Data analysis methods and equipment

Abstract

The present disclosure provides a method and device for analyzing medical big data. The method comprises: receiving physical signs and symptoms represented by a vector h; locating the subspace where the patient is located in the medical big data using the physical signs and symptoms as features, wherein a matrix D is used to represent the case set in the big data, D=[D1,D2,……,D M ], D i Represents the i-th subspace, 1≤i≤M, and the positioning includes solving the formula #imgabs0#; and analyzing the probability P1 that the patient is in a specific subspace by analyzing the semantic consistency of the subspace. In addition, the probability P2 that the patient is in a specific node can be analyzed based on the evidence transfer score on the medical knowledge graph. The probability P that the patient is in a specific subspace or node can also be determined according to the following formula: #imgabs1#, where #imgabs2# is a reconciliation parameter, 0<#imgabs3#<1. Through these schemes, not only the accuracy of the analysis can be improved; but also the patient's condition can be analyzed according to the symptoms at the first time, so that the patient can be examined in a targeted manner, thereby reducing costs and improving efficiency.

Description

[0001] This application is a divisional application of the invention patent application with application date of January 19, 2017, application number 201710038539.X, and invention name “Data Analysis Method and Device”. Technical Field

[0002] The present disclosure relates to a data analysis method and device, and more specifically, to a data analysis method and device based on medical big data and medical knowledge graph. Background Art

[0003] The latest research from the Advisory Board Company shows that in the next decade, the cost of medical institutions for the general public will increase by 5% each year. Therefore, in order to survive and develop, medical institutions should find ways to reduce costs by 20%. An effective way to achieve this goal is to use big data analysis technology to detect major diseases as early as possible.

[0004] It is well known that if the warning signs of major diseases can be detected early, the treatment process will be much simpler and cheaper than if they are detected later, and the patient's recovery effect will be better. According to the EMC report, 22% of medical institutions use data analysis technology to improve the early detection rate of major diseases. Some medical institutions in the United States have saved considerable costs and improved the quality of patient care through data analysis: Meriter Health Services in Madison, Wisconsin, USA, deployed a business intelligence solution that integrates data from analytical systems and electronic health records (EHRs) to provide administrators and clinicians with a large amount of information that is instructive for practical work. The rich information provides accurate benchmark data for Meriter's plastic surgeons and provides a basis for doctors to choose implants that are more suitable for patients. With the help of this information, hospitals can use medical expenses more efficiently-Meriter saved nearly $1 million in just 8 months after using data analysis.

[0005] At present, the common practice of predicting diseases based on medical data is to extract features from the data, such as blood pressure values, blood sugar values ​​and other test indicators, and then train a prediction model (such as a logistic regression model, etc.) through machine learning methods, and finally use the model for prediction. However, this approach has two shortcomings: 1) It is necessary to establish a model for each disease, and when a patient visits a doctor, each model needs to be used for prediction, which is very inefficient; 2) Since the prediction of different diseases usually requires different test indicators, patients need to undergo some unnecessary tests, which increases medical costs. At the same time, this approach usually uses test indicators as features and cannot make full use of other useful information such as patients' self-reported symptoms. Summary of the invention

[0006] In order to solve or alleviate at least one of the above-mentioned defects in the prior art, it is desirable to provide a new data prediction and analysis method and device.

[0007] According to one aspect, a method for analyzing medical big data is provided, comprising:

[0008] Receiving physical signs and symptoms of the patient and representing the physical signs and symptoms with a vector h;

[0009] The subspace where the patient is located in the medical big data is located using the physical signs and symptoms as features, where the matrix D is used to represent the case set in the big data, D = [D1, D2, …, D M ], D i represents the i-th subspace, 1≤i≤M, and the positioning includes the formula Solve, X is a coefficient vector used to represent the distribution of vector h in each subspace; and

[0010] The probability P1 that the patient is in a specific subspace is analyzed by analyzing the semantic consistency of the subspace where the sign and symptom vector h is located.

[0011] In one embodiment, locating the subspace of the patient in the medical big data based on the physical signs and symptoms further includes:

[0012] By using the sparse solution method, the formula Solving to determine the subspace in which the patient is located; and

[0013] Based on the value of each element in the coefficient vector X, it is determined in which subspace the largest portion of h is located, thereby determining in which subspace the probability of being located is the highest.

[0014] In one embodiment, the sparse solution includes:

[0015] Solution , where X satisfies ,in is the L1 paradigm, is the L2 paradigm,

[0016] The solution obtained x* The cases corresponding to the dimensions whose median values ​​are not zero constitute the subspace where h is located.

[0017] In one embodiment, analyzing the semantic consistency of the subspace where the sign and symptom vector h is located further includes:

[0018] The semantic subspace and subspace D where h is located are calculated by the following formula: i Consistency:

[0019] ,

[0020] in , , is the error, Indicates that the coefficient vector X belongs to the subspace D i The dimension of is 1, and the rest of the dimensions are 0.

[0021] In one embodiment, the above method further comprises:

[0022] Output C1, C2, ..., C M , The maximum value among is taken as the probability P1 that the patient is in the corresponding subspace.

[0023] According to another aspect, a medical data analysis method based on a medical knowledge graph is provided, comprising:

[0024] Receive the patient's signs and symptoms;

[0025] Access the medical knowledge graph to obtain a partial medical knowledge graph related to the patient, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N} and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i Initial evidence score;

[0026] Based on the evidence transfer of the patient's physical signs and symptoms on the partial medical knowledge graph, the final evidence score p of each node is determined. t ;as well as

[0027] Based on the final evidence score p t The scores of each node in the is used to analyze the probability P2 that the patient is at a specific node.

[0028] In one embodiment, the final evidence score p of each node is determined t Further including:

[0029] The final evidence score of each node is determined by iterating through the following formula:

[0030] ,

[0031] Where d is the damping coefficient, 0 <d<1; , , w i,jDenote the edges e connecting each node v in V i and v j with weight w i,j .

[0032] In one embodiment, the termination condition of the iterative operation is that p t no longer changes or reaches the maximum number of iterations in the iterative operation.

[0033] In one embodiment, determining the final evidence score p of each node t further includes:[[]]

[0034] Determining the final evidence score of each node through the following formula:[[]]

[0035] ,

[0036] where d is the damping factor, 0 < d < 1; I is the identity matrix of N N, , , w i,j Denote the edges e connecting each node v in V i and v j with weight w i,j .

[0037] In one embodiment, analyzing the probability P2 that the patient is in a specific node based on the scores of each node in the final evidence score p t further includes:[[]]

[0038] Determining the probability of being in each node by calculating the percentage of the final score of each disease node in V in the sum of the final scores of all disease nodes in V; and

[0039] Outputting the maximum probability among the probabilities as the probability P2 that the patient is in a specific node.

[0040] According to another aspect, there is provided a medical data analysis method based on a medical knowledge graph, including:[[]]

[0041] Receiving the physical signs and symptoms of the patient;

[0042] Accessing the medical knowledge graph to obtain a partial medical knowledge graph related to the physical signs and symptoms of the patient, the partial medical knowledge graph including a plurality of nodes V = {v1, v 2, …, v N} and the initial evidence score of each node , where , each node v i represents a symptom or a related disease of the patient, represents the node v iThe initial evidence score of the node, the edge between the nodes reflects the semantic relevance between the nodes;

[0043] The final evidence score p of each node is determined based on the evidence propagation of the patient's physical signs and symptoms on the partial medical knowledge graph. t , wherein the evidence propagation of the portion of the medical knowledge graph includes a greater evidence score propagated from one node to another node with a higher semantic relevance than a greater evidence score propagated to another node with a lower semantic relevance; and

[0044] Based on the final evidence score p t The scores of each node in the is used to analyze the probability P2 that the patient is in a specific disease node.

[0045] According to another aspect, a medical data analysis method is provided, comprising:

[0046] Receive the patient's signs and symptoms;

[0047] Using the physical signs and symptoms as characteristics to locate the subspace where the patient is located in the medical big data;

[0048] The probability P1 that the patient is in a specific subspace is analyzed by analyzing the semantic consistency of the subspace where the patient's physical signs and symptoms are located;

[0049] Analyze the probability P2 that the patient is at a specific node based on the evidence transfer score of the patient's physical signs and symptoms on the medical knowledge graph; and

[0050] The probability P of the patient being in a specific subspace or node is determined based on the probability P1 and the probability P2:

[0051] ,in is the harmonic parameter, 0< <1.

[0052] According to another aspect, a medical data analysis method is provided, comprising:

[0053] Receive the patient's signs and symptoms;

[0054] Use vector h to represent the signs and symptoms;

[0055] The subspace where the patient is located in the medical big data is located using the physical signs and symptoms as features, where the case set in the big data is represented by a matrix D and the matrix D constitutes the semantic space of known cases, D = [D1, D2, …, D M ], wherein the subspace is a subspace of the semantic space consisting of known cases of a disease, and D irepresents the i-th subspace consisting of known cases of the i-th disease, 1≤i≤M, and the positioning includes the formula Solve, where X is the coefficient vector to be solved, which is used to represent the distribution of vector h in each subspace, and the cases corresponding to the dimensions whose values ​​in X are not zero constitute the subspace where vector h is located;

[0056] The probability P1 that the patient is in a specific subspace is analyzed by analyzing the consistency between the subspace where the sign and symptom vector h is located and each subspace;

[0057] Access the medical knowledge graph to obtain a partial medical knowledge graph related to the patient's physical symptoms, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N} and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i The initial evidence score of the node, the edge between the nodes reflects the semantic relevance between the nodes;

[0058] The final evidence score p of each node is determined based on the evidence propagation of the patient's physical signs and symptoms on the partial medical knowledge graph. t , wherein the evidence propagation of the portion of the medical knowledge graph includes a greater evidence score propagated from one node to another node with a higher semantic relevance than a greater evidence score propagated to another node with a lower semantic relevance; and

[0059] Based on the final evidence score p t The probability P2 that the patient is at a specific disease node is analyzed by using the scores of each node in ; and

[0060] Determine the probability P of a patient belonging to a specific disease based on probability P1 and probability P2:

[0061] ,in is the harmonic parameter, 0< <1.

[0062] According to another aspect, a medical big data analysis device is provided, comprising:

[0063] A receiving unit configured to receive physical signs and symptoms of a patient and represent the physical signs and symptoms with a vector h;

[0064] The subspace positioning unit is configured to locate the subspace of the patient in the medical big data based on the physical signs and symptoms, wherein the case set in the big data is represented by a matrix D, D = [D1, D2, …, D M ], D i represents the i-th subspace, 1≤i≤M, and the positioning includes the formula Solve, X is a coefficient vector used to represent the distribution of vector h in each subspace; and

[0065] The first analysis unit is configured to analyze the probability P1 that the patient is in a specific subspace by analyzing the semantic consistency of the subspace where the sign and symptom vector h is located.

[0066] In one embodiment, the subspace positioning unit is further configured to:

[0067] By using the sparse solution method, the formula Solving to determine the subspace in which the patient is located; and

[0068] Based on the value of each element in the coefficient vector X, it is determined in which subspace the largest portion of h is located, thereby determining in which subspace the probability of being located is the highest.

[0069] In one embodiment, the sparse solution includes:

[0070] Solution , where X satisfies ,in is the L1 paradigm, is the L2 paradigm,

[0071] The solution obtained x* The cases corresponding to the dimensions whose median values ​​are not zero constitute the subspace where h is located.

[0072] In one embodiment, the first analysis unit is further configured to:

[0073] The semantic subspace and subspace D where h is located are calculated by the following formula: i Consistency:

[0074] ,

[0075] in , , is the error, Indicates that the coefficient vector X belongs to the subspace D i The dimension of is 1, and the rest of the dimensions are 0.

[0076] In one embodiment, the above device further comprises:

[0077] The first output unit is configured to output C1, C2, ..., C M , The maximum value among is taken as the probability P1 that the patient is in the corresponding subspace.

[0078] According to another aspect, a medical data analysis device based on a medical knowledge graph is provided, comprising:

[0079] a receiving unit configured to receive physical signs and symptoms of the patient;

[0080] An access unit is configured to access the medical knowledge graph to obtain a partial medical knowledge graph related to the patient, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N} and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i Initial evidence score;

[0081] The second analysis unit is configured to determine the final evidence score p of each node based on the evidence transfer score of the patient's physical signs and symptoms on the partial medical knowledge graph. t ;as well as

[0082] The determination unit is configured to determine the final evidence score p t The scores of each node in the is used to analyze the probability P2 that the patient is at a specific node.

[0083] According to another aspect, a medical data analysis device based on a medical knowledge graph is provided, comprising:

[0084] a receiving unit configured to receive physical signs and symptoms of the patient;

[0085] An access unit is configured to access a medical knowledge graph to obtain a partial medical knowledge graph related to the patient's physical symptoms, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N} and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i The initial evidence score of the node, the edge between the nodes reflects the semantic relevance between the nodes;

[0086] A second analysis unit, configured to determine the final evidence score p of each node based on the propagation of evidence of the patient's signs and symptoms on the partial medical knowledge graph t , where the propagation of evidence of the partial medical knowledge graph includes that the evidence score propagated from one node to another node with a high semantic relevance is greater than the evidence score propagated to another node with a low semantic relevance; and

[0087] A determination unit, configured to analyze the probability P2 that the patient is at a specific disease node based on the scores of each node in t p.

[0088] In one embodiment, the second analysis unit further includes:

[0089] A calculation unit, configured to perform iterative operations through the following formula to determine the final evidence score of each node:

[0090] ,

[0091] where d is a damping factor, 0 < d < 1; , , w i,j represents the edge e i connecting each node v j and v i,j in V.

[0092] In one embodiment, the termination condition of the iterative operation is that p t no longer changes or reaches the maximum number of iterations in the iterative operation.

[0093] In one embodiment, the second analysis unit further includes:

[0094] A calculation unit, configured to determine the final evidence score of each node through the following formula:

[0095] ,

[0096] where d is a damping factor, 0 < d < 1; I is the identity matrix of N N, , , w i,j represents the edge e i connecting each node v j and v i,j in V.

[0097] In one embodiment, the determination unit is further configured to:

[0098] Determine the probability of being at each node by calculating the percentage of the final score of each disease node in V to the sum of the final scores of all disease nodes in V; and

[0099] The maximum probability among the probabilities is output as the probability P2 of being at the corresponding node.

[0100] According to another aspect, there is provided a medical data analysis device, comprising:

[0101] a receiving unit configured to receive physical signs and symptoms of the patient;

[0102] A subspace positioning unit, configured to locate the subspace where the patient is located in the medical big data based on the physical signs and symptoms;

[0103] A first analysis unit is configured to analyze the probability P1 of being in a specific subspace by analyzing the semantic consistency of the subspace where the patient's physical signs and symptoms are located;

[0104] A second analysis unit is configured to analyze the probability P2 that the patient is at a specific node based on the evidence transfer score of the patient's physical signs and symptoms on the medical knowledge graph; and

[0105] A reconciliation unit is configured to determine a probability P that the patient is in a specific subspace or node based on probability P1 and probability P2:

[0106] ,in is the harmonic parameter, 0< <1

[0107] According to another aspect, a medical big data analysis device is provided, comprising:

[0108] a memory configured to store computer-executable instructions; and

[0109] A processor coupled to the memory is configured to execute the computer executable instructions so that the processor performs any one of the methods described above.

[0110] According to another aspect, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the instructions are executed by a computing device, the computing device will perform any one of the methods described above.

[0111] Some embodiments of the present disclosure propose to use medical big data to analyze the patient's condition. The scheme uses the patient's physical signs and symptoms as features, and can find the patient's subspace in the medical big data based on the patient's symptoms, and analyze the patient's condition through the semantic consistency of the subspace. Compared with the prior art, the scheme does not need to establish a model for each disease, and has high prediction efficiency. At the same time, by using the patient's physiological phenomena (symptoms) for analysis, the patient's condition can be analyzed based on the symptoms at the first time, allowing the patient to be examined in a targeted manner, thereby reducing costs and improving efficiency.

[0112] Other embodiments propose methods and devices for analyzing diseases using medical knowledge graphs. The method and device can analyze the patient's condition by transmitting relevant evidence in the medical knowledge graph through the symptoms of the disease. The knowledge graph is established based on medical knowledge. The embodiments can use explicit knowledge or information in the knowledge graph to analyze the disease, analyze the patient's condition based on symptoms at the first time, and allow the patient to undergo targeted examinations, thereby reducing costs and improving efficiency.

[0113] Some other embodiments propose methods and devices for comprehensively analyzing diseases using medical big data and medical knowledge graphs. The method and device use the patient's physical signs and symptoms as features. On the one hand, it can find the subspace where the patient is located in the medical big data based on the patient's symptoms, and analyze the patient's illness through the semantic consistency of the subspace. On the other hand, it can also transmit relevant evidence in the medical knowledge graph through the symptoms of the disease and then analyze the patient's illness. Finally, it can also analyze the information from both aspects and output the final conclusion or result. Compared with previous practices, the embodiment can comprehensively utilize the implicit laws, knowledge or information in large-scale medical big data and the explicit knowledge or information in the knowledge graph for disease analysis, thereby improving the accuracy of the analysis; at the same time, the patient's physiological phenomena (symptoms) are used for analysis, and the patient's condition can be analyzed based on the symptoms at the first time, allowing the patient to conduct targeted examinations, thereby reducing costs and improving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments. It should be appreciated that the drawings described below only relate to some embodiments, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, and the other drawings are also within the scope of the present invention.

[0115] Figure 1 is a flowchart of a data analysis method based on medical big data and medical knowledge graph according to one embodiment;

[0116] Figure 2 A schematic diagram showing several subspace examples within a big data space;

[0117] Figure 3 shows an example of graph-based evidence propagation in a medical knowledge graph;

[0118] Figure 4 A schematic diagram showing a simplified example of a medical knowledge graph;

[0119] Figure 5 is a schematic structural diagram of a data analysis device according to an embodiment;

[0120] Figure 6 An example computing device is illustrated that may be used to implement one or more embodiments. DETAILED DESCRIPTION

[0121] The following will be combined with the accompanying drawings to clearly and completely describe the technical solutions of some embodiments of the present disclosure, so that the purposes, technical solutions and advantages of some embodiments can be more clearly understood. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments in the present disclosure, a person of ordinary skill in the art can obtain other embodiments, and all of these obtained embodiments belong to the scope of protection of the present invention. It should be pointed out that although this article uses medical big data and medical knowledge graphs as examples for explanation, those skilled in the art know that the present invention can also be applied to the analysis of other types of big data and knowledge graphs, so as to determine which subspace or type of a certain object belongs to in the big data or which node or category in the knowledge graph. It should also be pointed out that, as pointed out above, the present invention is not limited to the medical field, so all the methods provided in the present disclosure are not methods directly used for the diagnosis and treatment of diseases.

[0122] Figure 1 FIG. 1 is a flow chart of a data analysis method based on medical big data and medical knowledge graph according to an embodiment. Figure 1 As shown, the embodiment comprehensively utilizes implicit knowledge in medical big data and explicit knowledge in medical knowledge graph to analyze diseases. Specifically, the method includes three parts: analysis based on medical big data ( Figure 1 The part contained in the dotted box in the middle), analysis based on medical knowledge graph ( Figure 1 The part contained in the right dotted box) and the output of the first two parts of information ( Figure 1 Step S8 in (1): Final result output). It should be noted that although Figure 1The illustrated embodiment includes all three parts, but those skilled in the art will appreciate that the two parts, analysis based on medical big data and analysis based on medical knowledge graphs, can be implemented separately as independent technical solutions.

[0123] The following will refer to Figure 1 Let's introduce the principles and specific implementation methods of these three parts in detail.

[0124] Analysis based on medical big data

[0125] This section will introduce in detail the basic idea and specific implementation of the medical big data analysis module. The basic idea of ​​this module is that patients with similar symptoms may suffer from the same disease; patients with the same disease are very likely to have similar characteristics. This is consistent with the real world. For example, in 1998, Lloyd Minor, Dean of the Stanford University School of Medicine, and his colleagues reported a rare disease for the first time in the world - "superior semicircular canal rupture syndrome". Patients with this disease will experience symptoms such as dizziness and abnormal sensitivity to sound. This was originally a very common academic discovery, but many patients in the world who could not find the cause of the disease for many years or who struggled to try treatment plans in other departments were finally diagnosed and treated by searching for relevant symptom information.

[0126] This part includes 4 steps: S1. Receive the patient's physical signs and symptoms; S2. Search the subspace where the patient is located in the medical big data; S3. Analyze the semantic consistency of the subspace; S4. Output the analysis result P1 based on medical big data, that is, the probability P1 that the patient is in a specific subspace.

[0127] First, in step S1, the patient's physical signs and symptoms are received. The main function of this step is to collect the patient's self-reported symptoms during the consultation process, such as dizziness, headache, etc.

[0128] Then, the method proceeds to step S2, searching the subspace where the patient is located in the medical big data. The main function of this step is to find the subspace where the patient is located in the medical big data based on the collected symptoms. Since this embodiment is based on medical big data for disease analysis, a large number of cases are required, such as confirmed cases in various hospitals over the years, which corresponds to Figure 1 The “Case Big Data” module in .

[0129] This disclosure uses the notation D = [D1, D2, ..., D M ] to represent the case set. Assuming that it contains M diseases, then D i (1≤i≤M) represents the i-th disease, where D i It can be expressed as D i = [D i,1 , D i,2, ……,D i,K ], which means that the i-th disease contains K cases, D i,j (1≤i≤M, 1≤j≤K) represents the jth case of the i-th disease. Each case consists of a series of corresponding feature vectors (such as symptoms), so the matrix D constitutes a semantic space of confirmed cases, and D i (1≤i≤M) constitutes a subspace within the semantic space. It should be pointed out that although medical big data is used as an example for explanation here, the present invention can also be applied to the analysis of other types of big data, so as to determine which subspace or category within the big data an object belongs to. In a more general example, big data may include multiple subspaces, each representing a plurality of categories, each category may have corresponding instances, and each instance may have multiple features. In this way, through the embodiments provided by the present invention, the features of an input object can be used to locate or search the subspace where the object is located, and the principle is exactly the same as locating the subspace where the patient is located.

[0130] For a new patient h, assume that he has disease D i According to the basic idea of ​​this module: patients with the same disease are very likely to have similar symptoms, that is, the new patient h can be expressed as D i The linear combination of cases in

[0131]

[0132] Among them, α i,j (1≤j≤K) is the correlation coefficient. For example, for the disease "hypertension", the symptoms in case 1 are "dizziness, nausea, palpitations and shortness of breath", the symptoms in case 2 are "palpitations and shortness of breath, tinnitus, and numbness of limbs", and the symptoms in case 3 are "dizziness, nausea, tinnitus, palpitations and shortness of breath", and the symptoms of the new patient are "palpitations and shortness of breath, numbness of limbs", then "new patient = case 1 + case 2 - case 3".

[0133] As a more general explanation, we collect all the symptoms in the case set D and set them to set S. Let the number of symptoms in S be |S|. We use a column vector of |S| dimension to represent a case. For example, for cases 1, 2, and 3, S = {dizziness, nausea, palpitations, shortness of breath, tinnitus, numbness of limbs} T , S contains a total of 5 symptoms. We set dizziness as the first dimension in the vector, and so on, limb numbness is the fifth, so different symptoms in the case correspond to different dimensions of the vector. The dimension value corresponding to the symptom that the patient has is set to 1, and the dimension value corresponding to the symptom that does not appear is set to 0. So case 1=[1,1,1,0,0] T , case 2 = [0,0,1,1,1] T, case 3 = [1,1,1,1,0] T , where the superscript T represents the transpose of the matrix. New patient h=[0,0,1,0,1] T = [1,1,1,0,0] T + [0,0,1,1,1] T - [1,1,1,1,0] T That is, "new patient = case 1 + case 2 - case 3". Assuming that the disease "hypertension" corresponds to the i-th disease, then D i,1 =[1,1,1,0,0] T ;D i,2 =[0,0,1,1,1] T , D i,3 =[1,1,1,1,0] T h = D i,1 + D i,2 -D i,3 In this example, the case vector is defined as a column vector, but those skilled in the art will appreciate that the case vector may also be a row vector, but accordingly, the dimensions of the case set and other vectors should also be transposed accordingly.

[0134] For simplicity and convenience, the above expression can be expressed in a matrix. Assume D i = [D i,1 , D i,2 ,……, D i,K ], x i =[α i,1 , α i,2 , ..., α i,K ] T , where the superscript T represents the transpose of the matrix, then

[0135]

[0136] From the above discussion, we can see that each disease type can be represented as a subspace composed of the known cases it contains, and a case belonging to this disease type can be composed of a linear combination of the basis of the corresponding subspace. Figure 2 A schematic diagram showing several subspace examples within a big data space;

[0137] Based on the above discussion, given a case set D, we can determine the disease of patient h by finding the subspace of patient h in D. Let D = [D1, D2, …, D M ], then the subspace where patient h is located can be obtained by formula (1),

[0138] (1)

[0139] Where D = [D1, D2, …, D M ]=[ D 1,1 ,D 1,2 , …,D i,1 , D i,2 , ……, D M,K ] is a matrix containing all the diseases in D i (1≤i≤M), where each element D i,j (1≤i≤M, 1≤j≤K) corresponds to a case in D, that is, the jth case of the i-th disease, and each case is a column vector.

[0140] Where X is a column vector, X = [x1 T , x2 T , …,x i T ,…, x M T ] T , its dimension is not M, but the number of cases in D, that is, the dimension is the same as [D 1,1 ,D 1,2 , …,D i,1 , D i,2 , ……, D M,K ] For example, suppose D contains two diseases, disease 1 contains a cases, and disease 2 contains b cases, then the dimension of X is a+b.

[0141] However, since the space of D is very large, there are usually many X that can satisfy equation (1), that is, there are multiple subspaces that can reconstruct h. In order to solve this problem, the present invention adopts a sparse solution method (using the least number of cases to reconstruct h). The advantage of using a sparse solution method is that it can reduce the impact of "noise" data, making the model have good robustness. The specific solution is as follows:

[0142] (2)

[0143] in is the L1 normal form, which is used to find the sum of the absolute values ​​of all elements. is the L2 norm, which is used to find the sum of squares of all elements. ε is a given parameter. X=[α 1,1 , α 1,2 , ..., α i,K , …] T is the coefficient to be solved. The x obtained by solving equation (2) is *The cases corresponding to the dimensions whose median values ​​are not zero constitute the subspace where h is located. In this way, a column vector X is required, which makes the sum of the absolute values ​​of the elements in X the smallest, and each element in the vector obtained by the matrix product DX is very close to the patient's symptoms. i,j The specific value of is solved by using the stochastic gradient descent method through equation (2).

[0144] For the above example, the new patient h = [0,0,1,0,1] T = [1,1,1,0,0] T + [0,0,1,1,1] T - [1,1,1,1,0] T This is an ideal example. In reality, due to the limitation of computational accuracy, α i,j It is not the optimal solution. For example, in this example, α 1,1 =0.8, so we have 0.8*[1,1,1,0,0] T + [0,0,1,1,1] T -[1,1,1,1,0] T = [0.2,0.2,0.8,0,1]. At this time, Dx-h=[0.2,0.2,0.8,0,1] T -[0,0,1,0,1] T = [0.2,0.2,-0.2,0,0] T ,but If we set =0.2, then Still holds true. This is to reduce the impact of "noise". i,j When they are not equal, the operation can still be performed. Moreover, when solving the problem through formula (2), theoretically, α is not required. i,j are equal, they are just a weight of the vector.

[0145] Then, the method proceeds to step S3 to perform subspace semantic consistency analysis. The main function of this step is to analyze the probability of h being in a specific subspace by analyzing the semantic consistency of the subspace where h is located. Indicates that the coefficient vector X belongs to the subspace D i A column vector with a dimension of 1 and other dimensions of 0, whose dimension is the number of all cases in D, that is, the dimension and [D 1,1 ,D 1,2 , …,D i,1 , D i,2 , ……, D M,K ] is the same. Then the corresponding subspace D in vector hi The semantic components of , then h can be expressed as ,in Based on the above representation, the present invention converts the semantic subspace where h is located into the subspace D i The consistency of is defined as:

[0146] (3)

[0147] in, is the square of the L2 norm.

[0148] After performing the subspace semantic consistency analysis in step S3, the result based on the analysis of medical big data is output in step S4. That is, output C1, C2, ..., C M , The maximum value in is taken as the probability P1 that the patient is in the subspace corresponding to the maximum value. At the same time, the subspace or category corresponding to the maximum value is finally determined as the subspace or category where the patient is located.

[0149] The main function of step S4 is to output the analysis results based on medical big data. , we can see from the definition of formula (3) , C i It reflects that h is in the subspace D i The size of the possibility, that is, the corresponding Figure 1 The probability P1 in which This reflects that h is not in any of the previous subspaces D1-D M This is because C i The larger the value, the more it indicates that the vectors constituting h contain vectors belonging to the subspace D. i The more cases there are, that is, h is located in D i The more parts of the subspace, the more they belong to the subspace D i For example, Figure 2 In the example, it is assumed that there are three subspaces in the known case space, corresponding to different shapes: four-pointed star nodes, triangle nodes, and hexagonal star nodes, corresponding to the first three dimensions of C, and the last dimension is the error . Round nodes represent newly arrived patients. Figure 2The two circles shown represent two linear combinations used to represent new patients. The first one can represent new patients using only the subspace represented by the four-pointed star node, while the second one requires all three subspaces to represent new patients. It can be clearly seen that for the left graph, C=[1,0,0,0], that is, the patient may be in the subspace represented by the four-pointed star. For the right graph, C=[0.25, 0.375, 0.375, 0], it is difficult to analyze which subspace the patient is in or which category he belongs to.

[0150] Analysis based on medical knowledge graph

[0151] Next, let’s introduce Figure 1 The second part of the flowchart shown is the analysis based on the medical knowledge graph. This section will introduce the basic idea and specific implementation of the analysis module based on the medical knowledge graph. Figure 3 An example of graph-based evidence propagation in a medical knowledge graph is shown. In a medical knowledge graph, nodes represent diseases or symptoms (in more general examples, they represent categories and features), and the edges between nodes reflect the semantic relevance between nodes. The basic idea of ​​analysis based on medical knowledge graphs is that the evidence score obtained by propagating symptoms (or diseases) to diseases (or symptoms) with high semantic relevance is greater than that of diseases (or symptoms) with low semantic relevance. For example, Figure 3 As shown in the figure, assuming that the weight score of "migraine without aura" is 1, it can spread 0.7 evidence scores to the symptom "nausea" and 0.3 evidence scores to the symptom "photophobia". Intuitively explained, the probability of migraine without aura patients experiencing nausea symptoms is 70%, and the probability of photophobia symptoms is 30%, that is, the cases of nausea symptoms in migraine without aura patients are more than the cases of photophobia symptoms. The probability of "rabies" patients experiencing photophobia symptoms is 80%.

[0152] It should be noted that the present invention can also be applied to other knowledge graphs with similar structures to the medical knowledge graph to analyze and determine the node or category to which an object belongs. The principle is exactly the same as that of the solution based on the medical knowledge graph.

[0153] The following will refer to Figure 1 The flowchart of Figure 1 details the graph-based evidence propagation method.

[0154] In step S5, the method performs graph-based evidence propagation. That is, the probability P2 of which disease node the patient is in is analyzed based on the evidence transfer score on the medical knowledge graph. The purpose of this step is to analyze the patient's condition through the patient's symptoms combined with the explicit medical knowledge in the medical knowledge graph. Specifically, given the initial evidence score of the symptom, the evidence score is propagated in the medical knowledge graph according to the basic idea that "the evidence score of the symptom (or disease) propagated to the disease (or symptom) with high semantic relevance is greater than that of the disease (or symptom) with low semantic relevance", until the evidence scores of all nodes no longer change or change very little.

[0155] In one embodiment, assume that V = {v1, v 2, …, v N} represents the set of vertices or nodes in the medical knowledge graph, E={…, e i,j , …} represents the set of edges between nodes, where e i,j Represents node v i and v j The edges between them, where 1≤i≤N, 1≤j≤N, i≠j. W is the set of edge weights, where w i,j Represents edge e i,j The weight on the node. Assume that the initial evidence score of the node is ,in , Represents node v i For example, for Figure 3 For example, assuming that the patient has symptoms of "nausea" and "photophobia", the initial evidence score can be set to p0["nausea", "photophobia", "migraine without aura", "rabies"] = [0.5, 0.5, 0, 0], that is, in the absence of any prior knowledge, the distribution that meets the conditions and has the largest entropy is used as the initial evidence score.

[0156] set up is the evidence score of each node after t iterations. Based on the basic idea that "the evidence score of a symptom (or disease) propagating to a disease (or symptom) with high semantic relevance is greater than that of a disease (or symptom) with low semantic relevance", the evidence score of each node after the t+1th iteration is:

[0157] (4)

[0158] Where d is the damping coefficient (0 <d<1); M (i) represents the node v i A collection of connected nodes. For example, Figure 3In the example in , assume that the initial evidence score of each node is p0[“nausea”, “photophobia”, “migraine without aura”, “rabies”] = [0.5, 0.5, 0, 0], and the edge weight is Figure 3 As shown, after one round of iteration, the evidence score of "nausea" is (1-d) 0.5+0 0.7=0.1 (assuming d=0.8), the evidence score for "photophobia" is 0.2 0.5+0 0.8=0.1, the evidence score for “migraine without aura” is , the score of "rabies" can be calculated by the same logic Then keep iterating formula (4) until the termination condition is reached, that is, p t No more changes occur or the maximum number of iterations is reached.

[0159] For simplicity and efficiency, formula (4) can be transformed into a matrix representation as follows:

[0160] (5)

[0161] According to Markov theory, It must exist. Therefore, taking the limit on both sides of formula (5) we have

[0162] (6)

[0163] Where I is N The identity matrix of N, p i The value of node v i The final evidence score of each node can be directly calculated according to formula (6). For example, Figure 3 The final evidence score of each node in is p[“nausea”, “photophobia”, “migraine without aura”, “rabies”] = [0.25, 0.30, 0.27, 0.18]. The higher the final evidence score of a disease node, the greater the probability that the patient is in the node. Therefore, the score of the disease node can be normalized to obtain the probability that the patient is in the node, corresponding to Figure 1 For example, Figure 3 In the example, the probability that the patient is in the node "migraine without aura" is , the probability of being at the node "rabies" is .

[0164] Figure 4The diagram shows a simplified example of a medical knowledge graph. The large circle in the middle represents a category in the knowledge graph, in this case a disease, and the nodes directly connected to it represent the relationship between the category and other features, such as causes, symptoms, and treatments. The outermost circles represent the corresponding features, which in this case are symptoms, causes, and treatments. For the sake of simplicity, Figure 4 The edge weights are omitted.

[0165] The method then proceeds to step S7, outputting the probability P2 of the patient being in a certain disease node or belonging to a certain category obtained based on the medical knowledge graph analysis.

[0166] Finally, the method proceeds to step S8 and outputs the final result. That is, the probability that the patient belongs to a certain category is determined based on P1 obtained from step S4 and P2 obtained from step S7. The function of this step is to give the final analysis result by combining the analysis based on medical big data and the analysis based on the medical knowledge graph. Specifically, the scores of the two are combined using a linear weighting method. Specifically, the probability that the patient belongs to a certain category is calculated or determined by the following formula:

[0167] ,

[0168] in is the harmonic parameter, 0< <1, used to adjust the weight of the two analysis methods. When the data and quality of medical big data are high, and the accuracy of the analysis based on medical big data is high, Increase (for example, = 0.7, etc.); On the contrary, when the amount of medical big data is insufficient or the quality is not high, and the analysis accuracy based on medical big data is not high, we should make full use of the medical knowledge in the medical knowledge graph for analysis. Lower (if setting =0.2, etc.).

[0169] Figure 5 FIG. 1 is a schematic diagram of a data analysis device according to an embodiment. In this example, the data analysis device performs analysis based on medical big data and medical knowledge graphs. Figure 1 similar, Figure 5The data analysis device 500 shown also includes three parts, a big data analysis device (as shown in the dotted box on the left), a medical data analysis device (the part included in the upper right dotted box plus the receiving unit 510), and a reconciliation part for reconciling the output results of the above two parts (as shown in the lower right dotted box). Those skilled in the art will understand that the big data analysis device and the medical data analysis device can be implemented independently as independent devices. It should also be pointed out that the big data analysis device can also be used to analyze other big data to determine the subspace or category of an object in the big data space, and the medical data analysis device can also analyze other types of knowledge graphs with structures similar to medical knowledge graphs to determine the category node or category of an object in the knowledge graph.

[0170] like Figure 5 As shown, the big data analysis device may include: a receiving unit 510, a subspace positioning unit 520, a first analysis unit 530, and an optional first output unit 540. The receiving unit 510 may be configured to receive the patient's physical signs and symptoms and represent the physical signs and symptoms with a vector h. The subspace positioning unit 520 may be configured to locate the subspace where the patient is located in the medical big data based on the physical signs and symptoms. The matrix D is used to represent the case set in the big data, D = [D1, D2, ..., D M ], D i represents the i-th subspace, 1≤i≤M. The positioning may include the following formula Solve, X is a coefficient vector, which is used to represent the distribution of vector h in each subspace. The first analysis unit 530 can be configured to analyze the probability P1 that the patient is in a specific subspace or category by analyzing the semantic consistency of the subspace where the sign and symptom vector h is located.

[0171] In one embodiment, the subspace positioning unit 520 may be further configured to determine the subspace where the patient is located by solving the following formula using a sparse solution method: The subspace positioning unit 520 may be further configured to determine, based on the value of each element in the coefficient vector X, in which subspace the largest portion of h is located, thereby determining in which subspace the probability of being located is the largest.

[0172] In one embodiment, the sparse solution method includes solving , where X satisfies ,in is the L1 paradigm, is the L2 norm. The solution obtained is x* The cases corresponding to the dimensions whose median values ​​are not zero constitute the subspace where h is located.

[0173] In one embodiment, the first analysis unit 520 may be further configured to calculate the semantic subspace where h is located and the subspace D by the following formula: i Consistency:

[0174] ,

[0175] in , , is the error, Indicates that the coefficient vector X belongs to the subspace D i A column vector with one dimension of 1 and the remaining dimensions of 0.

[0176] In one embodiment, the optional first output unit 540 may be configured to output C1, C2, ..., C M , The maximum value among is taken as the probability P1 that the patient is in the corresponding subspace or belongs to the corresponding category.

[0177] The medical data analysis device may include a receiving unit 510, an access unit 550, a second analysis unit 560, and a determination unit 570. The access unit 550 may be configured to access a medical knowledge graph to obtain a partial medical knowledge graph related to a patient, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N} and the initial evidence score of each node ,in , each node v i represents a symptom of a patient or a related disease (or in more general cases a characteristic or a class), Represents node v i The second analysis unit 560 may be configured to determine the final evidence score p of each node based on the evidence transfer score of the patient's physical signs and symptoms on the partial medical knowledge graph. t The determination unit 570 may be configured to determine the final evidence score p t The probability P2 of the patient being in a specific node or category is analyzed by the scores of each node in the second analysis unit 550. In one embodiment, the second analysis unit 550 may further include a calculation unit. The calculation unit may be configured to perform iterative operations through the following formula to determine the final evidence score of each node:

[0178] ,

[0179] Where d is the damping coefficient, 0 <d<1; , , wi,j denote the edges e i connecting nodes v j and v i,j in V. In one embodiment, the termination condition of the iterative operation is that p t no longer changes or the maximum number of iterations is reached in the iterative operation..

[0180] In another embodiment, the calculation unit may be configured to determine the final evidence score of each node through the following formula:

[0181] ,

[0182] where d is the damping factor, 0 < d < 1; I is the identity matrix of N N, , , w i,j denote the edges e i connecting nodes v j and v i,j in V.

[0183] In one embodiment, the determination unit 570 may be further configured to: calculate the percentage of the final score of each disease node in V in the sum of the final scores of all disease nodes in V to obtain the probability of each node; and output the maximum probability among the probabilities as the probability P2 of being in the corresponding node or category. Accordingly, the node and category are determined as the node or category to which the patient belongs.

[0184] The reconciliation part may include a reconciliation unit 580 and a final result output unit 590. The reconciliation unit 580 may be configured to determine the probability P of the patient being in a specific subspace or node or category based on the probability P1 and the probability P2:

[0185] ,

[0186] where is the reconciliation parameter, 0 < < 1. The final result output unit 590 may be configured to output the probability P as the probability of the patient being in the corresponding subspace or node or category.

[0187] It should be noted that Figure 5 the data analysis device 500 shown may execute Figure 1 any method steps in the method shown. Since the principle of the present disclosure is the same as that of the analysis method, those skilled in the art can obtain other details about the data analysis device 500 from the description of the method, and the details of the data analysis device 500 and its components executing the above method and its steps will not be repeated here.

[0188] Figure 6 An example computing device 600 is illustrated that can be used to implement one or more embodiments. In particular, devices according to some embodiments can be implemented in the example computing device 600. As shown, the example computing device 600 includes one or more processors 610 or processing units, one or more computer-readable media 620 that can include one or more memories 622, one or more displays 640 for displaying content to a user, one or more input / output (I / O) interfaces 650 for input / output (I / O) devices, one or more communication interfaces 660 for communicating with other computing devices or communication devices, and a bus 630 that allows different components and devices to communicate with each other.

[0189] Computer readable media 620, display 640, and / or one or more I / O devices may be included as part of computing device 600, or may alternatively be coupled to computing device 600. Bus 630 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. Bus 630 may include a wired and / or wireless bus.

[0190] The one or more processors 610 are not limited in any way in terms of the materials from which they are formed or the processing mechanisms employed therein. For example, a processor may be composed of one or more semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In this context, processor executable instructions may be electrically executable instructions. The memory 622 represents memory / storage capacity associated with one or more computer-readable media. The memory 622 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disk, magnetic disk, etc.). The memory 622 may include fixed media (e.g., RAM, ROM, fixed hard disk drive, etc.) and removable media (e.g., flash memory drive, removable hard disk drive, optical disk, etc.).

[0191] One or more input / output interfaces 650 allow a user to input commands and information to computing device 600, and likewise allow information to be presented to the user and / or presented to other components or devices using various input / output devices. Examples of input devices include a keyboard, a touch screen display, a cursor control device (e.g., a mouse), a microphone, a scanner, etc. Examples of output devices include a display device (e.g., a monitor or projector), speakers, a printer, a network card, etc.

[0192] The communication interface 660 allows communication with other computing devices or communication devices. The communication interface 660 has no restrictions on the communication technology it adopts. The communication interface 660 may include a wired communication interface such as a local area network communication interface and a wide area network communication interface, and may also include a wireless communication interface, such as an infrared, Wi-Fi or Bluetooth communication interface.

[0193] Various techniques are described herein in the general context of software, hardware (fixed logic circuits), or program modules. Generally, the program modules include routines, programs, objects, elements, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The implementation of these modules and techniques can be stored on a certain form of computer-readable medium or transmitted via the computer-readable medium. Computer-readable media can include a variety of available media or media that can be accessed by a computing device.

[0194] The specific modules, functions, components and techniques described herein can be implemented in software, hardware, firmware and / or a combination thereof. The computing device 600 can be configured to execute specific instructions and / or functions corresponding to the software and / or hardware modules implemented on the computer-readable medium. The instructions and / or functions can be executed / operated by a manufacturing product (e.g., one or more computing devices 600 and / or processor 610) to implement the technology described herein. Such technology includes but is not limited to the example processes described herein. Therefore, the computer-readable medium can be configured to store or provide instructions for implementing the above-mentioned different technologies when accessed by one or more devices described herein.

[0195] Although some embodiments of the present invention are described in detail above with reference to the accompanying drawings, it will be understood by those of ordinary skill in the art that the above detailed description is only for the purpose of explaining the present invention, and the present invention is by no means limited to the above specific implementation methods. Based on the specific description and teachings of these embodiments herein, those of ordinary skill in the art can make various modifications, additions, substitutions and variations to these specific implementation methods without departing from the protection scope of the present invention, that is, these modifications, additions, substitutions and variations should all be covered within the protection scope of the present invention. The protection scope of the present invention shall be subject to the protection scope of the claims. The specific features and behaviors described above are disclosed as example forms of implementing the claims.

[0196] It should be noted that the above embodiments are only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed. The internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the functions of one module can be completed by multiple modules, and the functions of the multiple modules can also be integrated into one module.

[0197] The present application uses terms such as "first", "second", "third", etc. When there is no additional context, the use of such terms is not intended to imply an order, and in fact they are only used for identification purposes. For example, the phrases "first analysis unit" and "second analysis unit" do not necessarily mean that the first analysis unit performs operations or performs processing before the second analysis unit in time. In fact, these phrases are only used to identify different analysis units.

[0198] In the claims, any reference signs placed between brackets shall not be construed as limiting the claim. The terms "comprise" or "comprising" do not exclude the presence of elements or steps other than those listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. In a device or system claim enumerating several means, one or more of these means can be embodied in the same item of hardware. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

Claims

1. A medical data analysis method based on a medical knowledge graph, comprising: Receive the patient's signs and symptoms; Access the medical knowledge graph to obtain a partial medical knowledge graph related to the patient's physical symptoms, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N } and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i The initial evidence score of the node, the edge between the nodes reflects the semantic relevance between the nodes; The final evidence score p of each node is determined based on the evidence propagation of the patient's physical signs and symptoms on the partial medical knowledge graph. t , wherein the evidence propagation of the portion of the medical knowledge graph includes a greater evidence score propagated from one node to another node with a higher semantic relevance than a greater evidence score propagated to another node with a lower semantic relevance; and Based on the final evidence score p t The scores of each node in the is used to analyze the probability P2 that the patient is in a specific disease node.

2. According to the method of claim 1, determine the final evidence score p of each node t Further including: The final evidence score of each node is determined by iterating through the following formula: ， Where d is the damping coefficient, 0 <d<1; , , w i,j Represents the connection of each node v in V i and v j The edge i,j The weight of .

3. The method according to claim 2, wherein the termination condition of the iterative operation is that p t No more changes occur or the maximum number of iterations is reached.

4. According to the method of claim 1, determining the final evidence score p of each node t Further including: The final evidence score of each node is determined by the following formula: ， where d is the damping coefficient, 0 < d < 1; I is the identity matrix of N , , w i,j represents the edge e i connecting nodes v j and v i,j in V, and the weight of e 5. The method according to claim 1, based on the final evidence score p t The probability P2 of a patient being in a specific disease node is analyzed by using the scores of each node in the further includes: The probability of being at each disease node is determined by calculating the percentage of the final score of each disease node in V to the sum of the final scores of all disease nodes in V; as well as The maximum probability among the probabilities is output as the probability P2 that the patient is at a specific disease node.

6. A medical data analysis method, comprising: Receive the patient's signs and symptoms; Use vector h to represent the signs and symptoms; The subspace where the patient is located in the medical big data is located using the physical signs and symptoms as features, where the case set in the big data is represented by a matrix D and the matrix D constitutes the semantic space of known cases, D = [D1, D2, …, D M ], wherein the subspace is a subspace of the semantic space consisting of known cases of a disease, and D i represents the i-th subspace consisting of known cases of the i-th disease, 1≤i≤M, and the positioning includes the formula Solve, where X is the coefficient vector to be solved, which is used to represent the distribution of vector h in each subspace, and the cases corresponding to the dimensions whose values ​​in X are not zero constitute the subspace where vector h is located; The probability P1 that the patient is in a specific subspace is analyzed by analyzing the consistency between the subspace where the sign and symptom vector h is located and each subspace; Access the medical knowledge graph to obtain a partial medical knowledge graph related to the patient's physical symptoms, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N } and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i The initial evidence score of the node, the edge between the nodes reflects the semantic relevance between the nodes; The final evidence score p of each node is determined based on the evidence propagation of the patient's physical signs and symptoms on the partial medical knowledge graph. t , wherein the evidence propagation of the portion of the medical knowledge graph includes a greater evidence score propagated from one node to another node with a higher semantic relevance than a greater evidence score propagated to another node with a lower semantic relevance; and Based on the final evidence score p t The probability P2 that the patient is at a specific disease node is analyzed by using the scores of each node in ; and Determine the probability P of a patient belonging to a specific disease based on probability P1 and probability P2: ,in is the harmonic parameter, 0< <1; Among them, locating the subspace of the patient in the medical big data based on the physical signs and symptoms further includes: By using the sparse solution method, the formula Solving to determine the subspace in which the patient is located; and Based on the value of each element in the coefficient vector X, it is determined in which subspace the largest portion of h is located, thereby determining in which subspace the probability of being located is the highest.

7. The method according to claim 6, wherein the sparse solution method comprises: Solution , where X satisfies ,in is the L1 paradigm, is the L2 norm, ε is a given parameter, The solution obtained x* The cases corresponding to the dimensions whose median values ​​are not zero constitute the subspace where h is located.

8. According to the method of claim 6 or 7, the semantic subspace where h is located and the subspace D are calculated by the following formula: i Consistency: ， in , , is the error, h i is the i-th subspace D in the vector h i The semantic components of Indicates that the coefficient vector X belongs to the subspace D i The dimension of is 1, and the rest of the dimensions are 0.

9. The method according to claim 8, further comprising: Output C1, C2, ..., C M , The maximum value among is taken as the probability P1 that the patient is in the corresponding subspace.

10. The method according to claim 6, determining the final evidence score p of each node t Further including: The final evidence score of each node is determined by iterating through the following formula: ， Where d is the damping coefficient, 0 <d<1; , , w i,j Represents the connection of each node v in V i and v j The edge i,j The weight of .

11. The method according to claim 10, wherein the termination condition of the iterative operation is that p t No more changes occur or the maximum number of iterations is reached.

12. The method according to claim 6, determining the final evidence score p of each node t Further including: The final evidence score of each node is determined by the following formula: ， where d is the damping coefficient, 0 < d < 1; I is the identity matrix of N of N, , , w i,j represents the edge e i connecting nodes v j and v i,j in V, and the weight of e is w 13. The method according to claim 6, based on the final evidence score p t The probability P2 of a patient being in a specific disease node is analyzed by using the scores of each node in the further includes: The probability of being at each disease node is determined by calculating the percentage of the final score of each disease node in V to the sum of the final scores of all disease nodes in V; as well as The maximum probability among the probabilities is output as the probability P2 that the patient is at a specific disease node.

14. A medical data analysis device based on a medical knowledge graph, comprising: a receiving unit configured to receive physical signs and symptoms of the patient; An access unit is configured to access a medical knowledge graph to obtain a partial medical knowledge graph related to the patient's physical symptoms, wherein the partial medical knowledge graph includes a plurality of nodes V={v1, v 2, …, v N } and the initial evidence score of each node ,in , each node v i Indicates a symptom of the patient or a related disease, Represents node v i The initial evidence score of the node, the edge between the nodes reflects the semantic relevance between the nodes; The second analysis unit is configured to determine the final evidence score p of each node based on the evidence propagation of the patient's physical signs and symptoms on the partial medical knowledge graph. t , wherein the evidence propagation of the portion of the medical knowledge graph includes a greater evidence score propagated from one node to another node with high semantic relevance than the evidence score propagated to another node with low semantic relevance; as well as The determination unit is configured to determine the final evidence score p t The scores of each node in the is used to analyze the probability P2 that the patient is in a specific disease node.

15. The device according to claim 14, wherein the second analyzing unit further comprises: The computing unit is configured to perform iterative calculations through the following formula to determine the final evidence score of each node: ， Where d is the damping coefficient, 0 <d<1; , , w i,j Represents the connection of each node v in V i and v j The edge i,j The weight of .

16. The device according to claim 15, wherein the termination condition of the iterative operation is that p t No more changes occur or the maximum number of iterations is reached.

17. The device according to claim 14, wherein the second analyzing unit further comprises: The computing unit is configured to determine the final evidence score of each node using the following formula: ， where d is the damping coefficient, 0 < d < 1; I is the identity matrix of N of N, , , w i,j represents the edge e i connecting nodes v j and v i,j in V, and the weight of e is w 18. The device according to claim 14, wherein the determining unit is further configured to: Determine the probability of being at each node by calculating the percentage of the final score of each disease node in V to the sum of the final scores of all disease nodes in V; and The maximum probability among the probabilities is output as the probability P2 of being at a specific disease node.

19. A computing device comprising: a memory configured to store computer executable instructions; as well as A processor coupled to the memory is configured to execute the computer executable instructions so that the processor performs the method as claimed in any one of claims 1-13.

20. A computer-readable storage medium having computer-readable instructions stored thereon, which, when executed by a computing device, cause the computing device to perform the method of any one of claims 1-13.

Images