Intelligent neurosurgery tumor patient nursing system

Abstract

The invention belongs to the field of intelligent nursing, and particularly relates to an intelligent neurosurgery tumor patient nursing system which comprises a patient monitoring module, a data intelligent analysis module, an early warning notification module, a remote support module, an information management module and a user interface module. Feature extraction and analysis are carried out on the brain image and the sign data of the patient, an intelligent analysis result is obtained, and the symptom change of the patient is found in time; the system creatively adopts a cross-modal multimode graph neural network comparative learning method, graph construction is performed on sign data and brain images of a patient, cross-modal graph features are subjected to fine-grained feature coding in respective modals through a graph neural network, cross-modal feature comparison loss is designed, and the cross-modal feature comparison loss is obtained. The incidence relation between the brain image of the patient and each piece of vital sign data is integrated through a comparative learning method, intelligent prediction analysis of patient symptoms is achieved, an accurate patient symptom early warning analysis result is provided for medical staff, and medical care efficiency and quality are improved.

Description

Technical Field

[0001] The invention belongs to the field of intelligent nursing, and in particular is an intelligent neurosurgery tumor patient nursing system. Background Art

[0002] Neurosurgical tumor patient care is a specialized nursing field that aims to provide comprehensive care and support for patients with nervous system tumors. The main goals of care include alleviating symptoms and monitoring disease changes. Nurses need to regularly monitor patients' condition changes, evaluate treatment effectiveness, and promptly detect and deal with any complications. Under the traditional nursing model, medical staff have limited real-time monitoring and feedback mechanisms for the vital signs and condition of patients with nervous system tumors, and are unable to promptly detect and respond to changes in the patient's condition. Existing computer-assisted monitoring methods are usually limited to single modality data such as brain MRI images or heart rate graphs. There is no way to combine the patient's medical image information with various vital sign information to achieve intelligent analysis of the patient's symptoms. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention creatively proposes an intelligent neurosurgical tumor patient care system. In view of the limited real-time monitoring and feedback mechanism of the vital signs and condition of patients with nervous system tumors by medical staff, and the inability to timely discover and respond to changes in the patient's condition, the present invention extracts and analyzes the patient's brain image and vital sign data through a neural network to obtain intelligent analysis results, which can more timely discover changes in patient symptoms; in view of the problem that the existing computer-assisted monitoring methods are usually limited to a single modality data and cannot combine the patient's medical image information with various vital sign information for symptom analysis, the system of the present invention creatively adopts a cross-modal multi-modal graph neural network comparative learning method, constructs a graph with the patient's vital sign data and brain image, and uses a graph neural network to perform fine-grained feature encoding of cross-modal graph features within each modality, and designs feature contrast loss between cross-modalities, and integrates the correlation between the patient's brain image and various vital sign data by a comparative learning method, so as to realize intelligent prediction and analysis of patient symptoms, provide medical staff with more accurate patient symptom warning analysis results, and improve medical care efficiency and quality.

[0004] The present invention provides an intelligent neurosurgery tumor patient care system, which includes a patient monitoring module, a data intelligent analysis module, an early warning notification module, a remote support module, an information management module and a user interface module;

[0005] The patient monitoring module collects brain images of all neurosurgery tumor patients through brain magnetic resonance imaging equipment, collects patient vital sign data through heart rate, blood pressure, and body temperature monitoring instruments, and sends the brain images and vital sign data to the data intelligent analysis module;

[0006] The data intelligent analysis module adopts a cross-modal multi-modal graph neural network comparative learning method to compare and learn the physical sign data, brain images and corresponding symptoms of all patients, obtain symptom warning analysis information of patients who need care through the comparative learning results, and send the symptom warning analysis information to the warning notification module;

[0007] The warning notification module sets the warning threshold. Once the warning analysis information exceeds the healthy range, the system automatically issues an alarm.

[0008] The remote support module provides a remote medical support function, allowing experts to remotely view patient symptom warning analysis information and provide nursing advice;

[0009] The information management module records the patient's brain images, vital signs data and early warning analysis information for medical staff to review at any time;

[0010] The user interface module provides an intuitive and easy-to-use user interface for medical staff and patients to view data, receive notifications and interact.

[0011] Furthermore, in the data intelligent analysis module, the cross-modal multi-modal graph neural network comparative learning method specifically includes the following steps:

[0012] Step S1: construct triples and multimodal sets, represent the brain image, physical sign data and corresponding symptoms of the i-th patient as a triple Xi={Pi,Ci,Yi}, where P represents the patient's brain image, C represents the patient's physical sign data, and Y represents the patient's corresponding symptom label, and construct the triples of all patients into a multimodal set X={X1,X2,...,Xn};

[0013] Step S2: feature extraction, using an encoder to extract features from the brain images and vital sign data in the multimodal set to obtain image features Fp and data features Fc of all patients;

[0014] Step S3: Graph construction: all image features Fp and data features Fc are used as graph nodes to construct modal graphs Gp and Gc respectively, and the adjacency matrix is ​​constructed by the K-nearest neighbor graph method. and : ;

[0015] In the formula, and Represents two adjacent nodes in the modal graph Gp or Gc, Represents features based on Euclidean distance and The K adjacent index set of Representative features and The connectivity relationship between the corresponding graph nodes, 1 represents connected, and 0 represents disconnected;

[0016] Step S4: Based on the modality-based graph attention coding, the GAT model is used to extract features of the modality graphs Gp and Gc to obtain the attention coding feature vectors of the modality graphs Gp and Gc;

[0017] Feature S5: Feature connection, the attention encoding feature vectors of the modal graphs Gp and Gc are respectively combined with the image feature Fp and the data feature Fc to obtain the graph connection feature and the data connection feature;

[0018] Step S6: Feature activation: The graph connection feature and the data connection feature are processed by a multi-layer perceptron. The results of the two feature processing are element-wise multiplied with the two features before processing to obtain the graph activation feature and the data activation feature: ;

[0019] In the formula, Represents graph connection features and data connection features, represents multi-layer perceptron processing, Element-wise multiplication, Represents graph activation features and data activation features;

[0020] Step S7: cross-modal feature fusion, adding the graph activation feature and the data activation feature at the element level to obtain the cross-modal feature;

[0021] Step S8: Similarity calculation, constructing a similarity matrix for cross-modal features, where the elements in the matrix represent the feature similarity between two patients: S ij = Z i ∙ ( Z j ) T , ∀ i,j ∈ [1,n] ;

[0022] In the formula, Representing patients and The feature similarity between them is the element in the similarity matrix. Representative Cross-modal characteristics of each patient, Representative Cross-modal characteristics of each patient, The transpose of the matrix representing the eigenvectors, represents the total number of patients;

[0023] Step S9: contrast learning and loss optimization, calculating contrast loss through adjacency matrix and similarity matrix, optimizing contrast loss, obtaining the final similarity comparison result of symptoms between patients, predicting the symptoms of patients who need care through the similarity comparison result, and obtaining symptom warning analysis information, specifically including the following steps:

[0024] Step S91: mask matrix design, designing positive mask matrix and negative mask matrix: ; ; ;

[0025] In the formula, is the designed threshold function, Representative Matrix The elements in , Represents the adjacency matrix and , and Represent the positive mask matrix and the negative mask matrix respectively;

[0026] Step S92: Calculation of positive and negative scores: ; ; ;

[0027] In the formula, represents the similarity matrix, and is the positive and negative pair in the calculation of positive and negative scores, that is, the intermediate variable, Represents the symptom label, Represents the preset parameters, and are the elements in the positive and negative pairs, represents a positive fraction, represents a negative score;

[0028] Step S93: Calculation of positive and negative losses: ; ;

[0029] In the formula, It is a very small preset value to prevent the positive and negative scores from being 0 The numerical calculation problem of the function, and Represent positive loss and negative loss respectively;

[0030] Step S94: Add the positive loss and the negative loss to obtain the total loss, and calculate the cross entropy loss;

[0031] Step S95: Comparative loss calculation: ;

[0032] In the formula, represents the cross entropy loss, is the total loss, To adjust the parameters, is the contrast loss;

[0033] Step S96: Optimize the comparison loss to obtain the final similarity comparison result of the symptoms between patients. Based on the similarity comparison result, predict and analyze the symptoms of the patients who need care to obtain symptom warning analysis information.

[0034] The beneficial results achieved by the present invention using the above scheme are as follows:

[0035] (1) The present invention creatively proposes an intelligent neurosurgery tumor patient care system. In view of the limited real-time monitoring and feedback mechanism of the vital signs and condition of patients with nervous system tumors by medical staff, and the inability to timely detect and respond to changes in the patient's condition, the present invention uses a neural network to extract and analyze the features of the patient's brain image and vital sign data, obtains intelligent analysis results, and can detect changes in the patient's symptoms more timely;

[0036] (2) In view of the problem that existing computer-assisted monitoring methods are usually limited to single modality data and cannot combine the patient's medical image information with various vital signs information to perform symptom analysis, the system of the present invention creatively adopts a cross-modal multi-modal graph neural network comparative learning method, which constructs a graph based on the patient's vital sign data and brain image, uses a graph neural network to encode the cross-modal graph features into fine-grained features within each modality, and designs a cross-modal feature contrast loss. The system uses a comparative learning method to integrate the correlation between the patient's brain image and various vital signs data, realizes intelligent predictive analysis of patient symptoms, provides medical staff with more accurate patient symptom warning analysis results, and improves the efficiency and quality of medical care. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A module diagram of an intelligent neurosurgery tumor patient care system provided by the present invention;

[0038] Figure 2Schematic diagram of the process of a cross-modal multi-modal graph neural network comparative learning method.

[0039] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] Example 1, see Figure 1 , the present invention provides an intelligent neurosurgery tumor patient care system, including a patient monitoring module, a data intelligent analysis module, an early warning notification module, a remote support module, an information management module and a user interface module;

[0042] The patient monitoring module collects brain images of all neurosurgery tumor patients through brain magnetic resonance imaging equipment, collects patient vital sign data through heart rate, blood pressure, and body temperature monitoring instruments, and sends the brain images and vital sign data to the data intelligent analysis module;

[0043] The data intelligent analysis module adopts a cross-modal multi-modal graph neural network comparative learning method to compare and learn the physical sign data, brain images and corresponding symptoms of all patients, obtain symptom warning analysis information of patients who need care through the comparative learning results, and send the symptom warning analysis information to the warning notification module;

[0044] The warning notification module sets the warning threshold. Once the warning analysis information exceeds the healthy range, the system automatically issues an alarm.

[0045] The remote support module provides a remote medical support function, allowing experts to remotely view patient symptom warning analysis information and provide nursing advice;

[0046] The information management module records the patient's brain images, vital signs data and early warning analysis information for medical staff to review at any time;

[0047] The user interface module provides an intuitive and easy-to-use user interface for medical staff and patients to view data, receive notifications and interact.

[0048] By executing the above operations, in order to address the problem that medical staff have limited real-time monitoring and feedback mechanisms for the vital signs and condition of patients with nervous system tumors, and are unable to promptly detect and respond to changes in the patient's condition, the present invention uses a neural network to extract and analyze features of the patient's brain images and vital sign data, obtains intelligent analysis results, and can detect changes in patient symptoms more promptly.

[0049] Example 2, see Figure 2 This embodiment is based on the above embodiment. In the data intelligent analysis module, the cross-modal multi-modal graph neural network comparative learning method specifically includes the following steps:

[0050] Step S1: construct triples and multimodal sets, represent the brain image, physical sign data and corresponding symptoms of the i-th patient as a triple Xi={Pi,Ci,Yi}, where P represents the patient's brain image, C represents the patient's physical sign data, and Y represents the patient's corresponding symptom label, and construct the triples of all patients into a multimodal set X={X1,X2,...,Xn};

[0051] Step S2: feature extraction, using an encoder to extract features from the brain images and vital sign data in the multimodal set to obtain image features Fp and data features Fc of all patients;

[0052] Step S3: Graph construction: all image features Fp and data features Fc are used as graph nodes to construct modal graphs Gp and Gc respectively, and the adjacency matrix is ​​constructed by the K-nearest neighbor graph method. and : ;

[0053] In the formula, and Represents two adjacent nodes in the modal graph Gp or Gc, Represents features based on Euclidean distance and The K adjacent index set of Representative features and The connectivity relationship between the corresponding graph nodes, 1 represents connected, and 0 represents disconnected;

[0054] Step S4: Based on the modality-based graph attention coding, the GAT model is used to extract features of the modality graphs Gp and Gc to obtain the attention coding feature vectors of the modality graphs Gp and Gc;

[0055] Feature S5: Feature connection, the attention encoding feature vectors of the modal graphs Gp and Gc are respectively combined with the image feature Fp and the data feature Fc to obtain the graph connection feature and the data connection feature;

[0056] Step S6: feature activation, the graph connection feature and the data connection feature are processed by a multi-layer perceptron, and the processing results of the two features are respectively element-wise multiplied with the two features before processing to obtain the graph activation feature and the data activation feature;

[0057] Step S7: cross-modal feature fusion, adding the graph activation feature and the data activation feature at the element level to obtain the cross-modal feature;

[0058] Step S8: Similarity calculation, constructing a similarity matrix for cross-modal features, where the elements in the matrix represent the feature similarity between two patients: S ij = Z i ∙ ( Z j ) T , ∀ i,j ∈ [1,n] ;

[0059] In the formula, Representing patients and The feature similarity between them is the element in the similarity matrix. Representative Cross-modal characteristics of each patient, Representative Cross-modal characteristics of each patient, The transpose of the matrix representing the eigenvectors, represents the total number of patients;

[0060] Step S9: Contrastive learning and loss optimization. The contrast loss is calculated through the adjacency matrix and the similarity matrix. The final similarity comparison result of the symptoms between patients is obtained by optimizing the contrast loss. The symptoms of the patients who need care are predicted through the similarity comparison result to obtain symptom warning analysis information.

[0061] By performing the above operations, in order to address the problem that existing computer-assisted monitoring methods are usually limited to a single modality data and cannot combine the patient's medical image information with various vital signs information to perform symptom analysis, the system of the present invention creatively adopts a cross-modal multi-modal graph neural network comparative learning method, which constructs a graph with the patient's vital sign data and brain image, uses the graph neural network to encode the cross-modal graph features in each modality, and designs the feature contrast loss between cross-modalities. The comparative learning method is used to integrate the correlation between the patient's brain image and various vital signs data, realize intelligent predictive analysis of patient symptoms, provide medical staff with more accurate patient symptom warning analysis results, and improve the efficiency and quality of medical care.

[0062] Embodiment 3: This embodiment is based on the above embodiment. In step S9, the contrastive learning and loss optimization specifically include the following steps:

[0063] Step S91: mask matrix design, designing positive mask matrix and negative mask matrix: ; ; ;

[0064] In the formula, is the designed threshold function, Representative Matrix The elements in , Represents the adjacency matrix and , and Represent the positive mask matrix and the negative mask matrix respectively;

[0065] Step S92: Calculation of positive and negative scores: ; ; ;

[0066] In the formula, represents the similarity matrix, and is the positive and negative pair in the calculation of positive and negative scores, that is, the intermediate variable, Represents the symptom label, Represents the preset parameters, and are the elements in the positive and negative pairs, represents a positive fraction, represents a negative score;

[0067] Step S93: Calculation of positive and negative losses: ; ;

[0068] In the formula, It is a very small preset value to prevent the positive and negative scores from being 0 The numerical calculation problem of the function, and Represent positive loss and negative loss respectively;

[0069] Step S94: Add the positive loss and the negative loss to obtain the total loss, and calculate the cross entropy loss;

[0070] Step S95: Comparative loss calculation: ;

[0071] In the formula, represents the cross entropy loss, is the total loss, To adjust the parameters, is the contrast loss;

[0072] Step S96: Optimize the comparison loss to obtain the final similarity comparison result of the symptoms between patients. Based on the similarity comparison result, predict and analyze the symptoms of the patients who need care to obtain symptom warning analysis information.

[0073] Embodiment 4: This embodiment is based on the above embodiment. In step S4, the GAT model is used to extract features from the modal graphs Gp and Gc. Specifically, the following steps are included:

[0074] Initialize node features:

[0075] Initialize the feature vector for each node, where the node is all image features Fp and data features Fc;

[0076] Define the attention mechanism:

[0077] Define the attention function used to calculate the relationship weights between nodes, including the weight matrix and activation function;

[0078] Building an attention mechanism:

[0079] Create an attention mechanism for calculating the relationship between nodes, including calculating the similarity score between nodes and applying the softmax function for normalization to obtain the attention coefficient;

[0080] Update node representation:

[0081] Use the calculated attention coefficient to update the node representation, and obtain the new representation of each node by weighted summing the neighbor node features;

[0082] Aggregate neighbor information:

[0083] Aggregate the neighbor node features of a node by weighted summing of neighbor node features to update the representation of the node;

[0084] Nonlinear transformations:

[0085] Apply a nonlinear activation function after each node update to increase the expressiveness and nonlinearity of the model;

[0086] Repeat multiple times:

[0087] The node representation is gradually enriched through multiple rounds of attention mechanism updates.

[0088] Embodiment 5, this embodiment is based on the above embodiment, in step S6, the multi-layer perceptron processing specifically includes the following steps:

[0089] Design the multi-layer perceptron structure:

[0090] Define the structure of MLP, including input layer, 3 hidden layers and output layer. Each hidden layer contains 5 neurons, and each neuron is connected to all neurons in the previous layer.

[0091] Initialize model parameters: Initialize the weight parameters of MLP;

[0092] Choose an activation function:

[0093] The ReLU activation function is chosen to introduce nonlinearity between each hidden layer;

[0094] Forward propagation:

[0095] Passing data through multiple hidden layers, each layer performs weighted summation and activation function calculations until the output layer obtains the final result;

[0096] Backward Propagation:

[0097] Use the back-propagation algorithm to calculate the gradient of the loss function with respect to the model parameters, and then use methods such as gradient descent to update the model parameters to minimize the loss;

[0098] Train the model:

[0099] Iteratively train the model on the training data set and continuously optimize the model parameters;

[0100] Feature extraction:

[0101] In the trained MLP model, the output of the intermediate hidden layer is used as a further fine-grained feature representation.

[0102] Example 6: This example is based on the above example, and the details of the patient symptom labels used for system training are as follows:

[0103] Headache: persistent or severe headache;

[0104] Neurological symptoms: including paresthesia, muscle weakness, ataxia, vision problems, speech disorders, hearing loss, cranial nerve damage;

[0105] Seizures: Epilepsy caused by the tumor's interference with the brain's electrical activity;

[0106] Cognitive and behavioral changes: including memory loss, difficulty concentrating, decreased cognitive function, mood swings, and abnormal behavior;

[0107] Visual problems: Neurosurgical tumors compress or invade the optic nerve, causing blurred vision, visual field defects, and abnormal eye movements;

[0108] Hearing problems: Tumors in the inner ear or near cranial nerves can cause hearing loss, tinnitus, dizziness and other hearing problems;

[0109] Motor dysfunction: When the tumor affects the motor center or motor nerves, it causes symptoms such as muscle weakness and motor coordination disorders;

[0110] Imbalance and gait abnormalities: Tumors that affect balance receptors or cerebellum function may cause imbalance, gait abnormalities, or ataxia.

[0111] Wasting symptoms: long-term chronic pain, loss of appetite, weight loss, fatigue and other wasting symptoms.

[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0113] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0114] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent neurosurgery tumor patient care system, characterized by: It includes patient monitoring module, data intelligent analysis module, early warning notification module, remote support module, information management module and user interface module; The patient monitoring module collects brain images of all neurosurgery tumor patients through brain magnetic resonance imaging equipment, collects patient vital sign data through heart rate, blood pressure, and body temperature monitoring instruments, and sends the brain images and vital sign data to the data intelligent analysis module; The data intelligent analysis module adopts a cross-modal multi-modal graph neural network comparative learning method to compare and learn the physical sign data, brain images and corresponding symptoms of all patients, obtain symptom warning analysis information of patients who need care through the comparative learning results, and send the symptom warning analysis information to the warning notification module; The warning notification module sets the warning threshold. Once the warning analysis information exceeds the healthy range, the system automatically issues an alarm. The remote support module provides remote medical support functions; The information management module records the patient's brain images, vital sign data and early warning analysis information; The user interface module provides a user interface for medical staff and patients to view data and interact with information.

2. The intelligent neurosurgical tumor patient care system according to claim 1 is characterized by: In the data intelligent analysis module, the cross-modal multi-modal graph neural network comparative learning method specifically includes the following steps: Step S1: construct triples and multimodal sets, represent the brain image, physical sign data and corresponding symptoms of the i-th patient as a triple Xi={Pi,Ci,Yi}, where P represents the patient's brain image, C represents the patient's physical sign data, and Y represents the patient's corresponding symptom label, and construct the triples of all patients into a multimodal set X={X1,X2,...,Xn}; Step S2: feature extraction, using an encoder to extract features from the brain images and vital sign data in the multimodal set to obtain image features Fp and data features Fc of all patients; Step S3: Graph construction: All image features Fp and data features Fc are used as graph nodes to construct modal graphs Gp and Gc respectively, and the adjacency matrix is ​​constructed by the K-nearest neighbor graph method. and : ； In the formula, and Represents two adjacent nodes in the modal graph Gp or Gc, Represents features based on Euclidean distance and The K adjacent index set of Representative features and The connectivity relationship between the corresponding graph nodes, 1 represents connected, 0 represents disconnected; Step S4: Based on the modality graph attention coding, the GAT model is used to extract features of the modality graphs Gp and Gc to obtain the attention coding feature vectors of the modality graphs Gp and Gc; Feature S5: Feature connection, the attention encoding feature vectors of the modal graphs Gp and Gc are respectively combined with the image feature Fp and the data feature Fc to obtain the graph connection feature and the data connection feature; Step S6: feature activation, the graph connection feature and the data connection feature are processed by a multi-layer perceptron, and the processing results of the two features are respectively element-wise multiplied with the two features before processing to obtain the graph activation feature and the data activation feature; Step S7: cross-modal feature fusion, adding the graph activation feature and the data activation feature at the element level to obtain the cross-modal feature; Step S8: Similarity calculation, constructing a similarity matrix for cross-modal features, where the elements in the matrix represent the feature similarity between two patients: ； In the formula, Representing patients and The feature similarity between them is the element in the similarity matrix. Representative Cross-modal characteristics of each patient, Representative Cross-modal characteristics of each patient, The transpose of the matrix representing the eigenvectors, represents the total number of patients; Step S9: Contrastive learning and loss optimization. The contrast loss is calculated through the adjacency matrix and the similarity matrix. The final similarity comparison result of the symptoms between patients is obtained by optimizing the contrast loss. The symptoms of the patients who need care are predicted through the similarity comparison result to obtain symptom warning analysis information.

3. The intelligent neurosurgical tumor patient care system according to claim 2 is characterized by: In step S9, the contrastive learning and loss optimization specifically include the following steps: Step S91: mask matrix design, designing positive mask matrix and negative mask matrix: ； ； ； In the formula, is the designed threshold function, Representative Matrix The elements in , Represents the adjacency matrix and , and Represent the positive mask matrix and the negative mask matrix respectively; Step S92: Calculation of positive and negative scores: ； ； ； In the formula, represents the similarity matrix, and is the positive and negative pair in the calculation of positive and negative scores, that is, the intermediate variable, Represents the symptom label, Represents the preset parameters, and are the elements in the positive and negative pairs, represents a positive fraction, represents a negative score; Step S93: Calculation of positive and negative losses: ； ； In the formula, It is a very small preset value to prevent the positive and negative scores from being 0 The numerical calculation problem of the function, and Represent positive loss and negative loss respectively; Step S94: Add the positive loss and the negative loss to obtain the total loss, and calculate the cross entropy loss; Step S95: Comparative loss calculation: ； In the formula, represents the cross entropy loss, is the total loss, To adjust the parameters, is the contrast loss; Step S96: Optimize the comparison loss to obtain the final similarity comparison result of the symptoms between patients. Based on the similarity comparison result, predict and analyze the symptoms of the patients who need care to obtain symptom warning analysis information.

Images