Intelligent sleep disease diagnosis and treatment system based on dynamic knowledge graph

Abstract

The invention discloses an intelligent sleep disease diagnosis and treatment system based on a dynamic mapping knowledge domain, a sleep disorder corpus construction module is used for collecting related data of sleep disorders and constructing an original sleep disorder disease corpus, an entity and relationship extraction module is used for performing entity and relationship extraction on data in the corpus by using a BERT-BiLSTM-CRF model, and an entity and relationship extraction module is used for performing entity and relationship extraction on the data in the corpus. Then, a global knowledge graph and a personalized dynamic knowledge graph are constructed based on corpus data and pathological data, and a sleep disease diagnosis and treatment module achieves intelligent diagnosis and treatment of sleep diseases based on the global knowledge graph and the personalized dynamic knowledge graph. According to the scheme, the global knowledge graph and the personalized dynamic knowledge graph both adopt a multi-level architecture in construction, and the comprehensive weight and the time sequence weight are introduced, so that the association degree in the graph can be adjusted through new data, the association degree, the duration time and the level difference, the flexibility and the dynamism of a diagnosis system are enhanced, and real-time updating can be performed.

Description

Technical Field

[0001] The present invention belongs to the field of medical information, and specifically relates to an intelligent diagnosis and treatment system for sleep disorders based on a dynamic knowledge graph. Background Art

[0002] Sleep is a normal physiological process for people to maintain life activities. It can restore energy and promote physical and mental health. However, according to statistics, more than 300 million people in my country face problems related to sleep disorders. Studies have shown that sleep disorders increase the risk of cardiovascular diseases, neurodegenerative diseases, cognitive disorders and other diseases, and seriously affect people's life, study, work and physical and mental health. Therefore, early detection, early diagnosis and treatment intervention of sleep disorders are very important.

[0003] However, many people do not know enough about sleep disorders and do not realize that they may have sleep disorders, even if they have some clinical manifestations, such as shouting, punching and kicking at night, or because of the inconvenience of seeing a doctor, they have not yet sought professional medical treatment. At the same time, the diagnosis and treatment resources of hospitals are also very tight. Therefore, the development of a knowledge graph that can be applied to the diagnosis of sleep disorders in life can alleviate this problem to a large extent.

[0004] By establishing "entity-relationship-entity" triplets, the knowledge graph can effectively utilize the corresponding disease diagnosis and treatment data, build a sleep disorder diagnosis and reasoning architecture, establish the logical relationship between the clinical characteristics of various sleep disorders, link the clinical characteristics to each other, and finally provide a preliminary diagnosis and clinical recommendations for the corresponding sleep disorders.

[0005] At present, disease diagnosis and treatment systems based on knowledge graphs generally rely more on public medical treatment data sets and lack effective targeting of diseases in a single field. Secondly, in terms of disease prediction, the existing intelligent diagnosis and treatment systems for sleep diseases lack depth and cannot effectively mine the connections between different entities in the disease knowledge graph, making it impossible to establish accurate disease diagnosis and treatment predictions. The ever-changing medical environment also poses new challenges to the dynamic updating of knowledge graphs. Summary of the invention

[0006] In view of the above problems, the object of the present invention is to provide an intelligent diagnosis and treatment system for sleep disorders based on a dynamic knowledge graph.

[0007] The specific technical solution for achieving the purpose of the present invention is:

[0008] An intelligent diagnosis and treatment system for sleep disorders based on a dynamic knowledge graph, including the following modules:

[0009] Sleep disorder corpus construction module: used to collect relevant data on sleep disorders, build an original sleep disorder disease corpus, and preprocess the data in the corpus;

[0010] Entity and relation extraction module: used to extract entities and relations from the data in the corpus using the BERT-BiLSTM-CRF model;

[0011] Global knowledge graph construction module: used to build a global knowledge graph based on the data in the corpus and the results of entity and relationship extraction;

[0012] Personalized dynamic knowledge graph construction module: used to build a personalized dynamic knowledge graph based on the constructed global knowledge graph and perform personalized completion on it;

[0013] Sleep disorder diagnosis and treatment module: Based on the global knowledge graph and personalized dynamic knowledge graph, intelligent diagnosis and treatment of sleep disorders can be realized.

[0014] Furthermore, the data in the original sleep disorder disease corpus are medical data and case data related to sleep disorder diseases, including clinical records, case reports, and medical literature;

[0015] The data types include structured data, semi-structured data and unstructured data;

[0016] The data preprocessing includes cleaning of original data, integration of data from different sources, data standardization, and word segmentation.

[0017] Furthermore, the global knowledge graph is constructed using medical data related to sleep disorders; the entity types involved include: sleep disorder diseases, symptoms and clinical manifestations, diagnostic criteria, examinations and tests, susceptibility factors, sleep disorder-related diseases, preventive measures and treatment recommendations, and complications;

[0018] Divide each entity into multiple levels, including:

[0019] The first layer: disease layer, which represents the main types of sleep disorders;

[0020] The second layer: symptom layer, which characterizes the relevant symptoms and clinical manifestations of sleep disorders;

[0021] The third layer: diagnostic layer, which represents the diagnostic criteria for sleep disorders;

[0022] The fourth layer: the inspection layer, which is a detection method for characterizing sleep disorders;

[0023] The fifth layer: Susceptibility layer, which characterizes the susceptible factors and susceptible complications of sleep disorders;

[0024] The sixth layer: diagnosis and treatment layer, which represents the diagnosis and treatment methods and treatment recommendations for sleep disorders;

[0025] Entities at each level are connected through specific relationships, and a comprehensive weight is assigned to the relationship between each entity. The strength of the relationship between entities is quantified through the comprehensive weight:

[0026] W 综合 =α·P 共现 +β·P 发生

[0027]

[0028]

[0029] Among them, P 共现 represents the co-occurrence probability of entity pairs, N 共现 Represents the number of times the entity pair co-occurs in the original sleep disorder disease corpus, N 全部 represents the number of all entity pairs in the corresponding level; P 发生 Represents the probability of the larger entity in the entity pair, N 因素 It represents the number of entities at the smaller level in the entity pair and the number of related entities at the other level. α and β represent the parameters for adjusting the weight, and α+β=1.

[0030] Furthermore, the association level between entities is determined based on the comprehensive weights between the entities:

[0031] When the comprehensive weight value exceeds a, the association level between entities is considered to be a strong association;

[0032] When the comprehensive weight value is between a and b, the association level between entities is considered to be medium;

[0033] When the comprehensive weight value is lower than b, the association level between entities is considered to be weak;

[0034] When entity A and entity B have a strong or moderate association, and entity B also has a strong or moderate association with entity C at another level, then entity A and entity C are considered to have a new relationship and are assigned a weight value:

[0035]

[0036] Among them, Sim(vec A ,vec C ) is the basic cosine similarity between entities A and C, max(Sim B association) is the maximum similarity between entity B and all its associated entities, W AB is the comprehensive weight between entity A and entity B, W BC is the comprehensive weight between entity B and entity C, δ 层级差is the level difference, indicating the level difference between entity A and entity C, vec A ,vec c is the word vector representation of entity A and entity C, ∥vec A ∥∥vec C ∥ indicates the length of the vector.

[0037] Furthermore, the personalized dynamic knowledge graph is initially constructed using case data of sleep disorders;

[0038] The types of entities involved include: sleep disorder diseases, symptoms and clinical manifestations, diagnostic criteria, examinations and tests, susceptibility factors, sleep disorder-related diseases, preventive measures and treatment recommendations, and complications;

[0039] Divide each entity into multiple levels, including:

[0040] The first layer: disease layer, which represents the main types of sleep disorders;

[0041] The second layer: symptom layer, which characterizes the relevant symptoms and clinical manifestations of sleep disorders;

[0042] The third layer: diagnostic layer, which represents the diagnostic criteria for sleep disorders;

[0043] The fourth layer: the inspection layer, which is a detection method for characterizing sleep disorders;

[0044] The fifth layer: Susceptibility layer, which characterizes the susceptible factors and susceptible complications of sleep disorders;

[0045] The sixth layer: diagnosis and treatment layer, which represents the diagnosis and treatment methods and treatment recommendations for sleep disorders;

[0046] After the personalized dynamic knowledge graph is initially constructed, the entity data corresponding to each entity in the knowledge graph in the global knowledge graph and the associated entity data are added to the personalized dynamic knowledge graph to complete the completion of the personalized dynamic knowledge graph.

[0047] Furthermore, in the process of completing the personalized dynamic knowledge graph, meaningful relationships are constructed between entities at different levels, and the initial comprehensive weight values ​​are retained;

[0048] The relationship types include: association relationships, which are used to describe the association between two entities, including the correlation between symptoms and clinical manifestations and specific sleep disorders, the correlation between diagnostic criteria and specific sleep disorders, the correlation between examinations and tests and specific sleep disorders, the correlation between clinical manifestations and diagnostic criteria, and the correlation between examinations and tests and diagnostic criteria;

[0049] Co-occurrence relationship, which is used to describe the relationship between multiple symptoms or diseases occurring simultaneously in the same patient;

[0050] diagnostic relationships, which describe the relationship between symptoms, test results, and the diagnosis of a specific sleep disorder;

[0051] Risk assessment relationships, which describe how a factor may increase the risk of a sleep disorder, how a lifestyle habit or preventive measure may reduce the risk of a sleep disorder, and the complications or other health problems that a sleep disorder may cause;

[0052] Recommended applicability is used to describe whether specific health education content and treatment recommendations are applicable to a certain sleep disorder.

[0053] Furthermore, the personalized dynamic knowledge graph is updated in real time according to the case data, and records the time when the entity is first introduced into the knowledge graph and the time when it is last updated;

[0054] Introduce temporal weight to update the comprehensive weight of the entity:

[0055] W 综合新 =W 综合原 +(1-W 综合原 )·z·P 时序

[0056]

[0057] Among them, T represents the duration of symptoms or factors in the knowledge graph, λ represents the constant that regulates the growth rate, and δ 层级差 is the level difference, indicating the difference in level depth between two entities; W 综合原 represents the original comprehensive weight of two entities based on the global multi-level knowledge graph, P 时序 represents the timing weight, z represents P 时序 Coefficient, age represents the patient's age, N temporal Indicates the number of entities introduced into the time series, N total represents the total number of entities in the user's personalized dynamic knowledge graph, T r Indicates the patient's recovery time, T t Indicates the duration of treatment for the patient, T i represents the duration of the patient's illness in months, and X represents the diagnostic score value of the acquired case data.

[0058] Furthermore, the global knowledge graph is updated in real time based on the update of the sleep disorder database;

[0059] The personalized dynamic knowledge graph is updated in real time based on the update of the global knowledge graph and the update of case data;

[0060] At the same time, new entities and relationships between entities that appear in the update of the personalized dynamic knowledge graph are fed back to the global knowledge graph.

[0061] Furthermore, in the sleep disorder diagnosis and treatment module:

[0062] First, based on the symptoms, living habits, susceptibility factors and related case data input by the user, a personalized dynamic knowledge graph belonging to the user is constructed using the personalized dynamic knowledge graph construction module. Then, intelligent diagnosis and treatment of sleep disorders are achieved based on the personalized dynamic knowledge graph and the global knowledge graph.

[0063] Furthermore, the sleep disorder diagnosis and treatment module includes a user interface, an event processing module, a logic judgment module, a result generation and treatment suggestion display module, a safety module, and a test module;

[0064] When the diagnosis is completed, the diagnosis and treatment results are displayed based on the sleep disorder diagnosis and treatment module, and charts or graphs are used to display complex data or diagnosis results to enhance the user's understanding and provide further explanations or suggestions;

[0065] The sleep disorder diagnosis and treatment module has good response speed and stability, can handle simultaneous access by a large number of users, and is also scalable to accommodate the increase in users or data volume;

[0066] The sleep disorder diagnosis and treatment module has fault tolerance and compatibility. The fault tolerance is used to ensure business continuity, and the compatibility is used to be compatible with different hardware, operating systems and software environments.

[0067] At the same time, the personal information entered by users will be protected and strictly comply with privacy regulations, allowing users to provide feedback on the diagnostic process and results to continuously improve the system.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] (1) The global knowledge graph and personalized dynamic knowledge graph in this solution are constructed using a multi-level architecture. Entities are divided into multiple levels by category, and each level is associated through specific relationships. The multi-level design ensures the clarity of the knowledge graph structure, facilitates the presentation of the association and causal relationship of disease knowledge layer by layer, and enables the diagnosis and treatment system to accurately locate the diagnosis and treatment path in a complex information network;

[0070] (2) This solution introduces comprehensive weight calculation in graph construction to quantify the strength of the relationship between entities. At the same time, a new cross-level relationship creation mechanism is designed. The new relationship creation mechanism ensures the flexibility of the knowledge graph in actual use. That is, the weight of each relationship in the graph is automatically adjusted according to new data and the degree of association, which enhances the adaptability of the system.

[0071] (3) This solution introduces temporal weights into the personalized dynamic knowledge graph, adjusts the temporal weights according to duration and level differences, and enhances the dynamics of the diagnosis system;

[0072] (4) This solution also designs an update feedback mechanism between the global and personalized maps. When the user's personalized data changes, the global map can also be adjusted in time, bringing higher intelligence performance to the diagnosis and treatment system in terms of information linkage.

[0073] The present invention is further described below in conjunction with specific implementation modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the architecture of the intelligent diagnosis and treatment system for sleep disorders based on a dynamic knowledge graph of the present invention. DETAILED DESCRIPTION

[0075] Example

[0076] The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. The described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment 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.

[0077] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an kind" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0078] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to those of ordinary skill in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be considered as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so that once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0079] Combination Figure 1 , an intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph, including the following modules:

[0080] Sleep disorder corpus construction module: used to collect relevant data on sleep disorders, build an original sleep disorder disease corpus, and preprocess the data in the corpus;

[0081] Specifically, the data in the original sleep disorder disease corpus are medical data and case data related to sleep disorder diseases, including clinical records, case reports, and medical literature;

[0082] The data types include structured data, semi-structured data and unstructured data;

[0083] The data preprocessing includes cleaning of original data, integration of data from different sources, data standardization, and word segmentation. Data from different sources are integrated to ensure data consistency and integrity. Data preprocessing, including data standardization and word segmentation, is performed to construct a high-quality sleep disorder disease corpus covering information such as diseases, symptoms, clinical manifestations, diagnostic criteria, examination tests, patient susceptibility factors, sleep disorder phase complications, and treatment recommendations.

[0084] Entity and relation extraction module: used to extract entities and relations from the data in the corpus using the BERT-BiLSTM-CRF model;

[0085] Entity relationship extraction is an important basic task in building a knowledge graph. First, for a public medical treatment dataset, the BERT-BiLSTM-CRF model is used to extract entities and relationships. BERT is a bidirectional encoder derived from Transformer, which can effectively represent unstructured text data as vectors. In the BERT model, each input word is initially converted into a high-dimensional vector, and this conversion involves three embedding methods. The first is word embedding, which is used to capture the basic meaning of the word; the second is paragraph embedding, which helps to distinguish different parts within a sentence; and the last is position embedding, which is used to record the specific position of each word in the sentence. These three embedding vectors are added together to form the final vector representation of each word.

[0086] When processing these input representations, BERT uses a self-attention mechanism that allows the model to consider all words when encoding a sentence. Self-attention is implemented using the following formula:

[0087]

[0088] Among them, Q, K, and V are query, key, and value matrices, respectively. These matrices are transformed through different weight matrices.k is the dimension of the key vector.

[0089] To further enhance the expressiveness of the model, BERT implements a multi-head attention mechanism, which allows the model to process information in multiple subspaces in parallel and learn the features of input data from different perspectives. The specific implementation is:

[0090] MultiHead(Q,K,V)=Concat(head1,...,head h )W O

[0091]

[0092] in, are different weight matrices used to map different Q, K, V to different representation spaces, W O is the output linear transformation weight matrix.

[0093] The advantage of the BiLSTM structure is that it can capture the contextual information of each word in the sentence at the same time, so it is very effective in processing long-distance dependencies in medical text data. For an input sentence sequence. X = {x1, x2, ..., x n}, each x i Represents the word vector of the i-th word in the sentence.

[0094] The calculation formula of forward LSTM is as follows:

[0095] Forget Gate:

[0096]

[0097] Input Gate:

[0098]

[0099] New memory cell candidate value:

[0100]

[0101] Update memory status:

[0102]

[0103] Output Gate:

[0104]

[0105] Update hidden state:

[0106]

[0107] The calculation formula of reverse LSTM is as follows:

[0108] Forget Gate:

[0109]

[0110] Input Gate:

[0111]

[0112] New memory cell candidate value:

[0113]

[0114] Update memory status:

[0115]

[0116] Output Gate:

[0117]

[0118] Update hidden state:

[0119]

[0120] Among them, f t is the forget gate vector in LSTM, W f is the weight matrix, U f is the weight matrix of the previous hidden state, b f is the bias term, σ is the sigmoid activation function, i t is the input gate vector in LSTM, W i , U i 、b i are the weight matrix and bias term of the input gate, respectively. is a candidate memory cell in LSTM, c t is the core memory cell state in LSTM, o t is the output gate vector in LSTM, h t is the hidden state of the LSTM at the current time step.

[0121] In BiLSTM, represents the input sequence {x1,x2,…,x n}The forward hidden state calculated from left to right, Represents the input sequence {x n ,x n-1 ,…,x1} is the reverse hidden state computed from right to left.

[0122] h t is the final hidden state at the current time step in the BiLSTM:

[0123]

[0124] CRF helps improve the accuracy of named entity recognition tasks by optimizing the score of the entire tag sequence, taking into account the mutual transition probability between tags. The vector input to the CRF layer is the hidden state vector h generated by BiLSTM t The task of the CRF layer is to calculate the global optimal score of the label sequence based on the hidden state vector at each time step and the dependency between the previous and next labels.

[0125] The score s(X,Y) of the CRF layer can be expressed as the sum of the position score and the transition score:

[0126]

[0127] Find the most likely tag sequence by scoring.

[0128] The background of named entity recognition is completed based on the BERT-BiLSTM-CRF model, and relationship recognition is completed by extension. For all entities obtained in the previous step, entity pairs are constructed, and the entity pair representations are input and output into the relationship classifier, and finally the relationship category of the entity pair is output.

[0129] Secondly, the structured and semi-structured case data provided by relevant brain hospitals usually contain clear fields, such as basic patient information, symptoms, diagnosis and treatment methods, etc. They can be directly located and extracted through regular expressions or template matching.

[0130] For data extracted from different sources, it is necessary to ensure that the data format is unified and related entities are aligned.

[0131] Global knowledge graph construction module: used to build a global knowledge graph based on the data in the corpus and the results of entity and relationship extraction:

[0132] Specifically, the global knowledge graph is constructed using medical data related to sleep disorders; the entity types involved include: sleep disorder diseases (accounting for 20% in this embodiment, such as insomnia, sleep apnea syndrome, periodic limb movement disorder, etc.), symptoms and clinical manifestations (accounting for 20% in this embodiment, such as difficulty falling asleep, snoring, early awakening, etc.), diagnostic criteria (accounting for 20% in this embodiment, medical standards and guidelines for diagnosing sleep disorders, such as relevant entries in DSM-5 and ICD-10, etc.), examinations and tests (accounting for 10% in this embodiment, specific methods and tools for diagnosing sleep disorders, including but not limited to polysomnography PSG, portable sleep monitors, mobile phone programs, Questionnaire survey, etc.), susceptible factors (accounting for 7% in this embodiment, various causes and risk factors leading to sleep disorders, including but not limited to population, age, drug side effects, living habits, environmental factors, social and psychological factors, etc.), sleep disorder-related diseases (accounting for 8% in this embodiment, including but not limited to other diseases or health problems closely related to sleep disorders, such as depression, anxiety, Parkinson's disease, etc.), preventive measures and treatment suggestions (accounting for 10% in this embodiment, suggestions and measures for preventing related sleep disorders, such as healthy sleep habits, stress management, and medical advice, etc.), complications (accounting for 5% in this embodiment, other health problems that may be caused by sleep disorders, such as high blood pressure, heart disease, depression, etc.);

[0133] Divide each entity into multiple levels, including:

[0134] The first layer: disease layer, which represents the main types of sleep disorders;

[0135] The second layer: symptom layer, which characterizes the relevant symptoms and clinical manifestations of sleep disorders;

[0136] The third layer: diagnostic layer, which represents the diagnostic criteria for sleep disorders;

[0137] The fourth layer: the inspection layer, which is a detection method for characterizing sleep disorders;

[0138] The fifth layer: Susceptibility layer, which characterizes the susceptible factors and susceptible complications of sleep disorders;

[0139] The sixth layer: diagnosis and treatment layer, which represents the diagnosis and treatment methods and treatment recommendations for sleep disorders;

[0140] Entities at each level are connected through specific relationships, and a comprehensive weight is assigned to the relationship between each entity. The strength of the relationship between entities is quantified through the comprehensive weight:

[0141] W 综合 =α·P 共现 +β·P 发生

[0142]

[0143] The extracted data is presented as entity pairs. The two entity pairs are distributed at different levels. The extracted data is not deduplicated. 共现 Represents the co-occurrence probability of entity pairs, examining the frequency of occurrence of entity pairs, N 共现 Represents the number of times the entity pair co-occurs in the original sleep disorder disease corpus, N 全部 Indicates the number of all entity pairs in the corresponding level. This data indicates the proportion of a specific entity pair in the undeduplicated data.

[0144] P 发生 N represents the probability of the occurrence of the larger-level entity in the entity pair. If we want to represent that the smaller-level entity corresponds to multiple larger-level entities, N 因素 Indicates the number of entities at the smaller level in the entity pair and the number of entities at the other level that are related. This part of the data is deduplicated in the graph. α and β represent the parameters for adjusting the weights. α+β=1. In the current method, the co-occurrence probability is given priority to reflect the commonality of sleep disorder treatment. α=0.6, β=0.4;

[0145] For some relationships, there is no obvious other logical relationship. Only the co-occurrence probability is used as the comprehensive weight representation. The association level between entities is determined according to the comprehensive weight between entities. In this embodiment, the entities are divided into the following three levels:

[0146] When the comprehensive weight value exceeds a, which is set to 0.7 in this embodiment, the association level between the entities is considered to be a strong association;

[0147] When the comprehensive weight value is between a and b, which is set to 0.4-0.7 in this embodiment, the association level between the entities is considered to be medium association;

[0148] When the comprehensive weight value is lower than b, which is set to 0.4 in this embodiment, the association level between entities is considered to be weakly associated;

[0149] At the same time, a new relationship creation mechanism is also constructed. When entity A and entity B have a strong or medium association relationship, and entity B also has a strong or medium association relationship with entity C at another level, then entity A and entity C are considered to have a new relationship. The newly created relationship will not appear at the same level. The new relationship must span different levels to avoid generating redundant relationships at the same level. The new relationship is generally weakly associated and is assigned a weight value:

[0150]

[0151] Among them, Sim(vec C ,vec C) is the basic cosine similarity between entities A and C, max(Sim B association) is the maximum similarity between entity B and all its associated entities, W AB is the comprehensive weight between entity A and entity B, W BC is the comprehensive weight between entity B and entity C, δ 层级差 is the level difference, indicating the level difference between entity A and entity C, vec A ,vec C is the word vector representation of entity A and entity C, ∥vec A ∥∥vec C ∥ indicates the length of the vector;

[0152] Personalized dynamic knowledge graph construction module: used to build a personalized dynamic knowledge graph based on the constructed global knowledge graph and perform personalized completion on it;

[0153] The personalized dynamic knowledge graph is initially constructed using case data of sleep disorders. First, user case data is collected based on case data from relevant brain hospitals, and user susceptibility factor data is collected using relevant consultation systems. The relevant susceptibility factor data include, but are not limited to: user living habits, sleep duration, sleep patterns, environmental factors, work pressure, anxiety and depression, etc. After a series of data are integrated, a personalized dynamic knowledge graph for users is constructed:

[0154] The types of entities involved include: sleep disorder diseases, symptoms and clinical manifestations, diagnostic criteria, examinations and tests, susceptibility factors, sleep disorder-related diseases, preventive measures and treatment recommendations, and complications;

[0155] Divide each entity into multiple levels, including:

[0156] The first layer: disease layer, which represents the main types of sleep disorders;

[0157] The second layer: symptom layer, which characterizes the relevant symptoms and clinical manifestations of sleep disorders;

[0158] The third layer: diagnostic layer, which represents the diagnostic criteria for sleep disorders;

[0159] The fourth layer: the inspection layer, which is a detection method for characterizing sleep disorders;

[0160] The fifth layer: Susceptibility layer, which characterizes the susceptible factors and susceptible complications of sleep disorders;

[0161] The sixth layer: diagnosis and treatment layer, which represents the diagnosis and treatment methods and treatment recommendations for sleep disorders;

[0162] After the personalized dynamic knowledge graph is initially constructed, the entity data corresponding to each entity in the knowledge graph in the global knowledge graph and the associated entity data are added to the personalized dynamic knowledge graph to complete the completion of the personalized dynamic knowledge graph.

[0163] In the process of completing the personalized dynamic knowledge graph, construct meaningful relationships between entities at different levels and retain the initial comprehensive weight values;

[0164] The relationship types include: association relationships, which are used to describe the association between two entities, including the correlation between symptoms and clinical manifestations and specific sleep disorders, the correlation between diagnostic criteria and specific sleep disorders, the correlation between examinations and tests and specific sleep disorders, the correlation between clinical manifestations and diagnostic criteria, and the correlation between examinations and tests and diagnostic criteria;

[0165] Co-occurrence relationship, which is used to describe the relationship between multiple symptoms or diseases occurring simultaneously in the same patient;

[0166] diagnostic relationships, which describe the relationship between symptoms, test results, and the diagnosis of a specific sleep disorder;

[0167] Risk assessment relationships, which describe how a factor may increase the risk of a sleep disorder, how a lifestyle habit or preventive measure may reduce the risk of a sleep disorder, and the complications or other health problems that a sleep disorder may cause;

[0168] Recommended applicability, used to describe specific health education content and treatment recommendations applicable to a certain sleep disorder;

[0169] At the same time, it is necessary to distinguish and identify the completed knowledge graph entity relationships from the confirmed entity relationships known to the user, and give clear labels to the entity relationships known to the user.

[0170] The personalized dynamic knowledge graph is updated in real time according to the case data, and records the time points when the entity is first introduced into the knowledge graph and the last updated, which is used to mark the change of the entity's state in the time dimension and introduce new time series weights. For entities with longer durations such as symptoms or susceptibility factors, their weights should be gradually increased, and the time series weights are introduced to update the comprehensive weights of the entities:

[0171] W 综合新 =W 综合原 +(1-W 综合原 )·z·P 时序

[0172]

[0173] Among them, T represents the duration of symptoms or factors in the knowledge graph, λ represents the constant that regulates the growth rate, and δ 层级差 is the level difference, indicating the difference in level depth between two entities; W 综合原 represents the original comprehensive weight of two entities based on the global multi-level knowledge graph, P 时序 represents the timing weight, z represents P 时序 Coefficient, age represents the patient's age, N temporal Indicates the number of entities introduced into the time series, N total represents the total number of entities in the user's personalized dynamic knowledge graph, T r Indicates the patient's recovery time, T t Indicates the duration of treatment for the patient, T i represents the duration of the patient's illness in months, and X represents the diagnostic score value of the acquired case data.

[0174] According to the user's real-time health status and case data, the comprehensive weight can be dynamically updated in combination with the global knowledge graph to build a personalized dynamic knowledge graph for the user. At the same time, the user's personalized knowledge graph can be improved after entity classification for different newly appearing entities in the user's case. Similarly, a new relationship creation mechanism is constructed. When there is a strong or medium association relationship between two entities A and B, and entity B also has a strong or medium association relationship with entity C at another level, a new relationship is created for entity A and entity C.

[0175] In addition, the global knowledge graph is updated in real time based on the update of the sleep disorder database;

[0176] The personalized dynamic knowledge graph is updated in real time based on the update of the global knowledge graph and the update of case data;

[0177] At the same time, the new entities and relationships between entities that appear in the update of the personalized dynamic knowledge graph are fed back to the global knowledge graph;

[0178] Sleep disorder diagnosis and treatment module: Based on the global knowledge graph and personalized dynamic knowledge graph, it realizes intelligent diagnosis and treatment of sleep disorders;

[0179] In the sleep disorder diagnosis and treatment module:

[0180] The user is guided step by step through questions or options, each question or option is associated with the previous one to form a logical process, receiving symptoms, living habits and susceptibility factors as well as related case data input by the user, and using the personalized dynamic knowledge graph construction module to build a personalized dynamic knowledge graph belonging to the user, and based on the personalized dynamic knowledge graph and the global knowledge graph, intelligent diagnosis and treatment of sleep disorders can be achieved.

[0181] The sleep disease diagnosis and treatment module includes a user interface, an event processing module, a logic judgment module, a result generation and treatment suggestion display module, a safety module, and a test module. The user can use the system to diagnose the disease conveniently and accurately. The system interface has clear navigation, concise layout, and intuitive options to help the user understand the current location and how to return to the main menu or the previous menu, avoiding excessive cluttered information to disturb the user, and avoiding the use of overly professional medical terms, or providing easy-to-understand explanations for professional terms;

[0182] When the diagnosis is completed, the diagnosis and treatment results are displayed based on the sleep disorder diagnosis and treatment module, and charts or graphs are used to display complex data or diagnosis results to enhance the user's understanding and provide further explanations or suggestions;

[0183] The sleep disorder diagnosis and treatment module has good response speed and stability, can handle simultaneous access by a large number of users, and is also scalable to accommodate the increase in users or data volume;

[0184] The sleep disorder diagnosis and treatment module has good fault tolerance to ensure business continuity and is compatible with different hardware, operating systems and software environments;

[0185] At the same time, the personal information entered by users will be protected and strictly comply with privacy regulations, allowing users to provide feedback on the diagnostic process and results to continuously improve the system.

[0186] In this embodiment, a male user, aged 45, recently developed symptoms such as insomnia, frequent awakenings and daytime sleepiness, and had life habits and susceptibility factors such as high-pressure work, staying up late and family history of depression. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph of this solution was used;

[0187] First, in the initially created global knowledge graph, new relationships will be constructed based on the comprehensive weights between different entities to improve the knowledge graph; for example, there is a moderate correlation between "high-pressure work" and "anxiety disorder" (weight 0.62), and "anxiety disorder" and "anxiety insomnia" have a strong correlation (weight 0.78), thus deriving a new relationship that "high-pressure work" may cause "anxiety insomnia". The comprehensive weight of the new relationship will be calculated according to the formula, and the new comprehensive weight will be calculated to be 0.28. This new relationship will be incorporated into the subsequent personalized knowledge graph to further provide a basis for diagnosis.

[0188] The sleep disorder intelligent diagnosis and treatment system receives relevant information provided by the user through the user system, including symptoms, lifestyle habits (such as staying up late, drinking coffee) and susceptibility factors (such as family history of depression, other complications), etc. At the same time, if the user's case data is retrieved, it will be further supplemented and updated;

[0189] The system passes this information to the personalized dynamic knowledge graph construction module to build the user's personalized dynamic knowledge graph. The existing knowledge graph will be matched with the entities in the global knowledge graph, other related entities will be introduced, the comprehensive weights between entities will be calculated, the strength of association between different entities will be determined, and for specific entities, timestamps will be attached to facilitate subsequent updates of timing weights; for example, there is a strong correlation between a user's high-pressure work and insomnia, and anxiety disorder as a possible complication will also be considered in the diagnosis process.

[0190] Based on this information, the sleep disease intelligent diagnosis and treatment system will provide users with intelligent diagnosis by combining the comprehensive weights of different entities, preliminarily judging it as "anxiety insomnia", and recommending corresponding examinations (such as psychological assessment, blood tests, etc.) and treatment plans (such as cognitive behavioral therapy and relaxation training) based on the user's living habits and susceptibility factors. The system also provides personalized health education suggestions, such as adjusting work and rest, reducing caffeine intake, etc., and displays the relationship between symptoms and treatment through charts to help users understand the diagnosis and treatment process.

[0191] When the user uses the system, the system calculates the initial time series weight update factor based on the duration of the user's symptoms (such as 3 months) and the duration of treatment: Assuming that the user's symptoms last for 3 months, the system calculates the time series factor as 0.95 according to the formula. This factor will be applied to the association weight between the symptoms and other related entities (such as anxiety disorder, staying up late, etc.); for the strong association between "anxiety insomnia" and "anxiety disorder", after the time series update, the association weight is adjusted from 0.78 to 0.96, indicating that anxiety disorder occupies a more important position in the diagnosis of user insomnia;

[0192] If the user reports that the treatment is not effective during the treatment, the system will adjust the temporal weights based on the user's feedback and increase the weights between related entities, so that the correlation between anxiety disorders can be maintained or even enhanced. At the same time, if the user changes his or her lifestyle (such as reducing caffeine intake), the system will establish new entity relationships based on the new data and recalculate the temporal weights. Ultimately, the user's personalized dynamic knowledge graph will be continuously adjusted as the treatment progresses and the feedback information changes, and will be synchronized to the global sleep disorder disease knowledge graph in real time.

[0193] User feedback will be used to optimize diagnosis and treatment recommendations. At the same time, the updated data will be fed back into the global knowledge graph to further improve the system's overall knowledge base and achieve continuously improved and refined intelligent diagnosis and treatment services.

[0194] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. An intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph, characterized in that: Includes the following modules: Sleep disorder corpus construction module: used to collect relevant data on sleep disorders, build an original sleep disorder disease corpus, and preprocess the data in the corpus; Entity and relation extraction module; Used to extract entities and relations from the data in the corpus using the BERT-BiLSTM-CRF model; Global knowledge graph construction module: used to build a global knowledge graph based on the data in the corpus and the results of entity and relationship extraction; Personalized dynamic knowledge graph construction module: used to build a personalized dynamic knowledge graph based on the constructed global knowledge graph and perform personalized completion on it; Sleep disorder diagnosis and treatment module: Based on the global knowledge graph and personalized dynamic knowledge graph, intelligent diagnosis and treatment of sleep disorders can be realized.

2. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 1 is characterized in that: The data in the original sleep disorder disease corpus are medical data and case data related to sleep disorder diseases, including clinical records, case reports, and medical literature; The data types include structured data, semi-structured data and unstructured data; The data preprocessing includes cleaning of original data, integration of data from different sources, data standardization, and word segmentation.

3. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 1 is characterized in that: The global knowledge graph is constructed using medical data related to sleep disorders; the entity types involved include: sleep disorder diseases, symptoms and clinical manifestations, diagnostic criteria, examinations and tests, susceptibility factors, sleep disorder-related diseases, preventive measures and treatment recommendations, and complications; Divide each entity into multiple levels, including: The first layer: disease layer, which represents the main types of sleep disorders; The second layer: symptom layer, which characterizes the relevant symptoms and clinical manifestations of sleep disorders; The third layer: diagnostic layer, which represents the diagnostic criteria for sleep disorders; The fourth layer: the inspection layer, which is a detection method for characterizing sleep disorders; The fifth layer: Susceptibility layer, which characterizes the susceptible factors and complications of sleep disorders; The sixth layer: diagnosis and treatment layer, which represents the diagnosis and treatment methods and treatment recommendations for sleep disorders; Entities at each level are connected through specific relationships, and a comprehensive weight is assigned to the relationship between each entity. The strength of the relationship between entities is quantified through the comprehensive weight: W 综合 =α·P 共现 +β·P 发生 Among them, P 共现 represents the co-occurrence probability of entity pairs, N 共现 Represents the number of times the entity pair co-occurs in the original sleep disorder disease corpus, N 全部 represents the number of all entity pairs in the corresponding level; P 发生 Represents the probability of the larger entity in the entity pair, N 因素 It represents the number of entities at the smaller level in the entity pair and the number of related entities at the other level. α and β represent the parameters for adjusting the weight, and α+β=1.

4. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 3 is characterized in that: Determine the association level between entities based on the comprehensive weights between the entities: When the comprehensive weight value exceeds a, the association level between entities is considered to be a strong association; When the comprehensive weight value is between a and b, the association level between entities is considered to be medium; When the comprehensive weight value is lower than b, the association level between entities is considered to be weak; When entity A and entity B have a strong or moderate association, and entity B also has a strong or moderate association with entity C at another level, then entity A and entity C are considered to have a new relationship and are assigned a weight value: Among them, Sim(vec A ,vec C ) is the basic cosine similarity between entities A and C, max(Sim B association) is the maximum similarity between entity B and all its associated entities, W AB is the comprehensive weight between entity A and entity B, W BC is the comprehensive weight between entity B and entity C, δ 层级差 is the level difference, indicating the level difference between entity A and entity C, vec A ,vec C is the word vector representation of entity A and entity C, ∥vec A ∥∥vec C ∥ indicates the length of the vector.

5. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 3 is characterized in that: The personalized dynamic knowledge graph is initially constructed using case data of sleep disorders; The types of entities involved include: sleep disorder diseases, symptoms and clinical manifestations, diagnostic criteria, examinations and tests, susceptibility factors, sleep disorder-related diseases, preventive measures and treatment recommendations, and complications; Divide each entity into multiple levels, including: The first layer: disease layer, which represents the main types of sleep disorders; The second layer: symptom layer, which characterizes the relevant symptoms and clinical manifestations of sleep disorders; The third layer: diagnostic layer, which represents the diagnostic criteria for sleep disorders; The fourth layer: the inspection layer, which is a detection method for characterizing sleep disorders; The fifth layer: Susceptibility layer, which characterizes the susceptible factors and complications of sleep disorders; The sixth layer: diagnosis and treatment layer, which represents the diagnosis and treatment methods and treatment recommendations for sleep disorders; After the personalized dynamic knowledge graph is initially constructed, the entity data corresponding to each entity in the knowledge graph in the global knowledge graph and the associated entity data are added to the personalized dynamic knowledge graph to complete the completion of the personalized dynamic knowledge graph.

6. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 5 is characterized in that: In the process of completing the personalized dynamic knowledge graph, construct meaningful relationships between entities at different levels and retain the initial comprehensive weight values; The relationship types include: association relationships, which are used to describe the association between two entities, including the correlation between symptoms and clinical manifestations and specific sleep disorders, the correlation between diagnostic criteria and specific sleep disorders, the correlation between examinations and tests and specific sleep disorders, the correlation between clinical manifestations and diagnostic criteria, and the correlation between examinations and tests and diagnostic criteria; Co-occurrence relationship, which is used to describe the relationship between multiple symptoms or diseases occurring simultaneously in the same patient; diagnostic relationships, which describe the relationship between symptoms, test results, and the diagnosis of a specific sleep disorder; Risk assessment relationships, which describe how a factor may increase the risk of a sleep disorder, how a lifestyle habit or preventive measure may reduce the risk of a sleep disorder, and the complications or other health problems that a sleep disorder may cause; Recommended applicability is used to describe whether specific health education content and treatment recommendations are applicable to a certain sleep disorder.

7. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 6 is characterized in that: The personalized dynamic knowledge graph is updated in real time according to the case data, and records the time when the entity is first introduced into the knowledge graph and the time when it is last updated; Introduce temporal weight to update the comprehensive weight of the entity: IN 综合新 =In 综合原 +(1-W 综合原 ) with P 时序 Among them, T represents the duration of symptoms or factors in the knowledge graph, λ represents the constant that regulates the growth rate, and δ 层级差 is the level difference, indicating the difference in level depth between two entities; W 综合原 represents the original comprehensive weight of two entities based on the global multi-level knowledge graph, P 时序 represents the timing weight, z represents P 时序 Coefficient, age represents the patient's age, N temporal Indicates the number of entities introduced into the time series, N total represents the total number of entities in the user's personalized dynamic knowledge graph, T r Indicates the patient's recovery time, T t Indicates the duration of treatment for the patient, T i represents the duration of the patient's illness in months, and X represents the diagnostic score value of the acquired case data.

8. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 1 is characterized in that: The global knowledge graph is updated in real time based on the update of the sleep disorder database; The personalized dynamic knowledge graph is updated in real time based on the update of the global knowledge graph and the update of case data; At the same time, new entities and relationships between entities that appear in the update of the personalized dynamic knowledge graph are fed back to the global knowledge graph.

9. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 1 is characterized in that: In the sleep disorder diagnosis and treatment module: First, based on the symptoms, living habits, susceptibility factors and related case data input by the user, a personalized dynamic knowledge graph belonging to the user is constructed using the personalized dynamic knowledge graph construction module. Then, intelligent diagnosis and treatment of sleep disorders are achieved based on the personalized dynamic knowledge graph and the global knowledge graph.

10. The intelligent diagnosis and treatment system for sleep disorders based on dynamic knowledge graph according to claim 9 is characterized in that: The sleep disease diagnosis and treatment module includes a user interface, an event processing module, a logic judgment module, a result generation and treatment suggestion display module, a safety module, and a test module; When the diagnosis is completed, the diagnosis and treatment results are displayed based on the sleep disorder diagnosis and treatment module, and charts or graphs are used to display complex data or diagnosis results to enhance the user's understanding and provide further explanations or suggestions; The sleep disorder diagnosis and treatment module has good response speed and stability, can handle simultaneous access by a large number of users, and is also scalable to accommodate the increase in users or data volume; The sleep disorder diagnosis and treatment module has fault tolerance and compatibility. The fault tolerance is used to ensure business continuity, and the compatibility is used to be compatible with different hardware, operating systems and software environments. At the same time, the personal information entered by users will be protected and strictly comply with privacy regulations, allowing users to provide feedback on the diagnostic process and results to continuously improve the system.

Images