Neurogenic bladder personalized health management method and system based on knowledge enhancement

Abstract

The invention discloses a knowledge enhancement-based neurogenic bladder personalized health management method and system, and relates to the technical field of chronic disease management, and the method comprises the steps: obtaining the illness state information of a neurogenic bladder patient, and obtaining text representation and knowledge graph representation according to the preprocessed illness state information; generating disease condition comprehensive information representation of the patient through splicing and fusion; constructing a chronic disease management scheme library of the neurogenic bladder, performing graph processing on the chronic disease management scheme library, constructing a health management scheme recommendation model based on a graph neural network, taking disease condition comprehensive information representation of the patient as model input, and generating a personalized health management scheme of the neurogenic bladder patient; and according to the personalized health management scheme, regularly performing scheme evaluation, and based on an evaluation score, judging the applicability of the personalized health management scheme. According to the method, the condition characterization of the patient is enhanced by means of massive information knowledge of the knowledge graph, and intelligent recommendation and construction of personalized schemes of chronic diseases of neurogenic bladder patients are enriched.

Description

Technical Field

[0001] The present invention relates to the technical field of chronic disease management, and more specifically, to a knowledge-enhanced personalized health management method and system for neurogenic bladder. Background Art

[0002] Bladder and urethral dysfunction caused by damage to the central nervous system or peripheral nerves that control urination is called neurogenic bladder, and urinary incontinence or urinary retention is one of its most common symptoms. Neurogenic bladder is a lower urinary tract dysfunction caused by a disorder in the neural regulatory mechanism. Usually, patients can only be diagnosed with neurogenic bladder if they have neuropathy. Neurogenic bladder usually has different clinical manifestations depending on the degree and location of the neuropathy. In addition, it can cause a variety of long-term complications, the most serious of which are upper urinary tract damage and renal failure.

[0003] In terms of treatment, artificial intelligence can help develop personalized treatment plans. For example, in the field of neurogenic bladder, most patients face chronic diseases that require long-term management, and each patient's condition is different, so it is impossible to cover everyone with a unified plan. Usually, a phased assessment is conducted based on the patient's specific situation, and an individualized treatment plan is developed. Among them, artificial intelligence plays an important role. It can adjust the treatment plan in real time according to the evaluation results, thereby generating a personalized plan suitable for the patient. After the plan is generated by artificial intelligence, it can also actively remind the patient to perform bladder management, such as scheduled urination, delayed urination, pelvic floor muscle training, etc., and achieve good treatment effects through these behavioral intervention methods. Therefore, how to use the massive information knowledge of the knowledge graph to improve the intelligent recommendation of personalized chronic disease plans for patients with neurogenic bladder, and develop a set of professional and detailed nursing plans to improve patient prognosis is an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a personalized health management method and system for neurogenic bladder based on knowledge enhancement, with the aim of improving the intelligent recommendation and construction of personalized chronic disease solutions for patients with neurogenic bladder with the help of the massive information knowledge of the knowledge graph.

[0005] The first aspect of the present invention provides a method for personalized health management of neurogenic bladder based on knowledge enhancement, comprising the following steps:

[0006] Acquire the condition information of the patient with neurogenic bladder, preprocess the condition information, extract descriptive text data as a text representation of the condition information according to the preprocessed condition information, and import the preprocessed condition information into a knowledge graph to generate a knowledge graph representation of the condition information;

[0007] The text representation of the medical condition information and the knowledge graph representation are combined and integrated to generate a comprehensive information representation of the patient's condition;

[0008] A chronic disease management solution library for neurogenic bladder is constructed based on historical chronic disease management data of neurogenic bladder, and the chronic disease management solution library is represented by a graph to construct a heterogeneous graph;

[0009] A health management plan recommendation model is constructed based on a graph neural network to learn and represent the heterogeneous graph, and the comprehensive information representation of the patient's condition is used as the model input to generate a personalized health management plan for patients with neurogenic bladder;

[0010] The personalized health management plan is evaluated, and the applicability of the personalized health management plan to the patient is determined based on the evaluation score.

[0011] In this solution, the condition information of the patient with neurogenic bladder is obtained and the condition information is preprocessed, specifically:

[0012] Obtaining the data of the entire process of medical treatment according to the medical records of patients with neurogenic bladder, standardizing and desensitizing the data of the entire process of medical treatment, and dividing the data of clinical events, diagnostic events, drug events and surgical events according to the processed data of the entire process of medical treatment;

[0013] Characterizing the condition information of patients with neurogenic bladder according to data sets corresponding to events with different attributes, extracting quantifiable data from the data sets, removing significant error values ​​in the quantifiable data through data cleaning, and marking abnormal data based on the cleaned quantifiable data;

[0014] A correlation calculation is performed based on the marked abnormal data to obtain complication information related to the abnormal data, possible complication labels are generated to supplement the condition information, and the supplemented condition information is adaptively preprocessed.

[0015] In this solution, descriptive text data is extracted from the preprocessed disease information as a text representation of the disease information, specifically:

[0016] Initialize the RoBerta pre-trained model, extract the descriptive text data of different attribute events from the pre-processed disease information, import the descriptive text data into the model, and construct a multi-layer bidirectional encoder network through Transformer to encode the descriptive text data;

[0017] The contextual semantic information is integrated through a multi-layer encoder structure, and the word vector representation describing the text data is obtained through the output of the hidden layer. The word vector representation is input into the BiGRU layer for forward processing and reverse processing to obtain the text semantic features;

[0018] A multi-head attention mechanism is introduced to calculate the acquired text semantic features using scaled dot product attention to obtain the importance of the model to the text semantic features at different positions, and the information of each attention head is spliced ​​to obtain the output of the multi-head attention mechanism;

[0019] The output of the multi-head attention mechanism is imported into the feedforward neural network to obtain the medical entity labels corresponding to the different word vector representations in the descriptive text data, and the word vector representation annotated with the label is used as the text representation of the disease information.

[0020] In this solution, the preprocessed disease information is imported into the knowledge graph to generate a knowledge graph representation of the disease information, specifically:

[0021] Obtain structured data of events with different attributes from the preprocessed disease information, access the knowledge graph related to neurogenic bladder, and determine the entities and relationships corresponding to the structured data of events with different attributes in the knowledge graph;

[0022] The TransR model is used to map entities and relations to different semantic spaces. The vector representation of entities and relations is obtained based on the triples in the knowledge graph. For each entity vector, the graph convolutional network is used to obtain the initial node embedding vector.

[0023] Other entity vectors belonging to the same attribute event are regarded as neighbor entities, and the neighbor entities are used for aggregation update to obtain the final embedding representation of each entity vector. Biased random walk sampling is used to obtain the path information containing the patient entity vector, and the knowledge graph representation containing the candidate disease type is obtained based on the path information.

[0024] In this solution, a chronic disease management solution library for neurogenic bladder is constructed based on the historical chronic disease management data of neurogenic bladder, and the chronic disease management solution library is represented by a graph to construct a heterogeneous graph, specifically:

[0025] Obtain historical chronic disease management examples of neurogenic bladder, preprocess the historical chronic disease management examples, and extract disease indicators and disease types in the historical chronic disease management examples;

[0026] Construct a chronic disease management solution library for neurogenic bladder, match the disease indicators, health management solutions and disease types at different stages in each instance, and store them in the chronic disease management solution library;

[0027] In the chronic disease management solution library, the disease type is used to divide the data, and the health management solutions and disease indicators adopted by different disease types are obtained. The disease types, disease indicators and health management solutions are used as nodes, and the historical interaction relationships between the nodes are used for graph representation to construct a heterogeneous graph.

[0028] In this solution, a health management solution recommendation model is constructed based on a graph neural network to learn and represent the heterogeneous graph, and the comprehensive information representation of the patient's condition is used as the model input to generate a personalized health management solution for patients with neurogenic bladder, specifically:

[0029] Construct a health management plan recommendation model based on a graph neural network, determine relevant disease indicator nodes and candidate disease type nodes in the heterogeneous graph according to the comprehensive information representation of the patient's disease, and construct a neighbor matrix based on the disease indicator nodes and the graph structure;

[0030] Graph convolution is used to aggregate node features using a neighbor matrix. After feature aggregation using the mean aggregation method, the aggregated features are embedded and concatenated with the features of the disease indicator node to obtain the embedded representation of the disease indicator node.

[0031] The embedded representation of the disease index node under the disease type neighbor matrix is ​​used to calculate the inner product with the candidate disease type node to obtain the patient's disease type;

[0032] Perform similarity calculation in the chronic disease management program library according to the patient's condition type, select a health management program that meets the similarity standard, obtain the development rate of the condition type in the chronic disease management instance corresponding to the health management program, use the development rate to set the initial weight of the health management program, and select a candidate health management program node according to the initial weight;

[0033] Construct a graph structure to construct a neighbor matrix of the patient's condition type node, introduce a graph attention mechanism to obtain the attention weight of each neighbor node in the neighbor matrix through nonlinear transformation, use the attention weight to perform weighted aggregation to implement node update, and obtain the embedded representation of the patient's condition type node;

[0034] The embedded representation of the patient's condition type node and the representation of the candidate health management plan node are calculated internally, and the corresponding health management plan nodes are selected according to the inner product calculation results to generate a personalized health management plan for patients with neurogenic bladder.

[0035] In this scheme, the personalized health management scheme is evaluated, and the applicability of the personalized health management scheme to the patient is determined based on the evaluation score, specifically:

[0036] Obtain the average change rate of different disease types at different stages within a preset time step in the chronic disease management program library, select health management programs with a rate greater than the average change rate, train the multi-layer perceptron based on the selected health management programs as training samples, and obtain the benchmark value of treatment and nursing effects;

[0037] Import the current patient's personalized health management plan into the trained multi-layer perceptron to obtain the predicted value of the treatment and nursing effect, calculate the residual between the predicted value of the treatment and nursing effect and the benchmark value of the treatment and nursing effect, and use the residual as the evaluation score of the health management plan;

[0038] The average change rate is used as a threshold to obtain the change rate of the evaluation score within a preset time step after the current patient implements the personalized health management plan. When the change rate is greater than the threshold, it is considered that the personalized health management plan is not suitable for the patient.

[0039] The second aspect of the present invention provides a personalized health management system for neurogenic bladder based on knowledge enhancement, including a patient data acquisition module, a RoBerta pre-trained representation module, a knowledge graph pre-trained representation module, a representation fusion module, a chronic disease management program library module, a health management program recommendation module and a health management program evaluation module;

[0040] The patient data acquisition module obtains the condition information of the neurogenic bladder patient according to various condition indicators, and pre-processes the condition information;

[0041] The RoBerta pre-trained representation module uses the RoBerta pre-trained model to obtain the disease text data and generate a text representation of the disease information;

[0042] The knowledge graph pre-training representation module uses the knowledge graph in the field of neurogenic bladder to construct a knowledge graph pre-training representation model to generate a knowledge graph representation of the disease information;

[0043] The representation fusion module fuses the text representation of the disease information and the knowledge graph representation through a splicing operation to generate a knowledge-enhanced comprehensive information representation of the disease;

[0044] The chronic disease management program library module constructs a chronic disease management program library for neurogenic bladder based on historical chronic disease management data of neurogenic bladder and different combinations of disease indicators;

[0045] The health management plan recommendation module represents the chronic disease management plan library in a graph, constructs a heterogeneous graph, learns the heterogeneous graph based on a graph convolutional network, takes the comprehensive information representation of the patient's condition as input, generates a personalized health management plan for patients with neurogenic bladder, and displays it visually;

[0046] The health management program evaluation module performs program evaluation according to the personalized health management program, and determines the applicability of the personalized health management program to the patient based on the evaluation score.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention utilizes knowledge graphs to enhance the patient's condition information, characterize the patient's comprehensive condition representation, improve the accuracy of identifying the condition type of patients with neurogenic bladder, and introduce graph learning concepts to recommend personalized health management plans for patients, solving the problem that the treatment and nursing effects of traditional chronic disease management plans are affected by the subjective factors of doctors.

[0049] The present invention constructs a health management recommendation model for neurogenic bladder based on a graph convolutional neural network to achieve intelligent recommendation of personalized plans. Through scientific chronic disease management plans, the occurrence of complications in patients is reduced, the quality of life is improved, the patients' understanding of neurogenic bladder is enhanced, and their self-care ability is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0051] Figure 1 A flow chart of the personalized health management method for neurogenic bladder based on knowledge enhancement of the present invention is shown;

[0052] Figure 2 A flow chart showing generation of a knowledge graph representation of disease information is shown;

[0053] Figure 3 A flow chart showing generation of a personalized health management plan for a patient with neurogenic bladder is shown;

[0054] Figure 4 A block diagram of the knowledge-enhanced personalized health management system for neurogenic bladder according to the present invention is shown. DETAILED DESCRIPTION

[0055] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0057] Figure 1 A flow chart of the knowledge-enhanced personalized health management method for neurogenic bladder of the present invention is shown.

[0058] like Figure 1 As shown, in the first embodiment of the present invention, a personalized health management method for neurogenic bladder based on knowledge enhancement is provided, comprising:

[0059] S102, obtaining condition information of a patient with neurogenic bladder, preprocessing the condition information, extracting descriptive text data as a text representation of the condition information according to the preprocessed condition information, and importing the preprocessed condition information into a knowledge graph to generate a knowledge graph representation of the condition information;

[0060] S104, splicing and fusing the text representation of the medical condition information and the knowledge graph representation to generate a comprehensive information representation of the patient's medical condition;

[0061] S106, constructing a chronic disease management solution library for neurogenic bladder based on historical chronic disease management data of neurogenic bladder, and graphing the chronic disease management solution library to construct a heterogeneous graph;

[0062] S108, constructing a health management plan recommendation model based on a graph neural network to learn and represent the heterogeneous graph, taking the comprehensive information representation of the patient's condition as a model input, and generating a personalized health management plan for patients with neurogenic bladder;

[0063] S110, evaluating the personalized health management plan, and determining the suitability of the personalized health management plan for the patient based on the evaluation score.

[0064] It should be noted that the data of the whole process of medical treatment is obtained according to the medical record data of patients with neurogenic bladder, including the patient's physical indicators, laboratory test data, etc. The whole process data of medical treatment is standardized and desensitized, and the patient is represented by a numerical ID; the time of medical treatment, examination, discharge, etc. in the whole process data of medical treatment is processed by a desensitization function, and only the patient's medical treatment process is displayed. According to the processed whole process data of medical treatment, the data of clinical events, diagnostic events, drug events and surgical events are divided; according to the data sets corresponding to different attribute events, the condition information of patients with neurogenic bladder is characterized, quantifiable data is extracted from the data sets, and significant error values ​​in the quantifiable data are removed by data cleaning, and abnormal data is marked based on the cleaned quantifiable data, for example, whether the patient's blood pressure data meets the normal blood pressure range is judged according to the patient's blood pressure data, and when it does not meet the normal blood pressure range, abnormal blood pressure data (hypertension or hypotension) is marked. Based on the marked abnormal data, the correlation is calculated in combination with the historical medical record data, and the complication information related to the abnormal data is obtained in the historical medical record data, and the possible complication labels are generated to supplement the condition information, and the supplemented condition information is adaptively preprocessed for model analysis. Preferably, possible complication labels can be obtained through a large language model. By importing abnormal data in the patient's medical record data into the large language model, a comprehensive risk assessment of the patient's condition can be performed, and the patient's possible complication labels can be predicted based on the laws of occurrence and development of complication diseases.

[0065] The RoBERTa model uses dynamic masking technology to process the input text sequence to generate new masking strategies, so as to learn to adapt to different masking strategies and understand different language representations during continuous training. RoBERTa uses byte pair encoding as a new word segmentation method, which increases the size and coverage of the vocabulary. Initialize the RoBerta pre-training model, extract the descriptive text data of different attribute events from the pre-processed disease information, import the descriptive text data into the model, and construct a multi-layer bidirectional encoder network through Transformer to encode the descriptive text data; integrate contextual semantic information through a multi-layer encoder structure, obtain the word vector representation of the descriptive text data through the output of the hidden layer, input the word vector representation into the BiGRU layer for forward and reverse processing, extract sentence features from both forward and reverse perspectives to obtain text semantic features; input the text semantic features obtained by the BiGRU network into the fully connected layer, and introduce the Dropout mechanism to prevent overfitting of the neural network.

[0066] In order to further extract contextual features, emphasize the importance of different words and fine-tune the model, a multi-head attention mechanism is introduced to calculate the acquired text semantic features using scaled dot product attention, and different linear transformations are performed in a single attention head to obtain the importance of the model to the text semantic features at different positions. Its parameters can be learned and calculated to ensure that relevant features are learned from different subspaces. The attention results output by all attention heads are merged and spliced ​​to obtain the output of the multi-head attention mechanism; the output of the multi-head attention mechanism is imported into a feedforward neural network, and the feedforward neural network is used to learn the mapping relationship between labels and word vectors, and the association and dependency between labels are learned, and constraint rules are obtained to correct the matching of abnormal labels, and the medical entity labels corresponding to different word vector representations in the descriptive text data are obtained. The word vector representation with label annotation is used as the text representation of the disease information. The text representation of the disease information reflects the basic disease of patients with neurogenic bladder, and the disease indicators that can clearly represent the disease are determined, and its embedded representation under the disease indicators is obtained.

[0067] Figure 2 A flowchart for generating a knowledge graph representation of disease information is shown.

[0068] According to an embodiment of the present invention, the preprocessed disease information is imported into the knowledge graph to generate a knowledge graph representation of the disease information, specifically:

[0069] S202, obtaining structured data of events with different attributes from the preprocessed disease information, accessing the knowledge graph in the field related to neurogenic bladder, and determining entities and relationships corresponding to the structured data of events with different attributes in the knowledge graph;

[0070] S204, using the TransR model to map entities and relationships to different semantic spaces, obtaining vector representations of entities and relationships based on triples in the knowledge graph, and obtaining an initial node embedding vector for each entity vector using a graph convolutional network;

[0071] S206, taking other entity vectors belonging to the same attribute event as neighbor entities, using neighbor entities for aggregation update to obtain the final embedding representation of each entity vector, using biased random walk sampling to obtain path information containing patient entity vectors, and obtaining a knowledge graph representation containing candidate disease types based on the path information.

[0072] It should be noted that TransR introduces a relation-specific transformation matrix, customizes the representation space for each relation, and more accurately captures the semantic relationship between entities and relations. TransR embeds entities and relations in the knowledge graph related to neurogenic bladder, effectively introduces the rich semantic structure information of the knowledge graph into the recommendation model, and obtains the entity vector representation corresponding to the structured data of different attribute events. At the same time, in order to add the structural information of the knowledge graph related to neurogenic bladder in the entity embedding representation, the convolution operation is iteratively performed through GCN to aggregate the high-order neighbor information of the node, and the structural information of the knowledge graph is integrated into the vector representation of the node. In addition, in the knowledge graph related to neurogenic bladder, the entity vector corresponding to the structured data of different attribute events is randomly selected as the starting entity, and the condition type is used as the terminal entity for biased random walks. Based on the path length threshold, K paths corresponding to the entity vectors of the structured data of all patients' different attribute events are collected, and the terminal entities of the K paths are statistically obtained. Through massive knowledge reasoning, the candidate condition type of the patient is obtained, and the condition type includes pathogenesis and degree of condition. In addition, path sampling is performed according to biased random walks to further verify the patient's possible complication labels and explore the possibility of complications in patients. The text representation of the patient's condition information and the representation based on the knowledge graph are spliced ​​and fused to obtain a comprehensive information representation of the patient's condition. With the help of the massive information knowledge of the knowledge graph, the knowledge of the patient's condition information is enhanced to clarify the preliminary judgment of the patient's condition information.

[0073] It should be noted that historical chronic disease management examples of neurogenic bladder are obtained, the historical chronic disease management examples are preprocessed, and the disease indicators and disease types in the historical chronic disease management examples are extracted; a chronic disease management program library for neurogenic bladder is constructed, and the disease indicators, health management programs and disease types at different stages in each example are matched and stored in the chronic disease management program library; the disease type is used in the chronic disease management program library for data division, and the health management programs and disease indicators used for different disease types are obtained. The disease types, disease indicators and health management programs are used as nodes, and the historical interaction relationship between the nodes is used for graph representation. If there is an interaction between any two nodes of the disease type, disease indicator and health management program, that is, the relationship between the two nodes is 1, and if there is no interaction, it is 0. After the matrix construction is completed, normalization is performed to prevent the occurrence of gradient disappearance and other situations during training; the corresponding heterogeneous graph is generated according to the interaction matrix.

[0074] Figure 3 A flow chart for generating a personalized health management plan for patients with neurogenic bladder is shown.

[0075] According to an embodiment of the present invention, a health management plan recommendation model is constructed based on a graph neural network to learn and represent the heterogeneous graph, and the comprehensive information representation of the patient's condition is used as the model input to generate a personalized health management plan for patients with neurogenic bladder, specifically:

[0076] S302, constructing a health management plan recommendation model based on a graph neural network, determining relevant condition indicator nodes and candidate condition type nodes in the heterogeneous graph according to the comprehensive information representation of the patient's condition, and constructing a neighbor matrix based on the condition indicator nodes and the graph structure;

[0077] S304, using graph convolution to aggregate node features using a neighbor matrix, and after performing feature aggregation using a mean aggregation method, the aggregated features are embedded and concatenated with the features of the disease indicator node to obtain an embedded representation of the disease indicator node;

[0078] S306, performing inner product calculation on the embedded representation of the condition index node under the condition type neighbor matrix and the candidate condition type node to obtain the patient's condition type;

[0079] S308, performing similarity calculation in a chronic disease management program library according to the patient's condition type, selecting a health management program that meets the similarity standard, obtaining a development rate of the condition type in the chronic disease management instance corresponding to the health management program, using the development rate to set an initial weight of the health management program, and selecting a candidate health management program node according to the initial weight;

[0080] S310, constructing a graph structure to construct a neighbor matrix of the patient's condition type node, introducing a graph attention mechanism to obtain the attention weight of each neighbor node in the neighbor matrix through nonlinear transformation, using the attention weight to perform weighted aggregation to implement node update, and obtaining an embedded representation of the patient's condition type node;

[0081] S312, perform an internal machine calculation on the embedded representation of the patient's condition type node and the representation of the candidate health management solution node, select the corresponding health management solution node according to the inner product calculation result, and generate a personalized health management solution for patients with neurogenic bladder.

[0082] It should be noted that the nursing goal of the health management program for patients with neurogenic bladder is to reduce the occurrence of complications such as urinary tract infection and hydronephrosis in patients. Improve the patient's bladder urine storage and urination function, improve the quality of life, enhance the patient's and family's understanding of neurogenic bladder, and improve self-care ability. The nursing treatment in the health management program mainly includes life care, diet conditioning, bladder training, drug therapy, psychological care, and health education. The graph neural network is used to process the graph structure of the heterogeneous graph corresponding to the chronic disease management program library, and to better capture the relationship between the node and the graph structure information by aggregating the information of the neighbor nodes of each node. Preferably, the CTR estimation model is used to calculate the inner product of the embedding representations of different nodes, and the inner product is used to characterize the degree of matching between the two. When the health management plan recommendation model determines the patient's health management plan, the development rate of the disease type under different health management plans in the database is determined, and the initial weight of the health management plan is set using the development rate. When the disease progresses more and more seriously, its initial weight is smaller, indicating that the corresponding health management plan is not good. When the disease progresses stably or improves, the initial weight of the corresponding health management plan is larger. By setting the initial weight to characterize the recommendation degree of different health management plans, candidate health management plans are selected, the efficiency of neighbor aggregation of the candidate graph attention mechanism is improved, and the recommendation accuracy of the model is improved.

[0083] The personalized health management plan for the patient is forwarded to medical staff for confirmation and improvement, and is confirmed item by item with the patient through visualization. When the personalized health management plan for the patient contains content that does not match the disease type or is inapplicable, new types and plans are added based on the feedback from medical staff, and the improved health management plan is updated to the chronic disease management plan library.

[0084] The average change rate of disease types at different stages within a preset time step is obtained from the chronic disease management program library, and health management programs with a rate of change greater than the average rate of change are screened. The multi-layer perceptron is trained based on the screened health management programs as training samples to obtain a benchmark value for treatment and nursing effects; the personalized health management program of the current patient is imported into the trained multi-layer perceptron to obtain a predicted value for treatment and nursing effects, and the residual between the predicted value for treatment and nursing effects and the benchmark value for treatment and nursing effects is calculated, and the residual is used as an evaluation score for the health management program; the average rate of change is used as a threshold to obtain the rate of change of the evaluation score within a preset time step after the current patient implements the personalized health management program; when the rate of change is greater than the threshold, it is deemed that the personalized health management program is not suitable for the patient.

[0085] The second aspect of the present invention provides a knowledge-enhanced personalized health management system for neurogenic bladder 4, comprising a patient data acquisition module 401, a RoBerta pre-trained representation module 402, a knowledge graph pre-trained representation module 403, a representation fusion module 404, a chronic disease management program library module 405, a health management program recommendation module 406 and a health management program evaluation module 407;

[0086] The patient data acquisition module obtains the condition information of the neurogenic bladder patient according to various condition indicators, and pre-processes the condition information;

[0087] The RoBerta pre-trained representation module uses the RoBerta pre-trained model to obtain the disease text data and generate a text representation of the disease information;

[0088] The knowledge graph pre-training representation module uses the knowledge graph in the field of neurogenic bladder to construct a knowledge graph pre-training representation model to generate a knowledge graph representation of the disease information;

[0089] The representation fusion module fuses the text representation of the disease information and the knowledge graph representation through a splicing operation to generate a knowledge-enhanced comprehensive information representation of the disease;

[0090] The chronic disease management program library module constructs a chronic disease management program library for neurogenic bladder based on historical chronic disease management data of neurogenic bladder and different combinations of disease indicators;

[0091] The health management plan recommendation module represents the chronic disease management plan library in a graph, constructs a heterogeneous graph, learns the heterogeneous graph based on a graph convolutional network, takes the comprehensive information representation of the patient's condition as input, generates a personalized health management plan for patients with neurogenic bladder, and displays it visually;

[0092] The health management program evaluation module performs program evaluation according to the personalized health management program, and determines the applicability of the personalized health management program to the patient based on the evaluation score.

[0093] A third embodiment of the present invention provides a computer-readable storage medium, which includes a knowledge-enhanced personalized health management method program for neurogenic bladder. When the knowledge-enhanced personalized health management method program for neurogenic bladder is executed by a processor, the steps of the knowledge-enhanced personalized health management method for neurogenic bladder are implemented.

[0094] In the several embodiments provided in the present application, it should be understood that the disclosed method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0095] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0096] Alternatively, if the above-mentioned integrated module of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0097] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A personalized health management method for neurogenic bladder based on knowledge enhancement, characterized in that: The following steps are involved: Acquire the condition information of the patient with neurogenic bladder, preprocess the condition information, extract descriptive text data as a text representation of the condition information according to the preprocessed condition information, and import the preprocessed condition information into a knowledge graph to generate a knowledge graph representation of the condition information; The text representation of the medical condition information and the knowledge graph representation are combined and integrated to generate a comprehensive information representation of the patient's condition; A chronic disease management solution library for neurogenic bladder is constructed based on historical chronic disease management data of neurogenic bladder, and the chronic disease management solution library is represented by a graph to construct a heterogeneous graph; A health management plan recommendation model is constructed based on a graph neural network to learn and represent the heterogeneous graph, and the comprehensive information representation of the patient's condition is used as the model input to generate a personalized health management plan for patients with neurogenic bladder; The personalized health management plan is evaluated, and the applicability of the personalized health management plan to the patient is determined based on the evaluation score.

2. According to claim 1, a method for personalized health management of neurogenic bladder based on knowledge enhancement is characterized in that: Obtaining the condition information of the patient with neurogenic bladder and preprocessing the condition information, specifically: Obtaining the data of the entire process of medical treatment according to the medical records of patients with neurogenic bladder, standardizing and desensitizing the data of the entire process of medical treatment, and dividing the data of clinical events, diagnostic events, drug events and surgical events according to the processed data of the entire process of medical treatment; Characterizing the condition information of patients with neurogenic bladder according to data sets corresponding to events with different attributes, extracting quantifiable data from the data sets, removing significant error values ​​in the quantifiable data through data cleaning, and marking abnormal data based on the cleaned quantifiable data; A correlation calculation is performed based on the marked abnormal data to obtain complication information related to the abnormal data, possible complication labels are generated to supplement the condition information, and the supplemented condition information is adaptively preprocessed.

3. A method for personalized health management of neurogenic bladder based on knowledge enhancement according to claim 1, characterized in that: Extract descriptive text data from the preprocessed disease information as the text representation of the disease information, specifically: Initialize the RoBerta pre-trained model, extract the descriptive text data of different attribute events from the pre-processed disease information, import the descriptive text data into the model, and construct a multi-layer bidirectional encoder network through Transformer to encode the descriptive text data; The contextual semantic information is integrated through a multi-layer encoder structure, and the word vector representation describing the text data is obtained through the output of the hidden layer. The word vector representation is input into the BiGRU layer for forward processing and reverse processing to obtain the text semantic features; A multi-head attention mechanism is introduced to calculate the acquired text semantic features using scaled dot product attention to obtain the importance of the model to the text semantic features at different positions, and the information of each attention head is spliced ​​to obtain the output of the multi-head attention mechanism; The output of the multi-head attention mechanism is imported into the feedforward neural network to obtain the medical entity labels corresponding to the different word vector representations in the descriptive text data, and the word vector representation annotated with the label is used as the text representation of the disease information.

4. The method for personalized health management of neurogenic bladder based on knowledge enhancement according to claim 1, characterized in that: Import the preprocessed disease information into the knowledge graph to generate a knowledge graph representation of the disease information, specifically: Obtain structured data of events with different attributes from the preprocessed disease information, access the knowledge graph related to neurogenic bladder, and determine the entities and relationships corresponding to the structured data of events with different attributes in the knowledge graph; The TransR model is used to map entities and relations to different semantic spaces. The vector representation of entities and relations is obtained based on the triples in the knowledge graph. For each entity vector, the graph convolutional network is used to obtain the initial node embedding vector. Other entity vectors belonging to the same attribute event are regarded as neighbor entities, and the neighbor entities are used for aggregation update to obtain the final embedding representation of each entity vector. Biased random walk sampling is used to obtain the path information containing the patient entity vector, and the knowledge graph representation containing the candidate disease type is obtained based on the path information.

5. The method for personalized health management of neurogenic bladder based on knowledge enhancement according to claim 1, characterized in that: According to the historical chronic disease management data of neurogenic bladder, a chronic disease management solution library of neurogenic bladder is constructed, and the chronic disease management solution library is represented by a graph to construct a heterogeneous graph, specifically: Obtain historical chronic disease management examples of neurogenic bladder, preprocess the historical chronic disease management examples, and extract disease indicators and disease types in the historical chronic disease management examples; Construct a chronic disease management solution library for neurogenic bladder, match the disease indicators, health management solutions and disease types at different stages in each instance, and store them in the chronic disease management solution library; In the chronic disease management solution library, the disease type is used to divide the data, and the health management solutions and disease indicators adopted by different disease types are obtained. The disease types, disease indicators and health management solutions are used as nodes, and the historical interaction relationships between the nodes are used for graph representation to construct a heterogeneous graph.

6. The method for personalized health management of neurogenic bladder based on knowledge enhancement according to claim 1, characterized in that: A health management plan recommendation model is constructed based on a graph neural network to learn and represent the heterogeneous graph, and the comprehensive information representation of the patient's condition is used as the model input to generate a personalized health management plan for patients with neurogenic bladder, specifically: Construct a health management plan recommendation model based on a graph neural network, determine relevant disease indicator nodes and candidate disease type nodes in the heterogeneous graph according to the comprehensive information representation of the patient's disease, and construct a neighbor matrix based on the disease indicator nodes and the graph structure; Graph convolution is used to aggregate node features using a neighbor matrix. After feature aggregation using the mean aggregation method, the aggregated features are embedded and concatenated with the features of the disease indicator node to obtain the embedded representation of the disease indicator node. The embedded representation of the disease index node under the disease type neighbor matrix is ​​used to calculate the inner product with the candidate disease type node to obtain the patient's disease type; Perform similarity calculation in the chronic disease management program library according to the patient's condition type, select a health management program that meets the similarity standard, obtain the development rate of the condition type in the chronic disease management instance corresponding to the health management program, use the development rate to set the initial weight of the health management program, and select a candidate health management program node according to the initial weight; Construct a graph structure to construct a neighbor matrix of the patient's condition type node, introduce a graph attention mechanism to obtain the attention weight of each neighbor node in the neighbor matrix through nonlinear transformation, use the attention weight to perform weighted aggregation to implement node update, and obtain the embedded representation of the patient's condition type node; The embedded representation of the patient's condition type node and the representation of the candidate health management plan node are calculated internally, and the corresponding health management plan nodes are selected according to the inner product calculation results to generate a personalized health management plan for patients with neurogenic bladder.

7. The method for personalized health management of neurogenic bladder based on knowledge enhancement according to claim 1, characterized in that: The personalized health management plan is evaluated, and the applicability of the personalized health management plan to the patient is determined based on the evaluation score, specifically: Obtain the average change rate of different disease types at different stages within a preset time step in the chronic disease management program library, select health management programs with a rate greater than the average change rate, train the multi-layer perceptron based on the selected health management programs as training samples, and obtain the benchmark value of treatment and nursing effects; Import the current patient's personalized health management plan into the trained multi-layer perceptron to obtain the predicted value of the treatment and nursing effect, calculate the residual between the predicted value of the treatment and nursing effect and the benchmark value of the treatment and nursing effect, and use the residual as the evaluation score of the health management plan; The average change rate is used as a threshold to obtain the change rate of the evaluation score within a preset time step after the current patient implements the personalized health management plan. When the change rate is greater than the threshold, it is considered that the personalized health management plan is not suitable for the patient.

8. A knowledge-enhanced personalized health management system for neurogenic bladder, characterized in that: Implementing the knowledge-enhanced personalized health management method for neurogenic bladder as described in any one of claims 1 to 7, comprising a patient data acquisition module, a RoBerta pre-trained representation module, a knowledge graph pre-trained representation module, a representation fusion module, a chronic disease management program library module, a health management program recommendation module, and a health management program evaluation module; The patient data acquisition module obtains the condition information of the neurogenic bladder patient according to various condition indicators, and pre-processes the condition information; The RoBerta pre-trained representation module uses the RoBerta pre-trained model to obtain the disease text data and generate a text representation of the disease information; The knowledge graph pre-training representation module uses the knowledge graph in the field of neurogenic bladder to construct a knowledge graph pre-training representation model to generate a knowledge graph representation of the disease information; The representation fusion module fuses the text representation of the disease information and the knowledge graph representation through a splicing operation to generate a knowledge-enhanced comprehensive information representation of the disease; The chronic disease management program library module constructs a chronic disease management program library for neurogenic bladder based on historical chronic disease management data of neurogenic bladder and different combinations of disease indicators; The health management plan recommendation module represents the chronic disease management plan library in a graph, constructs a heterogeneous graph, learns the heterogeneous graph based on a graph convolutional network, takes the comprehensive information representation of the patient's condition as input, generates a personalized health management plan for patients with neurogenic bladder, and displays it visually; The health management program evaluation module performs program evaluation according to the personalized health management program, and determines the applicability of the personalized health management program to the patient based on the evaluation score.

Images