WO2023061377A1 - 一种多中心知识图谱联合决策支持方法与系统 - Google Patents

一种多中心知识图谱联合决策支持方法与系统 Download PDF

Info

Publication number
WO2023061377A1
WO2023061377A1 PCT/CN2022/124692 CN2022124692W WO2023061377A1 WO 2023061377 A1 WO2023061377 A1 WO 2023061377A1 CN 2022124692 W CN2022124692 W CN 2022124692W WO 2023061377 A1 WO2023061377 A1 WO 2023061377A1
Authority
WO
WIPO (PCT)
Prior art keywords
patient
information
chain
data
clinical
Prior art date
Application number
PCT/CN2022/124692
Other languages
English (en)
French (fr)
Inventor
李劲松
田雨
周天舒
尚勇
Original Assignee
浙江大学
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 浙江大学 filed Critical 浙江大学
Publication of WO2023061377A1 publication Critical patent/WO2023061377A1/zh

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
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/602—Providing cryptographic facilities or services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/64—Protecting data integrity, e.g. using checksums, certificates or signatures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • 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
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records

Definitions

  • the present invention relates to the technical field of medical knowledge graphs, in particular to a multi-center knowledge graph joint decision support method and system.
  • the main technical solution of the existing multi-center medical data application is to use multi-center fragmented clinical data for model modeling through distributed learning, so as to realize clinical decision support functions such as risk prediction and disease diagnosis.
  • clinical cohort construction standards and machine learning algorithms are distributed under a unified data structure; each hospital internally refers to the issued cohort construction standard, and establishes a local on the same data structure Study the cohort and use the same machine learning algorithm for local training; finally, the training results are aggregated to form a clinical decision-making model without the original data leaving the hospital.
  • the local data of the hospital is analyzed according to the training model, so as to realize the decision support function.
  • the purpose of the present invention is to address the deficiencies of the existing technology, and propose a multi-center knowledge graph joint decision support method and system, using medical knowledge graph technology and block chain technology, through the combination of local knowledge graph and on-chain synchronous graph, Realize the local semantic reasoning of clinical data and the summary of results on the chain, so that when the original medical data does not leave the hospital, rely on the knowledge graph technology to synthesize the fragmented cross-institutional medical data of patients, and based on deductive reasoning and evidence-based medicine, give a comprehensive Interpretable clinical decision support for complete patient clinical evidence.
  • the invention utilizes the electronic medical record-oriented knowledge map to effectively adapt the heterogeneous electronic medical record data structure and terminology system of multiple hospitals, and can realize the localized deployment and application of the decision support system in multiple hospitals, solving the problem of multi-center medical system.
  • Heterogeneity problem the present invention performs semantic reasoning on medical data through the local map, and synchronizes the clinical findings generated by reasoning with the original data through asymmetric encryption, and uses hash encryption to achieve patient identity matching, thereby ensuring During the entire decision support process, the original data does not leave the hospital, and no data exposure occurs, which solves the data privacy protection problem of multi-center data applications; the present invention effectively summarizes the data of the same patient in multiple hospitals by combining the local map and the on-chain map. In the reasoning process, the medical data covered by multiple centers can be effectively used to generate reliable clinical decision support, which solves the problem that the existing system can only use the data of a single hospital for analysis.
  • a multi-center knowledge map joint decision support method includes the following steps:
  • step (1) Based on the local knowledge graph and semantic triplet constructed in step (1), according to the reasoning requirements of clinical decision support, through the semantic reasoning method, infer the patient's clinical discovery information and generate the intermediate results of reasoning; according to the semantics of the local knowledge graph Structure, based on patient matching information, time window information of patient visits, and clinical discovery information obtained by reasoning, construct an on-chain subgraph for blockchain synchronization, and the on-chain subgraph isolates other patients' original medical information;
  • step (3) Based on the encrypted triplet file in step (3), construct a block chain synchronization data structure composed of three parts: block data group, triplet data group and communication data group, and perform point-to-point communication in multiple centers. data transmission;
  • Each center compares the local patient identity hash value with the patient identity hash value in the on-chain sub-graph according to the triplet information in the on-chain sub-graph obtained synchronously by the blockchain. Similarity to get patient matching results;
  • each center obtains the inference intermediate results of the matching patients in the sub-graph on the chain, based on the local patient information and inference intermediate results, combined with the local knowledge graph, and generates decision support results through semantic reasoning.
  • the medical standard terminology uses the OMOP term base; the local knowledge map builds a clinical decision support reasoning rule base based on doctor experience and medical literature, and the node elements in the rules conform to the knowledge map medical knowledge structure and OMOP term encoding system.
  • step (1) the semantic triplet of clinical data adopts the three-level clinical data semantic model of "patient-visit-diagnosis", and the electronic medical record data of each patient is constructed according to the clinical data semantic model. Modulus and value information entry.
  • step (2) according to the clinical discovery category of the patient’s medical records, the corresponding semantic inference rules are obtained based on the clinical decision support inference rule base. Obtain semantic reasoning results.
  • step (2) for all patient instances, visit instances and clinical discovery instances, the corresponding virtual instances required for on-chain synchronization are generated based on the local knowledge graph, where the virtual patient instances are simultaneously injected with the patient's unique identity information and non-unique Identity information, the virtual visit instance includes the time window information consisting of the start date of the visit and the end date of the visit, and the virtual clinical discovery instance includes the confirmation time of the clinical discovery, the category of the clinical discovery, and the positive and negative results of the clinical discovery.
  • the patient matching information includes the patient's unique identity information and non-unique identity information;
  • the unique identity information is the patient's ID card ID and medical insurance ID, which is encrypted using the SHA-256 hash algorithm , to generate a unique identity hash value;
  • the non-unique identity information is name, gender, birthday, address and work address, which is encrypted using the SimHash algorithm to generate a non-unique identity hash value.
  • the block data group includes block number, block hash value, Merkle tree root and time record stamp information, which are used for matching verification between multi-center block nodes;
  • the triplet data group is encrypted sub-graph triplet file data on the chain;
  • the communication data group records data flow status, database connection status, P2P node connection parameters and triplet file identification, and is used for block Chain node communication and process discrimination.
  • the present invention also provides a multi-center knowledge graph joint decision support system, which includes a local knowledge graph module, a distributed module and an on-chain module;
  • the local knowledge map module constructs a local knowledge map through medical literature and medical standard terminology, establishes clinical data as a semantic triplet of "patient-visit-diagnosis", and associates it with the medical knowledge in the local knowledge map; based on The constructed local knowledge map and semantic triplets, according to the reasoning requirements of clinical decision support, use the semantic reasoning method to infer the clinical discovery information of patients and generate intermediate results of reasoning; each center obtains the on-chain After matching the inference intermediate results of the patients in the subgraph, based on the local patient information and inference intermediate results, combined with the local knowledge map, the decision support results are generated through semantic reasoning;
  • the distributed module is used to construct subgraphs on the chain and perform patient matching
  • the construction of sub-graphs on the chain is specifically: based on the inference intermediate results generated by the local knowledge graph module, according to the semantic structure of the local knowledge graph, based on patient matching information, time window information of patient visits, and clinical discovery information obtained by reasoning, construct a block for The on-chain subgraph of the chain synchronization, and the triplet information is encrypted by the encryption module, and then transmitted to the on-chain module; the encryption module encrypts the patient matching information to generate an anonymized patient identity hash value, and converts the triplet Semantic structure information and clinical discovery information in the group are de-expressed through coding table mapping, and then triplet file encryption is performed;
  • Patient matching is specifically: each center obtains the triplet information in the sub-graph on the chain according to the block chain synchronization results of the modules on the chain, and compares the local patient identity hash value with the patient identity hash value in the sub-graph on the chain , get the patient matching result according to the similarity of the identity hash value;
  • the on-chain module constructs an area composed of block data group, triplet data group and communication data group through the data communication module.
  • Block chain synchronization data structure block chain synchronization through consensus mechanism, and point-to-point data transmission in multiple centers.
  • the distributed module obtains the triplet information used for distributed reasoning and reasoning intermediate result summary from the local knowledge map module through limited entity interaction, parses the information in the triplet and isolates the patient's original medical data information, and then perform triple reconstruction according to the semantic structure of the subgraph on the chain.
  • decision support results generated by the local knowledge graph module are transmitted to the distributed module and encrypted with triplet information, and then transmitted to the on-chain module for blockchain synchronization for decision support in other centers.
  • the present invention utilizes the electronic medical record knowledge map system and the OMOP (Observational Medical Outcomes Partnership) CDM (Common Data Model) standard terminology system to construct multi-center electronic medical record data into a clinical data model with unified medical concept coding and unified data semantic structure, In this way, multi-center medical data can be jointly reasoned under a unified and standardized data structure.
  • OMOP Observational Medical Outcomes Partnership
  • CDM Common Data Model
  • the invention utilizes blockchain technology to realize the synchronization of inference intermediate results of multi-center local knowledge graphs, and integrates fragmented clinical findings of patients in multi-centers by generating inference intermediate results isolated from data without exposing the original data , to construct a virtual diagnosis and treatment path, so as to realize clinical decision support based on complete patient data, and provide complete and more accurate clinical decision support under the premise of ensuring data security and privacy.
  • the present invention uses SHA-256 hash encryption algorithm, SimHash encryption algorithm and ECC asymmetric encryption algorithm to anonymize and compare patient identity information to ensure complete encryption of data in out-of-hospital links and matching links, and triple node structure information Code mapping and asymmetric encryption are performed to ensure the security of data during transmission and only authorized participating centers can decrypt data, effectively ensuring data security and privacy protection in the process of multi-center joint reasoning.
  • the present invention constructs scattered electronic medical record data as a patient information model centered on the patient diagnosis and treatment process, and supports patient-centered personalized semantic reasoning; it utilizes semantic technology in data interactivity and scalability
  • the advantage of flexibility makes the present invention have good adaptability and expansibility to the heterogeneous data of different hospitals.
  • the clinical recommendations based on knowledge reasoning based on knowledge maps are derived from clinical guidelines and physician experience in line with evidence-based medicine.
  • the reasoning process and reasons for the recommendations can be obtained retrospectively by constructing reasoning examples, so that clinical recommendations can be given at the same time.
  • the reasoning process and the reason for the suggestion can be explained, and the doctor's trust in the decision support suggestion can be improved.
  • Figure 1 is a schematic diagram of the process flow of the multi-center electronic medical record knowledge map joint decision support method
  • Fig. 2 is a schematic diagram of the knowledge map clinical data structure framework
  • Figure 3 is a schematic diagram of the multi-center knowledge map joint reasoning process
  • Figure 4 is a schematic diagram of the multi-center electronic medical record knowledge map joint decision support system structure.
  • a multi-center knowledge map joint decision support method and system is based on the electronic medical record knowledge map and blockchain technology, combined with hash encryption and asymmetric encryption, to realize multi-center data security in a secure environment.
  • the combined reasoning of the electronic medical record knowledge graph provides complete and accurate clinical decision support by integrating the multi-center fragmented clinical data of patients without leaving the original data in the hospital and without exposing privacy.
  • the electronic medical records in each hospital are converted into the form of semantic triples, and semantic reasoning is performed through the local knowledge map in the hospital, and intermediate results of reasoning such as clinical findings related to decision support are generated.
  • Medical discovery use distributed modules to build clinical diagnosis and treatment process information, semantic reasoning intermediate results and encrypted identity information into an encrypted sub-graph on the chain suitable for blockchain synchronization, while ensuring the isolation between the original data and the sub-graph; the system passes Blockchain technology synchronizes decentralized map nodes in multiple hospitals; through hash encryption matching, a complete patient cross-hospital diagnosis and treatment process is constructed while ensuring data privacy, and the intermediate results are mapped and reconstructed through semantic reasoning. Summarize and integrate the clinical findings of local map inference in multiple hospitals to generate comprehensive clinical decision support results.
  • the inventive method specifically comprises the steps:
  • the electronic medical record knowledge map is constructed through medical literature and medical standard terminology, specifically: building a knowledge map medical knowledge base through clinical guidelines, OMOP (Observational Medical Outcomes Partnership) terminology database, doctor experience and medical literature as knowledge sources; coded with OMOP terms Unified identification of medical concepts, and construction of the top-level framework of medical knowledge structure based on the OMOP general data model, and establishment of top-level semantic classes and semantic relationships; for specific disease information, add sub-categories, sub-relationships, instances and attribute information.
  • the knowledge map uses the OWL restrict language and Apache Jena Rules language to build a clinical decision support reasoning rule base.
  • the node elements in the rules conform to the knowledge map medical knowledge structure framework and the OMOP terminology coding system.
  • the clinical data is established as a semantic triplet based on the knowledge graph framework, and is associated with the medical knowledge in the knowledge graph of the electronic medical record.
  • the semantic triplet of clinical data adopts the three-level clinical data semantic model of "patient-visit-diagnosis and treatment".
  • the ontology node modeling and numerical information entry of the electronic medical record data records are carried out;
  • the semantic framework of the knowledge map clinical data is shown in Figure 2
  • the left side shows the semantic structure of clinical data of the local knowledge map, which contains complete clinical information of patients.
  • the case record instance is subordinate to the patient instance, and the examination instance, prescription instance, diagnosis instance and clinical discovery instance are subordinate to the case record instance, and through the corresponding
  • the class identifies the specific data type; the right side is the corresponding semantic structure of the intermediate results used for blockchain synchronous reasoning, using the same top-level architecture, only retaining the core intermediate results and anonymized patient identity information, and isolating other original medical data information.
  • Patient instances, medical visit record instances, and clinical discovery instances correspond one-to-one to on-chain patient instances, on-chain medical visit instances, and on-chain clinical discovery instances through semantic relationships.
  • the local knowledge graph is inferred to generate clinical findings isolated from the original data, which are used as inference intermediate results for aggregating decision support recommendations.
  • the local knowledge map generates corresponding instance nodes on the virtual chain for patient instances, medical treatment instances and clinical discovery instances; the instance nodes on the virtual chain correspond to local information, and only the minimum joint reasoning information and
  • the hashed encrypted patient identity information does not retain the source data information and reasoning process information of clinical findings, so it is used for synchronous joint reasoning on the multi-center chain.
  • Each center adopts the same knowledge graph framework, which can rebuild the virtual nodes on the chain into the local knowledge graph after encrypted patient matching, and use the intermediate results of reasoning to assist the local knowledge graph to make comprehensive semantic reasoning.
  • the local knowledge map G obtains the corresponding semantic rules from the semantic rule base of the knowledge map according to the reasoning requirements of clinical decision support, and uses the semantic reasoning method to infer the clinical discovery information based on the local patient information triplet, which is used for Generate inference intermediate result nodes.
  • the inference engine is based on triples (s, r, o) ⁇ G and inference rules get results.
  • the right side of the arrow is the newly created triplet (s r , o r ) ⁇ r r after satisfying the condition on the left; Under the conditions of all conditional triples and conditional operation relations on the side, the triplets on the right side are added to the map, so as to realize the triplet operation based on the condition.
  • s i , s r are triplet head nodes, r i , r r are semantic relations, o i , or r are triplet tail nodes, o j is numerical patient information, ⁇ j is constant condition or reasoning three Variable conditions in a tuple.
  • the reasoning condition l i ⁇ (s i , o i ) ⁇ r i can be: s i is an instance of patient i, r i is the semantic relationship "blood creatinine check", and o i is the value of the test result, then it is required Only when the patient has the result of the blood creatinine test can the condition be met;
  • the operation relationship f(o j , ⁇ j ) can be: o j is the value of the above test result, ⁇ j is the threshold value, and f is the operation relationship "greater than", then the blood creatinine test is required Only when the result is greater than the threshold can the condition be met;
  • the new triplet (s r , or r ) ⁇ r r can be: s r is an example of a pre-narrative patient, r r is the semantic relationship "there is a clinical finding", or r is abnormal blood creatinine, It is the clinical finding of new abnormal serum creatinine in patients
  • the generated clinical discovery information together with the medical information and patient information, according to the semantic structure of the knowledge map, is constructed as a sub-graph on the chain for synchronization, which is used for the information transfer of the multi-center knowledge map and assists the local joint reasoning of the multi-center knowledge map .
  • the local map For all patient instances, visit instances and clinical discovery instances p, v, c ⁇ G, the local map generates virtual instances p′, v′, c′ required for on-chain synchronization, where (p, p′) ⁇ : hasVirtualPatient, (v, v') ⁇ : hasVirtualVisit, (c, c') ⁇ : hasVirtualClinicalFinding; the virtual instance corresponds to the node in the local map, and the universal unique identification code of each instance is constructed based on the ISO/IEC11578:1996 standard.
  • the virtual patient instance p′ injects the patient’s unique identity information and non-unique identity information at the same time; the virtual visit instance only includes the time window information of the start date and the end date of the visit; the virtual clinical discovery instance only includes the confirmation time of clinical findings, clinical findings Positive and negative results of category and clinical findings. All virtual instances p', v', c' and their corresponding information are constructed on-chain subgraphs according to the on-chain semantic structure on the right side of Figure 2.
  • rdf:type is the affiliation relationship
  • Delete the hierarchical structure owl:subProperty of all relational properties remove the association with the root node owl:ObjectProperty of the relational property, that is, delete the triplet satisfying p ⁇ G:(p,owl:objectProperty) ⁇ rdf:type, Satisfy p top ⁇ G: (p, p top ) ⁇ owl: subProperty triplet to delete, where p top is the upper-
  • the local visit instance v establishes the triple relation (v, cl) ⁇ rdf:type with all clinical discovery classes cl that have completed reasoning, marking that the visit record has completed the corresponding reasoning, and for the next intermediate process reasoning request , the search feedback will be directly obtained from the clinical findings corresponding to the case, without repeated reasoning.
  • s 1 ⁇ name, gender, date of birth, phone number ⁇
  • s 2 ⁇ province of residence, city of residence, district of residence, other addresses of residence ⁇
  • s 3 ⁇ Province of work address, city of work address, district of work address, others of work address ⁇ .
  • the residential address and work address are obtained by word segmentation of the original string using the NLPIR natural language processing tool.
  • Construct weight vector Construct 1 ⁇ 4 weight vector for non-unique identity hash value h(d i ) The weight vector is used to generate the non-unique identity weighted labels.
  • Weighted operation construct an operation matrix for the input vector H
  • Each line of hash value h(d i ) in H constructs the corresponding operation array h′(d i ), for the jth digit b ij of h(d i ), j ⁇ [1,6] and h′( The jth digit b′ ij of d i ), j ⁇ [1, 6], we have For invalid data and data not included in the non-unique identifier, all the rows in the corresponding operation matrix are set to zero.
  • the encrypted triplet file is used to construct a block chain synchronization data structure consisting of three parts: block data group, triplet data group and communication data group.
  • the block data group includes block number, block hash value, Merkle tree root and time record stamp information, which is used for matching verification between multi-center block nodes;
  • the triple group data group is encrypted on the chain Subgraph triplet file data;
  • communication data group records system process status, database connection status, P2P node connection parameters and triplet file identification, which is used for blockchain node communication and process identification.
  • the blockchain data synchronization of each hospital uses libp2p technology to realize point-to-point data transmission between nodes, and the PoS (Proof of Stake) algorithm is used between nodes to realize the blockchain consensus mechanism to ensure the data consistency of each node and block accuracy.
  • PoS Proof of Stake
  • the triplet information in the on-chain subgraph obtained synchronously by the blockchain compare the identity hash value group of the local patient with the patient identity hash value group in the on-chain subgraph, and calculate the similarity result of the identity hash , two patients above the threshold will be matched as the same patient.
  • the specific process is:
  • H SHA (u i ) and H SHA (u′ i ) are hash values of ID number or medical insurance number
  • H sim ( s j ) and H sim (s′ j ) are SimHash secret values of name, gender, date of birth, phone feature group (basic information group), residential address, or work address.
  • Hospital A which initiates multi-center joint reasoning, obtains patient visit records B 1 and patient visit records C 1 of participating central hospitals B and Hospital C through blockchain synchronization and patient matching. And obtain the inference intermediate results obtained by local semantic reasoning clinical findings B 1 and clinical findings C 1 ; initiating hospital A integrates the patient information of its own hospital and the obtained inference intermediate results of hospital B and hospital C, and uses semantic reasoning methods to generate clinical decision support As a result, ancillary clinical tasks. Specifically:
  • the local knowledge map obtains the on-chain subgraphs synchronized by other hospitals. After the triplet data decryption and patient matching, the relevant triplet nodes are mapped and reconstructed in the local knowledge map, so that the synchronization information is recorded in In the local knowledge graph.
  • the triple node information of all subgraphs on the chain is mapped to the new node g chain in the local knowledge map, that is, ⁇ g chain , r, o> ⁇ g′, r′, o′>, so that the synchronized
  • the clinical findings obtained from patient consultation information and reasoning are recorded into the local knowledge graph for joint reasoning.
  • the institution that initiates multi-center joint reasoning obtains the reasoning intermediate result nodes of other centers, and performs triple mapping and reconstruction based on the patient matching results of the local knowledge graph.
  • the intermediate result nodes of multi-center reasoning are classified, aggregated and sorted according to the time of the visit instance.
  • the local knowledge map generates rule-based clinical decision support suggestions based on the multi-center reasoning intermediate results and local clinical data, including disease diagnosis suggestions, clinical examination suggestions, treatment plan suggestions, etc., for reasoning and doctor-oriented local clinical decision support disease risk warning.
  • the present invention also proposes a multi-center electronic medical record knowledge map joint decision support system suitable for deployment and application in a multi-center environment, which is used to assist doctors in making comprehensive clinical decisions.
  • the system framework is shown in Figure 4.
  • the system includes a local knowledge graph module, a distributed module and an on-chain module;
  • the local knowledge map module collects, constructs, saves and invokes the medical knowledge map and patient information model; reasoning the nodes in the knowledge map according to the semantic rules to generate decision support suggestions; through limited node interaction, transmit the information to the distributed module on the chain Synchronize triplet information, and obtain medical information synchronized by other centers from the distributed module.
  • the local knowledge map module constructs a local knowledge map through medical literature and medical standard terminology, establishes clinical data as a semantic triplet of "patient-diagnosis-diagnosis and treatment", and performs the same process with the medical knowledge in the local knowledge map.
  • the local knowledge map uses Apache Jena TDB2 as a database to store knowledge map nodes and data, and uses Apache Jena Fuseki2 as a map SPARQL terminal to query and modify triplet nodes.
  • the Hermit reasoning engine and Apache Jena reasoning engine based on the clinical decision support reasoning rule base to reason the relationship between the triples and infer the patient's clinical findings
  • each center obtains the inference intermediate results of the matching patients in the sub-graph on the chain, based on the local patient information and inference intermediate results, combined with the local knowledge graph, through semantic reasoning Method, using the SPARQL language to query the triples related to the patient's clinical decision support from the local knowledge graph, sorting out the decision support information and generating interface-based decision support information;
  • the distributed module is used to construct subgraphs on the chain and perform patient matching; the distributed module obtains triplet information for distributed reasoning and intermediate result summary from the local knowledge map system through limited entity interaction, and the triplet The information in is parsed and the triples are reconstructed according to the semantic structure of the subgraph on the chain.
  • the limited entity interaction function filters triplet nodes sent to the distributed module, all patient information triplets, diagnosis records, inspection records and prescription record nodes will be filtered, and other manually defined restricted nodes will also be filtered to prevent unauthorized Licensed data and raw data flow out of the hospital.
  • the construction of sub-graphs on the chain is specifically: based on the inference intermediate results generated by the local knowledge graph module, according to the semantic structure of the local knowledge graph, based on patient matching information, time window information of patient visits, and clinical discovery information obtained by reasoning, construct a block for The on-chain subgraph of the chain synchronization, and encrypt the triplet information through the encryption module, and transmit it to the on-chain module; the encryption module encrypts the patient matching information, and generates an anonymized patient identity hash value through the hash module , the semantic structure information and clinical discovery information in the triplet are de-expressed through the way of coding table mapping, and then the triplet file is encrypted;
  • the distributed module performs hash encryption and SimHash encryption on the patient identity information in the triplet synchronized on the chain, encodes and maps other structural elements, and performs asymmetric encryption on the triplet file, so as to realize the sub-graph 3 on the chain. Full encryption of tuples.
  • Patient matching is specifically: each center obtains the triplet information in the sub-graph on the chain according to the block chain synchronization results of the modules on the chain, and compares the local patient identity hash value with the patient identity hash value in the sub-graph on the chain , get the patient matching result according to the similarity of the identity hash value;
  • the on-chain module constructs an area composed of block data group, triplet data group and communication data group through the data communication module.
  • the block chain synchronization data structure is used to record the block state of the file and the state of the multi-center reasoning process, and synchronize the block chain through the consensus mechanism, and perform point-to-point data transmission in the multi-center.
  • the on-chain module uses blockchain technology to realize the information transfer of sub-graphs on the chain between multiple centers, and accurately synchronizes the intermediate process nodes required for multi-center knowledge graph reasoning between different sub-centers; the module also maintains an immutable on-chain Data circulation records support the traceability of the data request and sending process between nodes.
  • the blockchain node communication system uses libp2p technology to realize the point-to-point communication protocol between multi-center nodes.
  • the nodes are linked through the tcp protocol according to the node network address information, and at the same time determine the basic network transmission protocol and obtain the data public key; libp2p saves the node information in In the local database, multiple node paths are found through node routing to ensure transmission efficiency and stability when performing multi-center blockchain data synchronization.
  • the blockchain database system uses the MySQL database to store blockchain synchronization data, blockchain system data, and synchronization data mapping tables, which are used for the on-chain synchronization process operation of the multi-center knowledge graph.
  • the database also saves the blockchain log records to record all the synchronization information and operation behaviors on the chain. The log is verified synchronously on the chain to ensure that all operation records can be traced and queried throughout the process, and the records cannot be tampered with.
  • the interface of the module on the chain is written using the ASP.NET framework, which is used for system docking with the distributed modules of the local knowledge map, sending communication requests and test requests to other nodes, and providing operation record query services.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Computer Security & Cryptography (AREA)
  • Public Health (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Primary Health Care (AREA)
  • Computer Hardware Design (AREA)
  • Epidemiology (AREA)
  • Bioethics (AREA)
  • Databases & Information Systems (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

本发明公开了一种多中心知识图谱联合决策支持方法与系统，利用医学知识图谱技术和区块链技术，通过本地知识图谱和链上同步图谱相结合的方式，实现临床数据的本地语义推理和链上结果汇总，从而在原始医疗数据不出医院的情况下，依靠知识图谱技术综合患者碎片化的跨机构医疗数据，基于演绎推理和循证医学，给出包含完整患者临床证据的、可解释的临床决策支持。本发明将患者身份信息进行匿名化比对，保证数据在院外环节和匹配环节的完全加密，将三元组节点结构信息进行编码映射和非对称加密，保证数据在传输过程中的安全以及只有有权限的参与中心可以解密数据，有效保证了多中心联合推理过程中的数据安全和隐私保障。

Description

一种多中心知识图谱联合决策支持方法与系统 技术领域
本发明涉及医学知识图谱技术领域，尤其涉及一种多中心知识图谱联合决策支持方法与系统。
背景技术
临床实践中，许多患者会在多家医院或社区医院就诊；研究显示，20％-40％的患者于1年内会在平均2家不同的医院就诊。患者的跨机构就诊会造成医疗数据的碎片化、孤立化，造成单家医院内的患者病历记录不完整、不充分；患者医疗数据不全面容易引发临床医生做出不准确的临床决策，造成诊断不及时、治疗不恰当和重复医疗等行为，严重威胁医疗服务质量，增加群众医疗负担。现有研究发现，医疗数据不完整会对44％的患者造成负面影响，其中59.5％的患者因此发生诊疗不及时和重复诊疗行为；尤其是针对需要长期监测、长期管理的慢性疾病，医疗数据碎片化更容易导致对患者慢性疾病诊断和管理的不及时，影响疾病的知晓率、治疗率。然而医疗数据敏感性强，整合利用多中心碎片化的患者临床数据过程中，必须考虑到数据安全和隐私保护问题。因此，需要研究多中心碎片化电子病历数据的联合分析与决策支持技术，在保障数据隐私和安全的前提下，综合患者在多家医院的诊疗记录，全面分析患者的病情病史、治疗方法、过敏禁忌等信息，辅助临床医生对患者做出全面、准确、及时的临床决策，避免重复医疗，有效提升医疗服务质量水平。
现有的多中心医疗数据应用的主要技术方案是通过分布式学习的方式，利用多中心碎片化临床数据进行模型建模，从而实现风险预测、疾病诊断等临床决策支持功能。通过在多家医院内使用相同的医学数据格式标准，在统一的数据结构下分发临床队列构建标准和机器学习算法；各医院内部参照下发的队列构建标准，在相同的数据结构上建立本地的研究队列，并使用相同的机器学习算法进行本地训练；最终在原始数据不出医院的情况下，将训练结果汇总，形成临床决策模型。根据训练模型对医院本地数据进行分析，从而实现决策支持功能。
现有的利用多中心碎片化医疗数据进行决策支持的技术存在以下问题：(1)现有技术只在模型训练阶段应用多中心的医疗数据，训练完成的模型在临床实际应用中依然只能应用单家医院的患者数据进行分析，其决策支持结果仍然缺乏其他医院的医疗数据作为辅助，生成的临床决策在全面性和可靠性上依然存在问题。(2)基于分布式机器学习构建的决策支持模型，其结果主要以置信权重的形式体现，不能给出基于循证医学的演绎式决策支持结果，难以系统化、全面化的展示决策支持相关的患者疾病风险因素和临床证据，容易造成医生接受 度低。(3)基于循证医学的演绎式语义推理技术，其分布式算法主要用于分布式数据搜索和提升三元组推理速度为主；针对碎片化患者数据分析和应用的临床场景，缺少数据安全和隐私保护支持，无法在不汇总多中心原始数据的情况下进行语义推理，在医疗数据安全性上依然存在问题。
发明内容
本发明目的在于针对现有技术的不足，提出一种多中心知识图谱联合决策支持方法与系统，利用医学知识图谱技术和区块链技术，通过本地知识图谱和链上同步图谱相结合的方式，实现临床数据的本地语义推理和链上结果汇总，从而在原始医疗数据不出医院的情况下，依靠知识图谱技术综合患者碎片化的跨机构医疗数据，基于演绎推理和循证医学，给出包含完整患者临床证据的、可解释的临床决策支持。本发明利用面向电子病历的知识图谱，有效适配多家医院异构化的电子病历数据结构和术语体系，可以实现决策支持系统在多家医院的本地化部署和应用，解决了多中心医疗系统异构性难题；本发明通过本地图谱对医疗数据进行语义推理，将推理生成的与原始数据相分离的临床发现通过非对称加密进行区块链同步，利用哈希加密实现患者身份匹配，从而确保整个决策支持过程中原始数据不出医院、不发生数据暴露，解决了多中心数据应用的数据隐私保护问题；本发明通过本地图谱和链上图谱相结合的方式，有效汇总同一患者在多家医院的临床发现，从而在推理过程中有效利用患者多中心全覆盖的医疗数据，生成可靠的临床决策支持，解决了现有系统只能利用单家医院数据进行分析的问题。
本发明的目的是通过以下技术方案来实现的：一种多中心知识图谱联合决策支持方法，该方法包括如下步骤：
(1)通过医学文献以及医学标准术语集构建本地知识图谱；将临床数据构建为“患者-就诊-诊疗”的语义三元组，并同本地知识图谱中的医学知识进行关联；
(2)基于步骤(1)构建的本地知识图谱和语义三元组，根据临床决策支持的推理需求，通过语义推理方法，推理出患者临床发现信息，生成推理中间结果；根据本地知识图谱的语义结构，基于患者匹配信息、患者就诊的时间窗信息和推理获得的临床发现信息，构建用于区块链同步的链上子图，链上子图隔绝其他患者原始医疗信息；
(3)将患者匹配信息进行加密，生成匿名化患者身份哈希值；将语义三元组中的语义结构信息和临床发现信息通过编码表映射的方式进行信息去明示化；之后进行三元组文件加密；
(4)基于步骤(3)中加密后的三元组文件，构建由区块数据组、三元组数据组和通讯数据组三部分组成的区块链同步数据结构，在多中心进行点对点的数据传输；
(5)各中心根据区块链同步获得的链上子图中的三元组信息，比对本地患者身份哈希值 与链上子图中的患者身份哈希值，根据身份哈希值的相似度得到患者匹配结果；
(6)各中心根据患者匹配结果，获取链上子图中的匹配患者的推理中间结果，基于本地患者信息和推理中间结果，结合本地知识图谱，通过语义推理生成决策支持结果。
进一步地，步骤(1)中，医学标准术语集采用OMOP术语库；本地知识图谱基于医生经验和医学文献构建临床决策支持推理规则库，规则中的节点元素符合知识图谱医学知识结构和OMOP术语编码体系。
进一步地，步骤(1)中，临床数据的语义三元组采用“患者-就诊-诊疗”的三级临床数据语义模型，将每位患者的电子病历数据，按照临床数据语义模型进行本体节点建模和数值信息录入。
进一步地，步骤(2)中，根据患者就诊记录的所属临床发现类，基于临床决策支持推理规则库获得对应的语义推理规则，对于本地知识图谱，使用推理机基于语义三元组和语义推理规则获得语义推理结果。
进一步地，步骤(2)中，针对所有患者实例、就诊实例和临床发现实例，基于本地知识图谱生成链上同步所需的相应虚拟实例，其中虚拟患者实例同时注入患者的唯一身份信息和非唯一身份信息，虚拟就诊实例包含就诊起始日期和就诊终止日期构成的时间窗信息，虚拟临床发现实例包含临床发现确认的时间、临床发现所属的类别和临床发现的阳性阴性结果。
进一步地，步骤(3)中，所述患者匹配信息包括患者的唯一身份信息和非唯一身份信息；所述唯一身份信息为患者的身份证ID和医保ID，采用SHA-256哈希算法进行加密，生成唯一身份标识哈希值；所述非唯一身份信息为姓名、性别、生日、住址和工作地址，采用SimHash算法进行加密，生成非唯一身份标识哈希值。
进一步地，步骤(4)中，所述区块数据组包括区块编号、区块哈希值、默克尔树根和时间记录戳信息，用于多中心区块节点间的匹配校验；所述三元组数据组为经过加密的链上子图三元组文件数据；所述通讯数据组记录数据流程状态、数据库连接状态、P2P节点连接参数和三元组文件标识，用于区块链节点通讯和流程判别。
本发明还提供了一种多中心知识图谱联合决策支持系统，该系统包括本地知识图谱模块、分布式模块和链上模块；
所述本地知识图谱模块通过医学文献以及医学标准术语集构建本地知识图谱，将临床数据建立为“患者-就诊-诊疗”的语义三元组，并同本地知识图谱中的医学知识进行关联；基于构建的本地知识图谱和语义三元组，根据临床决策支持的推理需求，通过语义推理方法，推理出患者临床发现信息，生成推理中间结果；各中心根据分布式模块的患者匹配结果，获取链上子图中的匹配患者的推理中间结果后，基于本地患者信息和推理中间结果，结合本地 知识图谱，通过语义推理生成决策支持结果；
所述分布式模块用于构建链上子图和进行患者匹配；
构建链上子图具体为：基于本地知识图谱模块生成的推理中间结果，根据本地知识图谱的语义结构，基于患者匹配信息、患者就诊的时间窗信息和推理获得的临床发现信息构建用于区块链同步的链上子图，并通过加密模块将三元组信息加密后，并传输至链上模块；所述加密模块将患者匹配信息进行加密，生成匿名化患者身份哈希值，将三元组中语义结构信息和临床发现信息，通过编码表映射的方式进行信息去明示化，之后进行三元组文件加密；
患者匹配具体为：各中心根据链上模块的区块链同步结果，获得链上子图中的三元组信息，比对本地患者身份哈希值与链上子图中的患者身份哈希值，根据身份哈希值的相似度得到患者匹配结果；
所述链上模块根据分布式模块构建的链上子图中加密后的三元组文件，通过数据沟通模块构建为由区块数据组、三元组数据组和通讯数据组三部分组成的区块链同步数据结构，并通过共识机制进行区块链同步，在多中心进行点对点的数据传输。
进一步地，所述分布式模块通过有限实体交互从本地知识图谱模块中获取用于分布式推理和推理中间结果汇总的三元组信息，对三元组中的信息进行解析并隔绝患者原始医疗数据信息，之后按照链上子图的语义结构进行三元组重构。
进一步地，所述本地知识图谱模块生成的决策支持结果传输至分布式模块并进行三元组信息加密后，传递至链上模块进行区块链同步，用于其他中心的决策支持。
本发明的有益效果：
1、本发明利用电子病历知识图谱系统和OMOP(Observational Medical Outcomes Partnership)CDM(Common Data Model)标准术语体系，将多中心电子病历数据构建为统一医学概念编码、统一数据语义结构的临床数据模型，从而使得多中心医疗数据可以在统一的、标准化的数据结构下进行联合推理。本发明利用区块链技术实现多中心本地知识图谱的推理中间结果同步，在不暴露原始数据的情况下，通过生成与数据隔离的推理中间结果，将患者在多中心碎片化的临床发现进行整合，构建虚拟的诊疗路径，从而实现基于完整患者数据的临床决策支持，在保障数据安全和隐私的前提下给出完善的、准确度更高的临床决策支持。本发明使用SHA-256哈希加密算法、SimHash加密算法和ECC非对称加密算法，将患者身份信息进行匿名化比对，保证数据在院外环节和匹配环节的完全加密，将三元组节点结构信息进行编码映射和非对称加密，保证数据在传输过程中的安全以及只有有权限的参与中心可以解密数据，有效保证了多中心联合推理过程中的数据安全和隐私保障。
2、本发明通过构建患者信息模型，将分散的电子病历数据建构为以患者诊疗过程为中心 的患者信息模型，支撑以患者为中心的个性化语义推理；发挥语义技术在数据交互性和可扩展性的优势，使得本发明对不同医院的异构数据有较好的适应性和扩展性。同时，基于知识图谱知识推理得出的临床建议，来源均为符合循证医学的临床指南和医师经验，推理流程和建议原因通过构建推理实例可以追溯获取，从而能够在给出临床建议的同时给出推理过程和建议原因，提升医师对决策支持建议的信任度。
附图说明
图1为多中心电子病历知识图谱联合决策支持方法流程示意图；
图2为知识图谱临床数据结构框架示意图；
图3为多中心知识图谱联合推理过程示意图；
图4为多中心电子病历知识图谱联合决策支持系统结构示意图。
具体实施方式
以下结合附图对本发明具体实施方式作进一步详细说明。
如图1所示，本发明提供的一种多中心知识图谱联合决策支持方法与系统，基于电子病历知识图谱和区块链技术，结合哈希加密和非对称加密，实现多中心数据安全环境下的电子病历知识图谱联合推理，在原始数据不出医院、不暴露隐私的情况下，综合患者的多中心碎片化临床数据，提供完整、准确的临床决策支持。各医院院内电子病历转换为语义三元组形式，通过院内本地知识图谱进行语义推理，生成与决策支持相关的临床发现等推理中间结果，推理中间结果同原始数据相隔离，只表达基于原始数据的医学发现；利用分布式模块，将临床诊疗过程信息、语义推理中间结果和加密身份信息构建为适用于区块链同步的链上加密子图谱，同时保证原始数据与子图谱间的隔离；系统通过区块链技术，在多家医院中进行去中心化的图谱节点同步；通过哈希加密匹配，在保障数据隐私的状态下构建完整的患者跨院诊疗流程，并通过语义推理中间结果映射与重建，将多家医院本地图谱推理的临床发现进行汇总和整合，生成全面的临床决策支持结果。
本发明方法具体包括如下步骤：
(1)本地知识图谱的构建
电子病历知识图谱通过医学文献以及医学标准术语集构建，具体为：通过临床指南、OMOP(Observational Medical Outcomes Partnership)术语库、医生经验和医学文献为知识来源构建知识图谱医学知识库；以OMOP术语编码为医学概念进行统一标识，并基于OMOP通用数据模型对医学知识结构进行顶层框架构建，建立顶层语义类和语义关系；针对具体疾病的信息，添加子类、子关系、实例和属性信息。知识图谱运用OWL restrict语言和Apache Jena Rules语言构建临床决策支持推理规则库，规则中的节点元素符合知识图谱医学知识结构框架 和OMOP术语编码体系。
临床数据基于知识图谱框架建立为语义三元组，并同电子病历知识图谱中的医学知识进行关联。临床数据的语义三元组采用“患者-就诊-诊疗”的三级临床数据语义模型。每一位患者的个人信息、就诊记录、诊断记录、检查记录和处方记录，按照临床数据语义模型对电子病历数据记录进行本体节点建模和数值信息录入；知识图谱临床数据语义框架如图2所示，左侧为本地知识图谱的临床数据语义结构，包含完整患者临床信息，就诊记录实例从属于患者实例，检查实例、处方实例、诊断实例和临床发现实例从属于就诊记录实例，并通过对应的类标识具体的数据类型；右侧为对应的用于区块链同步推理中间结果语义结构，采用相同的顶层架构，仅保留核心中间结果和匿名化患者身份信息，并隔绝其他原始医疗数据信息，患者实例、就诊记录实例和临床发现实例通过语义关系一一对应链上患者实例、链上就诊实例和链上临床发现实例。
本地知识图谱经过推理生成与原始数据相隔离的临床发现，作为推理中间结果，用于汇总决策支持建议。本地知识图谱根据多中心联合推理需要，针对患者实例、就诊实例和临床发现实例，生成对应的虚拟链上实例节点；虚拟链上实例节点同本地信息相对应，只保留最低限度的联合推理信息和经过哈希加密的患者身份信息，不保留临床发现的来源数据信息和推理流程信息，从而用于多中心链上同步的联合推理。各个中心采用相同的知识图谱框架，可以将链上虚拟节点经过加密患者匹配后重建到本地知识图谱中，利用推理中间结果辅助本地知识图谱做出全面的语义推理。
(2)生成推理中间结果节点和链上子图
(2.1)本地知识图谱G根据临床决策支持的推理需求，从知识图谱语义规则库中获取对应的语义规则，通过语义推理方法，基于本地的患者信息三元组，推理出临床发现信息，用于生成推理中间结果节点。推理机基于三元组(s，r，o)∈G和推理规则
Figure PCTCN2022124692-appb-000001
获得结果。其中s为三元组头节点、r为语义关系、o为尾节点；推理规则中，箭头左侧为条件三元组l i≡(s i，o i)∈r i和条件运算关系f(o j，τ j)，i＝1...n，j＝1...k，箭头右侧为满足左侧条件后新建的三元组(s r，o r)∈r r；满足左侧所有条件三元组和条件运算关系的条件下，在图谱中添加右侧的三元组，从而实现基于条件的三元组操作。其中s i，s r为三元组头节点，r i，r r为语义关系，o i，o r为三元组尾节点，o j为数值类患者信息，τ j为常量条件或推理三元组中的变量条件。
举例说明：推理条件l i≡(s i，o i)∈r i可以为：s i为患者i的实例、r i为语义关系“血肌酐检查”、o i为检查结果的值，则要求患者具有血肌酐检查结果才能满足条件；运算关系f(o j，τ j)可以为：o j为上述检查结果的值、τ j为阈值、f为运算关系“大于”，则要求血肌酐检查结果 大于阈值才能满足条件；新建三元组(s r，o r)∈r r可以为：s r为前叙患者实例、r r为语义关系“存在临床发现”、o r为血肌酐异常，即为患者新建血肌酐异常的临床发现。在上述例子中，要求患者具有血肌酐检查结果，且大于一定阈值的条件下，给患者加上血肌酐异常的三元组标签。
(2.2)生成的临床发现信息连同就诊信息和患者信息，按照知识图谱的语义结构，构建为链上子图进行同步，用于多中心知识图谱的信息传递，辅助多中心知识图谱的本地联合推理。针对所有患者实例、就诊实例和临床发现实例p，v，c∈G，本地图谱生成链上同步所需的虚拟实例p′，v′，c′，其中(p，p′)∈：hasVirtualPatient，(v，v′)∈：hasVirtualVisit，(c，c′)∈：hasVirtualClinicalFinding；虚拟实例同本地图谱中的节点相对应，基于ISO/IEC11578:1996标准构建每一个实例的通用唯一识别码。虚拟患者实例p′同时注入患者的唯一身份信息和非唯一身份信息；虚拟就诊实例只包含就诊起始日期和就诊终止日期的时间窗信息；虚拟临床发现实例只包含临床发现确认的时间、临床发现所属的类别和临床发现的阳性阴性结果。所有虚拟实例p′，v′，c′及其对应的信息按照如图2中右侧的链上语义结构进行链上子图的构建。对于生成的中间过程节点：对所有节点g与本体根节点owl:thing之间的从属关系进行删除，即满足g∈G：(g，owl：thing)∈rdf：type的三元组进行删除，其中rdf:type为从属关系；删除所有临床发现节点cl同顶层临床发现类、患者类和就诊类之间的关联，即满足cl∈{：ClinicalFinding，：Patient，：Visit}的三元组进行删除；对所有关系属性的层级结构owl:subProperty进行删除，去除同关系属性根节点owl:ObjectProperty的关联，即满足p∈G：(p，owl：objectProperty)∈rdf：type的三元组进行删除，满足p top∈G：(p，p top)∈owl：subProperty的三元组进行删除，其中p top为上层属性关系。从而保证中间过程节点的简易性，降低链上同步的数据量。
(2.3)本地的就诊实例v建立同所有完成推理的临床发现类cl的三元组关系(v，cl)∈rdf：type，标记该就诊记录已经完成相应推理，对于下一次的中间过程推理请求，将直接从就诊实例对应的临床发现结果中进行搜索反馈，不需要进行重复推理。
(3)三元组信息加密
(3.1)针对患者的唯一身份信息和非唯一身份信息，使用SHA-256哈希加密算法和SimHash加密算法进行身份信息加密。任一患者唯一身份信息三元组<p，r，u>，满足患者节点(p，：Patient)∈rdf：type，r∈{：cardID，：insurID}为身份证ID和医保ID关系，则对三元组字符串u使用SHA-256哈希算法进行加密，生成唯一身份标识哈希值H SHA(u i)。无效数据和未包含数据置零。
(3.2)任一患者非唯一身份信息三元组<p，r，s>，满足患者节点(p，：Patient)∈rdf：type，r∈{：patientName，：gender，：birth，：address，：workAddrwss}为姓名、性别、生日、住址、工作地址，则对三元组字符串s使用SimHash算法进行加密，生成非唯一身份标识哈希值H sim(s i)。无效数据和未包含数据置零。具体过程为：
构建输入向量：将患者的非唯一身份标识分为3组输入向量：s 1＝{姓名，性别，生日，电话}，s 2＝{住址省，住址市，住址区，住址其他}，s 3＝{工作地址省，工作地址市，工作地址区，工作地址其他}。其中住址和工作地址使用NLPIR自然语言处理工具对原字符串进行分词获得。对于每组非唯一身份标识构建输入参数d i，i＝1...4；对每一个输入参数d i，使用哈希算法获得6位哈希值h(d i)＝{k 1，...，k 6}，其中k 1～k 6∈{0，1}；将所有4个输入参数获得的哈希数组构建为4×6的SimHash输入向量
Figure PCTCN2022124692-appb-000002
构建权重向量：针对非唯一身份标识哈希值h(d i)，构建1×4权重向量
Figure PCTCN2022124692-appb-000003
权重向量用于生成非唯一身份标识加权标签。通过使用140名患者在两家医院的非唯一身份标签进行测试评估，计算3组输入向量的最佳匹配权重，权重值分别为为W s1＝{4，5，4，1}，W s2＝{4，3，2，3}，W s3＝{4，3，2，3}。
加权运算：针对输入向量H构建运算矩阵
Figure PCTCN2022124692-appb-000004
H中的每一行哈希数值h(d i)构建对应的运算数组h′(d i)，对h(d i)的第j位数字b ij，j∈[1，6]和h′(d i)的第j位数字b′ ij，j∈[1，6]，有
Figure PCTCN2022124692-appb-000005
非唯一身份标识中的无效数据和未包含的数据，则其对应的运算矩阵中的行全部置零。计算加权矩阵
Figure PCTCN2022124692-appb-000006
生成标签：对加权矩阵H w进行列加和
Figure PCTCN2022124692-appb-000007
Q i为加权计算结果，根据对H s的第j位数字q j，j∈[1，6]，获得SimHash加密结果H sim(s)；其中H sim(s)的第j位数字
Figure PCTCN2022124692-appb-000008
(3.3)汇总患者唯一身份标识哈希值和非唯一身份标识哈希值，生成患者P的身份哈希值组H P＝{H SHA(u i)，H sim(s j)}，i＝1，2，j＝1，2，3。
(3.4)针对三元组中语义结构信息和临床发现信息，通过编码表映射的方式进行信息去明示化；所有的语义类、属性关系和数值关系资源，根据编码映射表，将三元组中的资源名 称替换为资源编码。
(3.5)完成信息加密的三元组文件，根据同步需求，进行全文件的加密，保证链上同步过程中的数据安全；针对一对多的同步需求，即单中心向多个中心同步决策支持推理需求和中间过程节点的情况，三元组文件通过KP-ABE属性基加密算法进行加密；针对一对一的同步需求，即单中心向另一个中心同步决策支持推理中间结果的情况，三元组文件通过ECC算法进行非对称加密。
(4)区块链同步
用加密后的三元组文件，构建为由区块数据组、三元组数据组和通讯数据组3部分组成的区块链同步数据结构。区块数据组包括区块编号、区块哈希值、默克尔树根和时间记录戳信息，用于多中心区块节点间的匹配校验；三元组数据组为经过加密的链上子图三元组文件数据；通讯数据组记录系统流程状态、数据库连接状态、P2P节点连接参数和三元组文件标识，用于区块链节点通讯和流程判别。
各个医院的区块链数据同步采用libp2p技术实现节点间的点对点的数据传输，节点间使用PoS(Proof of Stake，权益证明)算法实现区块链共识机制，保证各个节点的数据一致性和区块准确性。
(5)患者匹配
根据区块链同步获得的链上子图中的三元组信息，比对本地患者的身份哈希值组与链上子图中的患者身份哈希值组，计算身份哈希的相似度结果，超过阈值的两个患者将被匹配为同一患者。具体过程为：
本地知识图谱中患者P的身份哈希值组为H P＝{H SHA(u i)，H sim(s j)}，链上子图中患者P′的身份哈希值组为H P′＝{H SHA(u′ i)，H sim(s′ j)}；其中H SHA(u i)和H SHA(u′ i)为身份证号或医保号的哈希密值，H sim(s j)和H sim(s′ j)为姓名、性别、生日、电话特征组(基础信息组)，居住地址，或工作地址的SimHash密值。
(5.1)对每一项配对的哈希密值，分别计算其相似度S i(H SHA(u i)，H SHA(u′ i))，i＝1，2及S j(H sim(s j)，H sim(s′ j))，j＝1，2，3。针对唯一身份标识哈希密值的相似度采用完全相同比较法进行计算，即
Figure PCTCN2022124692-appb-000009
针对非唯一身份标识哈希密值的相似度采用汉明距离法进行计算，对于哈希密值H sim(s j)和H sim(s′ j)，其密值的每一位数字为b k∈H sim(s j)和b′ k∈H sim(s′ j)，k＝1...6；哈希密值的相似度
Figure PCTCN2022124692-appb-000010
(5.2)对所有哈希密值相似度进行加权求和，获得密值配对权重和Sum＝∑w iS i+∑w jS j， 其中w i和w j为相似度权重，身份证哈希密值权重w 1＝1.0，医保卡号哈希密值权重w 2＝0.4，基础信息组SimHash密值权重w 3＝0.3，住址SimHash密值权重w 4＝0.15，工作地址SimHash密值权重w 5＝0.15。权重值基于140名患者在两家医院的非唯一身份标签进行测试评估获得，上述权重为获得最佳匹配准确度的权重测试结果。
(5.3)患者匹配分为两种情况：当患者的身份证哈希密值匹配时，即患者的匹配密值配对权重和Sum≥1.0时，则判断两名患者为相同患者，此时只保留权重和大于等于1的匹配项，将所有匹配项认定为相同患者，并将其信息整合为同一个患者实例记录到本地知识图谱；当患者无身份证哈希密值或未能匹配身份证哈希密值时，若患者匹配的密值配对权重和Sum＞0.65时，则判断两名患者为相似患者，此时保留权重和最高的3个匹配项用于作为备选患者进行多中心联合推理。两种权重值阈值基于140名患者在两家医院的非唯一身份标签进行测试评估获得，上述权重为获得最佳匹配准确度的权重测试结果。
(6)多中心联合推理汇总
多中心联合推理汇总过程如图3所示，发起多中心联合推理的医院A，通过区块链同步和患者匹配获取参与中心医院B和医院C的患者就诊记录B 1和患者就诊记录C 1，并获取本地语义推理获得的推理中间结果临床发现B 1和临床发现C 1；发起医院A综合本院的患者信息和获取的医院B和医院C的推理中间结果，利用语义推理方法生成临床决策支持结果，辅助临床任务。具体为：
(6.1)本地知识图谱获取其他医院同步来的链上子图，经过三元组数据解密和患者匹配后，相关的三元组节点在本地知识图谱中进行映射和重建，从而将同步信息记录到本地知识图谱中。对于已经匹配的患者实例p和链上子图中的虚拟患者实例p′满足(p，p′)∈：hasVirtualPatient，在本地知识图谱G中构建链上子图中患者p′的三元组节点g chain∈G：(g cgain，g′)∈{：hasVirtualVisitFrom，：hasVirtualClinicalFindingFrom}，即按照链上子图中的信息节点构建本地图谱的对应节点；其中(p′，g′)∈{：hasVisit，：hasClinicalFinding}，g′为链上子图节点，即链上子图中患者p′的各项就诊信息和临床发现信息。所有链上子图的三元组节点信息映射到本地知识图谱中的新建节点g chain中，即<g chain，r，o>≡<g′，r′，o′>，从而将同步来的患者就诊信息与推理获得的临床发现记录到本地知识图谱中用于联合推理。
(6.2)本地知识图谱完成链上子图整合后，根据链上子图的观测起始时间t s和观测终止时间t e，确定符合观测时间段的就诊实例v∈G：(v，t)∈：hasVisitTime∧(t≥t s)∧(t≤t e)，就诊实例v包含本地就诊记录和链上同步获得的虚拟就诊记录。根据虚拟就诊记录<v′，rdf：type，cl>的所属临床发现类cl，执行对应的语义规则推理。
(6.3)发起多中心联合推理的机构获取其他中心的推理中间结果节点，并根据本地知识图谱的患者匹配结果进行三元组映射和重建。多中心推理中间结果节点按照就诊实例的时间进行分类汇总和分类排序。本地知识图谱基于多中心汇总的推理中间结果和本地的临床数据，生成基于规则临床决策支持建议，包括疾病诊断建议、临床检查建议、治疗方案建议等，用于本地临床决策支持的推理和面向医生的疾病风险预警。
本发明还提出了一种适用于多中心环境部署应用的多中心电子病历知识图谱联合决策支持系统，用于辅助医生做出全面的临床决策，系统框架如图4所示。该系统包括本地知识图谱模块、分布式模块和链上模块；
本地知识图谱模块对医学知识图谱和患者信息模型进行采集、构建、保存和调用；对知识图谱中的节点按照语义规则进行推理，生成决策支持建议；通过有限节点交互，向分布式模块传输链上同步用的三元组信息，并从分布式模块获取其他中心同步获得的医学信息。具体为：所述本地知识图谱模块通过医学文献以及医学标准术语集构建本地知识图谱，将临床数据建立为“患者-就诊-诊疗”的语义三元组，并同本地知识图谱中的医学知识进行关联；本地知识图谱使用Apache Jena TDB2作为数据库，保存知识图谱节点和数据，使用Apache Jena Fuseki2作为图谱SPARQL终端，负责三元组节点的查询和修改。基于构建的本地知识图谱和语义三元组，根据临床决策支持的推理需求，使用Hermit推理机和Apache Jena推理机基于临床决策支持推理规则库，对三元组关系进行推理，推理出患者临床发现信息，生成推理中间结果；各中心根据分布式模块的患者匹配结果，获取链上子图中的匹配患者的推理中间结果后，基于本地患者信息和推理中间结果，结合本地知识图谱，通过语义推理方法，使用SPARQL语言从本地知识图谱中查询与患者临床决策支持相关的三元组，对其中的决策支持信息进行整理和并生成界面化的决策支持信息；
所述分布式模块用于构建链上子图和进行患者匹配；分布式模块通过有限实体交互从本地知识图谱系统中获取用于分布式推理和中间结果汇总的三元组信息，对三元组中的信息进行解析并按照链上子图的语义结构进行三元组重构。有限实体交互功能过滤向分布式模块输送的三元组节点，所有患者信息三元组、诊断记录、检查记录和处方记录节点将被过滤，其他人工定义的限制节点也将被过滤，防止未经许可的数据和原始数据流出医院。
构建链上子图具体为：基于本地知识图谱模块生成的推理中间结果，根据本地知识图谱的语义结构，基于患者匹配信息、患者就诊的时间窗信息和推理获得的临床发现信息构建用于区块链同步的链上子图，并通过加密模块将三元组信息加密后，并传输至链上模块；所述加密模块将患者匹配信息进行加密，通过哈希模块生成匿名化患者身份哈希值，将三元组中语义结构信息和临床发现信息，通过编码表映射的方式进行信息去明示化，之后进行三元组 文件加密；
分布式模块对上链同步的三元组中的患者身份信息进行哈希加密和SimHash加密，对其他结构性元素进行编码映射，对三元组文件进行非对称加密，从而实现链上子图三元组的全加密化。
患者匹配具体为：各中心根据链上模块的区块链同步结果，获得链上子图中的三元组信息，比对本地患者身份哈希值与链上子图中的患者身份哈希值，根据身份哈希值的相似度得到患者匹配结果；
所述链上模块根据分布式模块构建的链上子图中加密后的三元组文件，通过数据沟通模块构建为由区块数据组、三元组数据组和通讯数据组三部分组成的区块链同步数据结构，用于记录文件的区块状态和多中心推理流程状态，并通过共识机制进行区块链同步，在多中心进行点对点的数据传输。
链上模块使用区块链技术实现链上子图在多中心间的信息传递，将多中心知识图谱推理所需的中间过程节点在不同分中心间进行准确同步；模块同时维护不可篡改的链上数据流通记录，支持对各个节点间数据请求和发送过程的追溯。
区块链节点通讯系统采用libp2p技术实现多中心节点间的点对点通讯协议，节点间根据节点网络地址信息通过tcp协议进行链接，同时确定基础网络传输协议并获取数据公钥；libp2p将节点信息保存在本地数据库中，在执行多中心区块链数据同步时通过节点路由寻找多条节点路径以保证传输效率和稳定性。
区块链数据库系统使用MySQL数据库存储区块链同步数据、区块链系统数据和同步数据映射表，这些数据用于多中心知识图谱的链上同步流程运行。数据库同时保存区块链日志记录，对所有上链同步信息和操作行为进行记录，该日志进行链上同步校验，确保所有操作记录全程可追溯查询，且记录不可篡改。
链上模块的接口使用ASP.NET框架编写，用于同本地知识图谱的分布式模块进行系统对接，向其他节点发送通讯请求和测试请求，并提供操作记录查询服务。
上述实施例用来解释说明本发明，而不是对本发明进行限制，在本发明的精神和权利要求的保护范围内，对本发明作出的任何修改和改变，都落入本发明的保护范围。

Claims (9)

  1. 一种多中心知识图谱联合决策支持方法，其特征在于，该方法包括如下步骤：
    (1)通过医学文献以及医学标准术语集构建本地知识图谱；将临床数据构建为“患者-就诊-诊疗”的语义三元组，并同本地知识图谱中的医学知识进行关联；
    (2)基于步骤(1)构建的本地知识图谱和语义三元组，根据临床决策支持的推理需求，通过语义推理方法，推理出患者临床发现信息，生成推理中间结果；根据本地知识图谱的语义结构，基于患者匹配信息、患者就诊的时间窗信息和推理获得的临床发现信息，构建用于区块链同步的链上子图，链上子图隔绝其他患者原始医疗信息；所述患者匹配信息包括患者的唯一身份信息和非唯一身份信息；所述唯一身份信息为患者的身份证ID和医保ID，采用SHA-256哈希算法进行加密，生成唯一身份标识哈希值；所述非唯一身份信息为姓名、性别、生日、住址和工作地址，采用SimHash算法进行加密，生成非唯一身份标识哈希值；
    (3)将患者匹配信息进行加密，生成匿名化患者身份哈希值；将语义三元组中的语义结构信息和临床发现信息通过编码表映射的方式进行信息去明示化；之后进行三元组文件加密；
    (4)基于步骤(3)中加密后的三元组文件，构建由区块数据组、三元组数据组和通讯数据组三部分组成的区块链同步数据结构，在多中心进行点对点的数据传输；
    (5)各中心根据区块链同步获得的链上子图中的三元组信息，比对本地患者身份哈希值与链上子图中的患者身份哈希值，根据身份哈希值的相似度得到患者匹配结果；
    (6)各中心根据患者匹配结果，获取链上子图中的匹配患者的推理中间结果，基于本地患者信息和推理中间结果，结合本地知识图谱，通过语义推理生成决策支持结果。
  2. 根据权利要求1所述的一种多中心知识图谱联合决策支持方法，其特征在于，步骤(1)中，医学标准术语集采用OMOP术语库；本地知识图谱基于医生经验和医学文献构建临床决策支持推理规则库，规则中的节点元素符合知识图谱医学知识结构和OMOP术语编码体系。
  3. 根据权利要求1所述的一种多中心知识图谱联合决策支持方法，其特征在于，步骤(1)中，临床数据的语义三元组采用“患者-就诊-诊疗”的三级临床数据语义模型，将每位患者的电子病历数据，按照临床数据语义模型进行本体节点建模和数值信息录入。
  4. 根据权利要求2所述的一种多中心知识图谱联合决策支持方法，其特征在于，步骤(2)中，根据患者就诊记录的所属临床发现类，基于临床决策支持推理规则库获得对应的语义推理规则，对于本地知识图谱，使用推理机基于语义三元组和语义推理规则获得语义推理结果。
  5. 根据权利要求1所述的一种多中心知识图谱联合决策支持方法，其特征在于，步骤(2)中，针对所有患者实例、就诊实例和临床发现实例，基于本地知识图谱生成链上同步所需的 相应虚拟实例，其中虚拟患者实例同时注入患者的唯一身份信息和非唯一身份信息，虚拟就诊实例包含就诊起始日期和就诊终止日期构成的时间窗信息，虚拟临床发现实例包含临床发现确认的时间、临床发现所属的类别和临床发现的阳性阴性结果。
  6. 根据权利要求1所述的一种多中心知识图谱联合决策支持方法，其特征在于，步骤(4)中，所述区块数据组包括区块编号、区块哈希值、默克尔树根和时间记录戳信息，用于多中心区块节点间的匹配校验；所述三元组数据组为经过加密的链上子图三元组文件数据；所述通讯数据组记录数据流程状态、数据库连接状态、P2P节点连接参数和三元组文件标识，用于区块链节点通讯和流程判别。
  7. 一种多中心知识图谱联合决策支持系统，其特征在于，该系统包括本地知识图谱模块、分布式模块和链上模块；
    所述本地知识图谱模块通过医学文献以及医学标准术语集构建本地知识图谱，将临床数据建立为“患者-就诊-诊疗”的语义三元组，并同本地知识图谱中的医学知识进行关联；基于构建的本地知识图谱和语义三元组，根据临床决策支持的推理需求，通过语义推理方法，推理出患者临床发现信息，生成推理中间结果；各中心根据分布式模块的患者匹配结果，获取链上子图中的匹配患者的推理中间结果后，基于本地患者信息和推理中间结果，结合本地知识图谱，通过语义推理生成决策支持结果；
    所述分布式模块用于构建链上子图和进行患者匹配；
    构建链上子图具体为：基于本地知识图谱模块生成的推理中间结果，根据本地知识图谱的语义结构，基于患者匹配信息、患者就诊的时间窗信息和推理获得的临床发现信息构建用于区块链同步的链上子图，并通过加密模块将三元组信息加密后，并传输至链上模块；所述患者匹配信息包括患者的唯一身份信息和非唯一身份信息；所述唯一身份信息为患者的身份证ID和医保ID，采用SHA-256哈希算法进行加密，生成唯一身份标识哈希值；所述非唯一身份信息为姓名、性别、生日、住址和工作地址，采用SimHash算法进行加密，生成非唯一身份标识哈希值；所述加密模块将患者匹配信息进行加密，生成匿名化患者身份哈希值，将三元组中语义结构信息和临床发现信息，通过编码表映射的方式进行信息去明示化，之后进行三元组文件加密；
    患者匹配具体为：各中心根据链上模块的区块链同步结果，获得链上子图中的三元组信息，比对本地患者身份哈希值与链上子图中的患者身份哈希值，根据身份哈希值的相似度得到患者匹配结果；
    所述链上模块根据分布式模块构建的链上子图中加密后的三元组文件，通过数据沟通模块构建为由区块数据组、三元组数据组和通讯数据组三部分组成的区块链同步数据结构，并 通过共识机制进行区块链同步，在多中心进行点对点的数据传输。
  8. 根据权利要求7所述的一种多中心知识图谱联合决策支持系统，其特征在于，所述分布式模块通过有限实体交互从本地知识图谱模块中获取用于分布式推理和推理中间结果汇总的三元组信息，对三元组中的信息进行解析并隔绝患者原始医疗数据信息，之后按照链上子图的语义结构进行三元组重构。
  9. 根据权利要求7所述的一种多中心知识图谱联合决策支持系统，其特征在于，所述本地知识图谱模块生成的决策支持结果传输至分布式模块并进行三元组信息加密后，传递至链上模块进行区块链同步，用于其他中心的决策支持。
PCT/CN2022/124692 2021-10-13 2022-10-11 一种多中心知识图谱联合决策支持方法与系统 WO2023061377A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111190634.4A CN113643821B (zh) 2021-10-13 2021-10-13 一种多中心知识图谱联合决策支持方法与系统
CN202111190634.4 2021-10-13

Publications (1)

Publication Number Publication Date
WO2023061377A1 true WO2023061377A1 (zh) 2023-04-20

Family

ID=78426662

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/124692 WO2023061377A1 (zh) 2021-10-13 2022-10-11 一种多中心知识图谱联合决策支持方法与系统

Country Status (2)

Country Link
CN (1) CN113643821B (zh)
WO (1) WO2023061377A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117438079A (zh) * 2023-12-19 2024-01-23 北京万方医学信息科技有限公司 循证知识抽提及辅助临床决策的方法及介质
CN117476218A (zh) * 2023-12-27 2024-01-30 长春中医药大学 一种基于临床知识图谱的中医妇科护理辅助决策系统

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113643821B (zh) * 2021-10-13 2022-02-11 浙江大学 一种多中心知识图谱联合决策支持方法与系统
CN113921141B (zh) * 2021-12-14 2022-04-08 之江实验室 一种个体慢病演进风险可视化评估方法及系统
CN113990495B (zh) * 2021-12-27 2022-04-29 之江实验室 一种基于图神经网络的疾病诊断预测系统
CN114363354B (zh) * 2021-12-30 2022-09-30 海南大学 基于dikwp模型的区块链共识方法
CN114741521A (zh) * 2022-03-03 2022-07-12 清华大学 临床研究数据的集成方法、装置、电子设备及存储介质
CN114445207B (zh) * 2022-04-11 2022-07-26 广东企数标普科技有限公司 基于数字人民币的税务管理系统
CN116126976B (zh) * 2023-04-06 2023-07-04 之江实验室 一种数据同步的方法、装置、存储介质及电子设备
CN116665913B (zh) * 2023-07-13 2023-10-13 之江实验室 一种跨机构的患者匹配系统和方法
CN116994686B (zh) * 2023-09-26 2023-12-15 北斗云方(北京)健康科技有限公司 数据驱动的临床决策支持系统和方法

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180204111A1 (en) * 2013-02-28 2018-07-19 Z Advanced Computing, Inc. System and Method for Extremely Efficient Image and Pattern Recognition and Artificial Intelligence Platform
CN109471948A (zh) * 2018-11-08 2019-03-15 威海天鑫现代服务技术研究院有限公司 一种老年健康领域知识问答系统构建方法
CN111370127A (zh) * 2020-01-14 2020-07-03 之江实验室 一种基于知识图谱的跨科室慢性肾病早期诊断决策支持系统
CN111767410A (zh) * 2020-06-30 2020-10-13 平安国际智慧城市科技股份有限公司 临床医疗知识图谱的构建方法、装置、设备及存储介质
WO2020236832A1 (en) * 2019-05-23 2020-11-26 Healthpointe Solutions, Inc. System and method for using a blockchain to manage knowledge in a healthcare ecosystem
CN112037920A (zh) * 2020-08-31 2020-12-04 康键信息技术（深圳）有限公司 医疗知识图谱构建方法、装置、设备及存储介质
CN112700865A (zh) * 2021-01-07 2021-04-23 重庆中肾网络科技有限公司 一种基于综合推理的智能分诊方法
WO2021139282A1 (zh) * 2020-06-24 2021-07-15 平安科技（深圳）有限公司 医疗领域知识图谱构建方法、装置、设备及存储介质
WO2021178112A1 (en) * 2020-03-02 2021-09-10 Healthpointe Solutions, Inc. Systems and methods for using a distributed ledger to manage knowledge in a healthcare ecosystem
CN113643821A (zh) * 2021-10-13 2021-11-12 浙江大学 一种多中心知识图谱联合决策支持方法与系统

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180204111A1 (en) * 2013-02-28 2018-07-19 Z Advanced Computing, Inc. System and Method for Extremely Efficient Image and Pattern Recognition and Artificial Intelligence Platform
CN109471948A (zh) * 2018-11-08 2019-03-15 威海天鑫现代服务技术研究院有限公司 一种老年健康领域知识问答系统构建方法
WO2020236832A1 (en) * 2019-05-23 2020-11-26 Healthpointe Solutions, Inc. System and method for using a blockchain to manage knowledge in a healthcare ecosystem
CN111370127A (zh) * 2020-01-14 2020-07-03 之江实验室 一种基于知识图谱的跨科室慢性肾病早期诊断决策支持系统
WO2021178112A1 (en) * 2020-03-02 2021-09-10 Healthpointe Solutions, Inc. Systems and methods for using a distributed ledger to manage knowledge in a healthcare ecosystem
WO2021139282A1 (zh) * 2020-06-24 2021-07-15 平安科技（深圳）有限公司 医疗领域知识图谱构建方法、装置、设备及存储介质
CN111767410A (zh) * 2020-06-30 2020-10-13 平安国际智慧城市科技股份有限公司 临床医疗知识图谱的构建方法、装置、设备及存储介质
CN112037920A (zh) * 2020-08-31 2020-12-04 康键信息技术（深圳）有限公司 医疗知识图谱构建方法、装置、设备及存储介质
CN112700865A (zh) * 2021-01-07 2021-04-23 重庆中肾网络科技有限公司 一种基于综合推理的智能分诊方法
CN113643821A (zh) * 2021-10-13 2021-11-12 浙江大学 一种多中心知识图谱联合决策支持方法与系统

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117438079A (zh) * 2023-12-19 2024-01-23 北京万方医学信息科技有限公司 循证知识抽提及辅助临床决策的方法及介质
CN117438079B (zh) * 2023-12-19 2024-03-12 北京万方医学信息科技有限公司 循证知识抽提及辅助临床决策的方法及介质
CN117476218A (zh) * 2023-12-27 2024-01-30 长春中医药大学 一种基于临床知识图谱的中医妇科护理辅助决策系统
CN117476218B (zh) * 2023-12-27 2024-03-08 长春中医药大学 一种基于临床知识图谱的中医妇科护理辅助决策系统

Also Published As

Publication number Publication date
CN113643821A (zh) 2021-11-12
CN113643821B (zh) 2022-02-11

Similar Documents

Publication Publication Date Title
WO2023061377A1 (zh) 一种多中心知识图谱联合决策支持方法与系统
WO2021143779A1 (zh) 一种基于知识图谱的跨科室慢性肾病早期诊断决策支持系统
US20240119176A1 (en) Systems and methods for computing with private healthcare data
US20230044294A1 (en) Systems and methods for computing with private healthcare data
KR101873926B1 (ko) 빅데이터 기반 의료자문 서비스 제공 방법
Dasaklis et al. Blockchain meets smart health: Towards next generation healthcare services
Wimmer et al. A multi-agent system to support evidence based medicine and clinical decision making via data sharing and data privacy
CN111552734A (zh) 用户画像的生成方法及装置、计算机设备、存储介质
CN117238458B (zh) 基于云计算的重症护理跨机构协同平台系统
US20210057064A1 (en) Systems and methods for federated searching and retrieval of medical records across disparate databases
Demirbaga et al. MapChain: A blockchain-based verifiable healthcare service management in IoT-based big data ecosystem
Sarkar et al. A conceptual distributed framework for improved and secured healthcare system
Leng et al. Application of hyperledger in the hospital information systems: A survey
Thantilage et al. Healthcare data security and privacy in Data Warehouse architectures
Zhang et al. [Retracted] Design and Application of Electronic Rehabilitation Medical Record (ERMR) Sharing Scheme Based on Blockchain Technology
Canterberry et al. The patient-centered outcomes research network antibiotics and childhood growth study: implementing patient data linkage
Poonguzhali et al. A framework for electronic health record using blockchain technology
US20210248621A1 (en) Methods and systems for tracking the transfer of a cannabis article
US20220101961A1 (en) Systems and methods for matching medical records for patients across disparate medical providers to facilitate continuity of care
Shashi Leveraging Blockchain-based electronic health record systems in healthcare 4.0
Zhan et al. Medical record encryption storage system based on Internet of Things
Rydningen et al. Advantages and opportunities of the IOTA tangle for health data management: a systematic mapping study
Li et al. A Blockchain-Based Personal Health Knowledge Graph for Secure Integrated Health Data Management
Mammadova et al. Formation of Unified Digital Health Information Space in Healthcare 4.0 Environment and interoperability issues
Thirumahal et al. Semantic integration of heterogeneous healthcare data based on hybrid root linked health record ontology

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: 22880314

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE