A method and system for generating diagnosis and treatment decisions for critical and severe diseases

Abstract

The present invention relates to the field of medical technology, and specifically discloses a method and system for generating diagnosis and treatment decisions for acute and critical diseases, wherein the method comprises: S1, constructing a knowledge network for acute and critical diseases according to knowledge entities, relationships and attributes; S2, extracting key medical features from historical medical records, and constructing a medical record feature set for each medical record according to the key medical features; S3, calculating the similarity between different medical record feature sets, aggregating similar medical records according to the similarity, and generating a similar medical record set; S4, encoding the medical features through an ERNIE-Health model and performing semantic alignment to obtain encoded medical record features with semantic representation; S5, indexing the encoded medical record features into an engine Elasticsearch, so as to obtain similar historical medical records through similarity retrieval in Elasticsearch.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a method and system for generating diagnosis and treatment decisions for critical and severe diseases. Background Art

[0002] Traditional emergency diagnosis often relies on a single type of data, such as a doctor's physical examination, blood tests and routine imaging examinations, which has limitations. For some diseases, a single type of data may not be enough to provide comprehensive information, resulting in inaccurate or timely diagnosis results, or even misdiagnosis.

[0003] Therefore, there is an urgent need for a method that can provide diagnosis and treatment recommendations based on multiple data sources to support doctors in making decisions. Summary of the invention

[0004] In response to the above problems, the purpose of the present invention is to provide a method and system for generating diagnosis and treatment decisions for critical and emergency diseases. By comprehensively utilizing different types of data, the patient's condition can be more comprehensively assessed, the accuracy and timeliness of diagnosis can be improved, and the patient's quality of life can be better treated.

[0005] The present invention provides a method for generating diagnosis and treatment decisions for critical and severe diseases, comprising:

[0006] S1, obtaining medical data related to critical and emergency diseases, extracting knowledge entities, relationships and attributes in the medical data, and constructing a knowledge network of critical and emergency diseases based on the knowledge entities, relationships and attributes;

[0007] S2, extracting key medical features from historical medical records, and constructing a medical record feature set for each medical record based on the key medical features;

[0008] S3, calculating the similarity between different medical record feature sets, and aggregating similar medical records according to the similarity to generate a similar medical record set; the similar medical record set includes multiple similar medical records and the medical features corresponding to each medical record;

[0009] S4, encoding the medical features and performing semantic alignment through the ERNIE-Health model to obtain coded medical record features with semantic representation; each of the coded medical record features has multiple associated medical records in the same semantic space;

[0010] S5, indexing the encoded medical record features into the engine Elasticsearch, so as to obtain similar historical medical records through similarity retrieval in Elasticsearch.

[0011] In a possible implementation, the S1 further includes:

[0012] Cleaning, deduplication, formatting and normalizing the medical data to obtain normalized medical data;

[0013] The knowledge entities, relationships and attributes in the standardized medical data are extracted, and a knowledge network of critical and emergency diseases is constructed based on the knowledge entities, relationships and attributes.

[0014] In a possible implementation, the S2 further includes:

[0015] Standardizing the key medical features to obtain standardized key medical features;

[0016] A medical record feature set is constructed based on the standardized key medical features.

[0017] In a possible implementation, S3 includes:

[0018] The similarity between different medical record feature sets is calculated by using the Jaccard coefficient of different medical record feature sets.

[0019] In a possible implementation, S3 further includes:

[0020] The Jaccard coefficient of the medical record feature set is calculated according to the following formula:

[0021] ;

[0022] ;

[0023] in, D it For medical records D i In Features t The following medical record feature set, D jt For medical records D i and D j In Features t The following medical record feature set, w p is the first p values, J t For medical records D i and medical records D j In Features t The Jaccard coefficient under f Dit∪Djt ( w p ) is the central value of the medical record featurew p The weight of f Dit ( w p ) and f Djt ( w p ) The first p Value w p In medical records D i Features t The weight or frequency under .

[0024] In a possible implementation, S3 further includes:

[0025] The similarity between medical record feature sets is calculated according to the following formula:

[0026] ;

[0027] when sim ( D i, D j ) is greater than the preset threshold X, the similar case set { D i, D j, s ij};

[0028] in, sim ( D i, D j ) is the medical record D i and medical records D j The similarity of J t For medical records D i and medical records D j In Features t The Jaccard coefficient under n is the feature number, s ij For medical records D i and medical records D j The similarity between .

[0029] In a possible implementation, the S4 further includes:

[0030] Encoding the medical features through the ERNIE-Health model to obtain encoded medical features;

[0031] Calculating the characteristic distance between the medical records in the similar medical record set according to the encoded medical characteristics; the characteristic distance changes dynamically;

[0032] The ERNIE-health model is optimized according to the loss value of the feature distance.

[0033] In a possible implementation, the S4 further includes:

[0034] The characteristic distance between two medical records in a similar medical record set is calculated according to the following formula:

[0035] ;

[0036] in, d ij For medical records D i and medical records D j The characteristic distance between D i is the medical record number, D it For medical records D i In Features t The following feature set, D ite For medical records D i Medical characteristics encoded by the ERNIE-Health model D it , D jte For medical records D j Medical features encoded by the ERNIE-Health model D jt , t is a positive integer, n is the number of medical features in the medical record feature set, and L is the linear transformation.

[0037] In a possible implementation, the S4 further includes: calculating the loss value according to the following loss function:

[0038] ;

[0039] in, μ is a hyperparameter,n is the number of medical features in the medical record feature set, i , j , k is a positive integer, s ij For medical records D i and medical records D j The similarity between d ij The medical record feature set i and medical record feature set j The distance between d ik The medical record feature set i and medical record feature set k The distance between s ik For medical records D i and medical records D k The similarity between loss is a function that measures the difference between the model's predicted value and the actual value.

[0040] The present invention also provides a diagnosis and treatment decision generation system for critical and severe diseases, which is used to implement any of the above-mentioned diagnosis and treatment decision generation methods, including:

[0041] A knowledge network construction module is used to obtain medical data related to critical and emergency diseases, extract knowledge entities, relationships and attributes in the medical data, and construct a knowledge network of critical and emergency diseases based on the knowledge entities, relationships and attributes;

[0042] A medical record feature set construction module, used to extract key medical features from historical medical records, and construct a medical record feature set for each medical record based on the key medical features;

[0043] A similar medical record set generation module is used to calculate the similarity between different medical record feature sets, aggregate similar medical records according to the similarity, and generate a similar medical record set; the similar medical record set includes multiple similar medical records and the medical features corresponding to each medical record;

[0044] A coded medical record feature generation module is used to encode the medical features and perform semantic alignment through the ERNIE-Health model to obtain coded medical record features with semantic representation; each of the coded medical record features has multiple associated medical records in the same semantic space;

[0045] The retrieval engine module is used to index the coded medical record features into the engine Elasticsearch, so as to obtain similar historical medical records through similarity retrieval in Elasticsearch.

[0046] The method and system for generating diagnosis and treatment decisions for critical and severe diseases provided by the present invention collects data from various medical devices and systems, and cleans and standardizes the data to ensure the quality and consistency of the data; integrates data of different modalities together through feature extraction and data fusion models to form a unified feature space; uses historical data to train and optimize the model, inputs real-time patient data into the model, identifies potential critical and severe cases, and generates corresponding early warning signals. Use early warnings to notify medical personnel, and provide corresponding diagnosis and treatment recommendations to support doctors in making decisions; through the comprehensive use of different types of data, it is possible to more comprehensively evaluate the patient's condition, improve the accuracy and timeliness of diagnosis, better treat patients and improve their quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of a method for generating diagnosis and treatment decisions for critical and severe diseases provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following detailed description of the embodiments of the present invention is further described in detail in conjunction with the accompanying drawings and examples. The detailed description of the following embodiments and the accompanying drawings are used to exemplarily illustrate the principles of the present invention, but cannot be used to limit the scope of the present invention, that is, the present invention is not limited to the preferred embodiments described, and the scope of the present invention is defined by the claims.

[0049] In the description of the present invention, it should be noted that, unless otherwise specified, “plurality” means two or more than two; the terms “first”, “second”, etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance; for ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0050] Figure 1 A flow chart of a method for generating diagnosis and treatment decisions for critical illnesses provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides a method for generating diagnosis and treatment decisions for critical and severe diseases, comprising:

[0051] S1, obtain medical data related to critical and emergency diseases, extract knowledge entities, relationships and attributes in the medical data, and build a knowledge network of critical and emergency diseases based on the knowledge entities, relationships and attributes;

[0052] In one example, medical data includes electronic medical records, medical literature, and expert knowledge.

[0053] In a possible implementation, S1 also includes: cleaning, deduplication, formatting and normalizing the medical data to obtain normalized medical data; extracting knowledge entities, relationships and attributes in the normalized medical data, and constructing a knowledge network of critical and severe diseases based on the knowledge entities, relationships and attributes. The present invention can ensure the quality and consistency of data through data preprocessing.

[0054] In one possible implementation, the constructed knowledge network is stored in a graph database, and management and query interfaces are provided to ensure its accessibility and maintainability.

[0055] S2, extract key medical features from historical medical records, and construct a medical record feature set for each medical record based on the key medical features;

[0056] In one example, key medical characteristics include: symptoms, diagnosis, and treatment options.

[0057] In a possible implementation, S2 further includes: standardizing the key medical features to obtain standardized key medical features; and constructing a medical record feature set based on the standardized key medical features. Standardizing the key medical features can ensure the consistency and comparability of different medical record features, and improve the quality and comparison effect of medical record feature data.

[0058] S3, calculating the similarity between different medical record feature sets, and aggregating similar medical records according to the similarity to generate a similar medical record set;

[0059] Among them, the similar medical record set includes multiple similar medical records and the medical characteristics corresponding to each medical record; the similar medical record set has reference value and can assist in diagnosis and treatment decisions.

[0060] The present invention uses a supervised algorithm to learn the similarity function between medical records, and uses this function to determine the similarity between patient information and historical medical records. The most similar part of historical medical records is displayed to medical staff for reference. Patient information is divided into relevant medical record features, such as {gender, age, external causes of injury and poisoning, pathological diagnosis, drug allergy, blood type, medication record, ...} represented as T ={ t 1， t 2，…， t n}, n is the number of features. Finally, the Jaccard coefficients of all features under the two medical records are calculated and their average value is taken as the similarity between the two medical records.

[0061] In one possible implementation, S3 includes: calculating the similarity between different medical record feature sets by using the Jaccard coefficients of different medical record feature sets. The Jaccard coefficient is used to compare the similarity and difference between finite sample sets. The larger the Jaccard coefficient value, the higher the sample similarity. During the training process, the feature distance of similar medical records continues to decrease, while the feature distance of dissimilar medical records continues to increase.

[0062] In a possible implementation, S3 further includes: calculating the Jaccard coefficient of the medical record feature set according to the following formula:

[0063] ;

[0064] ;

[0065] in, D it For medical records D i In Features t The following medical record feature set, D jt For medical records D i and D j In Features t The following medical record feature set, w p is the first p values, J t For medical records D i and medical records D j In Features t The Jaccard coefficient under f Dit∪Djt ( w p ) is the central value of the medical record feature w p The weight of f Dit ( w p ) and f Djt ( w p ) The first p Value w p In medical records D i Features tThe weight or frequency under .

[0066] In a possible implementation, S3 further includes: calculating the similarity between the medical record feature sets according to the following formula:

[0067] ;

[0068] when sim ( D i, D j ) is greater than the preset threshold X, s ij 1, indicating medical record D i and medical records D j Similar, get similar case set { D i, D j, s ij};when sim ( D i, D j ) is less than or equal to the preset threshold value X, s ij 0, indicating medical records D i and medical records D j not similar;

[0069] in, sim ( D i, D j ) is the medical record D i and medical records D j The similarity of J t For medical records D i and medical records D j In Features t The Jaccard coefficient under n is the feature number, s ij For medical records D i and medical records D j The similarity between .

[0070] S4, encode the medical features through the ERNIE-Health model and perform semantic alignment to obtain the coded medical record features with semantic representation;

[0071] Each coded medical record feature has multiple associated medical records in the same semantic space; ensuring that different medical record features are compared in the same semantic space can improve the accuracy of medical record feature comparison and similarity calculation.

[0072] In a possible implementation, S4 also includes: encoding the medical features through the ERNIE-Health model to obtain the encoded medical features; calculating the feature distance between medical records in a similar medical record set based on the encoded medical features; the feature distance changes dynamically; and optimizing the ERNIE-health model based on the loss value of the feature distance.

[0073] The encoded medical features, namely the semantic vector representation, are semantic representations that conform to medical professional vocabulary.

[0074] In a possible implementation, S4 further includes: calculating the characteristic distance between two medical records in the similar medical record set according to the following formula:

[0075] ;

[0076] in, d ij For medical records D i and medical records D j The characteristic distance between D i is the medical record number, D it For medical records D i In Features t The following feature set, D ite For medical records D i Medical characteristics encoded by the ERNIE-Health model D it , D jte For medical records D j Medical features encoded by the ERNIE-Health model D jt , t is a positive integer, n is the number of medical features in the medical record feature set, and L is the linear transformation.

[0077] In a possible implementation, S4 further includes: calculating a loss value according to the following loss function:

[0078] ;

[0079] in, μ is a hyperparameter, n is the number of medical features in the medical record feature set, i , j , k is a positive integer, s ij For medical records D i and medical records D j The similarity between d ij The medical record feature set i and medical record feature set j The distance between d ik The medical record feature set i and medical record feature set k The distance between s ik For medical records D i and medical records D k The similarity between loss is a function that measures the difference between the model's predicted value and the actual value.

[0080] S5, indexing the coded medical record features into the engine Elasticsearch, so as to obtain similar historical medical records through similarity retrieval in Elasticsearch.

[0081] In one possible implementation, similar historical medical records are retrieved and analyzed, and presented to medical staff in a user-friendly manner, providing intuitive diagnosis and treatment references to assist medical staff in making decisions.

[0082] The present invention also provides a diagnosis and treatment decision generation system for critical and severe diseases, which is used to implement any of the above-mentioned diagnosis and treatment decision generation methods, including:

[0083] The knowledge network construction module is used to obtain medical data related to critical and emergency diseases, extract knowledge entities, relationships and attributes in the medical data, and construct a knowledge network of critical and emergency diseases based on the knowledge entities, relationships and attributes;

[0084] A medical record feature set construction module is used to extract key medical features from historical medical records and construct a medical record feature set for each medical record based on the key medical features;

[0085] A similar medical record set generation module is used to calculate the similarity between different medical record feature sets, aggregate similar medical records according to the similarity, and generate a similar medical record set; a similar medical record set includes multiple similar medical records and the medical features corresponding to each medical record;

[0086] The coded medical record feature generation module is used to encode medical features and perform semantic alignment through the ERNIE-Health model to obtain coded medical record features with semantic representation; each coded medical record feature has multiple associated medical records in the same semantic space;

[0087] The retrieval engine module is used to index the coded medical record features into the engine Elasticsearch, so as to obtain similar historical medical records through similarity retrieval in Elasticsearch.

[0088] The method and system for generating diagnosis and treatment decisions for critical and severe diseases provided by the present invention collects data from various medical devices and systems, and cleans and standardizes the data to ensure the quality and consistency of the data; integrates data of different modalities together through feature extraction and data fusion models to form a unified feature space; uses historical data to train and optimize the model, inputs real-time patient data into the model, identifies potential critical and severe cases, and generates corresponding early warning signals. Use early warnings to notify medical personnel, and provide corresponding diagnosis and treatment recommendations to support doctors in making decisions; through the comprehensive use of different types of data, it is possible to more comprehensively evaluate the patient's condition, improve the accuracy and timeliness of diagnosis, better treat patients and improve their quality of life.

[0089] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for generating diagnosis and treatment decisions for critical and severe diseases, characterized in that: include: S1, obtaining medical data related to critical and emergency diseases, extracting knowledge entities, relationships and attributes in the medical data, and constructing a knowledge network of critical and emergency diseases based on the knowledge entities, relationships and attributes; S2, extracting key medical features from historical medical records, and constructing a medical record feature set for each medical record based on the key medical features; S3, calculating the similarity between different medical record feature sets, and aggregating similar medical records according to the similarity to generate a similar medical record set; the similar medical record set includes multiple similar medical records and the medical features corresponding to each medical record; S4, encoding the medical features and performing semantic alignment through the ERNIE-Health model to obtain coded medical record features with semantic representation; each of the coded medical record features has multiple associated medical records in the same semantic space; S5, indexing the encoded medical record features into the engine Elasticsearch, so as to obtain similar historical medical records through similarity retrieval in Elasticsearch; The S4 further comprises: Encoding the medical features through the ERNIE-Health model to obtain encoded medical features; Calculating the characteristic distance between the medical records in the similar medical record set according to the encoded medical characteristics; the characteristic distance changes dynamically; Optimizing the ERNIE-health model according to the loss value of the feature distance; The characteristic distance between two medical records in a similar medical record set is calculated according to the following formula: Among them, d ij For medical record D i and medical record D j The characteristic distance between i is the medical record number, D it For medical record D i The feature set under feature t, D ite For medical record D i Medical features D encoded by the ERNIE-Health model it , D jte For medical record D j The medical features D encoded by the ERNIE-Health model jt , t is a positive integer, n is the number of medical features in the medical record feature set, and L is a linear transformation; The loss value is calculated according to the following loss function: Among them, μ is a hyperparameter, n is the number of medical features in the medical record feature set, i, j, k are positive integers, and s ij For medical record D i and medical record D j The similarity between ij is the distance between medical record feature set i and medical record feature set j, d ik is the distance between medical record feature set i and medical record feature set k, s ik For medical record D i and medical record D k The similarity between them is , and loss is a function that measures the difference between the model prediction value and the actual value.

2. The method for generating diagnosis and treatment decisions according to claim 1, characterized in that: The S1 further comprises: Cleaning, deduplication, formatting and normalizing the medical data to obtain normalized medical data; The knowledge entities, relationships and attributes in the standardized medical data are extracted, and a knowledge network of critical and emergency diseases is constructed based on the knowledge entities, relationships and attributes.

3. The method for generating diagnosis and treatment decisions according to claim 1, characterized in that: The S2 further includes: Standardizing the key medical features to obtain standardized key medical features; A medical record feature set is constructed based on the standardized key medical features.

4. The method for generating diagnosis and treatment decisions according to claim 1, characterized in that: The S3 includes: The similarity between different medical record feature sets is calculated by using the Jaccard coefficient of different medical record feature sets.

5. The method for generating diagnosis and treatment decisions according to claim 4, characterized in that: The S3 further includes: The Jaccard coefficient of the medical record feature set is calculated according to the following formula: Among them, D it For medical record D i The medical record feature set under feature t, D jt For medical record D i and D j The medical record feature set under feature t, w p is the pth value in the medical record feature set, J t For medical record D i and medical record D j The Jaccard coefficient under feature t, is the central value of the medical record feature w p The weight of and is the pth value w in the medical record feature set p In medical record D i The weight or frequency of feature t.

6. The method for generating diagnosis and treatment decisions according to claim 5, characterized in that: The S3 further includes: The similarity between medical record feature sets is calculated according to the following formula: When sim(D i ,D j ) is greater than the preset threshold X, the similar case set {D i ,D j ,s ij }; Among them, sim(D i ,D j ) is medical record D i and medical record D j The similarity of J t For medical record D i and medical record D j Jaccard coefficient under feature t, n is the number of features, s ij For medical record D i and medical record D j The similarity between .

7. A diagnosis and treatment decision-making system for critical and severe diseases, used to implement the diagnosis and treatment decision-making method as described in any one of claims 1 to 6, characterized in that: include: A knowledge network construction module is used to obtain medical data related to critical and emergency diseases, extract knowledge entities, relationships and attributes in the medical data, and construct a knowledge network of critical and emergency diseases based on the knowledge entities, relationships and attributes; A medical record feature set construction module, used to extract key medical features from historical medical records, and construct a medical record feature set for each medical record based on the key medical features; A similar medical record set generation module is used to calculate the similarity between different medical record feature sets, aggregate similar medical records according to the similarity, and generate a similar medical record set; the similar medical record set includes multiple similar medical records and the medical features corresponding to each medical record; A coded medical record feature generation module is used to encode the medical features and perform semantic alignment through the ERNIE-Health model to obtain coded medical record features with semantic representation; each of the coded medical record features has multiple associated medical records in the same semantic space; A search engine module, used to index the coded medical record features into the engine Elasticsearch, so as to obtain similar historical medical records through similarity search in Elasticsearch; The coded medical record feature generation module is also used for: Encoding the medical features through the ERNIE-Health model to obtain encoded medical features; Calculating the characteristic distance between the medical records in the similar medical record set according to the encoded medical characteristics; the characteristic distance changes dynamically; Optimizing the ERNIE-health model according to the loss value of the feature distance; The characteristic distance between two medical records in a similar medical record set is calculated according to the following formula: Among them, d ij For medical record D i and medical record D j The characteristic distance between i is the medical record number, D it For medical record D i The feature set under feature t, D ite For medical record D i Medical features D encoded by the ERNIE-Health model it , D jte For medical record D j The medical features D encoded by the ERNIE-Health model jt , t is a positive integer, n is the number of medical features in the medical record feature set, and L is a linear transformation; The loss value is calculated according to the following loss function: Among them, μ is a hyperparameter, n is the number of medical features in the medical record feature set, i, j, k are positive integers, and s ij For medical record D i and medical record D j The similarity between ij is the distance between medical record feature set i and medical record feature set j, d ik is the distance between medical record feature set i and medical record feature set k, s ik For medical record D i and medical record D k The similarity between them is , and loss is a function that measures the difference between the model prediction value and the actual value.

Images