A diagnostic data analysis method and system

Abstract

The present invention provides a diagnostic data analysis method and system, the method comprising: obtaining a medical knowledge subgraph corresponding to target diagnostic history data, and performing vector conversion on the medical knowledge subgraph to obtain a disease code embedding vector; based on a relational attention mechanism, a node attention mechanism, and a message passing mechanism, performing feature extraction according to the disease code embedding vector, and aggregating the extracted vector features to obtain a feature vector representation; performing diagnostic prediction on the feature vector representation to obtain diagnostic prediction data. The present invention combines a medical knowledge graph with a hierarchical attention relationship network constructed based on a relational attention mechanism, a node attention mechanism, and a message passing mechanism, and performs diagnostic prediction based on diagnostic history data, thereby obtaining more accurate diagnostic prediction result data.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a diagnostic data analysis method and system. Background Art

[0002] As a large number of electronic health records (EHR) are accumulated in clinics, the diagnostic data of these EHRs need to be grouped and organized. Medical knowledge graphs are constructed through medical knowledge or cases. Through visualization technology, the relationship between these diagnostic data can be displayed, thereby realizing relevant data mining and data analysis.

[0003] In the medical field, the relationship between various diseases and treatment methods is very complex. Especially when the diagnostic history data is missing or the description is unclear (for example, the text content is vague), it is necessary to predict the missing part or the unclear diagnostic data to obtain the diagnostic data analysis results. At present, the accuracy of predictive analysis of diagnostic data is low and needs to be further improved.

[0004] Therefore, there is an urgent need for a diagnostic data analysis method and system to solve the above problems. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a diagnostic data analysis method and system.

[0006] The present invention provides a diagnostic data analysis method, comprising:

[0007] Obtaining the medical knowledge sub-graph corresponding to the target diagnosis history data, and performing vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector;

[0008] Based on the relational attention mechanism, the node attention mechanism and the message passing mechanism, feature extraction is performed according to the disease code embedding vector, and the extracted vector features are aggregated to obtain a feature vector representation;

[0009] Performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data.

[0010] According to a diagnostic data analysis method provided by the present invention, before obtaining the medical knowledge sub-graph corresponding to the target diagnostic history data and performing vector conversion on the medical knowledge sub-graph to obtain the disease code embedding vector, the method further includes:

[0011] Get the diagnostic information text of the historical moment;

[0012] Convert the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text;

[0013] Based on the preset historical time period and the disease ICD code, target diagnosis historical data is obtained.

[0014] According to a diagnostic data analysis method provided by the present invention, the step of obtaining a medical knowledge sub-graph corresponding to target diagnostic history data and performing vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector includes:

[0015] According to the disease ICD code, extract the subgraph corresponding to the target diagnosis history data from the preset medical knowledge graph to obtain a medical knowledge subgraph;

[0016] The medical knowledge sub-graph is vectorized by using a pre-trained medical code encoder to obtain a disease code embedding vector;

[0017] Wherein, the pre-trained medical code encoder is constructed through a neural network model.

[0018] According to a diagnostic data analysis method provided by the present invention, before extracting features based on the disease code embedding vector based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, and aggregating the extracted vector features to obtain a feature vector representation, the method further includes:

[0019] Based on the diagnosis prediction results corresponding to different relationship types in the medical knowledge graph, a relational attention mechanism network is constructed;

[0020] Based on the importance of different neighbor nodes to the target node in the medical knowledge graph, a node attention mechanism network is constructed;

[0021] According to the messages transmitted by the edges corresponding to different relationship types in the medical knowledge graph, a message transmission mechanism network is constructed;

[0022] A hierarchical relational neural network is constructed based on the relational attention mechanism network, the node attention mechanism network and the message passing mechanism network to perform feature extraction on the disease code embedding vector through the hierarchical relational neural network.

[0023] According to a diagnostic data analysis method provided by the present invention, before performing diagnostic prediction on the feature vector representation to obtain diagnostic prediction data, the method further includes:

[0024] The pre-trained prediction model is fine-tuned by representing the sample feature vector corresponding to the sample disease code embedding vector to obtain a trained prediction model; wherein the pre-trained prediction model is constructed based on a long short-term memory neural network or a RETAIN model;

[0025] The step of performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data includes:

[0026] Using the trained prediction model, a diagnosis prediction is performed on the feature vector representation to obtain diagnosis prediction data;

[0027] The medical knowledge sub-graph is updated according to the diagnosis prediction data.

[0028] According to a diagnostic data analysis method provided by the present invention, after obtaining the diagnostic information text at the historical moment, the method further includes:

[0029] According to preset data screening rules, outliers in the diagnostic information text are removed to obtain target diagnostic information text;

[0030] The step of converting the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text includes:

[0031] The target diagnosis information text is subjected to ICD code conversion processing to obtain the disease ICD code corresponding to the target diagnosis information text.

[0032] The present invention also provides a diagnostic data analysis system, comprising:

[0033] A code encoding module is used to obtain the medical knowledge sub-graph corresponding to the target diagnosis history data, and perform vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector;

[0034] A feature extraction module, for extracting features from the disease code embedding vector based on a relational attention mechanism, a node attention mechanism, and a message passing mechanism, and aggregating the extracted vector features to obtain a feature vector representation;

[0035] The diagnostic data analysis and prediction module is used to perform diagnostic prediction on the feature vector representation to obtain diagnostic prediction data.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned diagnostic data analysis methods is implemented.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the diagnostic data analysis method as described above is implemented.

[0038] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned diagnostic data analysis methods.

[0039] The diagnostic data analysis method and system provided by the present invention combine the medical knowledge graph with a hierarchical attention relationship network constructed based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, and perform diagnostic prediction based on the diagnostic history data, thereby obtaining more accurate diagnostic prediction result data. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A schematic diagram of a process flow of a diagnostic data analysis method provided by the present invention;

[0042] Figure 2 A schematic diagram of the overall structure of the diagnostic data analysis model provided by the present invention;

[0043] Figure 3 A schematic diagram of the operation flow of the hierarchical relationship network based on state perception provided by the present invention;

[0044] Figure 4 A schematic diagram of the structure of the diagnostic data analysis system provided by the present invention;

[0045] Figure 5 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The present invention combines state-aware hierarchical relational neural networks with existing medical knowledge graphs and applies them to diagnostic prediction tasks based on electronic health record data, thereby obtaining more accurate diagnostic prediction results and providing more precise data support for the expansion of medical knowledge graphs.

[0048] Figure 1 A schematic diagram of the process of the diagnostic data analysis method provided by the present invention is shown in FIG. Figure 1 As shown, the present invention provides a diagnostic data analysis method, comprising:

[0049] Step 101, obtain the medical knowledge sub-graph corresponding to the target diagnosis history data, and perform vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector.

[0050] In the present invention, based on the user diagnosis information database constructed in the early stage, the diagnosis information text is obtained, and the diagnosis information text records the diagnosis information of the doctor, for example, the cause of the disease, the treatment method (such as drug treatment) and the clinical manifestations.

[0051] Optionally, before obtaining the medical knowledge sub-graph corresponding to the target diagnosis history data and performing vector conversion on the medical knowledge sub-graph to obtain the disease code embedding vector, the method further includes:

[0052] Get the diagnostic information text of the historical moment;

[0053] Convert the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text;

[0054] Based on the preset historical time period and the disease ICD code, target diagnosis historical data is obtained.

[0055] In the present invention, the diagnostic text information is first processed for data standardization. According to the International Classification of Diseases (ICD), the original diagnostic result in the diagnostic text is converted into ICD9 code through the relevant feature description of the disease in the diagnostic text information, that is, the disease ICD code corresponding to the diagnostic text is obtained, so as to align with the existing medical knowledge graph for subsequent diagnostic data prediction. It should be noted that, when necessary, the model training efficiency can be improved by converting the sample diagnostic text information into ICD code.

[0056] Furthermore, according to the predefined historical time period, the disease ICD codes corresponding to different historical moments are obtained, so as to construct the target diagnosis historical data. For example, by obtaining the target diagnosis historical data of the first two months, the disease ICD code corresponding to the diagnosis data of the third month is predicted through subsequent steps.

[0057] Furthermore, based on the target diagnosis history data, relevant entities and relationships are determined from the ICD code of the existing medical knowledge graph, so as to obtain the medical knowledge sub-graph corresponding to the target diagnosis history data, and then the medical knowledge sub-graph is vectorized to obtain the vector representation of the diagnosis data, that is, the disease code embedding vector.

[0058] Step 102, based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, feature extraction is performed according to the disease code embedding vector, and the extracted vector features are aggregated to obtain a feature vector representation;

[0059] In the present invention, a state-aware hierarchical relational neural network is constructed through a relational attention mechanism, a node attention mechanism, and a message passing mechanism. Then, the disease code embedding vector obtained in the above embodiment is input into the hierarchical relational neural network to obtain an updated diagnostic data vector representation, that is, the disease code embedding vector is feature extracted through the hierarchical relational neural network to obtain a feature vector representation.

[0060] Step 103: perform diagnosis prediction on the feature vector representation to obtain diagnosis prediction data.

[0061] In the present invention, the diagnosis prediction can be performed according to the feature vector representation through the existing prediction model, for example, the Long Short-Term Memory (LSTM) neural network model, the RETAIN model, so as to obtain the diagnosis prediction result corresponding to the target diagnosis history data at the future moment. Preferably, in the present invention, according to the ICD code corresponding to the diagnosis prediction data, it is expanded into the existing medical knowledge graph to obtain an updated medical knowledge graph.

[0062] The diagnostic data analysis method provided by the present invention combines the medical knowledge graph with a hierarchical attention relationship network constructed based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, and performs diagnostic prediction based on the diagnostic history data, thereby obtaining more accurate diagnostic prediction result data.

[0063] On the basis of the above embodiment, the step of obtaining the medical knowledge sub-graph corresponding to the target diagnosis history data and performing vector conversion on the medical knowledge sub-graph to obtain the disease code embedding vector includes:

[0064] According to the disease ICD code, extract the subgraph corresponding to the target diagnosis history data from the preset medical knowledge graph to obtain a medical knowledge subgraph;

[0065] The medical knowledge sub-graph is vectorized by using a pre-trained medical code encoder to obtain a disease code embedding vector;

[0066] Wherein, the pre-trained medical code encoder is constructed through a neural network model.

[0067] In the present invention, a knowledge graph extractor is used to extract a subgraph from an existing medical knowledge graph (such as SemMed) according to the disease ICD code corresponding to the target diagnosis history data, thereby obtaining a medical knowledge subgraph corresponding to the diagnosis history data. Furthermore, a medical code encoder is used to extract an embedding vector from the encoding matrix for each disease-related code in the medical knowledge subgraph, and a disease code embedding vector is generated for subsequent diagnosis prediction. In the present invention, a diagnosis data analysis model is constructed by a knowledge graph extractor, a medical code encoder, a hierarchical relational neural network and a prediction model. Figure 2 The overall structure diagram of the diagnostic data analysis model provided by the present invention can be referred to Figure 2 As shown in the figure, after the EHR data (i.e., diagnosis history data) and the medical knowledge graph (Medical KG) are input into the diagnosis data analysis model, the knowledge graph extractor (Personalized Graph Extraction) in the model is first used to obtain the subgraph corresponding to the EHR data from the medical knowledge graph. For example, according to the ICD code corresponding to the diagnosis history data, the corresponding entities (entity types such as Heart Failure, Drug Treatment, and Narrowed Arteries) and relationships (for example, symptoms are relieved by, symptoms are produced, etc.) are determined from the medical knowledge graph, so as to construct a medical knowledge subgraph based on these entities and relationships; then, the subgraph is converted into a disease code embedding vector through the medical code encoder, and then the feature of the disease code embedding vector is extracted through the hierarchical relational neural network; finally, the extracted feature vector is predicted through the prediction model to obtain diagnosis prediction data.

[0068] On the basis of the above embodiment, before extracting features based on the disease code embedding vector based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, and aggregating the extracted vector features to obtain the feature vector representation, the method further includes:

[0069] Based on the diagnosis prediction results corresponding to different relationship types in the medical knowledge graph, a relational attention mechanism network is constructed;

[0070] Based on the importance of different neighbor nodes to the target node in the medical knowledge graph, a node attention mechanism network is constructed;

[0071] According to the messages transmitted by the edges corresponding to different relationship types in the medical knowledge graph, a message transmission mechanism network is constructed;

[0072] A hierarchical relational neural network is constructed based on the relational attention mechanism network, the node attention mechanism network and the message passing mechanism network to perform feature extraction on the disease code embedding vector through the hierarchical relational neural network.

[0073] In the present invention, through the relation attention mechanism, the node attention mechanism and the message passing mechanism, a state-aware relation-level attention module (Stage-Aware Relation-Level Attention), a state-aware node-level attention module (Stage-Aware Node-Level Attention) and a relation-specific message passing module (Relation-Specific Message Passing) are respectively constructed, and then combined with a hierarchical message aggregation module (Hierarchical Message Aggregation), so as to construct a hierarchical relation network based on state awareness.

[0074] Specifically, in order to characterize the influence of different relationship types in the medical knowledge graph on the diagnosis prediction results, the present invention constructs a relationship attention mechanism to calculate the corresponding relationship-level attention coefficient, and the formula is:

[0075]

[0076]

[0077] Among them, R represents the set of all relationship types in the medical knowledge graph, and f t-1 represents the state vector returned by the prediction model in the previous time step, LeakyReLu is the activation function; w r and W r are the parameters to be learned.

[0078] Furthermore, in order to characterize the importance of different neighbor nodes to the target node, the present invention constructs a node attention mechanism to calculate the corresponding node-level attention coefficient, and the formula is:

[0079]

[0080]

[0081] Among them, h o represents the current expression of the target node o, h si represents the expression corresponding to the i-th edge of node s, w n and W n Parameters to be learned, N o Represents a collection of nodes in the medical knowledge graph.

[0082] Furthermore, in the process of message transmission, considering that the information transmitted by the edges corresponding to different relationships may be different, the present invention proposes a relationship-related message transmission mechanism. For the edge from node s to the target node o, assuming that the edge type is e, the transmitted message is:

[0083] msg(o,s)=W e h s ;

[0084] Among them, W e is a learnable parameter associated with edge type e, h s is the current expression of node s

[0085] Finally, the relationship-level attention coefficient and node-level attention coefficient calculated in the above embodiment, as well as the relationship-related transmission messages, are used to calculate the aggregated information for each node. The formula is:

[0086]

[0087] During the aggregation process, all the information of the nodes connected to the target node o through different relationship types are aggregated, thereby obtaining the updated expression of the target node o. Figure 3 This is a schematic diagram of the operation flow of the hierarchical relationship network based on state perception provided by the present invention. The specific operation process can be referred to Figure 3 shown.

[0088] In the present invention, a network layer in a state-aware hierarchical relational neural network is formed through a state-aware relation-level attention module, a state-aware node-level attention module, a relation-related message passing module and a hierarchical message aggregation module. Furthermore, multiple layers of the above network layers can be stacked, and residual connections are introduced between adjacent network layers:

[0089]

[0090] in, is the expression of the target node o in the lth layer.

[0091] Based on the above embodiment, before performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data, the method further includes:

[0092] The pre-trained prediction model is fine-tuned by representing the sample feature vector corresponding to the sample disease code embedding vector to obtain a trained prediction model; wherein the pre-trained prediction model is constructed based on a long short-term memory neural network or a RETAIN model;

[0093] The step of performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data includes:

[0094] Using the trained prediction model, a diagnosis prediction is performed on the feature vector representation to obtain diagnosis prediction data;

[0095] The medical knowledge sub-graph is updated according to the diagnosis prediction data.

[0096] In the present invention, the disease code embedding vector is passed through a multi-layer hierarchical attention relationship network to obtain an updated node expression and pass it to the prediction model. The neural network model parameters are updated through the loss function, and after the preset convergence conditions are met, a trained prediction model is obtained. At the same time, the prediction model returns the intermediate state vector to the hierarchical relationship network in the above embodiment. In the present invention, the prediction model can be trained in an end-to-end manner, and its loss function is:

[0097]

[0098] Among them, x1,x2,…,x T The diagnostic history data corresponding to the T diagnostic records of a patient. Since the operations involved in the model are all differentiable, the present invention can be updated by a standard gradient descent method.

[0099] Based on the above embodiment, after obtaining the diagnostic information text at the historical moment, the method further includes:

[0100] According to preset data screening rules, outliers in the diagnostic information text are removed to obtain target diagnostic information text;

[0101] The step of converting the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text includes:

[0102] The target diagnosis information text is subjected to ICD code conversion processing to obtain the disease ICD code corresponding to the target diagnosis information text.

[0103] In the present invention, based on prior knowledge, outlier points in user (patient) information data and disease diagnosis data are deleted to obtain target diagnosis information text. In one embodiment, the preset data screening rule is specifically: remove diagnosis information text with a diagnosis number of less than 3 times, and retain missing value data after cleaning the data.

[0104] In order to verify the implementation effect of the present invention, this embodiment is further described with experimental results on real data. The specific steps are as follows:

[0105] Step S1, training and test sample extraction rules: According to the standard division of the MIMIC-III dataset, the dataset is divided into a training set, a validation set, and a test set;

[0106] Step S2, using the training set for model training, and selecting the model with the best performance on the validation set as the final diagnostic data analysis model based on the prediction accuracy of the model on the validation set; wherein the diagnostic data analysis model is constructed by the knowledge graph extractor, medical code encoder, hierarchical relational neural network and prediction model provided in the above embodiment;

[0107] Step S3, the trained diagnostic data analysis model is tested on the test set, and the prediction precision and accuracy are evaluated. The prediction results show that the diagnostic data analysis model provided by the present invention has significantly improved prediction performance in terms of precision and accuracy compared to the existing models.

[0108] The following is a description of the diagnostic data analysis system provided by the present invention. The diagnostic data analysis system described below and the diagnostic data analysis method described above can be referenced to each other.

[0109] Figure 4 A schematic diagram of the structure of the diagnostic data analysis system provided by the present invention, such as Figure 4 As shown, the present invention provides a diagnostic data analysis system, including a code encoding module 401, a feature extraction module 402 and a diagnostic data analysis and prediction module 403, wherein the code encoding module 401 is used to obtain the medical knowledge sub-graph corresponding to the target diagnostic history data, and perform vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector; the feature extraction module 402 is used to perform feature extraction according to the disease code embedding vector based on a relational attention mechanism, a node attention mechanism and a message passing mechanism, and aggregate the extracted vector features to obtain a feature vector representation; the diagnostic data analysis and prediction module 403 is used to perform diagnostic prediction on the feature vector representation to obtain diagnostic prediction data.

[0110] The diagnostic data analysis system provided by the present invention combines the medical knowledge graph with a hierarchical attention relationship network constructed based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, and performs diagnostic prediction based on the diagnostic history data, thereby obtaining more accurate diagnostic prediction result data.

[0111] The system provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0112] Figure 5 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor (Processor) 501, a communication interface (Communications Interface) 502, a memory (Memory) 503 and a communication bus 504, wherein the processor 501, the communication interface 502, and the memory 503 complete mutual communication through the communication bus 504. The processor 501 can call the logic instructions in the memory 503 to execute the diagnosis data analysis method, which includes: obtaining the medical knowledge sub-graph corresponding to the target diagnosis history data, and performing vector conversion on the medical knowledge sub-graph to obtain the disease code embedding vector; based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, feature extraction is performed according to the disease code embedding vector, and the extracted vector features are aggregated to obtain a feature vector representation; diagnosis prediction is performed on the feature vector representation to obtain diagnosis prediction data.

[0113] In addition, the logic instructions in the above-mentioned memory 503 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and 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 perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0114] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the diagnostic data analysis method provided by the above methods, and the method includes: obtaining a medical knowledge sub-graph corresponding to the target diagnostic history data, and performing vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector; based on a relational attention mechanism, a node attention mechanism, and a message passing mechanism, performing feature extraction according to the disease code embedding vector, and aggregating the extracted vector features to obtain a feature vector representation; performing diagnostic prediction on the feature vector representation to obtain diagnostic prediction data.

[0115] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the diagnostic data analysis method provided in the above-mentioned embodiments, the method comprising: obtaining a medical knowledge sub-graph corresponding to the target diagnostic historical data, and performing vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector; based on a relational attention mechanism, a node attention mechanism, and a message passing mechanism, performing feature extraction according to the disease code embedding vector, and aggregating the extracted vector features to obtain a feature vector representation; performing diagnostic prediction on the feature vector representation to obtain diagnostic prediction data.

[0116] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0117] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A diagnostic data analysis method, characterized in that: include: Obtaining the medical knowledge sub-graph corresponding to the target diagnosis history data, and performing vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector; Based on the relational attention mechanism, the node attention mechanism and the message passing mechanism, feature extraction is performed according to the disease code embedding vector, and the extracted vector features are aggregated to obtain a feature vector representation; Performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data; Before obtaining the medical knowledge sub-graph corresponding to the target diagnosis history data and performing vector conversion on the medical knowledge sub-graph to obtain the disease code embedding vector, the method further includes: Get the diagnostic information text of the historical moment; Convert the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text; Based on the preset historical time period and the disease ICD code, obtaining target diagnosis historical data; The step of obtaining a medical knowledge sub-graph corresponding to the target diagnosis history data and performing vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector includes: According to the disease ICD code, extract the subgraph corresponding to the target diagnosis history data from the preset medical knowledge graph to obtain a medical knowledge subgraph; The medical knowledge sub-graph is vectorized by using a pre-trained medical code encoder to obtain a disease code embedding vector; Wherein, the pre-trained medical code encoder is constructed by a neural network model; Before extracting features based on the disease code embedding vector based on the relationship attention mechanism, the node attention mechanism and the message passing mechanism, and aggregating the extracted vector features to obtain the feature vector representation, the method further includes: Based on the diagnosis prediction results corresponding to different relationship types in the medical knowledge graph, a relational attention mechanism network is constructed; Based on the importance of different neighbor nodes to the target node in the medical knowledge graph, a node attention mechanism network is constructed; According to the messages transmitted by the edges corresponding to different relationship types in the medical knowledge graph, a message transmission mechanism network is constructed; A hierarchical relational neural network is constructed based on the relational attention mechanism network, the node attention mechanism network and the message passing mechanism network to perform feature extraction on the disease code embedding vector through the hierarchical relational neural network.

2. The diagnostic data analysis method according to claim 1, characterized in that: Before performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data, the method further includes: The pre-trained prediction model is fine-tuned by representing the sample feature vector corresponding to the sample disease code embedding vector to obtain a trained prediction model; wherein the pre-trained prediction model is constructed based on a long short-term memory neural network or a RETAIN model; The step of performing diagnosis prediction on the feature vector representation to obtain diagnosis prediction data includes: Using the trained prediction model, a diagnosis prediction is performed on the feature vector representation to obtain diagnosis prediction data; The medical knowledge sub-graph is updated according to the diagnosis prediction data.

3. The diagnostic data analysis method according to claim 1, characterized in that: After obtaining the diagnostic information text at the historical moment, the method further includes: According to preset data screening rules, outliers in the diagnostic information text are removed to obtain target diagnostic information text; The step of converting the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text includes: The target diagnosis information text is subjected to ICD code conversion processing to obtain the disease ICD code corresponding to the target diagnosis information text.

4. A diagnostic data analysis system, characterized in that: include: A code encoding module is used to obtain the medical knowledge sub-graph corresponding to the target diagnosis history data, and perform vector conversion on the medical knowledge sub-graph to obtain a disease code embedding vector; A feature extraction module, for extracting features from the disease code embedding vector based on a relational attention mechanism, a node attention mechanism, and a message passing mechanism, and aggregating the extracted vector features to obtain a feature vector representation; A diagnostic data analysis and prediction module, used to perform diagnostic prediction on the feature vector representation to obtain diagnostic prediction data; The diagnostic data analysis system is also used to: Get the diagnostic information text of the historical moment; Convert the diagnosis information text into an ICD code to obtain the disease ICD code corresponding to the diagnosis information text; Based on the preset historical time period and the disease ICD code, obtaining target diagnosis historical data; The code encoding module is specifically used for: According to the disease ICD code, extract the subgraph corresponding to the target diagnosis history data from the preset medical knowledge graph to obtain a medical knowledge subgraph; The medical knowledge sub-graph is vectorized by using a pre-trained medical code encoder to obtain a disease code embedding vector; Wherein, the pre-trained medical code encoder is constructed by a neural network model; The diagnostic data analysis system is also used to: Based on the diagnosis prediction results corresponding to different relationship types in the medical knowledge graph, a relational attention mechanism network is constructed; Based on the importance of different neighbor nodes to the target node in the medical knowledge graph, a node attention mechanism network is constructed; According to the messages transmitted by the edges corresponding to different relationship types in the medical knowledge graph, a message transmission mechanism network is constructed; A hierarchical relational neural network is constructed based on the relational attention mechanism network, the node attention mechanism network and the message passing mechanism network to perform feature extraction on the disease code embedding vector through the hierarchical relational neural network.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the diagnostic data analysis method according to any one of claims 1 to 3 is implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the diagnostic data analysis method according to any one of claims 1 to 3 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the diagnostic data analysis method according to any one of claims 1 to 3 is implemented.

Images