recognition system for mental illness, wherein the model graph convolutional neural network module comprises:

[0021] The graph convolution operator definition unit is used to preliminarily define the graph convolutional neural network through the graph inverse Fourier transform and the graph Laplacian matrix Z. The process is as follows:

[0022] Under the spectral graph convolution theory, an undirected graph is represented by the normalized graph Laplacian matrix Z as:

[0023]

[0024] Where A is the adjacency matrix of the graph, and D is the degree matrix of the graph. L is a real symmetric semi-positive definite square matrix, which can be decomposed into:

[0025] L=UΛU T

[0026] Where Λ is the diagonal matrix of eigenvalues ​​and U is the eigenvector. Specifically, the graph Fourier transform on the graph x can be defined as:

[0027]

[0028] The inverse Fourier transform of a graph is defined as:

[0029] x=Ux

[0030] The graph convolution operator is the product of the graph x and the filter in Fourier space, which is defined as:

[0031] x*Gg=U((U T x)⊙(U T x))x

[0032] The filter is represented as: g θ =diag(U T g)g∈R n ;

[0033] Then x*Gg means using filter g to perform spectral convolution on x;

[0034] The filter unit of the spectral graph convolution is used to further define the graph convolutional neural network by referring to the Chebyshev polynomial:

[0035] The spectral convolutional neural network assumes that the filter g θ is a matrix filled with learnable parameters The diagonal matrix of Contains the parameters of the k-th layer, mapping the i-th input feature to the j-th output feature; for the k-th layer, the output of the hidden layer is:

[0036]

[0037] The filters of spectral graph convolution are usually not local. They do not visit only the kth nearest neighbors of each node, which makes the complexity increase with the size of the graph and is not scalable. In order to achieve efficient calculation of node output, only neighbors that are no more than K hops away from the central node should be considered.

[0038]

[0039] Where T i (x) is the Chebyshev polynomial, θ i is the Chebyshev coefficient, is the normalized eigenvalue diagonal matrix, so the graph convolution becomes:

[0040]

[0041] The graph convolutional neural network definition unit is used to fully define the propagation rule of the graph convolutional neural network by introducing Chebyshev's first-order approximation and normalization:

[0042] Also because

[0043]

[0044] Finally, the graph convolution can be:

[0045]

[0046] So there is no need to calculate the eigenvector, it only depends on the node with the largest step from the central node, which means it is spatially localized.

[0047] The graph convolutional neural network model assumes λ max =2 and K = 1 introduce Chebyshev's first-order approximation, and the graph convolution can be written as:

[0048]

[0049] By normalizing, we can prevent overfitting and avoid gradient explosion or disappearance, so the final propagation rule is:

[0050]

[0051] GCN can aggregate the feature representations between neighboring nodes of graph data, learn features from edge weight matrices, and promote the expression of node features from low-level graphs to high-level graphs.

[0052] On the other hand, the present invention also provides an intelligent identification system for mental illness, wherein the diagnosis result output device comprises:

[0053] Input data module: used to input new patient data into the trained intelligent recognition model. The model will give the corresponding mental illness diagnosis results based on the input data characteristics; multiple models are used for integrated diagnosis, combined with the doctor's professional judgment for comprehensive analysis;

[0054] The personalized treatment module is used to provide personalized treatment recommendations to patients based on the diagnosis results.

[0055] The present invention also provides an intelligent identification system for mental illness, wherein the follow-up and monitoring stage device includes a follow-up control module and a doctor-patient reminder module;

[0056] Follow-up control module: After the mental illness intelligent identification system is identified, it will follow up and monitor the patients regularly, and regularly collect data on the patient's symptom changes and treatment effects;

[0057] The doctor-patient reminder module and the intelligent identification system for mental illness can send reminder messages to patients and doctors, urging them to take medication on time and undergo regular checkups, and provide doctors with information on the progress of their treatment.

[0058] The intelligent recognition system for mental illness provided by the present invention solves the defect of incomplete consideration of the physiological and psychological aspects of patients in the prior art by training the analyzed data using a residual neural network and a graph convolutional network combined recognition model, and achieves the following beneficial effects:

[0059] ResNet allows the use of larger learning rates and deeper networks, thereby speeding up convergence. Through deep networks, more complex feature representations can be learned, improving the performance of tasks such as image classification and target detection. ResNet networks can maintain good generalization capabilities even when they are very deep and will not overfit. The network structure of ResNet can be flexibly adjusted to adapt to different tasks and data sets by increasing or decreasing the number of residual blocks.

[0060] By combining two different types of network structures, a more robust and robust recognition model can be built, which can cope with complex and changeable intelligent recognition data and task requirements with ease; in addition, the good generalization ability of the residual neural network combined with the graph convolutional network's ability to process graph data can further improve the model's adaptability and generalization ability on different tasks and data sets. Residual neural networks are good at processing grid-like data and extracting its local features, while graph convolutional networks are good at processing graph-structured data and capturing the dependencies between nodes. Combining the two can fuse different types of feature representations, thereby improving the model's recognition ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1 This is a schematic diagram of the framework structure of an intelligent identification system for mental illness in Example 1 of the present invention;

[0063] Figure 2 This is a flow chart of an intelligent identification system for mental illness in Example 1 of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] like Figure 1-Figure 2 As shown, an intelligent identification system for mental illness provided by an embodiment of the present invention includes a data collection device, a data analysis device, a model building and training device, a diagnosis result output device, and a follow-up and monitoring device.

[0066] The data collection device includes a symptom information collection module, a behavior data collection module, and a medical record integration module.

[0067] The symptom information collection module is used to collect subjective symptom descriptions of patients through questionnaires, including emotional states such as anxiety, depression, mania, etc., cognitive functions (such as memory loss, inattention, etc.), sleep conditions (such as insomnia, dreaminess, drowsiness, etc.), and behavioral changes (such as increased or decreased activity, social withdrawal, etc.). Wearable devices or mobile applications are used to continuously monitor patients' physiological indicators, such as heart rate, blood pressure, body temperature, sleep quality, etc. These physiological data can provide objective physical status information and may be related to the occurrence and development of mental illness.

[0068] The behavioral data collection module is used to record the patient's behavior in a specific environment, such as facial expressions, body movements, and verbal expressions, with the help of video surveillance equipment. By analyzing these behavioral data, some potential symptoms of mental illness can be discovered, such as apathy, slow movements, and repeated speech. Collect patient activity data on social media, such as posted content, interaction frequency, emotional tendencies, etc. Although the accuracy of these data may be affected by many factors, they can be used as auxiliary information to help understand the patient's mental state and social behavior.

[0069] The medical record integration module is used to extract medical information such as the patient's medical history, diagnosis results, treatment plans, etc. from the hospital's electronic medical record system. This information is very important for a comprehensive understanding of the patient's health status and the development of mental illness. Share data with other medical institutions or health management platforms to obtain more patient health data to improve the accuracy and reliability of the intelligent recognition system.

[0070] The data analysis device includes a data preprocessing module, a data standardization module, a feature extraction module, and a physiological index processing module.

[0071] The data preprocessing module is used to clean and organize the collected data to remove noise and outliers. In physiological indicator data, noise may come from interference from the device itself, the influence of the external environment, and distortion during signal transmission. Through filtering, the interference of these noises on the data can be reduced, and the accuracy and reliability of the data can be improved.

[0072] The data normalization module is used to normalize data from different sources and scales so that they have the same mean and standard deviation. This helps improve the performance and stability of machine learning algorithms. For example, physiological indicator data can be normalized to a specific range to facilitate model comparison and analysis.

[0073] The feature extraction module is used to extract features related to mental illness from the preprocessed data. The feature extraction module includes a text preprocessing unit, a keyword extraction unit, and a text vectorization unit.

[0074] The text preprocessing unit is the first step to extract features such as keywords and sentiment tendencies, which mainly includes removing stop words, punctuation marks, numbers, etc., as well as lexicalization (converting words to lowercase, removing punctuation marks, etc.). This step helps reduce noise and improve the accuracy of subsequent processing.

[0075] The keyword extraction unit is used to extract keywords from the document using the statistical information of the words in the document. Usually, the text is preprocessed to obtain a set of candidate words, and then the keywords are obtained from the candidate set by eigenvalue quantification. Commonly used eigenvalue quantification indicators include part of speech, word frequency, inverse document frequency (IDF), relative word frequency, word length, word association information, etc., which are obtained after word segmentation and grammatical analysis. Among the existing keywords, the vast majority are nouns or gerunds, and the number of times the word appears in the text is counted, but the keywords obtained only by relying on word frequency have great uncertainty. For texts with relatively long lengths, this method will have a lot of noise.

[0076] The text vectorization unit is used to evaluate the importance of a word in a document. By calculating the TF-IDF value of each word, words with higher values ​​can be selected as keywords. TF-IDF is a commonly used text vectorization representation method that combines two factors: term frequency (TF) and inverse document frequency (IDF).

[0077] The physiological index processing module includes a mean processing unit, a variance processing unit, and a peak processing unit.

[0078] The mean processing unit is used to calculate the mean of physiological index data and make an overall evaluation of physiological indexes. The mean reflects the central trend of the data. The mean is the average of all values ​​in the data set. For example, the average heart rate, average blood pressure, etc. The mean calculation formula is:

[0079]

[0080] The variance processing unit is used to calculate the variance of physiological index data and evaluate the fluctuation of physiological indexes. Variance is a statistic that measures the degree of data dispersion. It indicates the degree of deviation between each value in the data set and the mean, such as the variability of heart rate and the fluctuation range of blood pressure. The larger the variance, the higher the degree of data dispersion, that is, the greater the difference between data points. The variance calculation formula is:

[0081]

[0082] The peak processing unit is used to calculate the peak value of physiological indicator data and detect the extreme conditions of physiological data. The peak value is the maximum value in the data set, which reflects the extreme conditions of the data. In the physiological indicator data, the peak value may correspond to the highest value of a physiological event, such as the highest value of heart rate during strenuous exercise, the highest value of blood pressure under stress, etc. By monitoring the peak value, we can detect abnormal or extreme conditions in time and take corresponding measures. The peak value calculation formula is:

[0083] P=max(x1,x2,…,x n )

[0084] The model building training device is used to build a model that can process a large amount of mental illness data, and uses the analyzed data to train the recognition model that combines the residual neural network and the graph convolutional network.

[0085] The amount of data that the intelligent recognition system for mental illness needs to process is usually very large to complete the analysis, such as a large number of medical records, patient self-descriptions, family descriptions, doctor observations, etc. Therefore, this embodiment uses a residual neural network (Res-Net) and a graph convolutional network (GCN) to combine the recognition model for training, establish a mental illness intelligent recognition model, and adjust the model parameters during the training process to improve the accuracy and generalization ability of the model.

[0086] The residual neural network module is used to extract data features through multiple convolutional layers, batch normalization layers, and nonlinear activation functions. The residual neural network solves the degradation phenomenon that occurs in the training of deep networks with too many layers: as the number of network layers increases, the accuracy gradually increases until saturation, and then as the number of network layers is increased, the accuracy decreases and degrades rapidly. In the residual neural network, the main path contains two 3x3 convolutional layers, each of which is followed by a batch normalization layer and a ReLU activation function. The stride of the second convolutional layer is 1 to ensure that the size of the output feature map is consistent with the input feature map. Figure 1 Add the input patient data directly to the output of the main path.

[0087] The graph convolutional neural network module is used to represent objects and scenes in the image as a graph structure, capture the relationship between objects and scenes, and improve recognition accuracy. GCN can regard the pixels in the image as nodes, appropriately map the entire image to a new discriminant space, and discriminate the attribute characteristics of the pixels based on the relationship between nodes. Hyperspectral images usually include hundreds of thousands of pixels. Treating all pixels as nodes of a graph structure will result in a huge amount of calculation. The graph convolutional neural network module includes: a graph convolution operator definition unit, a filter unit for spectral graph convolution, and a graph convolutional neural network definition unit.

[0088] The mechanism of neural structure search is applied to the architecture of graph convolutional neural network, and a graph convolutional network based on multi-scale feature search is proposed. The search method is used to seek superpixel features of the optimal scale, which can enhance the learning ability of superpixel features and combine with convolutional neural network to extract pixel-level features, thereby improving the accuracy of hyperspectral image classification.

[0089] The graph convolution operator definition unit is used to preliminarily define the graph convolutional neural network through the graph inverse Fourier transform and the graph Laplacian matrix Z:

[0090] Under the spectral graph convolution theory, an undirected graph is represented by the normalized graph Laplacian matrix Z as:

[0091]

[0092] Where A is the adjacency matrix of the graph, and D is the degree matrix of the graph. L is a real symmetric semi-positive definite square matrix, which can be decomposed into:

[0093] L=UΛU T

[0094] Where Λ is the diagonal matrix of eigenvalues ​​and U is the eigenvector. Specifically, the graph Fourier transform on the graph x can be defined as:

[0095]

[0096] The inverse Fourier transform of a graph is defined as:

[0097] x=Ux

[0098] The graph convolution operator is the product of the graph x and the filter in Fourier space, which is defined as

[0099] x*Gg=U((U T x)⊙(U T x))x

[0100] The filter is represented as: g θ =diag(U T g)g∈R n

[0101] Then x*Gg represents the spectral convolution of x using filter g.

[0102] The filter unit of the spectral graph convolution is used to further define the graph convolutional neural network by referring to the Chebyshev polynomial:

[0103] The spectral convolutional neural network assumes that the filter g θ is a matrix filled with learnable parameters The diagonal matrix of . Contains the parameters of the k-th layer, mapping the i-th input feature to the j-th output feature. For the k-th layer, the output of the hidden layer is:

[0104]

[0105] The filters of spectral graph convolution are usually not local, they do not visit only the kth nearest neighbors of each node, which makes the complexity increase with the size of the graph and is not scalable. In order to achieve efficient calculation of node output, only neighbors that are no more than K hops away from the central node should be considered.

[0106]

[0107] Where T i (x) is the Chebyshev polynomial, θ i is the Chebyshev coefficient, is the normalized eigenvalue diagonal matrix. So the graph convolution becomes:

[0108]

[0109] The graph convolutional neural network definition unit is used to fully define the propagation rules of the graph convolutional neural network by introducing Chebyshev's first-order approximation and normalization:

[0110] Also because

[0111]

[0112] Finally, the graph convolution can be:

[0113]

[0114] So there is no need to calculate the eigenvector, it only depends on the node with the largest step from the central node, which means it is spatially localized.

[0115] By assuming λ max =2 and K = 1 introduce Chebyshev's first-order approximation, and the graph convolution can be written as:

[0116]

[0117] By normalizing, we can prevent overfitting and avoid gradient explosion or disappearance, so the final propagation rule is:

[0118]

[0119] GCN can aggregate the feature representations between neighboring nodes of graph data, learn features from edge weight matrices, and promote the expression of node features from low-level graphs to high-level graphs.

[0120] By introducing shortcut connections into the GCN network structure, a new type of residual graph convolutional network (Res-GCN) was constructed, and the Res-GCN network model was applied to the classification of various brain diseases. The Res-GCN network contains 5 GCN layers, 3 batch normalization (BN) layers and a shortcut connection line. During the model training process, we tried to optimize the length of the feature vector of the nodes in the graph data, ranging from [1000-8000]. Based on the comparison of model classification performance, the node feature vector length finally selected was 5000.

[0121] The role of the BN layer in the Res-GCN network is to normalize the data before inputting into the GCN layer to 0 mean and 1 variance. The addition of the BN layer can accelerate the training and convergence of the entire network. The parameters of the graph convolution kernel in Res-GCN are initialized by uniformly distributed random sampling, and the activation function used in the entire network is ReLU. By adding a dropout mechanism rate of 0.3 to the Res-GCN network and adding a norm to the loss function, the generalization ability of the model is improved, and the problem of overfitting can be overcome to a certain extent. The addition of shortcut connections helps to strengthen the connection between shallow and deep graph data features, thereby improving the representation ability of graph data features and the classification ability of the network. The above structures and mechanisms are all aimed at improving the robustness of the Res-GCN network and its ability to identify three brain diseases, and ultimately achieving the four-class classification of schizophrenia patients, severe depression patients, bipolar disorder patients and normal individuals.

[0122] The diagnostic result output device is used to input new patient data into the trained intelligent recognition model, and the model gives the corresponding mental illness diagnosis result according to the input data characteristics. The diagnostic result output device includes a disease diagnosis module and a treatment suggestion module.

[0123] The disease diagnosis module is used to input new patient data into the trained intelligent recognition model. The model will give the corresponding mental illness diagnosis results based on the input data characteristics. The diagnosis result can be a specific disease type, such as depression, anxiety, schizophrenia, etc., or a disease risk level, such as high risk, medium risk, low risk, etc. In order to improve the accuracy and reliability of diagnosis, multiple models can be used for integrated diagnosis, or combined with the doctor's professional judgment for comprehensive analysis.

[0124] The treatment suggestion module is used to provide patients with personalized treatment suggestions based on the diagnosis results. Treatment suggestions can include aspects such as medication, psychotherapy, and lifestyle adjustments. For example, for patients with depression, antidepressant medication, cognitive behavioral therapy, exercise, and other treatment methods can be recommended.

[0125] The intelligent recognition system for mental illness can also provide doctors with treatment decision support, helping them to develop more scientific and reasonable treatment plans. For example, the system can recommend appropriate drug dosage and treatment time based on the patient's symptom characteristics and treatment response.

[0126] The follow-up and monitoring device is used to conduct regular follow-up and monitoring of patients after diagnosis, and to collect data on patients' symptom changes and treatment effects. The follow-up and monitoring stage device includes a follow-up control module and a doctor-patient reminder module.

[0127] The follow-up control module is used to follow up and monitor patients regularly after the mental illness intelligent identification system has completed the identification. It regularly collects patients' symptom changes and treatment effect data, and through the analysis of these data, timely adjusts the treatment plan to improve the treatment effect.

[0128] The English system of the doctor-patient reminder module sends reminder messages to patients and doctors, urging them to take medicine on time and have regular checkups. This improves patients' treatment compliance. At the same time, the system can also provide doctors with the progress of patients' treatment, making it easier for doctors to conduct remote management and guidance.

[0129] The intelligent recognition system for mental illness provided by the present invention solves the defect of incomplete consideration of the physiological and psychological aspects of patients in the prior art by training the analyzed data using a residual neural network and a graph convolutional network combined recognition model, and achieves the following beneficial effects:

[0130] ResNet allows the use of larger learning rates and deeper networks, thereby speeding up convergence. Through deep networks, more complex feature representations can be learned, improving the performance of tasks such as image classification and target detection. ResNet networks can maintain good generalization capabilities even when they are very deep and will not overfit. The network structure of ResNet can be flexibly adjusted to adapt to different tasks and data sets by increasing or decreasing the number of residual blocks.

[0131] By combining two different types of network structures, a more robust and robust recognition model can be built, which can cope with complex and changeable intelligent recognition data and task requirements with ease; in addition, the good generalization ability of the residual neural network combined with the graph convolutional network's ability to process graph data can further improve the model's adaptability and generalization ability on different tasks and data sets. Residual neural networks are good at processing grid-like data and extracting its local features, while graph convolutional networks are good at processing graph-structured data and capturing the dependencies between nodes. Combining the two can fuse different types of feature representations, thereby improving the model's recognition ability.

[0132] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent identification system for mental illness, characterized in that: include: Data collection device, used to collect symptom information, behavioral data, medical data of patients, and perform data integration; Data analysis device, used to clean and organize various collected data, remove noise and outliers, standardize data to a specific range, and extract data features; A model building training device is used to build a model and train a recognition model combining a residual neural network and a graph convolutional network using the analyzed data; The graph convolutional neural network module is used to represent objects and scenes in the image as graph structures and capture the relationship between objects and scenes; The graph convolutional neural network module includes: a graph convolution operator definition unit, a filter unit of spectral graph convolution, and a graph convolutional neural network definition unit; the graph convolution operator definition unit is used to preliminarily define the graph convolutional neural network through graph inverse Fourier transform and graph Laplace matrix Z; the filter unit of spectral graph convolution is used to further define the graph convolutional neural network by referring to Chebyshev polynomials; the graph convolutional neural network definition unit is used to completely define the propagation rules of the graph convolutional neural network by introducing Chebyshev's first-order approximation and normalization; A residual neural network module for extracting features through multiple convolutional layers, normalization layers, and nonlinear activation functions; The diagnostic result output device is used to input new patient data into the trained intelligent recognition model, and the model gives the corresponding mental illness diagnosis result according to the input data characteristics; Follow-up and monitoring devices are used to conduct regular follow-up and monitoring of patients after diagnosis, and to collect data on patients' symptom changes and treatment effects.

2. The intelligent identification system for mental illness according to claim 1, characterized in that: The patient's symptom information includes emotional state, sleep conditions, and behavioral changes; the patient's symptom information is detected through wearable devices; the patient's behavioral data in a specific environment is recorded through video monitoring equipment; and the patient's medical history, diagnosis results, and treatment plan medical data are extracted from the hospital's electronic medical record system.

3. The intelligent identification system for mental illness according to claim 1, characterized in that: The data analysis device includes: Data preprocessing module, used to clean and organize various collected data and remove noise and outliers; Data standardization module: used to standardize data from different sources and scales; The feature extraction module is used to extract features related to mental illness from the preprocessed data; The physiological index processing module is used to calculate the statistical characteristics of the physiological index data, such as mean, variance and peak value.

4. The intelligent identification system for mental illness according to claim 3, characterized in that: The feature extraction module includes: The text preprocessing unit is the first step to extract features such as keywords and sentiment, including removing stop words, punctuation marks, and numbers; A keyword extraction unit, used to extract keywords of a document using statistical information of words in the document; The text vectorization unit is used to evaluate the importance of a word in a document and select words with higher values ​​as keywords by calculating the TF-IDF value of each word.

5. The intelligent identification system for mental illness according to claim 3, characterized in that: The physiological index processing module includes: A mean processing unit is used to calculate the mean of physiological index data and make an overall evaluation of the physiological index; A variance processing unit is used to calculate the variance of physiological index data and evaluate the fluctuation of physiological indexes; The peak processing unit is used to calculate the peak value of the physiological index data and detect the extreme conditions of the physiological data.

6. The intelligent identification system for mental illness according to claim 1, characterized in that: The graph convolution operator definition unit uses the graph inverse Fourier transform and the graph Laplacian matrix Z to preliminarily define the graph convolutional neural network as follows: An undirected graph is represented by the normalized graph Laplacian matrix Z as: Where A is the adjacency matrix of the graph, D is the degree matrix of the graph, and L is a real symmetric semi-positive definite square matrix, which can be decomposed into: L=UΛU T Among them, Λ is the diagonal matrix of eigenvalues, and U is the eigenvector. Specifically, the graph Fourier transform on the graph x can be defined as: The inverse Fourier transform of a graph is defined as: x=Ux The graph convolution operator is the product of the graph x and the filter in Fourier space, which is defined as: x*Gg=U((U T x)⊙(U T x))x The filter is represented as: g θ =diag(U T g)g∈R n ; Then x*Gg represents the spectral convolution of x using filter g.

7. The intelligent identification system for mental illness according to claim 1, characterized in that: The filter unit of the spectral graph convolution further defines the graph convolutional neural network by referring to the Chebyshev polynomial: The graph convolutional neural network assumes that the filter g θ is a matrix filled with learnable parameters The diagonal matrix of Contains the parameters of the k-th layer, mapping the i-th input feature to the j-th output feature; for the k-th layer, the output of the hidden layer is: The filters of spectral graph convolution are not local. They do not visit only the kth nearest neighbors of each node, but only consider neighbors that are no more than K hops away from the central node. Where T i (x) is the Chebyshev polynomial, θ i is the Chebyshev coefficient, is the normalized eigenvalue diagonal matrix, so the graph convolution becomes: Where x*Gg represents the spectral convolution of x using filter g.

8. The intelligent identification system for mental illness according to claim 1, characterized in that: The graph convolutional neural network definition unit is used to fully define the propagation rule process of the graph convolutional neural network by introducing Chebyshev's first-order approximation and normalization: Define graph convolution as: By assuming λ max =2 and K = 1 introduce Chebyshev's first-order approximation, and the graph convolution is written as: The final propagation rules are: Where D is the degree matrix of the graph, and Z is a real symmetric semi-positive definite square matrix.

9. The intelligent identification system for mental illness according to claim 1, characterized in that: The diagnostic result output device includes: Input data module: used to input new patient data into the trained intelligent recognition model. The model will give the corresponding mental illness diagnosis results based on the input data characteristics; multiple models are used for integrated diagnosis, combined with the doctor's professional judgment for comprehensive analysis; The personalized treatment module is used to provide personalized treatment recommendations to patients based on the diagnosis results.

10. The intelligent identification system for mental illness according to claim 1, characterized in that: The follow-up and monitoring stage device includes a follow-up control module and a doctor-patient reminder module; Follow-up control module: After the mental illness intelligent identification system is identified, it will follow up and monitor the patients regularly, and regularly collect data on the patient's symptom changes and treatment effects; The doctor-patient reminder module and the intelligent identification system for mental illness can send reminder messages to patients and doctors, urging them to take medication on time and undergo regular checkups, and provide doctors with information on the progress of their treatment.

Images