Syndrome prediction method and system integrating multi-graph convolution and ontology knowledge

Abstract

The present invention provides a method and system for intelligent TCM syndrome differentiation that integrates multi-graph convolution and ontology knowledge. This method belongs to the technical field of disease syndrome prediction and involves acquiring clinical electronic medical record data and TCM knowledge base ontology data. A pre-trained syndrome prediction model is used to process the acquired clinical electronic medical record data and TCM knowledge base ontology data to obtain syndrome prediction results. The syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module, and an intelligent syndrome differentiation module. The present invention establishes a strategy for constructing symptom feature graphs, syndrome label graphs, and syndrome-symptom graphs that integrate clinical medical record data and TCM ontology knowledge, as well as a symptom and syndrome embedding representation learning technique that combines a multi-graph convolutional neural network with feature fusion. Ultimately, this method achieves accurate embedded representation of symptoms, syndromes, and patients, improving the prediction accuracy of the intelligent syndrome differentiation algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of disease symptom prediction, and in particular to a symptom prediction method and system integrating multi-graph convolution and ontology knowledge. Background Art

[0002] Intelligent TCM syndrome differentiation refers to using artificial intelligence technology to learn from patient medical records and predict patient symptoms, thereby simulating the process of TCM physicians diagnosing TCM syndromes. The core of its construction process is the formation of patient representation. The current TCM intelligent syndrome differentiation methods are as follows:

[0003] Yao Liang et al. fine-tuned the BERT language model using an unlabeled clinical corpus, trained a text classifier using Chinese characters from clinical text as input, and constructed a pre-trained language model for TCM (TCM-BERT). Ren Mucheng et al. collected a large-scale unlabeled text corpus in TCM and proposed a pre-trained language model, ZY-BERT, based on TCM. Chen Zhe et al. obtained expert knowledge from public TCM guidelines and textbooks and constructed a TCM-BERT-CNN model to implement end-to-end TCM syndrome differentiation. Wang Zeyuan et al. used multi-instance multi-task learning combined with a convolutional neural network (MIMT-CNN) for syndrome classification. This neural network takes local regions as input and consists of convolutional layers, fully connected layers, and maximum pooling layers. The sigmoid function in the syndrome prediction task is used for image representation learning and final result generation.

[0004] How to combine external knowledge and fully utilize symptom and syndrome information to construct an accurate representation of the patient, and then apply it to intelligent TCM syndrome differentiation to improve model performance, is a key issue in the field of intelligent assisted diagnosis and treatment. The data source for current TCM intelligent syndrome differentiation models is still mainly clinical electronic medical records. However, due to inaccurate doctor records or inaccurate patient statements, the accuracy and standardization of clinical data are poor. Existing intelligent syndrome differentiation methods do not consider incorporating external knowledge to improve the reliability and accuracy of the model. On the other hand, there are often complex correlations between symptoms, syndromes, and symptoms. In the process of TCM syndrome differentiation modeling, effectively exploring the complex internal correlations between them is conducive to improving the performance of intelligent syndrome differentiation models. However, existing work still does not fully utilize the complex relationships between symptoms and syndromes. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for predicting TCM syndromes by integrating multi-graph convolution and ontology knowledge. By integrating symptoms and symptom ontology knowledge as well as the co-occurrence relationship between syndromes and symptoms in clinical records, symptom maps, syndrome maps and symptom-syndrome maps are established. Deep multi-graph convolutional neural networks are used to learn the accurate representation of symptoms, syndromes and patients, so as to solve the problems of lack of external knowledge support and insufficient utilization of clinical information in existing intelligent syndrome differentiation methods, realize intelligent TCM syndrome differentiation, and solve at least one technical problem existing in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a syndrome prediction method integrating multi-graph convolution and ontology knowledge, comprising:

[0008] Obtain clinical electronic medical record data and TCM knowledge base ontology data;

[0009] The acquired clinical electronic medical record data and TCM knowledge base ontology data are processed using a pre-trained syndrome prediction model to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the prediction probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0010] Furthermore, a symptom map, a syndrome map and a symptom-syndrome map are established, including: based on clinical electronic medical record data, calculating co-occurrence relationships and constructing a co-occurrence network, thereby obtaining a symptom co-occurrence network, a symptom-syndrome co-occurrence network, and a syndrome co-occurrence network; based on the tree-shaped coding information of symptoms and syndromes, establishing the associations between symptoms and between syndromes respectively, to form a symptom ontology map and a syndrome ontology map; fusing the symptom ontology map and the symptom co-occurrence map to form a symptom feature map, fusing the syndrome ontology map and the syndrome co-occurrence map to form a syndrome label map, and using the symptom-syndrome co-occurrence map as a feature-label map.

[0011] Furthermore, with the symptom feature graph, syndrome label graph, and symptom-syndrome graph as input, a multi-graph convolutional neural network model is used to learn the embedded representations of symptom features and syndrome labels on each graph respectively, including: for the embedded representation of the symptom feature graph, the node-initialized embedding matrix of the symptom feature graph and the adjacency matrix of the feature graph are used as the input of the graph neural network to form the embedded representation of the (l+1)th layer; for the embedded representation of the syndrome label graph, the node-initialized embedding matrix of the syndrome label graph and the adjacency matrix of the feature graph are used as the input of the graph neural network to form the embedded representation of the (l+1)th layer; for the embedded representation of the feature-label graph, the node-initialized embedding matrix of the feature-label graph and the adjacency matrix of the feature graph are used as the input of the graph neural network to form the embedded representation of the (l+1)th layer.

[0012] Furthermore, for the embedding representation of the symptom feature graph, the symptom feature graph G Sym Initialize the node embedding matrix X Sym And the adjacency matrix A of the feature graph Sym As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula:

[0013]

[0014] in, is the degree matrix of the symptom feature map, is the weight matrix of the lth layer; through the learning of K convolutional layers, the symptom embedding feature representation of the symptom feature map is finally formed

[0015] Furthermore, for the embedding representation of the syndrome label graph, the syndrome label graph G Syn Initialize the node embedding matrix X Syn (i.e. a randomly initialized matrix) and the adjacency matrix A of the feature graph Syn As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula:

[0016]

[0017] in, is the degree matrix of the syndrome label graph, is the weight matrix of the lth layer; through the learning of K convolutional layers, the syndrome embedding feature representation of the syndrome label graph is finally formed

[0018] Furthermore, for the embedding representation of the feature-label map, the feature-label map G S-S =(X S-S ,A S-S )’s node initialization embedding matrix XS-S And the adjacency matrix A of the feature graph S-S As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula:

[0019]

[0020] in, is the degree matrix of the feature-label graph, is the weight matrix of the lth layer; through the learning of K convolutional layers, the node embedding feature representation of the feature-label graph is finally formed That is, forming the embedding representation of all symptom nodes in the feature-label graph and the embedded representation of all syndrome nodes

[0021] Furthermore, after embedding representation learning, the embedding representation of the symptom feature map is recorded as The semantic embedding representation of symptoms is denoted as The symptom embedding representation in the symptom-syndrome diagram is denoted as The embedded representation of syndrome is recorded as The embedded representation of the syndrome diagram is denoted as The semantic embedding representation of syndrome is recorded as Using the maximum pooling technology, we can achieve the organic fusion of symptom and syndrome embedding representation:

[0022]

[0023] Obtain embedded representation features of symptoms and syndromes that integrate graph structure and semantic information.

[0024] Furthermore, the final symptom is embedded into Multiply it with the patient's original Multi-hot embedding representation P to obtain the patient's embedding representation, which is recorded as The patient's symptoms The patient's final embedding representation is obtained through a multi-layer neural network At the same time, the syndrome is embedded in the expression After a multi-layer neural network, the final syndrome embedding representation is obtained Then, by embedding the patient Embedded with syndromes Perform multiplication calculations and pass the calculation results through the Sigmoid function to form the predicted probability value of each syndrome.

[0025] Furthermore, the loss function between the patient's true syndrome vector and the syndrome probability value predicted by the algorithm is calculated, and the entire neural network model is optimized with this loss function. The loss function is calculated as follows:

[0026]

[0027] Where N is the total number of samples; M is the total number of syndrome labels; y ij Indicates the true value of the jth syndrome label in the i-th sample. A value of 0 indicates that syndrome j is not the syndrome label of this sample, and a value of 1 indicates that syndrome j is the syndrome label of this sample. It represents the predicted probability value of the model for the jth syndrome label in the i-th sample.

[0028] In a second aspect, the present invention provides a syndrome prediction system integrating multi-graph convolution and ontology knowledge, comprising:

[0029] Acquisition module, used to obtain clinical electronic medical record data and TCM knowledge base ontology data;

[0030] A processing module is used to use a pre-trained syndrome prediction model to process the acquired clinical electronic medical record data and TCM knowledge base ontology data to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the prediction probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0031] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the method for predicting traditional Chinese medicine syndromes that integrates multi-graph convolution and ontology knowledge as described in the first aspect is implemented.

[0032] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the method for predicting traditional Chinese medicine syndromes that integrates multi-graph convolution and ontology knowledge as described in the first aspect.

[0033] In a fifth aspect, the present invention provides an electronic device comprising: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the method for predicting traditional Chinese medicine symptoms that integrates multi-graph convolution and ontology knowledge as described in the first aspect.

[0034] The beneficial effects of the present invention are as follows: a symptom feature graph, syndrome label graph, and syndrome-syndrome graph construction strategy that integrate clinical medical record data and traditional Chinese medicine ontology knowledge are established, as well as a symptom and syndrome embedding representation learning technology that combines multi-graph convolutional neural network with feature fusion, ultimately achieving accurate embedding representation of symptoms, syndromes, and patients, and improving the prediction accuracy of the intelligent syndrome differentiation algorithm.

[0035] Additional advantages of the present invention will be more clearly given in the following description or learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 This is a flow chart of the TCM syndrome prediction method that integrates multi-graph convolution and ontology knowledge as described in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0039] Those skilled in the art will understand that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0040] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0041] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0042] In the description of this specification, reference to the terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless otherwise contradictory.

[0043] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0044] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0045] The feature inputs for current intelligent TCM syndrome differentiation models primarily come from clinical electronic medical record data, focusing on patient symptoms. However, clinical data often suffers from poor accuracy and standardization, necessitating the introduction of external knowledge to support intelligent TCM syndrome differentiation. Furthermore, recent research on intelligent TCM syndrome differentiation has largely focused on patient symptoms, neglecting the rich knowledge inherent in symptom-syndrome and syndrome-syndrome relationships. This results in existing intelligent syndrome differentiation methods underutilizing syndrome information. Therefore, there is an urgent need to develop an intelligent TCM syndrome differentiation method that considers TCM ontology knowledge and fully utilizes symptom and syndrome information. The present invention proposes a method for intelligent syndrome differentiation in traditional Chinese medicine based on deep graph convolutional neural networks, which integrates the ontological knowledge of syndromes and symptoms, as well as the heterogeneous and complex relationships between symptoms and syndromes. That is, by integrating the ontological knowledge of symptoms and syndromes and clinical diagnosis and treatment data, symptom maps, syndrome maps and symptom-syndrome maps are constructed respectively, and then the feature learning ability of deep multi-graph convolutional neural networks is utilized to solve the problems of lack of external knowledge support and insufficient utilization of syndrome information in existing intelligent syndrome differentiation methods, so as to enrich the information representation of patients, thereby improving the performance of intelligent syndrome differentiation in traditional Chinese medicine and promoting the rapid development of clinical auxiliary diagnosis and treatment decision-making.

[0046] Example 1

[0047] In this embodiment 1, a syndrome prediction system integrating multi-graph convolution and ontology knowledge is first provided, including: an acquisition module for acquiring clinical electronic medical record data and traditional Chinese medicine knowledge base ontology data. A processing module is used to use a pre-trained syndrome prediction model to process the acquired clinical electronic medical record data and TCM knowledge base ontology data to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the prediction probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0048] In this embodiment, the above-mentioned system is used to implement a syndrome prediction method that integrates multi-graph convolution and ontology knowledge, including: obtaining clinical electronic medical record data and traditional Chinese medicine knowledge base ontology data; using a pre-trained syndrome prediction model to process the obtained clinical electronic medical record data and traditional Chinese medicine knowledge base ontology data to obtain syndrome prediction results.

[0049] Among them, the syndrome prediction model includes a graph construction module, an embedded representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the traditional Chinese medicine knowledge base and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedded representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use the multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the predicted probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0050] Establishing symptom maps, syndrome maps, and symptom-syndrome maps includes: calculating co-occurrence relationships and constructing co-occurrence networks based on clinical electronic medical record data, thereby obtaining symptom co-occurrence networks, symptom-syndrome co-occurrence networks, and syndrome co-occurrence networks; establishing associations between symptoms and between syndromes based on tree-shaped coding information of symptoms and syndromes, respectively, to form symptom ontology maps and syndrome ontology maps; fusing the symptom ontology map and the symptom co-occurrence map to form a symptom feature map, fusing the syndrome ontology map and the syndrome co-occurrence map to form a syndrome label map, and using the symptom-syndrome co-occurrence map as a feature-label map.

[0051] Taking the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, the multi-graph convolutional neural network model is used to learn the embedded representation of symptom features and syndrome labels on each graph, including: for the embedded representation of symptom feature graph, the symptom feature graph G is used as the input. Sym Initialize the node embedding matrix X Sym And the adjacency matrix A of the feature graph Sym As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula:

[0052]

[0053] in, is the degree matrix of the symptom feature map, is the weight matrix of the lth layer; through the learning of K convolutional layers, the symptom embedding feature representation of the symptom feature map is finally formed

[0054] For the embedding representation of syndrome label graph, the syndrome label graph G Syn Initialize the node embedding matrix X Syn (i.e. a randomly initialized matrix) and the adjacency matrix A of the feature graph Syn As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula:

[0055]

[0056] in, is the degree matrix of the syndrome label graph, is the weight matrix of the lth layer; through the learning of K convolutional layers, the syndrome embedding feature representation of the syndrome label graph is finally formed

[0057] For the embedding representation of the feature-label map, the feature-label map G S-S =(X S-s ,X S-S )’s node initialization embedding matrix X S-S And the adjacency matrix A of the feature graphS-S As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula:

[0058]

[0059] in, is the degree matrix of the feature-label graph, is the weight matrix of the lth layer; through the learning of K convolutional layers, the node embedding feature representation of the feature-label graph is finally formed That is, forming the embedding representation of all symptom nodes in the feature-label graph and the embedded representation of all syndrome nodes

[0060] After embedding representation learning, the embedding representation of the symptom feature map is recorded as The semantic embedding representation of symptoms is denoted as The symptom embedding representation in the symptom-syndrome diagram is denoted as The embedded representation of syndrome is recorded as The embedded representation of the syndrome diagram is denoted as The semantic embedding representation of syndrome is recorded as Using the maximum pooling technology, we can achieve the organic fusion of symptom and syndrome embedding representation:

[0061]

[0062] Obtain embedded representation features of symptoms and syndromes that integrate graph structure and semantic information.

[0063] Embed the final symptoms into Multiply it with the patient's original Multi-hot embedding representation P to obtain the patient's embedding representation, which is recorded as The patient's symptoms The patient's final embedding representation is obtained through a multi-layer neural network At the same time, the syndrome is embedded in the expression After a multi-layer neural network, the final syndrome embedding representation is obtained Then, by embedding the patient Embedded with syndromes Perform multiplication calculations and pass the calculation results through the Sigmoid function to form the predicted probability value of each syndrome.

[0064] Calculate the loss function between the patient's true syndrome vector and the syndrome probability value predicted by the algorithm, and use this loss function to optimize the entire neural network model. The loss function is calculated as follows:

[0065]

[0066] Where N is the total number of samples; M is the total number of syndrome labels; y ij Indicates the true value of the jth syndrome label in the i-th sample. A value of 0 indicates that syndrome j is not the syndrome label of this sample, and a value of 1 indicates that syndrome j is the syndrome label of this sample. It represents the predicted probability value of the model for the jth syndrome label in the i-th sample.

[0067] Example 2

[0068] like Figure 1 As shown, this embodiment proposes an intelligent TCM syndrome differentiation method that combines multi-graph convolution with ontology knowledge. Clinical electronic medical record data and TCM knowledge base ontology data are used as input. First, symptoms and symptom ontology knowledge as well as the co-occurrence relationship between syndromes and symptoms in clinical records are integrated to establish symptom maps, syndrome maps and symptom-syndrome maps. Then, a deep multi-graph convolutional neural network and a feature fusion module are used to learn and obtain accurate representations of symptoms, syndromes and patients. Through a multi-layer neural network, the predicted probability of each candidate syndrome is obtained, and the syndrome sequence sorted by probability is used as the final predicted syndrome output result.

[0069] First, based on clinical medical record data, we established symptom co-occurrence maps, syndrome co-occurrence maps, and symptom-syndrome co-occurrence maps. First, based on clinical electronic medical record data, we calculated co-occurrence relationships and constructed a co-occurrence network, thereby obtaining a symptom co-occurrence network, a symptom-syndrome co-occurrence network, and a syndrome co-occurrence network. Co-occurrence generally refers to the co-occurrence of words i and j within the same semantic structure in the same document. Based on this, we can construct a co-occurrence matrix and calculate the co-occurrence probability. The co-occurrence probability refers to the probability of word j appearing in the context of word i, expressed as follows.

[0070]

[0071] Among them, x ij Represents the frequency of word j appearing in word i as the central word in the co-occurrence matrix.

[0072] Based on the ontological knowledge of symptoms and syndromes, syndrome ontology and symptom ontology diagrams are constructed. Symptom ontology data is presented in a tree-like coding format, starting with the largest concept in the symptom domain and refining from top to bottom. For example, the larger concept "neurological symptoms" is coded as 01, "hearing impairment," as a type of neurological symptom, is coded as 01.01, "visual impairment," also a type of neurological symptom, is coded as 01.02, "deafness," as a type of hearing impairment, is coded as 01.01.01, "tinnitus," also a type of hearing impairment, is coded as 01.01.02, "visual impairment," as a type of visual impairment, is coded as 01.02.01, and "visual field loss," also a type of visual impairment, is coded as 01.02.02. Based on the tree-like coding information of symptoms and syndromes, associations between symptoms and between syndromes are established, thus forming symptom and syndrome ontology diagrams.

[0073] Finally, the symptom ontology map and the symptom co-occurrence map are fused to form the symptom feature map, the syndrome ontology map and the syndrome co-occurrence map are fused to form the syndrome label map, and the symptom-syndrome co-occurrence map is used as the feature-label map. The fusion of the symptom co-occurrence map and the symptom ontology map is introduced as an example: 1) Symptom dictionary table formation: Statistical symptom-symptom co-occurrence network N sym-co Nodes in the symptom ontology network N sym-on The nodes in the two categories are merged and deduplicated to form the symptom dictionary table s set , and re-form the symptom term number to form the nodes of the symptom homogeneity graph; 2) Symptom graph node alignment and fusion: retain the original two networks N sym-co With N sym-on The symptom node-edge relationship of the two networks is renamed according to the newly formed symptom network node number to align the nodes of the two symptom networks to the common symptom dictionary table s set , and merge the edge data of the two aligned symptom networks; 3) Symptom graph formation: Based on the merged symptom network data, duplicates are removed according to the edge relationship, and the symptom network after duplicate removal is the fused symptom feature graph G sym The construction method of the syndrome label graph is similar, that is, first forming the syndrome label dictionary table, and then aligning and merging the syndrome graph nodes to form the syndrome label graph G Syn The symptom-syndrome co-occurrence graph formed based on the co-occurrence relationship of clinical data is the feature-label graph G. S-S .

[0074] Using a symptom feature graph, a syndrome label graph, and a symptom-syndrome graph as input, a multi-graph convolutional neural network model learns the embedding representations of symptom features and syndrome labels on each graph. Multi-graph convolutional neural networks aggregate the features of surrounding nodes to generate a feature embedding for that node (symptom or syndrome). The learning process of a graph convolutional neural network (GCN) on a feature graph, label graph, and feature-label graph is explained as follows:

[0075] 1) For the embedded representation of symptom feature graph, the symptom feature graph G Syn Initialize the node embedding matrix X Sym (i.e. a randomly initialized matrix) and the adjacency matrix A of the feature graph Sym is the input of the graph neural network, and the embedding representation of the (l+1)th layer is formed by the following formula:

[0076]

[0077] in is the degree matrix of the symptom feature map, is the weight matrix of the lth layer. Through the learning of K convolutional layers, the symptom embedding feature representation of the symptom feature map is finally formed.

[0078] 2) For the embedded representation of syndrome label graph, the syndrome label graph G Syn Initialize the node embedding matrix X Sym (i.e. a randomly initialized matrix) and the adjacency matrix A of the feature graph Sym is the input of the graph neural network, and the embedding representation of the (l+1)th layer is formed by the following formula:

[0079]

[0080] in, is the degree matrix of the syndrome label graph, is the weight matrix of the lth layer. Through the learning of K convolutional layers, the syndrome embedding feature representation of the syndrome label graph is finally formed.

[0081] 3) For the embedding representation of the feature-label map, the feature-label map G S-S =(X S-S ,A S-S )’s node initialization embedding matrix X S-S (i.e. a randomly initialized matrix) and the adjacency matrix A of the feature graph S-S is the input of the graph neural network, and the embedding representation of the (l+1)th layer is formed by the following formula:

[0082]

[0083] in, is the degree matrix of the feature-label graph, is the weight matrix of the lth layer. Through the learning of K layers of convolutional layers, the node embedding feature representation of the feature-label graph is finally formed (i.e. forming an embedding representation of all symptom nodes in the feature-label graph and the embedded representation of all syndrome nodes ).

[0084] In addition, in order to enrich the information contained in the symptom embedding representation and syndrome embedding representation, we take symptoms and syndrome words as input, and combine them with the pre-trained language model BERT to learn the semantic embedding representations of symptoms and syndromes respectively: and

[0085] After embedding representation learning, the embedding representation of the symptom feature map is recorded as The semantic embedding representation of symptoms is denoted as The symptom embedding representation in the symptom-syndrome diagram is denoted as The embedded representation of syndrome is recorded as The embedded representation of the syndrome diagram is denoted as The semantic embedding representation of syndrome is recorded as The maximum pooling technology is used to achieve the organic fusion of symptom and syndrome embedding representation, as follows.

[0086]

[0087] After the above fusion process, we can finally obtain the embedded representation features of symptoms and syndromes that integrate graph structure and semantic information.

[0088] Embed the final symptoms into Multiply it with the patient's original Multi-hot embedding representation P to obtain the patient's embedding representation, which is recorded as It is expressed as follows.

[0089]

[0090] The patient's symptoms The patient's final embedding representation is obtained through a multi-layer neural network At the same time, the syndrome is embedded in the expression After a multi-layer neural network, the final syndrome embedding representation is obtained Then, by embedding the patient Embedded with syndromes Perform multiplication calculations and pass the calculation results through the Sigmoid function to form the predicted probability value of each syndrome. Finally, calculate the loss function of the patient's actual syndrome vector and the syndrome probability value predicted by the algorithm. This loss function is used to optimize the entire neural network model. The loss function of the actual syndrome and the predicted syndrome is calculated as follows:

[0091]

[0092] Among them, N is the total number of samples, M is the total number of syndrome labels, and y ij represents the true value of the jth syndrome label in the i-th sample (a value of 0 indicates that syndrome j is not the syndrome label of this sample, and a value of 1 indicates that syndrome j is the syndrome label of this sample), It represents the predicted probability value of the model for the jth syndrome label in the i-th sample. This loss is used as the objective function to guide the neural network to optimize the parameters and obtain the optimized intelligent syndrome differentiation neural network.

[0093] Example 3

[0094] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the method for predicting TCM syndromes by integrating multi-graph convolution with ontology knowledge as described above is implemented. The method includes:

[0095] Obtain clinical electronic medical record data and TCM knowledge base ontology data;

[0096] The acquired clinical electronic medical record data and TCM knowledge base ontology data are processed using a pre-trained syndrome prediction model to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the prediction probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0097] Example 4

[0098] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the above-mentioned TCM syndrome prediction method integrating multi-graph convolution with ontology knowledge, the method comprising:

[0099] Obtain clinical electronic medical record data and TCM knowledge base ontology data;

[0100] The acquired clinical electronic medical record data and TCM knowledge base ontology data are processed using a pre-trained syndrome prediction model to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the prediction probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0101] Example 5

[0102] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the above-mentioned method for predicting TCM syndromes by integrating multi-graph convolution with ontology knowledge. The method includes:

[0103] Obtain clinical electronic medical record data and TCM knowledge base ontology data;

[0104] The acquired clinical electronic medical record data and TCM knowledge base ontology data are processed using a pre-trained syndrome prediction model to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the prediction probability of each candidate syndrome through a multi-layer neural network, and use the syndrome sequence sorted by probability as the final predicted syndrome output result.

[0105] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0106] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0107] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0109] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solutions disclosed in the present invention without the need for creative work should be included in the scope of protection of the present invention.

Claims

1. A syndrome prediction method integrating multi-graph convolution and ontology knowledge, characterized in that: include: Obtain clinical electronic medical record data and TCM knowledge base ontology data; The acquired clinical electronic medical record data and TCM knowledge base ontology data are processed using a pre-trained syndrome prediction model to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to integrate symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use a multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the predicted probability of each candidate syndrome through a multi-layer neural network, and use the probability-sorted syndrome sequence as the final predicted syndrome output result; wherein, after the embedding representation learning, the embedded representation of the symptom feature graph is recorded as The semantic embedding representation of symptoms is denoted as The symptom embedding representation in the symptom-syndrome diagram is denoted as The embedded representation of syndrome is recorded as The embedded representation of the syndrome diagram is denoted as The semantic embedding representation of syndrome is recorded as Using the maximum pooling technology, we can achieve the organic fusion of symptom and syndrome embedding representation: Obtain embedded representation features of symptoms and syndromes that integrate graph structure and semantic information; Embed the final symptoms into Multiplying the patient's original Multi-hot embedding representation P to obtain the patient's symptom embedding representation, denoted as Embedding patient symptoms into representations The patient's final symptom embedding representation is obtained through a multi-layer neural network At the same time, the syndrome is embedded in the expression After a multi-layer neural network, the final syndrome embedding representation is obtained Then, by embedding the patient's symptoms into Embedded with syndromes Perform multiplication calculations and pass the calculation results through the Sigmoid function to form the predicted probability value of each syndrome.

2. The syndrome prediction method integrating multi-graph convolution and ontology knowledge according to claim 1 is characterized in that: Establishing symptom maps, syndrome maps, and symptom-syndrome maps includes: calculating co-occurrence relationships and constructing co-occurrence networks based on clinical electronic medical record data, thereby obtaining symptom co-occurrence networks, symptom-syndrome co-occurrence networks, and syndrome co-occurrence networks; establishing associations between symptoms and between syndromes based on tree-shaped coding information of symptoms and syndromes, respectively, to form symptom ontology maps and syndrome ontology maps; fusing the symptom ontology map and the symptom co-occurrence map to form a symptom feature map, fusing the syndrome ontology map and the syndrome co-occurrence map to form a syndrome label map, and using the symptom-syndrome co-occurrence map as a feature-label map.

3. The syndrome prediction method integrating multi-graph convolution and ontology knowledge according to claim 2 is characterized in that: Taking the symptom feature graph, syndrome label graph, and symptom-syndrome graph as input, a multi-graph convolutional neural network model is used to learn the embedded representations of symptom features and syndrome labels on each graph respectively, including: for the embedded representation of the symptom feature graph, the node-initialized embedding matrix of the symptom feature graph and the adjacency matrix of the feature graph are used as the input of the graph neural network to form the (l+1)-th layer embedded representation; for the embedded representation of the syndrome label graph, the node-initialized embedding matrix of the syndrome label graph and the adjacency matrix of the feature graph are used as the input of the graph neural network to form the (l+1)-th layer embedded representation; for the embedded representation of the feature-label graph, the node-initialized embedding matrix of the feature-label graph and the adjacency matrix of the feature graph are used as the input of the graph neural network to form the (l+1)-th layer embedded representation.

4. The syndrome prediction method integrating multi-graph convolution and ontology knowledge according to claim 3 is characterized in that: For the embedding representation of the symptom feature map, the symptom feature map G Sym Initialize the node embedding matrix X Sym And the adjacency matrix A of the feature graph Sym As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula: in, is the degree matrix of the symptom feature map, is the weight matrix of the lth layer; through the learning of K convolutional layers, the symptom embedding feature representation of the symptom feature map is finally formed 5. The syndrome prediction method integrating multi-graph convolution and ontology knowledge according to claim 3 is characterized in that: For the embedding representation of syndrome label graph, the syndrome label graph G Syn Initialize the node embedding matrix X Syn And the adjacency matrix A of the feature graph Syn As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula: in, is the degree matrix of the syndrome label graph, is the weight matrix of the lth layer; through the learning of K convolutional layers, the syndrome embedding feature representation of the syndrome label graph is finally formed 6. The syndrome prediction method integrating multi-graph convolution and ontology knowledge according to claim 3 is characterized in that: For the embedding representation of the feature-label map, the feature-label map G S-S =(X S-S ,A S-S )’s node initialization embedding matrix X S-S And the adjacency matrix A of the feature graph S-S As the input of the graph neural network, the embedding representation of the (l+1)th layer is formed by the following formula: in, is the degree matrix of the feature-label graph, is the weight matrix of the lth layer; through the learning of K convolutional layers, the node embedding feature representation of the feature-label graph is finally formed That is, forming the embedding representation of all symptom nodes in the feature-label graph and the embedded representation of all syndrome nodes 7. The syndrome prediction method integrating multi-graph convolution and ontology knowledge according to claim 6, characterized in that: Calculate the loss function between the patient's true syndrome vector and the syndrome probability value predicted by the algorithm, and use this loss function to optimize the entire neural network model. The loss function is calculated as follows: Where N is the total number of samples; M is the total number of syndrome labels; y ij Indicates the true value of the jth syndrome label in the i-th sample. A value of 0 indicates that syndrome j is not the syndrome label of this sample, and a value of 1 indicates that syndrome j is the syndrome label of this sample. It represents the predicted probability value of the model for the jth syndrome label in the i-th sample.

8. A syndrome prediction system integrating multi-graph convolution and ontology knowledge, characterized by: include: Acquisition module, used to obtain clinical electronic medical record data and TCM knowledge base ontology data; The processing module is used to process the acquired clinical electronic medical record data and TCM knowledge base ontology data using a pre-trained syndrome prediction model to obtain syndrome prediction results; wherein, the syndrome prediction model includes a graph construction module, an embedding representation learning module, a feature fusion module and an intelligent dialectical module; the graph construction module is used to fuse symptoms with the symptom ontology knowledge in the TCM knowledge base ontology and the co-occurrence relationship between syndromes and symptoms in clinical records to establish symptom graphs, syndrome graphs and symptom-syndrome graphs; the embedding representation learning module is used to use the symptom feature graph, syndrome label graph and symptom-syndrome graph as input, and use the multi-graph convolutional neural network model to learn the symptom features and the embedded representation of syndrome labels on each graph; the feature fusion module is used to use the maximum pooling technology to realize the organic fusion of the embedded representation of symptoms and syndromes, respectively, and obtain the embedded representation features of symptoms and syndromes that integrate the graph structure and semantic information; the intelligent dialectical module is used to obtain the predicted probability of each candidate syndrome through a multi-layer neural network, and use the probability-sorted syndrome sequence as the final predicted syndrome output result; wherein, after the embedding representation learning, the embedded representation of the symptom feature graph is recorded as The semantic embedding representation of symptoms is denoted as The symptom embedding representation in the symptom-syndrome diagram is denoted as The embedded representation of syndrome is recorded as The embedded representation of the syndrome diagram is denoted as The semantic embedding representation of syndrome is recorded as Using the maximum pooling technology, we can achieve the organic fusion of symptom and syndrome embedding representation: Obtain embedded representation features of symptoms and syndromes that integrate graph structure and semantic information; Embed the final symptoms into Multiplying the patient's original Multi-hot embedding representation P to obtain the patient's symptom embedding representation, denoted as Embedding patient symptoms into representations The patient's final symptom embedding representation is obtained through a multi-layer neural network At the same time, the syndrome is embedded in the expression After a multi-layer neural network, the final syndrome embedding representation is obtained Then, by embedding the patient's symptoms into Embedded with syndromes Perform multiplication calculations and pass the calculation results through the Sigmoid function to form the predicted probability value of each syndrome.

Images