Auxiliary decision support information generation system and auxiliary decision support information generation method

Abstract

The embodiment of the invention discloses an auxiliary decision support information generation system and an auxiliary decision support information generation method. A specific embodiment of the system comprises a server, each heterogeneous data source and at least one terminal, the server is configured to execute the following processing: acquiring a heart and kidney metabolic syndrome patient information table which comprises each patient identifier; creating and updating a heart and kidney metabolic syndrome knowledge graph; for each patient identifier, generating a standardized diagnosis and treatment event record information set; generating an initial diagnosis and treatment event space-time diagram; determining each generated initial diagnosis and treatment event time-space diagram as an initial diagnosis and treatment event time-space diagram set, and storing the initial diagnosis and treatment event time-space diagram set; generating a diagnosis and treatment event space-time diagram corresponding to the treatment data; generating auxiliary decision support information; and sending the auxiliary decision support information to the terminal. According to the embodiment, scheduling resources of the medicine taking robot are reduced.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular, to an auxiliary decision support information generation system and an auxiliary decision support information generation method. BACKGROUND

[0002] With the continuous evolution of medical informatization and multi-source data fusion technology, the extensive accumulation of massive heterogeneous medical data (such as electronic medical records, inspection and examination reports, wearable device monitoring data, etc.) provides a data basis for building a precise and personalized disease auxiliary decision system. However, Cardiovascular-Kidney-Metabolic syndrome (CKM) is a complex chronic disease involving the interaction of multiple organs of cardiovascular, kidney and metabolic system. Its diagnosis and treatment process needs to integrate multi-dimensional information across time and space, including patient medical history, diagnosis and treatment event sequence, physiological index dynamic change, etc. The auxiliary decision support information generation system is a system for generating auxiliary decision support information corresponding to the patient's visit data of Cardiovascular-Kidney-Metabolic syndrome. At present, the existing auxiliary decision support information generation system is usually based on a single data source or a static knowledge base to generate auxiliary decision support information corresponding to the patient's visit data of Cardiovascular-Kidney-Metabolic syndrome.

[0003] However, when generating auxiliary decision support information using the above system, the following technical problems often exist: Patients with Cardiovascular-Kidney-Metabolic syndrome have their own unique disease progression, treatment response, complications, etc. A system based on a single data source may not be able to reflect the long-term changes in the patient's condition (e.g., the patient's visit information is scattered in different medical institutions, different departments, and different information systems, and the system based on a single data source lacks information such as the patient's diagnosis and treatment history and disease trend). At the same time, a system based on a single data source or a static knowledge base often relies on pre-set rules and knowledge, which are usually limited and not updated with the patient's dynamic changes. For a complex disease like Cardiovascular-Kidney-Metabolic syndrome, knowledge and clinical guidelines are constantly updated over time and with research progress, leading to a mismatch between clinical decision recommendations and the patient's actual condition, resulting in low accuracy of the generated auxiliary decision support information. When a dispensing robot is dispatched based on auxiliary decision support information with low accuracy, the dispensed drugs may not meet the current patient's needs, wasting the dispensing robot's scheduling resources.

[0004] The above information disclosed in this BACKGROUND section is only for the purpose of enhancing the understanding of the background of the present inventive concepts, and therefore, it can contain information that does not form the prior art that is already known in this field to those skilled workers in this field. SUMMARY

[0005] This summary of the disclosure is presented in a simplified form to introduce some concepts that will be described in greater detail below in the DETAILED DESCRIPTION. This summary of the disclosure is not intended to identify key or essential features of the claimed technology nor is it intended to be used to limit the scope of the claimed technology.

[0006] Some embodiments of the present disclosure propose an auxiliary decision support information generation system and an auxiliary decision support information generation method to solve one or more of the technical intentions mentioned in the background section.

[0007] In a first aspect, some embodiments of the present disclosure provide an auxiliary decision support information generation system, comprising: a server, a plurality of heterogeneous data sources, and at least one terminal, wherein the server, the at least one heterogeneous data source, and the at least one terminal are communicatively connected; the server is configured to perform the following processing: obtaining a heart-kidney metabolic syndrome patient information table and a preset entity relationship information set, wherein the heart-kidney metabolic syndrome patient information table comprises a plurality of patient identifiers; creating and updating a heart-kidney metabolic syndrome knowledge graph based on the preset entity relationship information set; for each patient identifier, collecting at least one diagnosis and treatment event record information corresponding to the patient identifier from the plurality of heterogeneous data sources to generate a standardized diagnosis and treatment event record information set; for each generated standardized diagnosis and treatment event record information set, generating an initial diagnosis and treatment event spatio-temporal graph based on the standardized diagnosis and treatment event record information set; determining each generated initial diagnosis and treatment event spatio-temporal graph as an initial diagnosis and treatment event spatio-temporal graph set, and storing the initial diagnosis and treatment event spatio-temporal graph set; in response to receiving diagnosis data sent by one of the at least one terminal, generating a diagnosis and treatment event spatio-temporal graph corresponding to the diagnosis data based on the diagnosis data and the initial diagnosis and treatment event spatio-temporal graph set; generating auxiliary decision support information based on the diagnosis and treatment event spatio-temporal graph and the updated heart-kidney metabolic syndrome knowledge graph; sending the auxiliary decision support information to the terminal, and controlling a dispensing robot to dispense medicine based on the auxiliary decision support information.

[0008] In a second aspect, some embodiments of the present disclosure provide an auxiliary decision support information generation method, which comprises: obtaining a heart-kidney metabolic syndrome patient information table and a preset entity relationship information set, wherein the heart-kidney metabolic syndrome patient information table comprises respective patient identifiers; creating and updating a heart-kidney metabolic syndrome knowledge graph based on the preset entity relationship information set; for each patient identifier, collecting at least one diagnosis and treatment event record information corresponding to the patient identifier from respective heterogeneous data sources to generate a standardized diagnosis and treatment event record information set; for each generated standardized diagnosis and treatment event record information set, generating an initial diagnosis and treatment event spatio-temporal graph based on the standardized diagnosis and treatment event record information set; determining each generated initial diagnosis and treatment event spatio-temporal graph as an initial diagnosis and treatment event spatio-temporal graph set, and storing the initial diagnosis and treatment event spatio-temporal graph set; in response to receiving visit data sent by one of the at least one terminal, generating a diagnosis and treatment event spatio-temporal graph corresponding to the visit data based on the visit data and the initial diagnosis and treatment event spatio-temporal graph set; generating auxiliary decision support information based on the diagnosis and treatment event spatio-temporal graph and the updated heart-kidney metabolic syndrome knowledge graph; sending the auxiliary decision support information to the terminal, and controlling a dispensing robot to dispense medicine based on the auxiliary decision support information.

[0009] The above-described embodiments of this disclosure have the following beneficial effects: the auxiliary decision support information generation system of some embodiments of this disclosure reduces the scheduling resources of the medication dispensing robot. Specifically, the reason for the waste of scheduling resources for the medication dispensing robot is that patients with cardiorenal metabolic syndrome have their own unique disease progression, treatment response, complications, etc. Systems based on a single data source may not be able to reflect the long-term changes in the patient's condition (e.g., the patient's medical information is scattered across different medical institutions, different departments, and different information systems, and a single data source lacks information such as the patient's treatment history and the trend of disease changes). At the same time, systems based on a single data source or a static knowledge base often rely on pre-set rules and knowledge, which are usually limited and do not update with the dynamic changes of the patient. For complex diseases such as cardiorenal metabolic syndrome, knowledge and clinical guidelines are constantly updated with time and research progress, leading to a mismatch between clinical decision recommendations and the patient's actual condition. The accuracy of the generated auxiliary decision support information is low. When scheduling the medication dispensing robot to dispense medication based on auxiliary decision support information with low accuracy, it is easy to cause the medication to be dispensed to not meet the current needs of the patient, thus wasting the scheduling resources of the medication dispensing robot. Based on this, some embodiments of the auxiliary decision support information generation system disclosed herein include: a server, various heterogeneous data sources, and at least one terminal, wherein the server, the various heterogeneous data sources, and at least one terminal are communicatively connected; the server is configured to perform the following processing: acquiring a patient information table for cardiorenal metabolic syndrome and a preset entity relationship information set, wherein the patient information table for cardiorenal metabolic syndrome includes various patient identifiers. Thus, a patient information table for cardiorenal metabolic syndrome containing the identifiers corresponding to each patient with cardiorenal metabolic syndrome can be acquired. A preset entity relationship information set for creating a knowledge graph of renal metabolic syndrome is also acquired. Subsequently, based on the preset entity relationship information set, a knowledge graph for cardiorenal metabolic syndrome is created and updated. Thus, a knowledge graph for cardiorenal metabolic syndrome can be constructed and updated to reflect the latest clinical knowledge and guidelines, resulting in a knowledge graph for cardiorenal metabolic syndrome used for reasoning and generating auxiliary decision support information. Subsequently, for each patient identifier, at least one medical event record information corresponding to the patient identifier is collected from various heterogeneous data sources to generate a standardized medical event record information set. Therefore, multi-source heterogeneous data can be integrated from various heterogeneous data sources (e.g., different medical institutions, different departments, different information systems) to overcome the limitations of a single data source. This allows for the acquisition of more comprehensive information on at least one medical event record corresponding to a patient's identifier, thereby generating a standardized medical event record information set to understand the patient's medical history, disease progression trends, and other information. Subsequently, for each generated standardized medical event record information set, an initial spatiotemporal diagram of medical events is generated based on this standardized medical event record information set.Therefore, an initial spatiotemporal graph of treatment events can be obtained, structurally representing the temporal and spatial relationships of the patient's treatment events and characterizing the long-term trend of the condition. Next, each generated initial spatiotemporal graph of treatment events is defined as an initial spatiotemporal graph set, and this initial spatiotemporal graph set is stored. This allows the system to store the defined initial spatiotemporal graph sets so that they can be retrieved at any time during future treatment processes. Then, in response to receiving medical data sent by one of the at least one of the aforementioned terminals, a spatiotemporal graph of treatment events corresponding to the aforementioned medical data is generated based on the medical data and the initial spatiotemporal graph sets. Thus, by combining the current medical information (medical data) and historical spatiotemporal data (initial spatiotemporal graph sets), a spatiotemporal graph of treatment events reflecting the patient's latest condition, treatment history, and trend of condition changes can be constructed. Finally, based on the spatiotemporal graph of treatment events and the updated knowledge graph of cardiorenal metabolic syndrome, auxiliary decision support information is generated. Therefore, based on the spatiotemporal graph of diagnostic and treatment events reflecting the patient's latest condition and its changing trends, and the dynamically updated knowledge graph of cardiorenal metabolic syndrome, more tailored decision-making suggestions can be generated, improving the accuracy of the generated auxiliary decision support information. Subsequently, this auxiliary decision support information is sent to the aforementioned terminal, and based on this information, the medication dispensing robot associated with the terminal is controlled to retrieve medication. This allows for the transmission of more accurate auxiliary decision support information to the terminal, and the control of the medication dispensing robot based on this more accurate information, reducing the number of erroneous medication retrievals and thus minimizing the waste of scheduling resources for the medication dispensing robot. Attached Figure Description

[0010] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0011] Figure 1 This is an architecture diagram of an exemplary system for generating auxiliary decision support information based on this disclosure; Figure 2 Flowcharts of some embodiments of the method for generating auxiliary decision support information according to this disclosure. Detailed Implementation

[0012] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0013] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0014] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0015] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0016] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0017] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Figure 1 An exemplary system architecture 100 for a decision support information generation system to which some embodiments of the present disclosure may be applied is shown.

[0019] like Figure 1As shown, the system architecture 100 may include: a server 2, various heterogeneous data sources 1, and at least one terminal 3, wherein the server 2, the at least one heterogeneous data source 1, and the at least one terminal 3 are communicatively connected. Each of the at least one heterogeneous data source is communicatively connected to the server 2. Each of the at least one terminal 3 is communicatively connected to the server 2. The at least one heterogeneous data source 1 can be a data system from different sources or in different formats. For example, each of the at least one heterogeneous data source 1 can be, but is not limited to, one of the following: a medical system of a medical institution, or a departmental information system. The medical system of the medical institution can be a system used by the medical institution. The departmental information system can be a system or database used by different departments within the same hospital. Each of the at least one terminal 3 can be a terminal device used by a doctor in the department.

[0020] In some embodiments, the server 2 described above can be configured to perform the following steps: First, obtain a patient information table for cardiorenal metabolic syndrome and a preset entity relationship information set. The patient information table includes individual patient identifiers. In practice, the executing entity can obtain the patient information table and preset entity relationship information set from a preset database. The patient information table can be a data table used to record patient identifiers for patients with cardiorenal metabolic syndrome. The patient identifier can be a unique code (e.g., 0001) used to represent a patient with cardiorenal metabolic syndrome. Each preset entity relationship in the preset entity relationship information set can be textual information containing entities and relationships related to diseases, medications, symptoms, etc., associated with cardiovascular-kidney-metabolic syndrome (CKM). For example, the preset entity relationship information could be "Diabetes can lead to chronic kidney disease and worsen heart disease symptoms."

[0021] Second, based on the aforementioned pre-defined entity relationship information set, a knowledge graph of cardiorenal metabolic syndrome is created and updated. This knowledge graph can be a knowledge graph that organizes and displays knowledge and information related to cardiorenal metabolic syndrome through a graph structure.

[0022] In some optional implementations of certain embodiments, the server 2 described above can create and update the knowledge graph of cardiorenal metabolic syndrome based on the aforementioned preset entity relationship information set through the following steps: The first step is to perform the following steps for each preset entity relationship in the aforementioned preset entity relationship information set: The first sub-step involves performing entity recognition processing on the aforementioned preset entity relationship information to obtain entity recognition information. This entity recognition information includes at least one entity piece of information, and each entity piece of information includes an entity name and entity type information. The entity name can be the specific name or description of the entity. The entity type information can be the domain or category to which the entity belongs. For example, the entity name can include, but is not limited to, one of the following: disease entity names (e.g., "diabetes", "hypertension", "cardiorenal metabolic syndrome"), drug entity names (e.g., "metformin", "atorvastatin", "diuretics"), and symptom entity names (e.g., "fatigue", "dizziness", "edema"). The entity type information can be, but is not limited to, one of the following: disease entity (e.g., the entity type information corresponding to the entity name diabetes can be disease), drug entity, or symptom entity. In practice, the entity recognition information can be obtained by performing entity recognition processing on the aforementioned preset entity relationship information using a Hidden Markov Model. Optionally, the aforementioned execution entity can use a pre-trained NER model (such as a BERT-based model, a SpaCy NER model, etc.) to perform entity recognition on the pre-defined entity relationship information after word segmentation, mark the entities in the pre-defined entity relationship information, and determine their types (such as diseases, drugs, symptoms, etc.).

[0023] The second sub-step involves performing relation extraction processing on the at least one entity information to obtain extracted entity relation information. In practice, the server 2 can extract entity relation information corresponding to at least one entity information from the preset entity relation information using predefined rules and templates. For example, if one entity is "disease" and another entity is "drug," and there are words indicating a treatment relationship between them (such as "treatment," "relief," "control"), the relationship between them can be automatically determined. As an example, the at least one entity information could be "Entity information: diabetes, disease entity; Entity information: metformin, drug." The predefined rule could be that the relationship between an entity with the entity type information "disease" and an entity with the entity type information "drug" is "treatment," then the entity information containing the disease entity and the entity information containing the drug entity have a treatment relationship. The extracted entity relation information can be represented using JSON format data. For example, the extracted entity relation information could be "{ "head_entity": "diabetes", "tail_entity": "metformin", "relation_type": "treatment"}.

[0024] The third sub-step involves generating an entity relationship graph corresponding to the aforementioned preset entity relationship information, based on the extracted entity relationship information. In practice, the entity names and relationships in the extracted entity relationship information (JSON data) can be represented as nodes and edges in a graph using a graph database (such as Neo4j graph database) to obtain the entity relationship graph.

[0025] The second step is to generate a knowledge graph of cardiorenal metabolic syndrome based on the generated entity relationship graphs. In practice, the MERGE operation in the graph database can be used to merge the various entity relationship graphs into a unified graph as the knowledge graph of cardiorenal metabolic syndrome.

[0026] The third step involves retrieving data related to cardiorenal metabolic syndrome from a preset data source at predetermined time intervals as reference data to be updated. This reference data can be the latest data related to cardiorenal metabolic syndrome stored in the preset data source (e.g., data related to clinical knowledge and guidelines related to cardiorenal metabolic syndrome).

[0027] The fourth step involves inputting the obtained reference data to be updated into a pre-trained entity recognition model to obtain at least one entity information. The aforementioned NER model can be an NER model.

[0028] The fifth step is to input the aforementioned reference data to be updated and at least one entity information into the pre-trained relation extraction model to obtain at least one extracted entity relation information.

[0029] Step 6: Based on at least one of the extracted entity relationship information, update the aforementioned new cardiorenal metabolic syndrome knowledge graph. In practice, for each extracted entity relationship information, server 2 can use a graph database (such as Neo4j graph database) to represent the entity names and relationships in the extracted entity relationship information (JSON data) as nodes and edges in a graph, thus obtaining an entity relationship graph. Then, server 2 can merge the entity relationship graph into the cardiorenal metabolic syndrome knowledge graph to update the new cardiorenal metabolic syndrome knowledge graph.

[0030] Third, for each patient identifier, at least one medical event record information corresponding to the patient identifier is collected from various heterogeneous data sources to generate a standardized medical event record information set. Each medical event record information includes structured data and / or unstructured data. The structured data includes at least one sensitive information and original medical event data, and the unstructured data includes at least one medical examination image. The at least one sensitive information may include, but is not limited to, the following: mobile phone number, name, and ID card number. The original medical event data may be text information of a diagnostic record (e.g., vital signs: blood pressure: 160 / 100 mmHg, weight: 78 kg, medication record: medication taken: antihypertensive drug A (once daily)). Each medical examination image may be a medical image. In some optional implementations of some embodiments, the server 2 may collect at least one medical event record information corresponding to the patient identifier from various heterogeneous data sources through the following steps to generate a standardized medical event record information set: The first step is to perform the following steps for each of the at least one medical event record entries mentioned above: The first sub-step, in response to determining that the aforementioned medical event record information includes structured data, involves performing the following update steps on the aforementioned medical event record information: Sub-step one involves de-identifying at least one sensitive piece of information included in the structured data of the aforementioned medical event records, obtaining at least one de-identified piece of information. In practice, at least one sensitive piece of information included in the structured data can be identified by defining regular expressions (e.g., ID card number format, telephone number format, etc.). Then, K-anonymization technology can be used to fuzzify the identified at least one sensitive piece of information, obtaining at least one de-identified piece of information.

[0031] Sub-step two involves replacing at least one sensitive piece of information included in the structured data of the aforementioned medical event record information with at least one desensitized piece of data to update the aforementioned medical event record information.

[0032] Step three involves standardizing the original medical event data in the aforementioned medical event record information to update the information. In practice, a pre-defined terminology mapping information set can be used to map the terms in the original medical event data to standardized terms and units of measurement, thus updating the medical event record information. Specifically, Natural Language Processing (NLP) technology can be used to identify non-standard terms in the original medical event data and standardize them into international medical terms (e.g., SNOMEDCT or LOINC) by searching the mapping relationships represented by the pre-defined terminology mapping set, thus performing an initial update to the medical event record information. Then, server 2 can use a unit conversion algorithm to standardize the units of measurement in the initially updated medical event record information to standardized units, thus updating the medical event record information. For example, converting "mg / dL" in the medical event record information to "mmol / L".

[0033] The second sub-step involves, in response to determining that the aforementioned medical event record information includes unstructured data, desensitizing the aforementioned at least one medical test image to obtain at least one desensitized medical test image.

[0034] The third sub-step involves updating the unstructured data in the aforementioned medical event record information with at least one desensitized medical detection image to update the aforementioned medical event record information.

[0035] The second step is to identify at least one updated medical event record as a standardized medical event record information set.

[0036] In the process of adopting technical solutions to address the problems mentioned in the background section, the following issues often arise: Because medical examination images often contain sensitive data such as patient names, IDs, and dates, the "fixed area masking method" is commonly used to desensitize them. This method involves blurring or mosaicking fixed areas such as the image edges to desensitize the medical examination image. However, the fixed area masking method only processes certain areas of the image and cannot provide comprehensive privacy protection for all sensitive information (especially text information) in the image. If the sensitive data is located in a non-fixed area of ​​the image, incomplete desensitization or accidental desensitization may occur, resulting in poor desensitization effect.

[0037] Faced with the above-mentioned technical problems, the inventors decided to adopt the following solution: The first step is to perform the following desensitization process on each of the at least one medical examination image mentioned above: The first sub-step involves inputting the aforementioned medical examination image into a pre-trained text box region detection model to obtain at least one text box region information. This text box region detection model can be a YOLO series model that takes the medical examination image as input and at least one text box region information as input. Each of the at least one text box region information represents the position of a text box in the aforementioned medical examination image and may include various coordinate points.

[0038] The second sub-step involves determining at least one region image of the area containing at least one text box region information in the aforementioned medical examination image as at least one target region image.

[0039] The third sub-step involves performing the following steps for each of the at least one target region image mentioned above: Sub-step one involves extracting text information from the target region image to obtain the extracted text information. In practice, optical character recognition (OCR) technology can be used to extract the text information from the target region image to obtain the extracted text information.

[0040] Sub-step two involves performing privacy detection processing on the extracted text information to obtain privacy detection information. In practice, regular expressions (e.g., ID number format, telephone number format, etc.) can be defined to identify privacy data in the extracted text information, thus obtaining identified privacy data. In response to determining that the identified privacy data is empty, information indicating the absence of privacy data in the extracted text information is identified as privacy detection information. In response to determining that the identified privacy data is not empty, information indicating the presence of privacy data in the extracted text information is identified as privacy detection information.

[0041] In sub-step three, in response to determining that the above privacy detection information indicates the presence of privacy data in the above extracted text information, the text box area information corresponding to the above extracted text information is determined as the area information to be desensitized.

[0042] Sub-step four involves desensitizing the images in the aforementioned medical examination image that correspond to the information of the area to be desensitized, thereby updating the medical examination image. In practice, the server 2 can employ image blurring technology to desensitize the images in the aforementioned medical examination image that correspond to the information of the area to be desensitized, thereby updating the medical examination image.

[0043] The fourth sub-step involves identifying the updated medical test image as the initial desensitized medical test image.

[0044] The fifth sub-step involves inputting the initial desensitized medical test image into a pre-trained medical record card removal and content reconstruction model to obtain a desensitized medical test image. This medical record card removal and content reconstruction model can be a model that automatically removes the area containing the medical record card from the initial desensitized medical test image while simultaneously reconstructing the background content of the removed area. This model can include both a Mask R-CNN model and an image inpainting model. The Mask R-CNN model can be used to remove the area containing the medical record card from the initial desensitized medical test image. The image inpainting model can be used to reconstruct the background content after removing the area containing the medical record card from the image.

[0045] The above-mentioned technical solution and related content, as an inventive point of this disclosure, solve the technical problem of "poor desensitization effect of medical examination images". Factors leading to poor desensitization effect of medical examination images are often as follows: Since medical examination images are often burned with private data such as patient names, IDs, and dates, the "fixed region masking method" is usually used when desensitizing medical examination images. This involves directly blurring or mosaicking fixed areas such as image edges to desensitize the medical examination image. However, the fixed region masking method only processes certain areas of the image and cannot provide comprehensive privacy protection for all sensitive information (especially text information) in the image. If the private data is located in a non-fixed area of ​​the image, incomplete desensitization or false desensitization will occur, resulting in poor desensitization effect. Solving the above factors can improve the desensitization effect of medical examination images. To achieve this effect, firstly, for each of the above at least one medical examination image, the following desensitization process is performed: Step 1: The above medical examination image is input into a pre-trained text box region detection model to obtain at least one text box region information. Therefore, at least one text box region information can be detected in at least one text box region in a medical examination image, and all regions in the image that may contain text can be automatically and comprehensively located, regardless of whether they are located in a fixed or non-fixed region of the image. The second step is to determine at least one region image of the area containing the at least one text box region information in the aforementioned medical examination image as at least one target region image. The third step is to perform the following steps for each target region image: Sub-step one, extracting text information from the target region image to obtain extracted text information. This yields the text information of the region image, i.e., the region where the text box is located. Sub-step two, performing privacy detection processing on the extracted text information to obtain privacy detection information. This determines whether the extracted text information truly contains privacy data, yielding privacy detection information. Sub-step three, in response to determining that the privacy detection information indicates the presence of privacy data in the extracted text information, the text box region information corresponding to the extracted text information is determined as the region information to be desensitized. This identifies the region information to be desensitized for the region requiring desensitization. Step four involves desensitizing the images corresponding to the areas to be desensitized in the aforementioned medical examination images to update the medical examination images. This removes private data (such as patient information, examination dates, etc.) from the images, preventing leakage. Step four involves identifying the updated medical examination image as the initial desensitized medical examination image. Step five involves inputting the initial desensitized medical examination image into a pre-trained medical record card removal and content reconstruction model to obtain the desensitized medical examination image.Therefore, it is possible to remove medical record cards containing potentially private data from initially anonymized medical examination images, excluding text boxes, and to reconstruct the background content of the area where the medical record cards were located, thereby improving the visual integrity of the image after removing the medical record cards. Furthermore, because the text box region detection model achieves privacy data anonymization for all potential text box regions in the medical examination image, it solves the problem of incomplete or incorrect anonymization caused by the "fixed region masking method," thus improving the anonymization effect of medical examination images. Simultaneously, the use of a medical record card elimination and content reconstruction model to remove potentially private data-containing medical record cards from initially anonymized medical examination images, excluding text boxes, further enhances the anonymization effect of medical examination images.

[0046] Fourth, for each standardized medical event record information set generated, an initial spatiotemporal graph of medical events is generated based on the aforementioned standardized medical event record information set. Here, each standardized medical event record information set refers to each standardized medical event record information in the generated sets of standardized medical event record information.

[0047] In some optional implementations of certain embodiments, the server 2 described above can generate an initial spatiotemporal graph of medical events based on the standardized medical event record information set described above through the following steps: The first step involves extracting key information from each standardized medical event record in the aforementioned standardized medical event record information set to obtain medical event node information. This node information includes metadata, such as timestamps and node attribute information. The node attribute information includes node identifiers and attribute information. In practice, server 2 can utilize Natural Language Processing (NLP) technology to extract key information from the standardized medical events, obtaining timestamps, node identifiers, and attribute information. The standardized medical event record information can be the text information of the main content of the standardized medical event record information. The node identifier can be the name of the department the patient visited, and the attribute information can be the original medical event data. For example, the standardized medical event record information could be: "On June 10, 2024, at 09:00:00, the patient visited the endocrinology department. The original medical event data was: blood glucose: 8.0 mmol / L, blood pressure: 130 / 85 mmHg, chief complaint: the patient's first visit, complaining of polydipsia, polyuria, and weight loss for one week." The node attribute information mentioned above can be as follows: Node identifier: Endocrinology Clinic; Attribute information: Blood glucose: 8.0 mmol / L, Blood pressure: 130 / 85 mmHg; Chief complaint: The patient is visiting for the first time and complains of polydipsia, polyuria, and weight loss for one week.

[0048] The second step is to define the obtained diagnostic and treatment event node information as a diagnostic and treatment event node information set. Each diagnostic and treatment event node information can represent information related to the diagnostic and treatment event node.

[0049] The third step involves generating a time-series edge information set based on the aforementioned set of diagnostic and treatment event node information. In practice, the executing entity can sort the diagnostic and treatment event node information in the set from front to back according to the timestamps contained in the information, thus obtaining a sequence of diagnostic and treatment event node information. Then, server 2 can generate edge information between every two consecutive diagnostic and treatment event node information in the sequence as time-series edge information. This time-series edge information can be a directed edge indicating the temporal order of two consecutive diagnostic and treatment event node information (i.e., two diagnostic and treatment event node information that are sequentially related in the sequence).

[0050] The fourth step involves performing time-series classification on the aforementioned set of medical event node information to generate a set of same-visit edge information. In practice, the executing entity can group at least one medical event node information whose timestamps represent the same day into a set of medical event node information, based on the timestamps included in the information. Then, for each of the obtained sets of medical event node information, a same-visit edge is generated between every two medical event node information in the set using graph structure construction techniques, resulting in same-visit edge information. This same-visit edge information does not necessarily need to be graph data describing the relationship between two medical event node information.

[0051] The fifth step is to determine the set of information on the diagnosed event nodes, the set of information on the aforementioned temporal edges, and the set of information on the same visit edges as the spatiotemporal graph structure information.

[0052] Step 6: Based on the aforementioned spatiotemporal graph structure information, create an initial spatiotemporal graph of medical events. In practice, a pre-defined graph database (such as Neo4j) can be used to create nodes and edges (time-sequence edges and same-visit edges) related to the spatiotemporal graph structure information to obtain the initial spatiotemporal graph of medical events. The aforementioned initial spatiotemporal graph of medical events is a graph structure used to represent the standardized medical event records of a patient during the medical process, i.e., the patient's medical records and their temporal relationships. This graph contains multiple nodes and edges, where nodes represent medical events (such as diagnosis, treatment, examination, etc.), and edges represent the relationships between events (such as time sequence, same-visit, etc.).

[0053] Fifth, the generated spatiotemporal graphs of each initial treatment event are defined as an initial spatiotemporal graph set, and this initial spatiotemporal graph set is stored. In practice, the executing entity can store the initial spatiotemporal graph set in a preset database (e.g., the Neo4j database). Each initial spatiotemporal graph in the initial spatiotemporal graph set corresponds to one patient identifier among various patient identifiers.

[0054] Sixth, in response to receiving medical data sent by one of the at least one of the aforementioned terminals, a medical event spatiotemporal graph corresponding to the medical data is generated based on the medical data and the initial medical event spatiotemporal graph set. The medical data includes medical event record information and patient identifiers. The aforementioned update steps are performed on the medical data including the medical event record information to obtain updated medical event record information as the current standardized medical event record information. Key information extraction processing is performed on the current standardized medical event record information to generate a medical event node information set as the current medical event node information set. Based on the current medical event node information set, an initial medical event spatiotemporal graph corresponding to the current medical event node information set is created as the current medical event spatiotemporal graph. The server 2 can determine the patient identifier included in the medical event records as the query patient identifier. Then, the initial medical event spatiotemporal graph corresponding to the query patient identifier in the initial medical event spatiotemporal graph set can be determined as the query medical event spatiotemporal graph. Then, by employing a graph theory merging algorithm, the current treatment event spatiotemporal graph is integrated into the query treatment event spatiotemporal graph to obtain the treatment event spatiotemporal graph. It should be noted that server 2 can replace the query treatment event spatiotemporal graph in the stored initial treatment event spatiotemporal graph set with the aforementioned treatment event spatiotemporal graph to update the initial treatment event spatiotemporal graph set.

[0055] Seventh, based on the spatiotemporal graph of diagnosis and treatment events and the updated knowledge graph of cardiorenal metabolic syndrome, auxiliary decision support information is generated.

[0056] In some optional implementations of certain embodiments, the server 2 described above can generate auxiliary decision support information based on the spatiotemporal graph of diagnostic events and the updated knowledge graph of cardiorenal metabolic syndrome through the following steps: The first step is to determine the spatiotemporal graph structure information corresponding to the above-mentioned spatiotemporal graph of diagnosis and treatment events as the target spatiotemporal graph structure information.

[0057] The second step is to arrange the diagnosis and treatment event node information in the set of diagnosis and treatment event node information included in the above target spatiotemporal graph structure information according to the timestamp to obtain the diagnosis and treatment event node information sequence.

[0058] The third step is to arrange the attribute information of each node included in the above diagnosis and treatment event node information sequence to obtain the node attribute information sequence.

[0059] The fourth step involves inputting the aforementioned node attribute information sequence into the vital sign trend feature extraction layer of a pre-trained cross-modal feature extraction model to obtain vital sign trend feature information. The aforementioned staging risk and prognosis prediction model includes the aforementioned vital sign trend feature extraction layer, medical semantic feature extraction layer, and feature fusion layer. The aforementioned vital sign trend feature extraction layer can be an encoder layer. The aforementioned vital sign trend feature information can be an abstract feature representation characterizing the pattern and trend of changes in a patient's vital signs over time (e.g., the characteristic vital sign change trends of certain diseases).

[0060] The fifth step involves inputting the aforementioned set of diagnostic and treatment event node information into the aforementioned medical semantic feature extraction layer to obtain medical semantic feature extraction information. This medical semantic feature extraction layer can be a BioBERT model. The extracted medical semantic feature information can be an abstract feature representation reflecting the medical meaning, internal logic, and relationships of the diagnostic and treatment events. For example, the set of diagnostic and treatment event node information could be: "Node 1: Event type is 'Outpatient,' timestamp is '2024-07-10 09:00:00,' patient ID is 'P001,' symptom description is 'persistent cough for one week, accompanied by a small amount of white mucus sputum, no fever.' Node 2: Event type is 'Examination,' timestamp is '2024-07-10 10:00:00, examination item is 'Chest X-ray examination,' examination result is 'increased lung markings bilaterally.'" The extracted medical semantic feature information can be an abstract feature representation (such as a feature vector) representing that "the patient's 'chest X-ray examination' result shows 'increased lung markings bilaterally,' and the doctor diagnosed 'acute bronchitis' accordingly."

[0061] The sixth step involves inputting the aforementioned vital sign trend information and medical semantic feature extraction information into the feature fusion layer to obtain cross-modal patient feature information. Specifically, the feature fusion layer can concatenate the vital sign trend information and medical semantic feature extraction information to obtain cross-modal patient feature information.

[0062] Step 7: Based on the aforementioned cross-modal patient characteristic information and the aforementioned knowledge graph of cardiorenal metabolic syndrome, generate staging risk and prognostic prediction information, which includes disease stage information and prognostic risk information. In practice, the cross-modal patient characteristic information and the aforementioned knowledge graph of cardiorenal metabolic syndrome can be input into a pre-trained GraphSAGE model to obtain staging risk and prognostic prediction information. This staging risk and prognostic prediction information can represent the risk assessment of the stage of cardiorenal metabolic syndrome and information on future disease development (e.g., in the middle stage of cardiorenal metabolic syndrome, the risk of cardiovascular events within the next year is 30%, and the 5-year survival rate after standardized treatment is 70%).

[0063] Step 8: Based on the aforementioned staged risk and prognostic prediction information, generate personalized intervention reference information corresponding to the aforementioned medical data. In practice, the staged risk and prognostic prediction information can be input into a large-scale model in a medical vertical field (e.g., a medical large language model) to obtain personalized intervention reference information. This personalized intervention reference information can represent suggested medication based on the staged risk and prognostic prediction information. The personalized intervention reference information includes at least one drug identifier.

[0064] The ninth step is to identify the aforementioned staged risk and prognostic prediction information and the aforementioned individualized intervention reference information as auxiliary decision support information.

[0065] The above technical solution and its related content are combined to send the above-mentioned auxiliary decision support information to the above-mentioned terminal, and based on the above-mentioned auxiliary decision support information, control the medication dispensing robot associated with the above-mentioned terminal to dispense medication. As an inventive point of this disclosure, it solves the technical problem that "the accuracy of staged risk and prognosis prediction information is low, and the reference intervention plan represented by the generated individualized intervention reference information is uniform and lacks personalization, which in turn leads to the waste of scheduling resources of the medication dispensing robot." The factors that lead to the low accuracy of staged risk and prognosis prediction information, the uniformity of reference intervention plan represented by the generated individualized intervention reference information, the lack of personalization, and the waste of scheduling resources of the medication dispensing robot are often as follows: Cardiorenal metabolic syndrome is a complex chronic disease, and its disease progression is contained in a large number of multimodal and time-sequential diagnosis and treatment events of patients. Decision-making based on fragmented, static patient data lacking knowledge support struggles to simultaneously capture the long-term dynamic trends of physiological indicators and the rich medical semantics inherent in clinical texts. Furthermore, it lacks the ability to connect individual patient data with global medical knowledge, resulting in low accuracy in the staging risk and prognostic prediction information included in the decision support information. The generated personalized intervention reference information also presents generic intervention plans with weak individualization. Additionally, scheduling medication dispensing robots based on this inaccurate decision support information can easily lead to medications not meeting the patient's current needs, wasting the robot's scheduling resources. Addressing these factors can improve the accuracy of staging risk and prognostic prediction information, enhance the individualization of personalized intervention reference information, and reduce the waste of medication dispensing robot scheduling resources. To achieve this, firstly, the spatiotemporal graph structure information corresponding to the aforementioned diagnostic and treatment event spatiotemporal graph is determined as the target spatiotemporal graph structure information. This allows for the acquisition of the patient's target spatiotemporal graph structure information. Then, the diagnostic and treatment event node information in the diagnostic and treatment event node information set included in the above target spatiotemporal graph structure information is arranged according to timestamps to obtain a sequence of diagnostic and treatment event node information. This yields a sequence of diagnostic and treatment event node information used to generate a sequence of node attribute information. Next, the node attribute information included in the above sequence of diagnostic and treatment event node information is arranged to obtain a sequence of node attribute information. Then, the above sequence of node attribute information is input into the vital sign trend feature extraction layer of a pre-trained cross-modal feature extraction model to obtain vital sign trend feature information. This allows the use of a deep learning model to extract deep features representing the long-term evolution and dynamic patterns of the patient's physiological state. Next, the above set of diagnostic and treatment event node information is input into the above medical semantic feature extraction layer to obtain medical semantic feature extraction information. This allows the extraction of medical semantic features to generate cross-modal patient feature information. Finally, the above vital sign trend feature information and medical semantic feature extraction information are input into the above feature fusion layer to obtain cross-modal patient feature information.Therefore, based on the aforementioned cross-modal patient characteristic information and the aforementioned knowledge graph of cardiorenal metabolic syndrome, staging risk and prognostic prediction information can be generated. This involves reasoning between the patient's unique, deep cross-modal characteristics (i.e., cross-modal patient characteristic information) and the cardiorenal metabolic syndrome knowledge graph representing global medical knowledge (such as disease progression pathways and complication associations) to determine the patient's current disease stage and predict future risks (e.g., potential heart failure hospitalization in the second year). Then, based on the aforementioned staging risk and prognostic prediction information, personalized intervention reference information corresponding to the aforementioned medical data can be generated. This allows for the generation of personalized intervention reference information—information more aligned with the patient's condition and health status—improving the accuracy of staging risk and prognostic prediction information. Subsequently, based on the aforementioned staging risk and prognostic prediction information, personalized intervention reference information corresponding to the aforementioned medical data can be generated. Thus, based on more accurate staging risk and prognostic prediction information, unique personalized intervention reference information tailored to each patient can be generated, enhancing the personalization of the personalized intervention reference information. Next, the aforementioned staged risk and prognostic prediction information and the aforementioned individualized intervention reference information are identified as auxiliary decision support information. This auxiliary decision support information provides clinical physicians with support for decision-making. By sending this auxiliary decision support information to the aforementioned terminal and controlling the medication dispensing robot associated with the terminal based on this information, the medication dispensing robot can be controlled to dispense medication based on more accurate auxiliary decision support information, thereby reducing the number of erroneous dispensings and ultimately reducing the waste of scheduling resources for the medication dispensing robot.

[0066] Tenth, the aforementioned auxiliary decision support information is sent to the aforementioned terminal, and based on the aforementioned auxiliary decision support information, the medication dispensing robot associated with the aforementioned terminal is controlled to dispensing medication. In practice, the aforementioned executing entity can control the medication dispensing robot to retrieve at least one drug from the pharmacy corresponding to at least one drug identifier included in at least one drug identifier in the aforementioned individualized intervention reference information, and to deliver the retrieved medication to the department where the aforementioned terminal is located.

[0067] In some embodiments, the server 2 described above can be further configured as follows: Replace the initial diagnosis and treatment event spatiotemporal graph corresponding to the above diagnosis and treatment event spatiotemporal graph in the stored initial diagnosis and treatment event spatiotemporal graph set with the above diagnosis and treatment event spatiotemporal graph to update the stored initial diagnosis and treatment event spatiotemporal graph set.

[0068] In some embodiments, the server 2 described above can be further configured as follows: The first step, in response to receiving a spatiotemporal graph query request for diagnostic events from one of the aforementioned terminals, is to perform the following query steps: The first sub-step involves determining the aforementioned spatiotemporal graph query request information for the diagnosis and treatment events as query request information, which includes the query patient identifier.

[0069] The second sub-step involves retrieving the initial treatment event spatiotemporal graph or treatment event spatiotemporal graph corresponding to the queried patient identifier from the updated stored initial treatment event spatiotemporal graph set.

[0070] The second step is to send the obtained initial or spatiotemporal diagram of the diagnosis and treatment events to the aforementioned terminal.

[0071] Figure 2 A flowchart 200 illustrates some embodiments of a decision support information generation method according to the present disclosure, which applies a server included in the aforementioned decision support information generation system. The decision support information generation method includes the following steps: Step 201: Obtain the patient information table for cardiorenal metabolic syndrome and the preset entity relationship information set.

[0072] In some embodiments, the executing entity of the auxiliary decision support information generation method (e.g., the server included in the auxiliary decision support information generation system) can obtain a patient information table for cardiorenal metabolic syndrome and a preset entity relationship information set, wherein the aforementioned patient information table for cardiorenal metabolic syndrome includes individual patient identifiers.

[0073] Step 202: Based on the preset entity relationship information set, create and update the knowledge graph of cardiorenal metabolic syndrome.

[0074] In some embodiments, the aforementioned executing entity may create and update a knowledge graph of cardiorenal metabolic syndrome based on the aforementioned preset entity relationship information set.

[0075] Step 203: For each patient identifier, collect at least one medical event record information corresponding to the patient identifier from various heterogeneous data sources to generate a standardized medical event record information set.

[0076] In some embodiments, the aforementioned executing entity may collect at least one medical event record information corresponding to each patient identifier from various heterogeneous data sources to generate a standardized medical event record information set.

[0077] Step 204: For each standardized medical event record information set generated, an initial medical event spatiotemporal diagram is generated based on the standardized medical event record information set.

[0078] In some embodiments, the aforementioned executing entity may generate an initial spatiotemporal diagram of medical events based on each standardized medical event record information set generated.

[0079] Step 205: Determine the generated spatiotemporal graphs of each initial diagnosis and treatment event as the initial spatiotemporal graph set, and store the initial spatiotemporal graph set.

[0080] In some embodiments, the execution entity may determine the generated spatiotemporal graphs of each initial diagnosis and treatment event as an initial spatiotemporal graph set, and store the initial spatiotemporal graph set of the initial diagnosis and treatment events.

[0081] Step 206: In response to receiving medical data sent by one of the terminals at least one terminal, generate a spatiotemporal graph of medical events corresponding to the medical data based on the medical data and the initial spatiotemporal graph of medical events.

[0082] In some embodiments, the execution entity may, in response to receiving medical data sent by one of the at least one terminals, generate a spatiotemporal graph of medical events corresponding to the medical data based on the medical data and an initial spatiotemporal graph of medical events.

[0083] Step 207: Generate auxiliary decision support information based on the spatiotemporal graph of diagnosis and treatment events and the updated knowledge graph of cardiorenal metabolic syndrome.

[0084] In some embodiments, the aforementioned implementing entity may generate auxiliary decision support information based on the spatiotemporal graph of diagnostic and treatment events and the updated knowledge graph of cardiorenal metabolic syndrome.

[0085] Step 208: Send auxiliary decision support information to the terminal, and control the drug-dispensing robot associated with the terminal to dispense the drug based on the auxiliary decision support information.

[0086] In some embodiments, the execution entity may send the auxiliary decision support information to the terminal and, based on the auxiliary decision support information, control the drug-dispensing robot associated with the terminal to dispense the drug.

[0087] The above-described embodiments of this disclosure have the following beneficial effects: the auxiliary decision support information generation system of some embodiments of this disclosure reduces the scheduling resources of the medication dispensing robot. Specifically, the reason for the waste of scheduling resources for the medication dispensing robot is that patients with cardiorenal metabolic syndrome have their own unique disease progression, treatment response, complications, etc. Systems based on a single data source may not be able to reflect the long-term changes in the patient's condition (e.g., the patient's medical information is scattered across different medical institutions, different departments, and different information systems, and a single data source lacks information such as the patient's treatment history and the trend of disease changes). At the same time, systems based on a single data source or a static knowledge base often rely on pre-set rules and knowledge, which are usually limited and do not update with the dynamic changes of the patient. For complex diseases such as cardiorenal metabolic syndrome, knowledge and clinical guidelines are constantly updated with time and research progress, leading to a mismatch between clinical decision recommendations and the patient's actual condition. The accuracy of the generated auxiliary decision support information is low. When scheduling the medication dispensing robot to dispense medication based on auxiliary decision support information with low accuracy, it is easy to cause the medication to be dispensed to not meet the current needs of the patient, thus wasting the scheduling resources of the medication dispensing robot. Based on this, some embodiments of the auxiliary decision support information generation system disclosed herein include: a server, various heterogeneous data sources, and at least one terminal, wherein the server, the various heterogeneous data sources, and at least one terminal are communicatively connected; the server is configured to perform the following processing: acquiring a patient information table for cardiorenal metabolic syndrome and a preset entity relationship information set, wherein the patient information table for cardiorenal metabolic syndrome includes various patient identifiers. Thus, a patient information table for cardiorenal metabolic syndrome containing the identifiers corresponding to each patient with cardiorenal metabolic syndrome can be acquired. A preset entity relationship information set for creating a knowledge graph of renal metabolic syndrome is also acquired. Subsequently, based on the preset entity relationship information set, a knowledge graph for cardiorenal metabolic syndrome is created and updated. Thus, a knowledge graph for cardiorenal metabolic syndrome can be constructed and updated to reflect the latest clinical knowledge and guidelines, resulting in a knowledge graph for cardiorenal metabolic syndrome used for reasoning and generating auxiliary decision support information. Subsequently, for each patient identifier, at least one medical event record information corresponding to the patient identifier is collected from various heterogeneous data sources to generate a standardized medical event record information set. Therefore, multi-source heterogeneous data can be integrated from various heterogeneous data sources (e.g., different medical institutions, different departments, different information systems) to overcome the limitations of a single data source. This allows for the acquisition of more comprehensive information on at least one medical event record corresponding to a patient's identifier, thereby generating a standardized medical event record information set to understand the patient's medical history, disease progression trends, and other information. Subsequently, for each generated standardized medical event record information set, an initial spatiotemporal diagram of medical events is generated based on this standardized medical event record information set.Therefore, an initial spatiotemporal graph of treatment events can be obtained, structurally representing the temporal and spatial relationships of the patient's treatment events and characterizing the long-term trend of the condition. Next, each generated initial spatiotemporal graph of treatment events is defined as an initial spatiotemporal graph set, and this initial spatiotemporal graph set is stored. This allows the system to store the defined initial spatiotemporal graph sets so that they can be retrieved at any time during future treatment processes. Then, in response to receiving medical data sent by one of the at least one of the aforementioned terminals, a spatiotemporal graph of treatment events corresponding to the aforementioned medical data is generated based on the medical data and the initial spatiotemporal graph sets. Thus, by combining the current medical information (medical data) and historical spatiotemporal data (initial spatiotemporal graph sets), a spatiotemporal graph of treatment events reflecting the patient's latest condition, treatment history, and trend of condition changes can be constructed. Finally, based on the spatiotemporal graph of treatment events and the updated knowledge graph of cardiorenal metabolic syndrome, auxiliary decision support information is generated. Therefore, based on the spatiotemporal graph of diagnostic and treatment events reflecting the patient's latest condition and its changing trends, and the dynamically updated knowledge graph of cardiorenal metabolic syndrome, more tailored decision-making suggestions can be generated, improving the accuracy of the generated auxiliary decision support information. Subsequently, this auxiliary decision support information is sent to the aforementioned terminal, and based on this information, the medication dispensing robot associated with the terminal is controlled to retrieve medication. This allows for the transmission of more accurate auxiliary decision support information to the terminal, and the control of the medication dispensing robot based on this more accurate information, reducing the number of erroneous medication retrievals and thus minimizing the waste of scheduling resources for the medication dispensing robot.

[0088] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of technical features, but should also cover other technical solutions formed by arbitrary combinations of technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A decision support information generation system, comprising: A server, various heterogeneous data sources, and at least one terminal, wherein the server, the at least one heterogeneous data source, and the at least one terminal are communicatively connected; The server is configured to perform the following processes: Obtain a patient information table for cardiorenal metabolic syndrome and a preset entity relationship information set, wherein the patient information table for cardiorenal metabolic syndrome includes individual patient identifiers; Based on the preset entity relationship information set, a knowledge graph of cardiorenal metabolic syndrome is created and updated; For each patient identifier, at least one medical event record information corresponding to the patient identifier is collected from various heterogeneous data sources to generate a standardized medical event record information set; For each standardized medical event record information set generated, an initial medical event spatiotemporal graph is generated based on the standardized medical event record information set; Each generated initial diagnosis and treatment event spatiotemporal graph is defined as an initial diagnosis and treatment event spatiotemporal graph set, and the initial diagnosis and treatment event spatiotemporal graph set is stored. In response to receiving medical data sent by one of the at least one terminals, a spatiotemporal graph of medical events corresponding to the medical data is generated based on the medical data and an initial spatiotemporal graph of medical events. Based on the spatiotemporal graph of diagnosis and treatment events and the updated knowledge graph of cardiorenal metabolic syndrome, auxiliary decision support information is generated. The auxiliary decision support information is sent to the terminal, and based on the auxiliary decision support information, the drug-dispensing robot associated with the terminal is controlled to dispense the drug.

2. The auxiliary decision support information generation system according to claim 1, wherein, The spatiotemporal graph of the diagnosis and treatment events corresponds to an initial spatiotemporal graph in the initial spatiotemporal graph set of diagnosis and treatment events, and the server is further configured to: Replace the initial diagnosis and treatment event spatiotemporal graph corresponding to the diagnosis and treatment event spatiotemporal graph in the stored initial diagnosis and treatment event spatiotemporal graph set with the diagnosis and treatment event spatiotemporal graph to update the stored initial diagnosis and treatment event spatiotemporal graph set.

3. The auxiliary decision support information generation system according to claim 2, wherein, Each initial treatment event spatiotemporal graph or treatment event spatiotemporal graph in the updated initial treatment event spatiotemporal graph set has a corresponding patient identifier, and the server is further configured to: In response to receiving a spatiotemporal graph query request for diagnostic events from one of the at least one terminals, the following query steps are performed: The spatiotemporal graph query request information of the diagnosis and treatment event is determined as query request information, wherein the query request information includes the query patient identifier; Obtain the initial treatment event spatiotemporal graph or treatment event spatiotemporal graph corresponding to the queried patient identifier from the updated stored initial treatment event spatiotemporal graph set; The obtained initial or spatiotemporal diagram of diagnostic and treatment events is sent to the terminal.

4. The auxiliary decision support information generation system according to claim 1, wherein, Each of the at least one medical event record information includes structured data and / or unstructured data. The structured data includes at least one sensitive piece of information and original medical event data. The unstructured data includes at least one medical examination image. The server is further configured to: For each of the at least one medical event record entries, perform the following update steps: In response to determining that the medical event record information includes structured data, the following update steps are performed on the medical event record information: At least one sensitive piece of information included in the structured data of the medical event record information is desensitized to obtain at least one desensitized piece of information; The at least one sensitive piece of information included in the structured data of the medical event record information is replaced with the at least one desensitized piece of data to update the medical event record information; The original medical event data in the medical event record information is standardized in order to update the medical event record information; In response to determining that the medical event record information includes unstructured data, the at least one medical test image is desensitized to obtain at least one desensitized medical test image; The unstructured data in the medical event record information is updated with the at least one desensitized medical detection image to update the medical event record information; At least one updated medical event record is identified as a standardized medical event record information set.

5. The auxiliary decision support information generation system according to claim 1, wherein, The server is further configured to: For each preset entity relationship information in the preset entity relationship information set, perform the following steps: The preset entity relationship information is processed by entity recognition to obtain entity recognition information, wherein the entity recognition information includes at least one entity information, and each entity information in the at least one entity information includes entity name and entity type information; Perform relation extraction processing on the at least one entity information to obtain extracted entity relation information; Based on the extracted entity relationship information, an entity relationship graph corresponding to the preset entity relationship information is generated; Based on the generated entity relationship graphs, a knowledge graph of cardiorenal metabolic syndrome is generated. At preset time intervals, data related to cardiorenal metabolic syndrome are obtained from preset data sources of cardiorenal metabolic syndrome as reference data to be updated. The obtained reference data to be updated is input into the pre-trained entity recognition model to obtain at least one entity information; The reference data to be updated and the information of at least one entity are input into a pre-trained relation extraction model to obtain at least one extracted entity relation information. The new knowledge graph of cardiorenal metabolic syndrome is updated based on the extracted entity relationship information.

6. The auxiliary decision support information generation system according to claim 1, wherein, Each standardized medical event record in the standardized medical event record information set includes at least one de-identified data and standardized raw medical event data. The raw medical event data includes the treatment time, treatment identifier, and medical event information. The server is further configured to: For each standardized medical event record in the standardized medical event record information set, key information extraction processing is performed on the standardized medical event record information to obtain medical event node information. The medical event node information includes various metadata, which includes timestamps and node attribute information. The node attribute information includes node identifiers and attribute information. The obtained information on each diagnosis and treatment event node is defined as a set of diagnosis and treatment event node information. Based on the set of diagnostic and treatment event nodes, a time-series edge information set is generated; The information set of diagnosis and treatment event nodes is processed by time sequence classification to generate the information set of the same medical visit. The set of information on the diagnosed event nodes, the set of information on the temporal edges, and the set of information on the same visit edge are determined as the spatiotemporal graph structure information; Based on the spatiotemporal graph structure information, an initial spatiotemporal graph of diagnosis and treatment events is created.

7. A method for generating auxiliary decision support information, applied to a server included in the auxiliary decision support information generation system according to any one of claims 1-6, the method comprising: Obtain a patient information table for cardiorenal metabolic syndrome and a preset entity relationship information set, wherein the patient information table for cardiorenal metabolic syndrome includes individual patient identifiers; Based on the preset entity relationship information set, a knowledge graph of cardiorenal metabolic syndrome is created and updated; For each patient identifier, at least one medical event record information corresponding to the patient identifier is collected from various heterogeneous data sources to generate a standardized medical event record information set; For each standardized medical event record information set generated, an initial medical event spatiotemporal graph is generated based on the standardized medical event record information set; Each generated initial diagnosis and treatment event spatiotemporal graph is defined as an initial diagnosis and treatment event spatiotemporal graph set, and the initial diagnosis and treatment event spatiotemporal graph set is stored. In response to receiving medical data sent by one of the at least one terminals, a spatiotemporal graph of medical events corresponding to the medical data is generated based on the medical data and an initial spatiotemporal graph of medical events. Based on the spatiotemporal graph of diagnosis and treatment events and the updated knowledge graph of cardiorenal metabolic syndrome, auxiliary decision support information is generated. The auxiliary decision support information is sent to the terminal, and based on the auxiliary decision support information, the drug-dispensing robot associated with the terminal is controlled to dispense the drug.

Images