Intelligent medical record generation method and system

Abstract

The invention relates to an intelligent medical record generation method and system. The method comprises the following steps: extracting each piece of data in accessed multi-type data and constructing a time-based event; constructing a directed graph arrangement event; traversing the directed graph, and verifying whether an event conforms to a rule base or not by using the rule base containing causal logical reasoning in the traversing process; and traversing the directed graph based on the medical record generation condition, and sorting the traversing result based on the generation type to obtain a medical record document. According to the method, multi-source data are fused in the generation process, and cross-modal and cross-cycle long text integrated generation is achieved; performing real-time calling and verification on related knowledge based on the ternary knowledge base to realize dynamic enhancement; a time-event chain reasoning mechanism and fine-grained reference are realized based on a directed graph, so that the traceability of the time sequence, causal logic and information sources of disease course records is ensured, posterior verification and feedback after quality inspection are supported, and a high-quality medical record generation basis is provided.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent medical treatment, and particularly relates to a method and system for generating intelligent medical records. BACKGROUND

[0002] A medical record is a core medical document recording the entire diagnosis and treatment process, disease evolution and nursing measures of a patient. The diagnosis and treatment process, laboratory test results, physical sign entry and nursing records in daily medical records, stage summaries and discharge records are particularly important. These texts not only directly reflect the diagnosis and treatment, TCM syndrome differentiation and treatment, prescription modification and nursing differentiation and treatment process, but also are important evidence for medical insurance settlement, medical quality control, scientific research statistics and legal evidence. In the traditional mode, the above-mentioned content is mainly written by clinical physicians and nursing staff by hand, which has the problems of large writing workload, non-uniform terminology and format, information fragmentation, and difficulty in knowledge reuse. In addition, the data of admission records, medical orders, laboratory tests, nursing records and the like are scattered in multiple subsystems such as HIS, LIS, PACS and nursing systems, and physicians need to manually review, extract and integrate, which is easy to miss key information. In recent years, researchers have attempted to use natural language processing technology to alleviate the above-mentioned problems. Patent CN202311234567.8 proposes an electronic medical record named entity recognition method based on BERT-BiLSTM-CRF, which can extract entities such as syndromes, prescriptions and treatment methods from admission records, but only completes information extraction and does not involve automatic generation of medical record texts. Patent CN202410123456.7 uses a T5 model to generate a template for a western medicine discharge summary, the input of which is structured diagnosis and surgery coding, and the output of which is a fixed format western medicine course summary. However, this method does not consider the medical chain of differentiation-treatment-prescription-nursing, and lacks support for nursing records. In addition, existing methods are generally based on small and medium-sized parameter models, which are limited by the size of the field corpus and the efficiency of knowledge updating, making it difficult to accurately depict the complex mapping relationships existing in clinical practice, and lacking the ability to integrate multi-source heterogeneous data, resulting in insufficient accuracy, completeness and subject characteristics of the generated text. With the breakthrough of large language models in parameter quantity, context length and instruction following ability, they have shown near-human level performance in general text generation tasks; however, directly applying general large models to inpatient medical record generation still faces problems such as insufficient completeness of medical record text generation, lack of domain knowledge, insufficient modeling of long-range dependencies, high requirement for structuring, limited long-range dependency and causal reasoning ability, insufficient explainability and traceability, and lack of fine-grained reference mechanism in general models; although large language models have strong generation ability, their internal reasoning process is a black box that is difficult to trace, which is unacceptable in high-risk fields such as medicine; moreover, in fact, the vertical application complexity of large language models in the field of medical record generation is high, and directly throwing all clinical logic and data to the large model will lead to extremely complex prompt engineering, and the model needs a large amount of high-quality labeled data to learn the complex logic. Therefore, a large model medical record generation method is needed, which is oriented to specific academic scenarios, deeply integrates multi-source clinical data and external TCM knowledge, and has controllable generation and explainable verification capabilities, to improve medical record writing efficiency, standardization and clinical practicality; based on the above problems, the present application no longer requires the large model to understand the complete medical record from the chaotic data, but to pre-configure the graph structure representation and its rule base reasoning, so that the task of the large model changes from creating out of nothing to translating and polishing according to the graph, which conforms to the stable support ability of the general function of the current large model, reduces the uncertainty and illusion risk of the generation task, and makes the output result strictly controlled by the preset clinical logic. SUMMARY

[0003] In order to solve the above problems in the prior art, the present application provides an intelligent medical record generation method and system, the method comprising: Step S1: extracting and constructing time-based events for each data in the accessed multi-type data; the event attributes include time, category, content and source; Step S2: constructing a directed graph to arrange events, taking events as nodes in the graph, and setting a directed edge between two events that occur in sequence when the action relationship between them is met; The determination of the action relationship between the two events that occur in sequence is as follows: the action relationship includes the time effective relationship of the event and the category compatibility relationship of the event; a time window is defined for each event, and for the subsequent event, when the prior event occurs within the time window of the subsequent event, it is determined that the two events meet the time effective relationship; for the prior event, when the subsequent event meets the compatibility matrix or has a causal relationship with it, it is determined that the two events meet the category compatibility relationship; Step S3: traversing the directed graph, verifying whether the events conform to the rule base using the rule base containing the causal logic reasoning in the process of the traversing, splitting, merging and resolving contradiction for the events not conforming to the rule base; repeating the step until the directed graph is stable; Step S4: traversing the directed graph based on the medical record generation condition, arranging the traversing results based on the generation type to obtain the medical record document; the medical record generation condition contains the traversing mode, node type, source and / or limited node time range of the directed graph corresponding to the medical record document or the medical record document component; in the process of the traversing, only the path constituted by the nodes conforming to the medical record generation condition is determined as the traversing result; the generation type indicates the organization mode of the traversing result.

[0004] Further, the event attribute is organized in the form of a quadruple.

[0005] Further, data access is performed through the HIS interface or the CDR database, and multiple types of data generated during the patient's hospitalization in the EMR, LIS and PACS medical information systems are automatically obtained and time-synchronized.

[0006] Further, the step S3 specifically includes the following steps: Step S31: breadth-first traversing the directed graph and judging whether the in-out edge relationship of each node conforms to the rule base, making the directed graph conform to the rule base through splitting and / or merging operation; Step S32: depth-first traversing the directed graph and judging whether the sub-path where each node is located conforms to the rule base, making the directed graph conform to the rule base through resolving contradiction, splitting and / or merging operation; Step S33: judging whether the directed graph is stable, if yes, entering the next step, otherwise, returning to step S31; the stability of the directed graph means that all nodes conform to the rule base and no splitting, merging and resolving contradiction operation occurs in the previous traversing process.

[0007] Further, the step S31 specifically includes the following steps: Step S311: determining the current node; obtaining the current node and the parent node and child node meeting the time effective relationship with the current node; the current node is the first untraversed node in the process of the breadth-first traversing; Step S312: obtaining the event content of the current node and the child node and the parent node, judging whether the in-edge relationship between all parent nodes and the current node and the out-edge relationship between the current node and all child nodes conform to all related rules in the rule base of the causal logic reasoning, if yes, setting the current node as traversed and returning to step S311; otherwise, entering the next step; Step S313: If the current node corresponding event does not conform to the rule base, the event is split and / or merged based on the event content, and the directed graph is updated synchronously for the split and / or merge operation; after splitting, the nodes obtained after splitting inherit the parent-child relationship, and after merging, the merged nodes merge the parent-child relationship; the nodes obtained after splitting and the merged nodes are all set as untraversed; Step S314: Whether the traversal is ended; if not, return to step S311; if yes, proceed to the next step.

[0008] Further, the step S32 specifically comprises: obtaining a first unprocessed current node in the depth-first traversal process of the directed graph, obtaining a sub-path composed of all parent nodes conforming to the effective relationship of the current node event, judging whether the sub-path conforms to all related rules in the rule base of the causal logic reasoning, if not, performing contradiction elimination in the sub-path, and setting all nodes in the sub-path as unprocessed; repeating the step until all nodes conform to the rule base.

[0009] An intelligent medical record generation platform, the platform is used for realizing the intelligent medical record generation method.

[0010] An intelligent medical record generation server, comprising a processor, the processor and a memory are coupled, the memory stores program instructions, when the program instructions stored in the memory are executed by the processor, the intelligent medical record generation method is realized.

[0011] An intelligent medical record generation system, the system is used for realizing the intelligent medical record generation method.

[0012] A computer readable storage medium, comprising a program, when it runs on a computer, makes the computer execute the intelligent medical record generation method.

[0013] The beneficial effects of the present application include: (1) Based on multiple types of data, a directed graph composed of events and relationships is constructed; the directed graph is used as a pre-abstract layer of a large model, logical fragmentation processing is realized through a rule base, logical verification of the directed graph is realized, text generation is decoupled into independent modules, a collaborative paradigm of rule framing, intelligent filling and quality inspection is formed; the technical threshold and development and maintenance cost of the whole system are reduced, and the usability of medical record generation is improved; (2) Using event-based multi-type data atomic representation, multi-type data is represented in accordance with the appropriate size of the rule base through event splitting, merging and contradiction removal, and finally forming multi-type data representation based on event and reasoning logic. It supports personalized multi-type medical record generation based on directed graphs, links the medical record generation conditions, generation type and different components of the medical record document, and can be seamlessly connected with large language models and data flow feedback. It perfectly combines the accuracy of mature rule base with the flexibility of large model, which not only ensures professionalism, but also improves generation efficiency and quality. Attached Figure Description

[0014] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to unduly limit the invention. In the drawings: Figure 1 This is a schematic diagram of the intelligent medical record generation method provided by the present invention.

[0015] Figure 2 This is a schematic diagram of the intelligent medical record generation data stream feedback method provided by the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0017] This invention proposes an intelligent medical record generation method and system, as shown in the appendix. Figure 1 As shown, the method includes the following steps: Step S1: Data access is performed, extracting each data record from multiple data types and constructing a time-based event; the event attributes include time, category, content, and source; a ternary knowledge base is used to supplement and expand the knowledge of the formed event; the event attributes are organized in the form of quadruples; for example, event 1 is represented as a quadruple (2023-10-27 11:30, treatment, Yinqiao powder, one dose per day, HIS medical order); the supplementary content is shown in Table 1 below; of course, this supplementary content can be merged with the event quadruple, or exist as an additional part; Table 1: Examples of Supplementary Content Preferred method: Use quadruples to record and store events in a structured manner; Preferred approach: Data access is achieved through HIS interface or CDR database, automatically acquiring various types of data generated during patient hospitalization from medical information systems such as EMR, LIS, and PACS, and synchronizing them over time; key event information is extracted from the various data types, uniformly organized and stored to form a structured representation of the various data types in event form; further: key information in the events is extracted using a structured large model; after data access, data governance and standardization are performed, automatically indexing to the corresponding structured or unstructured templates according to the data source, and then performing standardized mapping of the data based on this, extracting key information from the time frame based on the standardized mapping results; Preferably, the multi-type data includes inpatient data such as diagnosis and treatment data, nursing records, syndrome differentiation and nursing assessment forms, nursing record forms, and nursing effect evaluation forms, and the data type can correspond to the event source; Preferably, the event types include symptoms, tests, nursing care, treatment, and diagnosis; Furthermore, this step also includes using a ternary knowledge base containing knowledge graphs, medical guidelines, and protocol knowledge to perform consistency processing such as terminology unification and format standardization coding on the extracted key information before filling it into the event representation; Step S2: Construct a directed graph to organize events, treating events as nodes in the graph. When two events that occur sequentially satisfy an interaction relationship, set a directed edge between them; the directed edge starts from the first event and points to the second event. The determination of a causal relationship between two events occurring sequentially specifically involves: the causal relationship between the events and the category compatibility relationship between the events; defining a time window for each event; for each subsequent event, determining that the two events satisfy a causal relationship when the preceding event is within the time window of the subsequent event; and for each preceding event, determining that the two events satisfy a category compatibility relationship when the subsequent event has a causal relationship with it or satisfies a compatibility matrix. Preferably, the causal relationship includes one or more of the following: triggering relationship, derivation relationship, execution relationship, evolution relationship, response relationship, feedback relationship, support relationship, guarantee relationship, coordination relationship, cooperation relationship, evaluation relationship, and monitoring relationship; Alternative: A compatibility matrix is ​​pre-set between event types to indicate the relationship between events. The compatibility matrix is ​​queried to determine whether two events satisfy the category compatibility relationship. For example, if the compatibility value in the compatibility matrix is ​​1, it is determined that the two events satisfy the compatibility matrix. Preferably, the time window for each event is the same or different. The length of the time window is related to the event type and / or content, and can be regarded as the maximum reasonable time interval preset for that type of event. For example, the time window from a symptom event to a test event may be short, such as within a few hours; the time window from a test event to a diagnosis event is in the middle, such as within 24 hours. Of course, the setting of the time window length is related to the specific classification granularity of the event. When the event classification granularity is coarse, a more fuzzy time window length can be set, and vice versa. Step S3: Traverse the directed graph. During the traversal, use a rule base containing causal logical reasoning to verify whether events conform to the rule base. For events that do not conform to the rule base, split, merge, and resolve contradictions. Repeat this step until the directed graph is stable. Specifically: Perform breadth-first traversal and depth-first traversal on the directed graph. During the traversal, use a rule base containing causal logical reasoning to verify whether events conform to the rule base. For events that do not conform to the rule base, split, merge, and resolve contradictions, and update the directed graph synchronously. Repeat this step on the updated directed graph until the directed graph is stable. Preferably, the stability of the directed graph means that all events conform to the rule base and no splitting, merging, or contradiction removal operations occur during the traversal process; Step S3 specifically includes the following steps: Step S31: Perform a breadth-first traversal of the directed graph and determine whether the inbound and outbound edge relationships of each node conform to the rule base. Make the directed graph conform to the rule base through splitting and / or merging operations; specifically including the following steps: Step S311: Determine the current node; obtain the current node and its parent and child nodes that satisfy the time validity relationship; the current node is the first untraversed node in the breadth-first traversal process; Preferred approach: Initially, set all nodes to an untraversed state; Preferred method: Set the starting node of the directed graph as the initial node of the current node; when there are multiple starting nodes, select any starting node as the initial node; the starting node is a node without any incoming edges. Step S312: Obtain the event content of the current node, its child nodes, and its parent nodes. Determine whether the incoming edge relationships between all parent nodes and the current node, as well as the outgoing edge relationships between the current node and all its child nodes, conform to all relevant rules in the rule base of causal logic reasoning. If so, set the current node as traversed and return to step S311; otherwise, proceed to the next step. The determination of whether the incoming edge relationship between all parent nodes and the current node conforms to all relevant rules in the rule base of causal logical reasoning is as follows: based on the event type and source, all relevant rules are obtained from the rule base; for each relevant rule, it is determined whether the event content set formed by all parent nodes and the event content of the current node satisfy the rule; if there is a rule that does not satisfy the rule, it is determined that it does not conform; otherwise, it is determined that it conforms. The determination of whether the outgoing edge relationship between the current node and all its child nodes conforms to all relevant rules in the rule base of causal logical reasoning is as follows: based on the event type and source, all relevant rules are obtained from the rule base; for each relevant rule, it is determined whether the event content of the current node and the event content set formed by all its child nodes satisfy the rule; if there is a rule that does not satisfy the rule, it is determined that it does not conform; otherwise, it is determined that it conforms. Step S313: If the event corresponding to the current node does not conform to the rule base, then split and / or merge the event based on the event content, and update the directed graph synchronously for the split and / or merge operation; after splitting, the nodes obtained after splitting inherit the parent-child relationship, and after merging, the nodes obtained after merging merge the parent-child relationship; set both the nodes obtained after splitting and the merged nodes as untraversed; Preferred: No repeated splitting and / or merging operations are performed; Preferred approach: Splitting and / or merging operations are performed based on the knowledge base; merging operations occur between sibling nodes, nodes without order relationships, and ordered nodes with a span less than or equal to the span threshold; when multiple selectable operations exist, random selection is performed. Preferably, the span threshold is a preset value, for example, 1 to 5. When the span threshold is 1, it corresponds to a parent-child relationship. Step S314: Has the traversal ended? If not, return to step S311; otherwise, proceed to the next step. Step S32: Perform a depth-first traversal of the directed graph and determine whether the sub-path of each node conforms to the rule base; make the directed graph conform to the rule base through contradiction removal, splitting, and / or merging operations; specifically: obtain the first unprocessed current node in the depth-first traversal process of the directed graph, obtain the sub-path formed by all parent nodes (or all parent-child nodes, or ordered nodes with a span less than or equal to the span threshold) that conform to the valid relationship of the current node's event, determine whether the sub-path conforms to all relevant rules in the rule base of causal logical reasoning, if not, perform contradiction removal (or splitting and / or merging operations) on the sub-path, and set all nodes (or the current node) in the sub-path to unprocessed; repeat this step until all nodes conform to the rule base; Preferred method: The contradiction removal involves removing nodes from the sub-path; after contradiction removal, a fully connected edge is set between the parent node and the child node; atomic nodes, i.e., indivisible nodes, are preferred for contradiction removal; of course, other selection methods can also be chosen; furthermore: feedback is provided before contradiction removal to obtain confirmation; the contradiction removal method can be manual and / or automatic modification, deletion, adjustment, etc. based on a ternary library (third-party knowledge base); Alternative: If the condition is not met, perform splitting, merging, and / or conflict resolution operations on the nodes in the sub-path and return to step S31; of course, the operation here also satisfies the non-repetition principle; Preferred method: During depth-first traversal, the starting node of the directed graph is set as the initial node of the current node; when there are multiple starting nodes, any starting node is selected as the initial node; the starting node is a node without any incoming edges. Step S33: Determine whether the directed graph is stable. If it is, proceed to the next step; otherwise, return to step S31. Step S4: Generate medical records based on the directed graph; specifically: traverse the directed graph based on the medical record generation conditions, and organize the traversal results based on the generation type to obtain the medical record document; the medical record generation conditions include the traversal method of the directed graph, node type, source and / or limiting the time range of nodes; during the traversal process, only the paths formed by nodes that meet the medical record generation conditions are determined; the generation type includes the organization method of the traversal results; Preferred: Different medical record document types and different medical record generation conditions and their generation type associations; Replaceable: Medical record documents contain multiple components based on their framework structure, each component being associated with medical record generation conditions and their generation type; The traversal of the directed graph based on the medical record generation conditions is as follows: starting from the starting node, traverse the graph structure, determine one or more paths that meet the medical record generation conditions, and organize the node attributes involved in the path according to the generation type. Preferably, the generation type includes vertical generation and horizontal generation methods; when using the vertical generation method, the nodes and their attributes in the path are arranged sequentially according to the time sequence; this generation type is suitable for causal reasoning and medical record presentation based on the specific needs of organizational structure and individual presentation; when using the horizontal generation method, the nodes and their attributes in the directed graph are organized according to the time range to form a medical record. Preferably, the generated medical records are used as pre-generated medical records, and post-feedback processing based on data flow is performed. The system supports generating various types of medical record documents, including admission records, initial progress notes, daily progress notes, and discharge summaries. Different types of medical records have different structural focuses and content depth requirements. Each type of medical record document supports different length options, such as concise, standard, and detailed versions. The concise version emphasizes core diagnostic and treatment information, the standard version maintains the integrity of the clinical record, and the detailed version includes all detailed clinical observations and reasoning processes. Each type of medical record document also supports highlighting different clinical dimensions, such as symptom evolution, treatment decision logic, efficacy evaluation analysis, and the effectiveness of nursing interventions. Each medical record document requires a time range limit and supports filtering events to be included in the medical record based on absolute time range or relative time stage. For different medical record document types, their corresponding medical record generation conditions are determined according to the framework structure of the standard medical record corresponding to the document. The event information required for each component of the corresponding medical record is extracted, so that each medical record segment can be traversed as needed to ensure that it obtains the most relevant clinical event support. For example, based on the medical record generation conditions, it can include a diagnostic reasoning chain from symptoms to diagnosis, a treatment response chain from treatment to efficacy, and a nursing intervention chain involving the correlation between nursing intervention and symptom improvement. Finally, using a standardized medical record template, the output of each generator is assembled according to the standardized medical record format, and the node attributes involved in the path are organized according to the generation type corresponding to each component. As attached Figure 2 As shown, from a data flow perspective, intelligent medical records require post-data flow feedback after generation. After generating the medical record, personalized corpus is filled in according to the requirements of inpatient medical record writing standards, such as summaries of recent examination and test results, summaries of recent nursing evaluation effects, and summaries of the process of syndrome differentiation and treatment and syndrome differentiation and nursing analysis. Furthermore, predictive filling is performed by combining prompt words and reinforcement learning technologies. Preferably, the generated medical record text is input into a large language model, and based on the large model, personalized corpus, prompt words, and few-shot support, the generated electronic medical record is optimized. Preferred method: To ensure the accuracy and compliance of the generated content, the generated electronic medical records are sent to the content quality control system for quality control of the generated electronic medical record content, and automatic verification is performed based on historical data; the quality control rules included in the content quality control system involve, but are not limited to, multiple dimensions such as completeness, timeliness, consistency, rationality and standardization; Preferred method: After quality control and verification based on similar medical records, the electronic medical record is sent to the user, and after waiting for the user's final confirmation and modification, the confirmed and modified electronic medical record is entered into the electronic medical record system; Example: After completing data access, event construction, and knowledge expansion (step S1), the system obtained the following 8 core events as inputs for subsequent processing; Based on the temporal sequence and type compatibility of events, a preliminary causal hypothesis graph is constructed. The initial directed graph is as follows: Event 1 (symptom) points to Event 2 (test) and Event 3 (symptom); Event 1 and Event 2 together point to Event 4 (diagnosis); Event 3 (symptom) points to Event 4 (diagnosis); Event 4 (diagnosis), as the core decision point, points to Event 5 (treatment) and Event 6 (nursing); Event 5 (treatment) points to Event 7 (symptom improvement) and Event 8 (improvement of test results); Event 6 (nursing) also points to Event 7 (symptom improvement). Then, based on the rule base, iterative verification, correction, and graph optimization were performed. Through breadth-first traversal and depth-first traversal, combined with the clinical causal logic rule base, the above directed graph was iteratively verified and corrected. In the first round of traversal, breadth-first traversal, it was found that the content of event 3, "cough, thick yellow phlegm", was a compound symptom, which violated the data normalization principle of "single event describing independent clinical facts". Therefore, a splitting operation was performed to split event 3 into two independent nodes: event 3a [10:30] symptom: "cough" and event 3b [10:30] symptom: "thick yellow phlegm". The edges in the original directed graph connected to event 3 were updated to event 3a and event 3b. In the second round of traversal, depth-first traversal was performed to deeply verify the key causal chain. In tracing "event 1 → When tracing the path "Event 4", it was found that the symptom "chills" is more typical of wind-cold common cold, which is slightly contradictory to the diagnosis of "wind-heat common cold". A contradiction-resolving operation was performed, correcting the content of Event 1 to the typical manifestation of wind-heat common cold: "fever of 38.5℃, aversion to wind". Simultaneously, a depth-first traversal verified the key paths "Event 4 (diagnosis) → Event 5 (treatment) → Event 7 / Event 8 (efficacy)," confirming the matching of syndrome differentiation and treatment between "wind-heat common cold" and "Yinqiao San", and that the treatment effect appeared within a reasonable time window, with complete and self-consistent logic. After completing the above corrections, the updated directed graph was re-traversed and verified. At this point, all event content conformed to terminology standards, and all key causal chains passed the verification of the clinical logic rule base, with no new anomalies found. Thus, the directed graph reached a stable state. Through the processing in steps S2 and S3, the chaotic original events were constructed and optimized into a logically rigorous and clearly causally related dynamic directed graph, laying the foundation for generating high-quality structured medical records. The directed graph contains the following edges: 1 → 2 (Symptom → Test); 1 → 3a (Symptom → Symptom); 1 → 3b (Symptom → Symptom); 1 → 4 (Symptom → Diagnosis); 2 → 4 (Test → Diagnosis); 3a → 4 (Symptom → Diagnosis); 3b → 4 (Symptom → Diagnosis); 4 → 5 (Diagnosis → Treatment); 4 → 6 (Diagnosis → Nursing); 5 → 7 (Treatment → Symptom); 5 → 8 (Treatment → Test); 6 → 7 (Nursing → Symptom); Preferred method: Use electronic medical records entered into the electronic medical record system for online learning and updating, and achieve self-learning and updating based on customer feedback through a dynamic learning mechanism; Based on the same inventive concept, the present invention also provides an intelligent medical record generation system, which is used to implement the above-mentioned intelligent medical record generation method; Based on the same inventive concept, the present invention also provides an intelligent medical record generation server, the server being used to implement the above-described intelligent medical record generation method; Based on the same inventive concept, the present invention also provides an intelligent medical record generation device, which is used to implement the above-mentioned intelligent medical record generation method; Based on the same inventive concept, the present invention also provides an intelligent medical record generation platform, which is used to implement the above-mentioned intelligent medical record generation method; A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of programming language, including assembly or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but does not necessarily, correspond to a file in a file system. A program can be stored as part of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to said program, or in multiple co-located files (e.g., a file storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on a single computer or on multiple computers located at a single site or distributed across multiple sites and interconnected by a communications network.

[0018] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0019] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0020] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a processFigure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0021] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0022] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for generating intelligent medical records, characterized in that, The method includes: Step S1: Extract and construct a time-based event for each data point from the multiple types of data received; the event attributes include time, category, content, and source; Step S2: Construct a directed graph to organize events, treating events as nodes in the graph. When two events that occur sequentially satisfy an interaction relationship, set a directed edge between them. The determination of the interaction relationship between two events occurring sequentially specifically involves: the interaction relationship including the temporal validity relationship and the category compatibility relationship of the events; defining a time window for each event; for a later event, if the earlier event is within the time window of the later event, the two events are determined to satisfy the temporal validity relationship; for an earlier event, if the later event has a causal relationship with it or satisfies the compatibility matrix, the two events are determined to satisfy the category compatibility relationship. Step S3: Traverse the directed graph. During the traversal, use a rule base containing causal logical reasoning to verify whether the events conform to the rule base. For events that do not conform to the rule base, split, merge, and resolve contradictions. Repeat this step until the directed graph is stable; Step S4: Traverse the directed graph based on the medical record generation conditions, and organize the traversal results based on the generation type to obtain the medical record document; the medical record generation conditions include the traversal method, node type, source and / or time range of the directed graph corresponding to the medical record document or its components. During the traversal, only the paths formed by nodes that meet the medical record generation conditions are determined as the traversal results; the generation type indicates how the traversal results are organized.

2. The intelligent medical record generation method according to claim 1, characterized in that, Event properties are organized using quadruples.

3. The intelligent medical record generation method according to claim 2, characterized in that, Data access is achieved through the HIS interface or CDR database, automatically acquiring various types of data generated during patient hospitalization in EMR, LIS, and PACS medical information systems, and synchronizing them over time.

4. The intelligent medical record generation method according to claim 3, characterized in that, Step S3 specifically includes the following steps: Step S31: Perform a breadth-first traversal on the directed graph and determine whether the inbound and outbound edge relationships of each node conform to the rule library. Make the directed graph conform to the rule library through splitting and / or merging operations. Step S32: Perform a depth-first traversal on the directed graph and determine whether the sub-path of each node conforms to the rule base. Make the directed graph conform to the rule base by resolving contradictions, splitting and / or merging operations. Step S33: Determine whether the directed graph is stable. If it is, proceed to the next step; otherwise, return to step S31. The stability of the directed graph means that all nodes conform to the rule base and no splitting, merging, or contradiction removal operations have occurred during the previous traversal.

5. The intelligent medical record generation method according to claim 4, characterized in that, Step S31 specifically includes the following steps: Step S311: Determine the current node; obtain the current node and its parent and child nodes that satisfy the time validity relationship; the current node is the first untraversed node in the breadth-first traversal process; Step S312: Obtain the event content of the current node, its child nodes, and its parent nodes. Determine whether the incoming edge relationships between all parent nodes and the current node, as well as the outgoing edge relationships between the current node and all its child nodes, conform to all relevant rules in the rule base of causal logic reasoning. If so, set the current node as traversed and return to step S311. Otherwise, proceed to the next step; Step S313: If the event corresponding to the current node does not conform to the rule base, then the event is split and / or merged based on the event content, and the directed graph is updated synchronously for the split and / or merge operation; After splitting, the resulting nodes inherit the parent-child relationship; after merging, the merged nodes retain the parent-child relationship; both the split-result nodes and the merged nodes are set to untraversed. Step S314: Has the traversal ended? If not, return to step S311; if yes, proceed to the next step.

6. The intelligent medical record generation method according to claim 5, characterized in that, Step S32 specifically involves: obtaining the first unprocessed current node during the depth-first traversal of the directed graph; obtaining the sub-path formed by all parent nodes that conform to the valid relationship of the current node's events; determining whether the sub-path conforms to all relevant rules in the rule base of causal logical reasoning; if not, resolving contradictions in the sub-path and setting all nodes in the sub-path as unprocessed; repeating this step until all nodes conform to the rule base.

7. A platform for generating intelligent medical records, characterized in that, The platform is used to implement the intelligent medical record generation method according to any one of claims 1-6.

8. A server for generating intelligent medical records, characterized in that, The system includes a processor coupled to a memory, the memory storing program instructions, which, when executed by the processor, implement the intelligent medical record generation method according to any one of claims 1-6.

9. An intelligent medical record generation system, characterized in that, The system is used to implement the intelligent medical record generation method according to any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, Includes a program that, when run on a computer, causes the computer to perform the intelligent medical record generation method according to any one of claims 1-6.

Images