WO2023096386A1 - Procédé et dispositif d'intégration de données médicales et support d'enregistrement lisible par ordinateur - Google Patents

Procédé et dispositif d'intégration de données médicales et support d'enregistrement lisible par ordinateur Download PDF

Info

Publication number
WO2023096386A1
WO2023096386A1 PCT/KR2022/018768 KR2022018768W WO2023096386A1 WO 2023096386 A1 WO2023096386 A1 WO 2023096386A1 KR 2022018768 W KR2022018768 W KR 2022018768W WO 2023096386 A1 WO2023096386 A1 WO 2023096386A1
Authority
WO
WIPO (PCT)
Prior art keywords
embedding
graph
nodes
data
node
Prior art date
Application number
PCT/KR2022/018768
Other languages
English (en)
Korean (ko)
Inventor
김영학
전태준
Original Assignee
재단법인 아산사회복지재단
울산대학교 산학협력단
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 재단법인 아산사회복지재단, 울산대학교 산학협력단 filed Critical 재단법인 아산사회복지재단
Publication of WO2023096386A1 publication Critical patent/WO2023096386A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9024—Graphs; Linked lists
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H70/00—ICT specially adapted for the handling or processing of medical references

Definitions

  • Embodiments relate to a medical data embedding method and apparatus for embedding medical data.
  • EMR electronic medical record
  • EMR is multi-model data including medical text and medical code (concept) such as ICD-10 code.
  • medical code such as ICD-10 code.
  • embedding is to vectorize each concept in order to computerize each word or code of the text.
  • an object of the embodiment is to provide a medical data embedding method and apparatus for embedding medical text data and medical code data having a relationship with each other in one embedding space.
  • the medical data embedding method of the embodiment includes the steps of text embedding medical text data into an embedding space, converting medical code data in an EMR into a graph, and initializing nodes of the graph using the text embedding; , sampling each node of the graph in which the nodes are initialized and nodes of a subgraph excluding the detailed diagnosis node in the graph in which the nodes are initialized, embedding the sampled nodes into the embedding space, and Learning the relationship between the embedded medical code data and the medical text data; embedding data including the text embedding and the initialized form of the graph nodes of the graph; and embedding data obtained by performing the relationship learning. It may include obtaining final embedding data using
  • the graph may include a large-grade node, a medium-grade node, a diagnosis node, and a detailed diagnosis node
  • the sub-graph may include a large-grade node, a medium-grade node, and a diagnosis node.
  • the nodes may be initialized with an average value of vectors of texts constituting the nodes.
  • nodes of the graph and nodes of the sub-graph are sampled using a return parameter and an in-out parameter used in the node2vec technique, and the nodes of the graph set the return parameter to the in-out parameter.
  • Sampling is performed by setting the out parameter to be greater than the out parameter, and the nodes of the subgraph may perform sampling by setting the return parameter to be smaller than the in-out parameter.
  • the sampled nodes may be embedded into the embedding space by using the encoding function, the decoding function, and the similarity function.
  • the obtaining of the final embedding data may be determined by Equation 1.
  • the medical data embedding apparatus of the embodiment includes a memory in which a control program for embedding medical data is stored, and a processor that executes the control program, wherein the processor performs text embedding of medical text data in an embedding space using a word2vec technique. and transforms the medical code data in the EMR into a graph, initializes the nodes of the graph using the text embedding, and excludes the nodes of the graph in which the nodes are initialized and the detailed diagnosis node in the graph in which the nodes are initialized.
  • Samples nodes of each embeds the nodes sampled by the Node2vec technique into the embedding space, performs relationship learning between the medical code data and the medical text data embedded in the embedding space, and Final embedding data may be obtained using the embedding data including the initialized form of the graph nodes and the embedding data obtained by performing the relationship learning.
  • the graph may include a large-grade node, a medium-grade node, a diagnosis node, and a detailed diagnosis node
  • the sub-graph may include a large-grade node, a medium-grade node, and a diagnosis node.
  • the processor may initialize the node of the graph with an average value of vectors of text constituting the node.
  • the processor samples a node of the graph and a node of the subgraph using a return parameter and an in-out parameter, and the node of the graph sets the return parameter to be greater than the in-out parameter, sampling, and the node of the subgraph may perform sampling by setting the return parameter smaller than the in-out parameter.
  • the processor may embed sampled nodes into the embedding space by using the encoding function, the decoding function, and the similarity function.
  • the final embedding data may be determined by Equation 1.
  • the embodiment is a computer readable recording medium storing a computer program, which, when executed by a processor, performs text embedding of medical text data into an embedding space using a word2vec technique, and medical treatment in EMR.
  • Converting code data into a graph initializing nodes of the graph using the text embedding, and generating nodes of the graph in which the nodes are initialized and sub-graphs excluding detailed diagnosis nodes in the graph in which the nodes are initialized Sampling each node, embedding the nodes sampled by the Node2vec technique into the embedding space, learning the relationship between the medical code data and the medical text data embedded in the embedding space;
  • the processor to perform an operation that includes obtaining final embedding data using embedding data including text embedding and an initialized form of the graph nodes of the graph, and embedding data obtained by performing the relation learning It may contain commands for
  • the embodiment is a computer program stored in a computer-readable recording medium, and the computer program, when executed by a processor, performs text embedding of medical text data into an embedding space using a word2vec technique; Converting code data into a graph, initializing nodes of the graph using the text embedding, and generating nodes of the graph in which the nodes are initialized and sub-graphs excluding detailed diagnosis nodes in the graph in which the nodes are initialized Sampling each node, embedding the nodes sampled by the Node2vec technique into the embedding space, learning the relationship between the medical code data and the medical text data embedded in the embedding space; To cause the processor to perform an operation that includes obtaining final embedding data using embedding data including text embedding and an initialized form of the graph nodes of the graph, and embedding data obtained by performing the relation learning It may contain commands for
  • the embodiment has an effect of embedding medical text data and medical code data into one embedding space.
  • the embodiment can effectively analyze EMR by performing embedding to have a relationship between medical text data and medical code data.
  • the embodiment can obtain data with a better embedding effect than when using an application by adjusting parameters.
  • FIG. 1 is a diagram illustrating a medical data embedding system according to an embodiment.
  • FIG. 2 is a diagram illustrating a medical data embedding device according to an embodiment.
  • FIG. 3 is a flowchart illustrating a medical data embedding method performed by a medical data embedding apparatus according to an embodiment.
  • FIG. 4 is a diagram illustrating a structure of nodes included in a graph used in a medical data embedding method according to an embodiment.
  • FIG. 5 is a diagram illustrating detailed attributes of nodes used in a medical data embedding method according to an embodiment.
  • FIG. 6 is a diagram illustrating parameters for nodes used in a method for embedding medical data according to an embodiment.
  • FIG. 7(a) and (b) are diagrams illustrating a main graph (MG) and a sub graph (SG) used in the medical data embedding method according to the embodiment, respectively.
  • FIGS. 8(a) and (b) are diagrams illustrating results of visualizing a specific node before and after graph embedding training in the medical data embedding method according to the embodiment, respectively.
  • FIG. 9 is a diagram illustrating literal names of medical code data and medical text data used in a medical data embedding method according to an embodiment.
  • FIG. 10 is a diagram illustrating a specific vector of medical code data and a specific vector of medical text data used in a medical data embedding method according to an embodiment.
  • FIG. 1 is a diagram illustrating a medical data embedding system according to an embodiment
  • FIG. 2 is a diagram illustrating a medical data embedding apparatus according to an embodiment.
  • a medical data embedding system may include a server 100 and a medical data embedding device 200 .
  • the server 100 may be a space in which medical data, for example, EMR data is stored.
  • the server 100 may be a server provided in a medical institution or a storage space for receiving EMR from a medical institution and storing the EMR.
  • the medical data embedding apparatus 200 may receive the EMR from the server and embed the medical text data and the medical code data into one embedding space based on the relationship between the medical text data and the medical code data.
  • the medical data embedding apparatus 200 may include a memory 210 , a communication unit 230 , and a processor 250 .
  • the memory 210 may store various data for overall operations of the medical data embedding apparatus 200, such as processing by the processor 250 or a control program for embedding medical data.
  • the memory 210 includes a plurality of application programs driven by the medical data embedding device 200, a plurality of application programs driven by the medical data embedding device 200, and data for operation of the medical data embedding device 200. and instructions.
  • EMR information received from the server 100 may be stored in the memory 210, but is not limited thereto.
  • the memory 210 may include magnetic storage media or flash storage media, but is not limited thereto.
  • the communication unit 230 is connected to the server 100 and may provide a communication interface capable of communicating with the server 100 using a plurality of communication methods.
  • the communication unit 230 may be a device including hardware and software necessary for transmitting/receiving a signal such as a control signal or a data signal with another network device through a wired/wireless connection.
  • EMR data may be stored in the memory 210 through the communication unit 230 .
  • the communication unit 230 may perform communication using a Low Power Wireless Network (LPWN) and a Low Power Wide Area Network (LPWAN) such as 3G, LTE, and 5G, as well as NB-IoT, LoRa, SigFox, and LTE-CAT1. there is.
  • LPWN Low Power Wireless Network
  • LPWAN Low Power Wide Area Network
  • the communication unit 230 may perform communication using a communication method using a wired local area network (LAN) as well as a wireless LAN such as WiFi 80211a/b/g/n. In addition to this, the communication unit 230 may perform communication with the server 100 using a communication method such as NFC or Bluetooth.
  • LAN local area network
  • WiFi 80211a/b/g/n a wireless LAN
  • the communication unit 230 may perform communication with the server 100 using a communication method such as NFC or Bluetooth.
  • the communication unit 230 may be omitted as a component of the medical data embedding device 200 .
  • the processor 250 may control the medical data embedding device 200 as a kind of central processing unit.
  • the processor 250 may include any type of device capable of processing data.
  • a 'processor' may refer to a data processing device embedded in hardware having a physically structured circuit to perform functions expressed by codes or instructions included in a program, for example.
  • a data processing device built into hardware, a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated (ASIC) circuit) and a processing device such as a field programmable gate array (FPGA), but is not limited thereto.
  • CPU central processing unit
  • ASIC application-specific integrated
  • FPGA field programmable gate array
  • FIG. 3 is a flowchart illustrating a medical data embedding method performed in a medical data embedding apparatus according to an embodiment
  • FIGS. 4 to 11 are diagrams for explaining detailed operations of the medical data embedding method according to an embodiment.
  • the medical data embedding apparatus 200 may perform text embedding (S100).
  • the medical data embedding apparatus 200 may embed medical text data in a Word2Vec method using Pubmed abstract, which is a large-scale medical corpus such as Wikipedia, and a relatively small EMR diagnosis node (eg, MIMIC-3 discharge note).
  • Pubmed abstract which is a large-scale medical corpus such as Wikipedia
  • EMR diagnosis node eg, MIMIC-3 discharge note
  • the medical data embedding apparatus 200 may convert medical code data in the EMR into a graph (S200).
  • Medical code data may be ICD-10 data, but is not limited thereto. ICD-10 codes can use up to 4 letters, except for Class U and Class V.
  • FIG. 4 is a diagram illustrating a structure of nodes included in a graph used in a medical data embedding method according to an embodiment.
  • graph (G) is a high-grade node (N1, eg: A00-B99), a middle-grade node (N2, eg: A00-A09), a diagnosis node (N3, eg: A00, A02) and detailed diagnosis node (N4, leaf node).
  • All nodes except the diagnosis node N3 and the detailed diagnosis node N4 are classifiers, and these nodes may mean categories to which the diagnosis node N3 and the detailed diagnosis node N4 belong. Both the diagnosis node N3 and detailed diagnosis node N4 may be used primarily in medical situations.
  • An ICD-10 graph can be constructed using a network package [Hagberg et al., 2008] that has a 'name' attribute representing the literal name of each node.
  • 10604 nodes may exist in the generated ICD-10 graph, but is not limited thereto.
  • the medical data embedding apparatus 200 may initialize nodes of the graph (S300).
  • the initialization can be initialized with the mean vector of each word in the attribute name using the text embedding performed in the previous step (eg Pubmed+EMR text embedding).
  • the initial value may be based on a pre-trained embedding value to prevent the code from being embedded only based on the ontology method.
  • FIG. 5 is a diagram illustrating detailed attributes of nodes used in a medical data embedding method according to an embodiment.
  • the infectious vector value is (0.0102, 0.5734 0.2122, ... , 0.2124), and the mononucleosis vector value is (-0.3435, 0.6423, ... . That is, it may be initialized to a value of (-0.1666, 0.6078, ... , 0.1683) for node B27.
  • the medical data embedding apparatus 200 may sample graph nodes (S400).
  • the medical data embedding device 200 may use a sampling technique of the node2vec embedding method.
  • FIG. 6 is a diagram illustrating parameters for nodes used in a method for embedding medical data according to an embodiment.
  • the node2vec embedding method may use a return parameter p and an in-out parameter q.
  • the return parameter p can be defined as the degree to return to the initial node
  • the in-out parameter q can be defined as the degree to move from the initial node to another node.
  • Parameters p and q depend on the graph size and graph characteristics, and the optimal parameters p and q can be obtained by grid search for a subset of p and q.
  • the medical data embedding apparatus 200 may sample graph nodes in two stages.
  • One of the biggest characteristics of the ICD-10 graph form is that about 80% of the nodes are located in detailed diagnosis nodes. In this case, if the entire graph is sampled with a pair of parameters p and q without phase partitioning, ontology learning for detailed diagnostic nodes is hardly performed.
  • the apparatus 200 for embedding medical data may perform sampling in two stages by dividing the graph between the diagnosis node and detailed diagnosis node layers.
  • FIG. 7(a) and (b) are diagrams illustrating a main graph (MG) and a sub graph (SG) used in the medical data embedding method according to the embodiment, respectively.
  • the apparatus 200 for embedding medical data includes a main graph MG including a large grade node, a middle grade node, a diagnosis node, and a detailed diagnosis node, and a detailed diagnosis node. Nodes may be sampled using the excluded subgraph (SG).
  • the medical data embedding apparatus 200 may set p greater than q in order to learn the ontology information of the main graph MG by inducing a graph depth first search.
  • p may be set to 10 5 and q may be set to 10 -5 .
  • the medical data embedding apparatus 200 may set q to be larger than p for the subgraph SG for a wide search such as a breadth first search.
  • p may be set to 0.5 and q may be set to 2.0.
  • the medical data embedding apparatus 200 may learn samples of nodes of the main graph (MG) and then samples of nodes of the sub graph (SG), but the order is not limited.
  • the medical data embedding apparatus 200 may embed nodes of the graph sampled after learning is completed into an embedding space (S500).
  • the apparatus 200 for embedding medical data may define an encoding function, a decoding function, and a similarity function in order to embed a node of a graph.
  • a lookup of the embedding matrix may be used as an encoding function.
  • the apparatus 200 for embedding medical data may reflect structural information in embedding and well training of sampled nodes by adopting decoding and similarity functions in the node2vec embedding method.
  • the main concept of the node2vec metric can be the probability that a target node u appears in a starting node v in a random walk of length w.
  • the decoding function for this method can be represented by Equation 1, and the similarity function can be represented by Equation 2.
  • Equation 1 DEC denotes a decoding function
  • Zu and Zv denote encoded node u and node v, respectively
  • e denotes a natural constant
  • T denotes matrix transposition.
  • Similarity means a similarity function
  • v) means a probability that a target node u appears in a starting node v.
  • FIGS. 8(a) and (b) are diagrams illustrating results of visualizing a specific node before and after graph embedding training in the medical data embedding method according to the embodiment, respectively.
  • the target node is the C22 node and the remaining nodes represent the 20 most similar ICD-10 nodes of the target node C22.
  • the medical data embedding apparatus 200 may perform relationship learning between medical text data and medical code data (S600).
  • FIG. 9 is a diagram illustrating literal names of medical code data and medical text data used in a medical data embedding method according to an embodiment.
  • the medical data embedding apparatus 200 may generate a sentence by connecting medical code data (target words) and literal names (context words) of medical text data.
  • the medical data embedding apparatus 200 may extract only positive samples in which the window size is set to the maximum length of a character.
  • FIG. 10 is a diagram illustrating a specific vector of medical code data and a specific vector of medical text data used in a medical data embedding method according to an embodiment.
  • samples may be trained using an Adam-type optimizer (artificial neural network).
  • Adam-type optimizer artificial neural network
  • the medical data embedding apparatus 200 may obtain final embedding data by using first embedding data including text embedding and an initialized form of graph nodes of the graph, and second embedding data obtained by performing relational learning. (S700).
  • the final embedding data can be obtained by Equation 3.
  • W i is the first embedding data
  • W f is the second embedding data
  • the relationship between medical code data and medical text data is The closer to zero, the looser it can be. On the other hand, the relationship between medical code data and medical text data is The closer to 1, the stronger it can be.
  • EMR was processed using the Word2Vec method, and the similarity between codes was digitized with cosine similarity using similar pair data of ICD-10 codes.
  • the Wikipedia search engine and EMR were converted to the Word2Vec method and then the similarity was confirmed.
  • the PubMed search engine and EMR were converted to the Word2Vec method and then the similarity was confirmed.
  • Comparative Example In Fig. 3 the Wikipedia search engine, PubMed search engine, and EMR were converted into the Word2Vec method and the similarity was confirmed.
  • the cosine similarity calculated in Comparative Example 1 was 0.711778
  • the cosine similarity calculated in Comparative Example 2 was 0.534551
  • the cosine similarity calculated in Comparative Example 3 was 0.682194.
  • Example 1 The cosine similarity was confirmed by setting the value to 0.9, and in Example 2 The cosine similarity was confirmed by setting the value to 0.8, and in Example 3, The cosine similarity was confirmed by setting the value to 0.5, and in Example 4, The value was set to 0.2 to confirm the cosine similarity.
  • the cosine similarity calculated in Example 1 was 0.870578
  • the cosine similarity calculated in Example 2 was 0.864785
  • the cosine similarity calculated in Example 3 was 0.830749
  • the cosine similarity calculated in Example 4 was 0.830749.
  • the cosine similarity obtained was 0.725989.
  • Example 1 As can be seen in FIG. 11, it can be seen that the examples according to the present invention show a higher degree of similarity than the comparative examples. In addition, among the embodiments according to the present invention, It can be seen that the best performance was obtained in Example 1, in which is set to 0.9.
  • Various embodiments of the present document are software (eg, machine-readable storage media) (eg, memory (internal memory or external memory)) including instructions stored in a storage medium readable by a machine (eg, a computer). : program).
  • a device is a device capable of calling a stored command from a storage medium and operating according to the called command, and may include an electronic device according to the disclosed embodiments.
  • the control unit may perform a function corresponding to the command directly or by using other components under the control of the control unit.
  • An instruction may include code generated or executed by a compiler or interpreter.
  • the device-readable storage medium may be provided in the form of a non-transitory storage medium.
  • non-temporary means that the storage medium does not contain a signal and is tangible, but does not distinguish whether data is stored semi-permanently or temporarily in the storage medium.
  • the method according to various embodiments disclosed in this document may be provided by being included in a computer program product.
  • a computer-readable recording medium storing a computer program includes: performing text embedding of medical text data in an embedding space using a word2vec technique; converting medical code data in an EMR into a graph; Initializing the nodes of the graph by using the text embedding, sampling the nodes of the graph in which the nodes are initialized and the nodes of the subgraph excluding the detailed diagnosis node in the graph in which the nodes are initialized, respectively, Node2vec technique embedding nodes sampled by , in the embedding space, learning the relationship between the medical code data and the medical text data embedded in the embedding space, and initializing the text embedding and the graph nodes of the graph. It may include instructions for causing a processor to perform a method including an operation for performing a step of obtaining final embedding data using the embedding data including the formatted form and the embedding data obtained by performing the relationship learning. there is.
  • a computer program stored in a computer readable recording medium includes the steps of text embedding medical text data into an embedding space using a word2vec technique, converting medical code data in an EMR into a graph, Initializing the nodes of the graph by using the text embedding, sampling the nodes of the graph in which the nodes are initialized and the nodes of the subgraph excluding the detailed diagnosis node in the graph in which the nodes are initialized, respectively, Node2vec technique embedding nodes sampled by , in the embedding space, learning the relationship between the medical code data and the medical text data embedded in the embedding space, and initializing the text embedding and the graph nodes of the graph. It may include instructions for causing a processor to perform a method including an operation for performing a step of obtaining final embedding data using the embedding data including the formatted form and the embedding data obtained by performing the relationship learning. there is.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pathology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

L'invention divulgue un procédé d'intégration de données médicales. Le procédé peut comprendre les étapes consistant à : intégrer du texte de données médicales dans un espace d'intégration ; convertir, en un graphique, des données de code médical dans un dossier médical électronique (DME) ; initialiser les nœuds du graphique en utilisant l'intégration de texte ; échantillonner chacun d'un nœud du graphique dans lequel les nœuds sont initialisés et d'un nœud d'un sous-graphique qui exclut un nœud de diagnostic détaillé dans le graphique dans lequel les nœuds sont initialisés ; intégrer les nœuds échantillonnés dans l'espace d'intégration ; apprendre la relation entre les données de code médical et les données de texte médical intégrées dans l'espace d'intégration ; et utiliser les données d'intégration, qui comprennent l'intégration de texte et une forme initialisée des nœuds de graphique du graphique et intégrer les données acquises par apprentissage de la relation, de sorte à acquérir des données d'intégration finales.
PCT/KR2022/018768 2021-11-24 2022-11-24 Procédé et dispositif d'intégration de données médicales et support d'enregistrement lisible par ordinateur WO2023096386A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR1020210163161A KR102588270B1 (ko) 2021-11-24 2021-11-24 의료 데이터 임베딩 방법 및 장치, 컴퓨터 판독 가능한 기록 매체 및 컴퓨터 프로그램
KR10-2021-0163161 2021-11-24

Publications (1)

Publication Number Publication Date
WO2023096386A1 true WO2023096386A1 (fr) 2023-06-01

Family

ID=86540142

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2022/018768 WO2023096386A1 (fr) 2021-11-24 2022-11-24 Procédé et dispositif d'intégration de données médicales et support d'enregistrement lisible par ordinateur

Country Status (2)

Country Link
KR (1) KR102588270B1 (fr)
WO (1) WO2023096386A1 (fr)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20190123609A (ko) * 2018-04-24 2019-11-01 네이버 주식회사 딥 어텐션 네트워크를 이용하여 환자 의료 기록으로부터 질병 예후를 예측하는 방법 및 시스템
KR20190139722A (ko) * 2018-06-08 2019-12-18 연세대학교 산학협력단 진단명 레이블링을 위한 딥러닝을 이용한 판독기록문으로부터 최종 진단명 추출 방법 및 장치
CN111382272A (zh) * 2020-03-09 2020-07-07 西南交通大学 一种基于知识图谱的电子病历icd自动编码方法
US20200293712A1 (en) * 2019-03-11 2020-09-17 Christopher Potts Methods, apparatus and systems for annotation of text documents
WO2021046536A1 (fr) * 2019-09-06 2021-03-11 F. Hoffmann-La Roche Ag Extraction et enrichissement d'informations automatisés dans un rapport de pathologie à l'aide d'un traitement de langage naturel
KR102329920B1 (ko) * 2021-06-04 2021-11-23 가천대학교 산학협력단 기계학습된 모델을 이용한 의료원문데이터를 증강하는 방법 및 시스템

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20190123609A (ko) * 2018-04-24 2019-11-01 네이버 주식회사 딥 어텐션 네트워크를 이용하여 환자 의료 기록으로부터 질병 예후를 예측하는 방법 및 시스템
KR20190139722A (ko) * 2018-06-08 2019-12-18 연세대학교 산학협력단 진단명 레이블링을 위한 딥러닝을 이용한 판독기록문으로부터 최종 진단명 추출 방법 및 장치
US20200293712A1 (en) * 2019-03-11 2020-09-17 Christopher Potts Methods, apparatus and systems for annotation of text documents
WO2021046536A1 (fr) * 2019-09-06 2021-03-11 F. Hoffmann-La Roche Ag Extraction et enrichissement d'informations automatisés dans un rapport de pathologie à l'aide d'un traitement de langage naturel
CN111382272A (zh) * 2020-03-09 2020-07-07 西南交通大学 一种基于知识图谱的电子病历icd自动编码方法
KR102329920B1 (ko) * 2021-06-04 2021-11-23 가천대학교 산학협력단 기계학습된 모델을 이용한 의료원문데이터를 증강하는 방법 및 시스템

Also Published As

Publication number Publication date
KR102588270B1 (ko) 2023-10-12
KR20230076335A (ko) 2023-05-31

Similar Documents

Publication Publication Date Title
US9619735B1 (en) Pure convolutional neural network localization
WO2020027540A1 (fr) Appareil et procédé de compréhension de langage naturel personnalisé
US11687570B2 (en) System and method for efficient multi-relational entity understanding and retrieval
WO2020262788A1 (fr) Système et procédé de compréhension de langage naturel
WO2012091400A1 (fr) Système et procédé de détection de logiciel malveillant dans un fichier sur la base d'une carte génétique de fichier
EP3828746A1 (fr) Systèmes et procédés de triage de vulnérabilités logicielles
WO2022131497A1 (fr) Appareil d'apprentissage et procédé de génération d'image, et appareil et procédé de génération d'image
WO2021215620A1 (fr) Dispositif et procédé pour générer automatiquement un sous-titre d'image spécifique au domaine à l'aide d'une ontologie sémantique
WO2019235828A1 (fr) Système de diagnostic de maladie à deux faces et méthode associée
WO2022059969A1 (fr) Procédé de pré-apprentissage de réseau neuronal profond permettant une classification de données d'électrocardiogramme
WO2020071854A1 (fr) Appareil électronique et son procédé de commande
WO2021157897A1 (fr) Système et procédé pour la compréhension et l'extraction efficaces d'une entité multi-relationnelle
CN113313241A (zh) 确定深度学习模型的张量信息的方法和计算装置
WO2022097927A1 (fr) Procédé de détection d'événement vidéo en direct sur la base d'interrogations en langage naturel, et appareil correspondant
WO2022255564A1 (fr) Procédé d'analyse de signal biologique
WO2022260293A1 (fr) Procédé de vectorisation de données médicales pour apprentissage automatique, dispositif de conversion de données et programme de conversion de données dans lesquels le procédé est mis en œuvre
WO2023096386A1 (fr) Procédé et dispositif d'intégration de données médicales et support d'enregistrement lisible par ordinateur
CN110414450A (zh) 关键词检测方法、装置、存储介质及电子设备
WO2022107964A1 (fr) Appareil et procédé de détection et de classification de code malveillant sur la base d'une matrice adjacente
WO2023048437A1 (fr) Procédé, programme et appareil d'entrainement et de déduction d'un modèle d'apprentissage profond sur la base de données médicales
WO2022139327A1 (fr) Procédé et appareil de détection d'énoncés non pris en charge dans la compréhension du langage naturel
WO2021157776A1 (fr) Dispositif, procédé et support d'enregistrement, commande d'enregistrement, pour prendre en charge l'apprentissage
WO2022004970A1 (fr) Appareil et procédé d'entraînement de points clés basés sur un réseau de neurones artificiels
WO2022108206A1 (fr) Procédé et appareil pour remplir un graphe de connaissances pouvant être décrit
WO2022102879A1 (fr) Appareil et procédé d'obtention d'une séquence de transaction vulnérable dans un contrat intelligent

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22899068

Country of ref document: EP

Kind code of ref document: A1