Medical data preloading method, computer device, medium, product

Abstract

The application discloses a medical data preloading method, a computer device, a medium and a product. The method comprises the following steps: determining the type of a patient medical record, calling a target strategy according to the type to extract a target entity from the patient medical record; constructing a multi-modal medical knowledge graph according to a multi-dimensional data source, and obtaining a plurality of association information associated with the target entity according to the graph; comprehensively considering a plurality of factors, respectively determining data urgency and weight, semantic correlation and weight, cache capacity factor and weight, then performing weighted summation to obtain a cache level score of the target entity and each association information, and performing dynamic preloading of the corresponding association information according to the score from high to low. The method can make dynamic decisions and dynamic preloading according to the importance of different data information, does not need to manually call multiple source data multiple times, effectively reduces the complexity and time cost of operation, and is simple and efficient.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to a medical data preloading method, a computer device, a medium and a product. BACKGROUND

[0002] In a traditional medical information system, common traditional medical data caching strategies mainly include least recently used (LRU) and least frequently used (LFU) algorithms, wherein the LRU strategy is based on the assumption that the data accessed recently is more likely to be accessed again in the future, and when the cache space is full, the data that has not been accessed for the longest time is preferentially evicted; the LFU strategy evicts the data with the lowest access frequency according to the historical access frequency of the data.

[0003] However, in the medical field, there is a close business association between various types of data, for example, CT images are usually associated with related test reports, and doctors often need to view these data with business contact at the same time when making a diagnosis, but the traditional LRU and LFU strategies cannot identify the business association between the data, and only rely on access time or access frequency for cache management, which leads to the need for doctors to manually retrieve multi-source data during the diagnosis process, increasing the complexity and time cost of the operation and reducing the diagnosis efficiency.

[0004] At the same time, the strategies disclosed in the prior art only rely on historical access frequency to determine the retention and eviction of cache content, without considering that medical diagnosis is a dynamic process; in actual diagnosis, real-time diagnosis results may affect the doctor's data needs; for example, after preliminary diagnosis, it is found that certain specific images or test reports need to be further viewed, but since the traditional strategy cannot dynamically adjust the cache content according to these real-time diagnosis results, the doctor may not be able to obtain the required data in time, affecting the accuracy and timeliness of the diagnosis.

[0005] In addition, medical data has the characteristics of multi-modal, including images, text, reports and other types, under the traditional caching strategy, these different modal data are often cached in isolation, lacking cross-modal semantic association; however, in actual medical diagnosis, doctors often need to analyze multiple modal data comprehensively to make accurate judgments, for example, doctors may need to combine image data and text reports to determine the condition, but due to isolated data caching, doctors have difficulty in quickly obtaining and integrating these cross-modal data, limiting the comprehensiveness and accuracy of the diagnosis. SUMMARY

[0006] Therefore, the medical data preloading method, computer device, medium and product provided by the embodiments of the present disclosure can solve the problems of the conventional cache method, such as semantic insensitivity, poor data service correlation, inability to dynamically adjust preloaded content, and poor subsequent diagnosis accuracy and poor diagnosis and treatment efficiency caused by multi-modal fragmentation, isolated cache of different types of data, and lack of cross-modal semantic correlation.

[0007] In a first aspect, the embodiments of the present disclosure provide a medical data preloading method, comprising:

[0008] determining a type of a patient medical record, calling a target strategy according to the type, and extracting a target entity from the patient medical record based on the target strategy;

[0009] constructing a multi-modal medical knowledge graph according to multi-dimensional data sources;

[0010] obtaining a plurality of correlation information associated with the target entity based on the medical knowledge graph; the plurality of correlation information includes correlation test information, correlation drug information, and correlation test information;

[0011] determining data urgency and data urgency weight according to the target entity and the correlation information;

[0012] determining semantic correlation degrees of each of the correlation information and the target entity and corresponding semantic correlation weights thereof;

[0013] determining cache capacity factors of each of the correlation information and corresponding cache capacity weights thereof;

[0014] weighting and summing the semantic correlation degrees, the data urgency, and the cache capacity factors based on the semantic correlation weights, the data urgency weight, and the cache capacity weight, to obtain cache level scores of the target entity and each of the correlation information;

[0015] performing dynamic preloading of corresponding correlation information in descending order of the cache level scores.

[0016] Optionally, the type of the patient medical record is an image type.

[0017] The calling of the target strategy according to the type and the extraction of the target entity from the patient medical record based on the target strategy comprises: calling an image processing large model to process the patient medical record and identify the target entity.

[0018] Optionally, the type of the patient medical record is an unstructured text type; and the unstructured text type is a medical record that lacks a fixed format and contains doctor's handwritten records and disease descriptions.

[0019] The target strategy is called according to the type, and a target entity is extracted from the patient medical record based on the target strategy, including:

[0020] The patient medical record is denoised, the denoised patient medical record is split into independent words or phrases, and each word or phrase is tagged with a part of speech to obtain a first text data set;

[0021] The BioBERT model is pre-trained using a large-scale medical corpus, and the first text data set is input into the trained BioBERT model to obtain the target entity.

[0022] Optionally, the type of the patient medical record is a semi-structured text type; the semi-structured text type is a medical record with a certain format and part of the content being a free description record;

[0023] The target strategy is called according to the type, and a target entity is extracted from the patient medical record based on the target strategy, including:

[0024] The patient medical record is field parsed to determine fixed fields and non-fixed fields containing diagnosis information;

[0025] The target entity is directly extracted from the fixed field;

[0026] The BioBERT model is pre-trained using a large-scale medical corpus, and the trained BioBERT model is used to identify the non-fixed field to obtain the target entity.

[0027] Optionally, the type of the patient medical record is a structured text type; the structured text type is a medical record with a fixed format and no free description record;

[0028] The target strategy is called according to the type, and a target entity is extracted from the patient medical record based on the target strategy, including: determining a target field storing diagnosis information according to the structure of the patient medical record, and extracting a target entity from the target field.

[0029] Optionally, the multi-modal medical knowledge graph is constructed based on multi-dimensional data sources, including: based on entity name-association test-association drug, a multi-modal medical knowledge graph is constructed based on multi-dimensional data sources;

[0030] The data source includes one or more of PubMed literature, NCCN clinical guidelines, hospital diagnosis and treatment rules, UMLS, and SNOMED CT, and the medical knowledge graph is a Neo4j knowledge graph.

[0031] Optionally, the data urgency acquisition method comprises:

[0032] determine a scene timeliness type according to the target entity; the scene timeliness type includes any one of a first type, a second type, and a third type, wherein scene emergency levels are sequentially ranked from large to small as the first type, the second type, and the third type;

[0033] obtain a timeliness weight of the scene timeliness type according to a preset scene timeliness weight list;

[0034] obtain a scene score of the scene timeliness type according to a preset label score list;

[0035] multiply the scene score and the timeliness weight to obtain an electronic label score;

[0036] determine an abnormal sign index according to the relevance information; wherein the sign index includes heart rate, blood oxygen, blood pressure, and respiratory rate;

[0037] determine a feature weight of all abnormal sign indexes based on a preset feature index weight list, and perform weighted summation on all abnormal sign indexes to obtain a target sign score;

[0038] determine a department emergency level score and a corresponding department weight according to the target entity and the relevance information;

[0039] multiply the department emergency level score and the department weight to obtain a department type score;

[0040] obtain an emergency degree score according to the electronic label score, the target sign score, the department type score, and corresponding electronic label preset weights, target sign preset weights, and department type preset weights;

[0041] the emergency degree score = electronic label score × electronic label preset weight + target sign score × target sign preset weight + department type score × department type preset weight;

[0042] perform normalization processing on the emergency degree score to obtain a data emergency degree.

[0043] Optionally, the determining of the semantic relevance degree of each of the relevance information and the target entity includes:

[0044] search, in the medical knowledge graph, a shortest path of the target entity to a corresponding target data entity in each of the relevance information and a hop number corresponding to the shortest path;

[0045] determine a maximum reference weight of a path in the knowledge graph;

[0046] determine a path weight of the shortest path;

[0047] The correlation degree score of each of the correlation information and the target entity is obtained based on a preset formula, and the correlation degree score=(path weight / max reference weight)×e (-0.5×跳数) ;

[0048] The correlation degree score is normalized to obtain semantic correlation degree of the correlation verification information and the target entity.

[0049] Optionally, the method for obtaining the cache capacity factor of each of the correlation information comprises:

[0050] The total storage space and the used storage space of the edge node are obtained periodically, and the storage space usage rate is calculated, and the calculation formula is: usage rate=used storage space / total storage space;

[0051] According to the calculated usage rate, the cache capacity factor is dynamically adjusted using a piecewise function:

[0052] When the usage rate is less than or equal to 70%, the cache capacity factor=1-(usage rate)2;

[0053] When the usage rate is greater than 70%, the cache capacity factor=(1-usage rate)3;

[0054] If the storage space information fails to be obtained, the cache capacity factor is set to a default value 0.5, and an error log is recorded; if the calculated cache capacity factor is negative, it is corrected to 0.

[0055] Optionally, the dynamic preloading of the corresponding correlation information according to the cache level score from high to low comprises:

[0056] The dynamic preloading of the corresponding correlation information according to the cache level score from high to low is pre-stored to the target edge node according to the first-in-first-out principle;

[0057] The management rule and the data retention rule of the corresponding correlation information are determined according to the cache level score;

[0058] The target edge node comprises a temporary queue and a protected queue, the proportion of the protected queue is 70% of the cache space, and the proportion of the temporary queue is 30% of the cache space.

[0059] Optionally, the management rule and the data retention rule of the corresponding correlation information are determined according to the cache level score, comprising:

[0060] All of the correlation information with the cache level score greater than a preset threshold value are obtained, and are recorded as first type target information;

[0061] The first type target information is stored to the protected queue in order according to the time of entering the queue.

[0062] configuring a data retention duration of the first type of target information; the data retention duration is T: T = preset basic duration + weight of corresponding target information x 12;

[0063] obtaining all the relevance information with a cache level score not greater than a preset threshold, denoted as second type of target information;

[0064] storing the second type of target information to the temporary queue in order according to the time of entering the queue;

[0065] configuring a data retention duration of the second type of target information as a preset duration.

[0066] In a second aspect, the present application discloses a medical data preloading system, comprising:

[0067] a data collection module configured to determine a type of a patient medical record, call a target strategy according to the type, and extract a target entity from the patient medical record based on the target strategy;

[0068] a graph construction module configured to construct a multi-modal medical knowledge graph according to multi-dimensional data sources;

[0069] an relevance information acquisition module configured to acquire a plurality of relevance information associated with the target entity based on the medical knowledge graph; the plurality of relevance information comprises association test information, associated drug information and associated test information;

[0070] a data urgency information acquisition module configured to determine data urgency and data urgency weight according to the target entity and the relevance information;

[0071] a semantic association information acquisition module configured to determine semantic association degree of each of the relevance information with the target entity and a corresponding semantic association weight;

[0072] a cache capacity information acquisition module configured to determine cache capacity factor of each of the relevance information and a corresponding cache capacity weight;

[0073] a calculation module configured to perform weighted summation on the semantic association degree, the data urgency and the cache capacity factor based on the semantic association weight, the data urgency weight and the cache capacity weight, to obtain cache level score of the target entity and each of the relevance information;

[0074] a preloading module configured to perform dynamic preloading of corresponding relevance information in order of descending cache level score.

[0075] In a third aspect, the present application discloses a computer device, adopting the following technical scheme:

[0076] The computer device comprises:

[0077] at least one processor; and,

[0078] a memory connected in communication with the at least one processor; wherein,

[0079] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the medical data preloading method of any one of the above.

[0080] In a fourth aspect, the embodiments of the present disclosure further provide a computer readable storage medium storing computer instructions for causing a computer to perform the medical data preloading method of any one of the above.

[0081] In a fifth aspect, the embodiments of the present disclosure further provide a computer program product comprising computer programs / instructions which, when executed by a processor, implement the steps of the method of any one of the above.

[0082] The medical data preloading method disclosed in the present application calls a target strategy according to a determined type of patient medical record, extracts a target entity from the patient medical record based on the target strategy, constructs a multi-modal medical knowledge graph based on multi-dimensional data sources, obtains a plurality of association information associated with the target entity based on the medical knowledge graph, and determines data urgency and data urgency weight according to the target entity and the association information. The association information includes association test information, association drug information and association test information. The semantic association degree of each association information and the corresponding semantic association weight of the target entity are determined. The cache capacity factor of each association information and the corresponding cache capacity weight are determined. The semantic association degree, the data urgency and the cache capacity factor are weighted and summed based on the semantic association weight, the data urgency weight and the cache capacity weight, to obtain the cache level score of the target entity and each association information. The corresponding association information is dynamically preloaded in the order from high to low according to the cache level score. The scheme can quickly obtain the required key information and reduce the time of information searching by accurately extracting the target entity, obtaining the association information and reasonably preloading. The scheme comprehensively considers the data urgency, the semantic association degree and the cache capacity and other factors, can reasonably allocate the cache resources of the system, avoid the waste of resources, and effectively improve the system operation efficiency and the data acquisition timeliness.

[0083] The above description is only a summary of the technical solutions of the present disclosure. In order to make the technical means of the present disclosure more clearly understood, the contents of the specification can be implemented, and in order to make the above and other purposes, characteristics and advantages of the present disclosure more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0084] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor.

[0085] Figure 1 The flowchart of the medical data preloading method provided by the embodiments of the present disclosure.

[0086] Figure 2 The flowchart of the target entity extraction method when the type of the patient medical record is an unstructured text type provided by the embodiments of the present disclosure.

[0087] Figure 3 The flowchart of the target entity extraction method when the type of the patient medical record is a semi-structured text type provided by the embodiments of the present disclosure.

[0088] Figure 4 The flowchart of the data urgency acquisition method provided by the embodiments of the present disclosure.

[0089] Figure 5 The flowchart of the method for determining the semantic correlation degree of each correlation information and the target entity provided by the embodiments of the present disclosure.

[0090] Figure 6 The flowchart of the method for acquiring the cache capacity factor of each correlation information provided by the embodiments of the present disclosure.

[0091] Figure 7 The flowchart of the method for dynamic preloading of the corresponding correlation information according to the sorting from high to low of the cache level score provided by the embodiments of the present disclosure.

[0092] Figure 8 The flowchart of the method for determining the management rule and the data retention rule of the corresponding correlation information according to the cache level score provided by the embodiments of the present disclosure.

[0093] Figure 9 The structural schematic diagram of a computer device provided by the embodiments of the present disclosure. DETAILED DESCRIPTION

[0094] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0095] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0096] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0097] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0098] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0099] Reference Figure 1 This application discloses a method for preloading medical data, including:

[0100] S100: Determine the type of patient medical record, invoke the target strategy according to the type, and extract the target entity from the patient medical record based on the target strategy.

[0101] Different types of medical records have different structures and key information. Invoking the target strategy according to the type can more accurately extract the target entity, improve the accuracy and efficiency of extraction, and avoid information omission or incorrect extraction caused by using a single strategy to process all medical records.

[0102] S200, constructing a multi-modal medical knowledge graph according to multi-dimensional data sources.

[0103] Specifically, based on entity name (i.e. disease name) - associated test - associated drug, a multi-modal medical knowledge graph is constructed by combining multi-dimensional data sources.

[0104] Among them, the data source includes one or more of PubMed literature, NCCN clinical guidelines, hospital diagnosis and treatment rules, UMLS, SNOMED CT; the medical knowledge graph is a Neo4j knowledge graph. In this embodiment, the information obtained from the multi-dimensional data source is integrated, with the entity name (i.e. disease name), the associated test, and the associated drug as the nodes, and the relationship (such as "cause" "treatment" and the like) between them as the edges, to construct a multi-modal medical knowledge graph.

[0105] The multi-modal medical knowledge graph integrates multi-dimensional medical information and can provide a more comprehensive and in-depth medical knowledge system, which can help doctors have a more comprehensive understanding of the disease and provide a rich knowledge base for subsequent acquisition of associated information.

[0106] S300, obtaining a plurality of associated information related to the target entity based on the medical knowledge graph; the plurality of associated information includes associated test information, associated drug information, and associated test information.

[0107] Among them, the associated test information can be test index information, the associated drug information can be historical treatment recommendation information, and the associated test information can include imaging test information.

[0108] Assuming that the target entity is "diabetes", through the medical knowledge graph, the associated test information such as blood glucose detection, glycosylated hemoglobin detection, etc. can be obtained; the associated drug information such as insulin, metformin, etc. can be obtained; the associated treatment information such as diet control, exercise therapy, etc. can be obtained. The knowledge graph will automatically associate these related information according to the relationship between the nodes, and use the medical knowledge graph to quickly and accurately obtain multi-aspect information related to the target entity, to provide more comprehensive reference for doctors and help to develop more reasonable treatment plan.

[0109] S400, determining data urgency and data urgency weight according to the target entity and the associated information.

[0110] The determination of the data urgency and the data urgency weight can prioritize and acquire urgent data, ensure that critical information can be quickly acquired in an emergency, and improve the timeliness of the close medical data preloading.

[0111] S500, determine the semantic correlation degree of each correlation information with the target entity and the corresponding semantic correlation weight.

[0112] The determination of the semantic correlation degree and the semantic correlation weight can more accurately measure the closeness of the correlation information to the target entity, and help to prioritize the information more closely associated with the target entity in subsequent preloading.

[0113] S600, determine the cache capacity factor of each correlation information and the corresponding cache capacity weight.

[0114] Considering the cache capacity factor and the cache capacity weight can reasonably allocate the cache resources of the system, and avoid affecting the system performance due to a large amount of cache space occupation.

[0115] In this embodiment, the order of S400, S500 and S600 is not limited, and all are within the protection scope of the present application.

[0116] S700, based on the semantic correlation weight, the data urgency weight and the cache capacity weight, the semantic correlation degree, the data urgency and the cache capacity factor are weighted and summed to obtain the cache level score of the target entity and each correlation information.

[0117] Wherein, the cache level score is S, , wherein, is the semantic correlation degree, is the semantic correlation weight, is the data urgency, is the data urgency weight, is the cache capacity factor, is the cache capacity weight.

[0118] S800, according to the cache level score from high to low, the dynamic preloading of the corresponding correlation information is carried out.

[0119] Specifically, the correlation information can be sorted according to the cache level score calculated in the foregoing; for information with high cache level score, such as commonly used therapeutic drug information with high score in the associated drug information, preloading is performed preferentially; when the medical system is idle, the correlation information is loaded into the cache in order according to the sorting, so that the doctor can quickly acquire when needed. Dynamic preloading can reasonably arrange the preloading order according to the priority of the information, ensure that the most important information is acquired and cached preferentially under limited system resources, improve the efficiency of the doctor acquiring information, and reduce the waiting time.

[0120] The medical data preloading method disclosed in the application calls a target strategy according to a determined type of a patient medical record, extracts a target entity from the patient medical record based on the target strategy, constructs a multi-modal medical knowledge graph based on multi-dimensional data sources, obtains a plurality of association information associated with the target entity based on the medical knowledge graph, and determines data urgency and data urgency weight according to the target entity and the association information. The plurality of association information includes association test information, association drug information and association test information. The semantic association degree of each association information and the corresponding semantic association weight of the target entity are determined. The cache capacity factor of each association information and the corresponding cache capacity weight are determined. The semantic association degree, the data urgency and the cache capacity factor are weighted and summed based on the semantic association weight, the data urgency weight and the cache capacity weight, and the cache level score of the target entity and each association information is obtained. The corresponding association information is dynamically preloaded in the order from high to low according to the cache level score. The scheme can quickly obtain the required key information and reduce the information search time by accurately extracting the target entity, obtaining the association information and reasonably preloading. The scheme comprehensively considers the data urgency, the semantic association degree and the cache capacity and the like, can reasonably allocate the cache resources of the system, avoids the waste of resources, and effectively improves the system operation efficiency and the data acquisition timeliness.

[0121] When the type of the patient medical record is an image type, the target strategy is called according to the type, and the target entity is extracted from the patient medical record based on the target strategy, including: calling an image processing large model to process the patient medical record and identifying the target entity.

[0122] When the type of the patient medical record is an unstructured text type, referring to Figure 2 , the method of “calling a target strategy according to a type and extracting a target entity from a patient medical record” in S100, that is, the extraction method of the target entity, includes:

[0123] A111, the patient medical record is subjected to denoising processing, the denoised patient medical record is split into independent words or phrases, and each word or phrase is labeled with a part of speech, to obtain a first text data set.

[0124] Specifically, the denoising processing includes: the patient medical record can contain some information irrelevant to the extraction of the target entity, such as medical record number, date, some repeated format words and the like. For example, in an outpatient medical record, the contents such as “medical record number: 123456, date: 2025-05-15” at the beginning are noise information that can be removed, and some garbled codes, special symbols and the like also need to be processed, such as removing symbols such as “@#$” that can appear in the medical record.

[0125] For word or phrase splitting, using a suitable word segmentation tool, the denoised medical record text is split into independent words or phrases. For example, "the patient has a cough, fever symptoms" will be split into "patient" "has" "cough" "," "fever" "symptoms".

[0126] For part-of-speech tagging, a part-of-speech tagging tool can be used to tag the part-of-speech of each split word or phrase. For example, "patient" is tagged as a noun, and "has" is tagged as a verb. In Chinese, a commonly used part-of-speech tagging tool can achieve this function. After tagging, the first text data set is formed.

[0127] In this step, irrelevant information can be removed to reduce interference in subsequent processing, so that the model focuses on the core content related to the target entity, improving the accuracy of extraction; splitting the text into independent units facilitates subsequent model analysis and processing of each word or phrase, which is a basic step in natural language processing; part-of-speech information can provide more semantic clues for the model, helping the model better understand the structure and meaning of the text, and thus more accurately extract the target entity.

[0128] A112, using a large-scale medical corpus to pre-train the BioBERT model, inputting the first text data set into the trained BioBERT model to obtain the target entity.

[0129] The large-scale medical corpus contains rich medical literature, case reports and other text data. For example, a large number of medical papers, clinical guidelines and other texts are collected from professional medical databases. These corpora are used to pre-train the BioBERT model, allowing the model to learn the language patterns and semantic information in the medical field. During the pre-training process, the model will continuously adjust its parameters to adapt to the characteristics of medical text. The first text data set obtained in the previous step is input into the pre-trained BioBERT model, which will analyze and process the input text, and according to the medical knowledge and language patterns it has learned, identify the target entity in it. For example, if the input text contains "the patient is diagnosed with pneumonia", the model will accurately extract "pneumonia" as the target entity.

[0130] BioBERT is a pre-trained language model designed specifically for the biomedical field, and pre-training with a large-scale medical corpus allows the model to fully learn the professional knowledge and language habits in the medical field, improving the model's performance in medical text processing. The first text data set obtained in the previous step has been denoised, segmented and tagged with part-of-speech, providing the model with more standardized and easier-to-process input, which helps the model more accurately extract the target entity.

[0131] In this embodiment, the unstructured text type is a medical record that lacks a fixed format and contains doctor's handwritten records and disease descriptions. Specifically, such medical records usually exist in the form of free text, containing doctor's handwritten records and disease descriptions, and lack fixed formats and structured information.

[0132] When the type of the patient's medical record is a semi-structured text type, referring to Figure 3 , the method of "calling the target strategy according to the type, and extracting the target entity from the patient's medical record based on the target strategy" in S100, i.e., the extraction method of the target entity, includes:

[0133] A121, field parsing is performed on the patient's medical record to determine fixed fields and non-fixed fields containing diagnosis information.

[0134] Classifying the fields in the medical record helps to use different extraction methods targetedly. The fixed fields are relatively simple and direct to extract due to their fixed position and name. The non-fixed fields need more complex processing methods, and classification can improve the extraction efficiency and accuracy.

[0135] Specifically, different types of patient medical records usually have some fixed formats and fields for recording specific information. For example, in the inpatient medical record, the fields of "discharge diagnosis" and "admission diagnosis" are fixed fields specifically used to record the patient's diagnosis information, and their position and name are relatively fixed. Through format analysis and experience summary of the medical record, these fixed fields can be identified.

[0136] In addition to the fixed fields, there may be some fields in the medical record that do not have fixed positions and names but may contain diagnosis information. For example, in the doctor's medical history record, there may be some diagnosis-related content mentioned in different paragraphs and expressions, which do not follow the fixed format and position and belong to the non-fixed field. For example, the doctor mentions "considering the possibility of upper respiratory tract infection in the patient" when recording the development of the disease, and the field where this content is located is not fixed.

[0137] A122, directly extracting the target entity from the fixed field.

[0138] The content of the fixed field has explicitness and normativity, and the operation of directly extracting the target entity is simple and fast, which can quickly obtain the diagnosis information explicitly recorded in the medical record and improve the efficiency of information extraction.

[0139] For example, for the "discharge diagnosis" field, if its content is "1. Hypertension 2. Coronary heart disease", "hypertension" and "coronary heart disease" can be directly extracted as target entities, because the content of these fixed fields is usually explicit diagnosis information, and direct extraction can obtain the key target entity.

[0140] A123, pre-train the BioBERT model using a large-scale medical corpus, and use the trained BioBERT model to identify non-fixed fields to obtain target entities.

[0141] BioBERT is a pre-trained language model designed specifically for the biomedical field. After pre-training using a large-scale medical corpus, it can fully grasp the professional knowledge and language habits of the medical field, laying the foundation for accurately identifying target entities in non-fixed fields. The content format of non-fixed fields is not standardized, and the position is not fixed. Traditional simple extraction methods are difficult to accurately identify target entities in non-fixed fields. The pre-trained BioBERT model has strong semantic understanding and recognition capabilities, which can effectively handle such complex situations and improve the accuracy of target entity extraction. For example, the content of the non-fixed field is "the patient has frequent cough and fever recently, combined with the examination results, suspected to have bronchitis". The model will identify "bronchitis" as the target entity based on its learned medical knowledge and language patterns.

[0142] In this embodiment, the semi-structured text type is a medical record with a certain format and part of the content is a free description record. There is a fixed field template, but some fields are filled in with free descriptions.

[0143] When the type of the patient's medical record is a structured text type, the target strategy is called according to the type, and the target entity is extracted from the patient's medical record based on the target strategy, including: determining the target field that stores the diagnosis information according to the structure of the patient's medical record, and directly extracting the target entity from the target field.

[0144] In this embodiment, the structured text type is a medical record with a fixed format and no free description record. Structured electronic medical records are presented in the form of tables, lists, etc., and the data has a clear field and format, and the diagnosis information is usually stored in a specific field.

[0145] Referring to Figure 4 , the data urgency acquisition method in S400 includes:

[0146] A110, determining the scene timeliness type according to the target entity; the scene timeliness type includes any one of the first type, the second type, and the third type.

[0147] Among them, the scene urgency degree is sequentially sorted from large to small as: the first type, the second type, and the third type.

[0148] By dividing the scene timeliness type by the target entity, the patient's condition can be initially classified according to the severity and urgency of the disease, providing a basis for subsequent calculation of the urgency, making the subsequent processing and evaluation more targeted.

[0149] Specifically, the first type can represent a critical type, the second type can represent an emergency type, and the third type can represent a routine type. For example, if the target entity is "acute myocardial infarction", since the disease is acute and the disease progresses rapidly, immediate treatment intervention is required, so the time effectiveness type of the scene is determined to be the first type; if the target entity is "pneumonia", the disease progresses relatively slowly and may need to be treated within a certain period of time, but it is not as urgent as acute myocardial infarction, and can be determined as the second type; if the target entity is "mild cold", the immediate threat to the patient's health is small, and the disease progresses slowly, and can be determined as the third type.

[0150] A120, obtaining a time effectiveness weight of the scene time effectiveness type according to a preset scene time effectiveness weight list.

[0151] The preset weight list assigns corresponding weights to different scene time effectiveness types, reflecting the importance of scenes of different emergency degrees in overall emergency assessment, so that subsequent calculations can more reasonably consider the scene time effectiveness factor.

[0152] Suppose that in the preset scene time effectiveness weight list, the time effectiveness weight corresponding to the first type of scene time effectiveness type is 0.6, the second type is 0.3, and the third type is 0.1. When the scene time effectiveness type is determined to be the first type in A110, the time effectiveness weight of 0.6 is obtained from the list.

[0153] A130, obtaining a scene score of the scene time effectiveness type according to a preset label score list.

[0154] The preset label score list quantifies the scores of different scene time effectiveness types, representing the emergency degree of the scene with specific numerical values, facilitating subsequent mathematical calculations and comprehensive assessments.

[0155] For example, in the preset label score list, the scene score corresponding to the first type of scene time effectiveness type is 70 points, the second type is 20 points, and the third type is 10 points. If the determined scene time effectiveness type is the second type, then the scene score of 20 points is obtained.

[0156] A140, multiplying the scene score by the time effectiveness weight to obtain an electronic label score.

[0157] By multiplying the scene score by the time effectiveness weight, the emergency degree of the scene and its importance in overall assessment are comprehensively considered, and a comprehensive electronic label score is obtained, which can more accurately reflect the influence of scene time effectiveness on data emergency degree.

[0158] A150, determining an abnormal sign index according to the relevance information; wherein the sign index includes heart rate, blood oxygen, blood pressure and respiratory rate.

[0159] The physical indicators are important basis for reflecting the physical condition of the patient. Determining the abnormal physical indicators can further assess the emergency degree of the disease from the physiological state of the patient, and provide data support for subsequent calculation of the target physical score.

[0160] Specifically, when the actual heart rate is greater than 120 bpm, the heart rate is determined as an abnormal physical indicator.

[0161] When the actual blood oxygen is less than 90%, the blood oxygen is determined as an abnormal physical indicator.

[0162] When the systolic pressure is greater than 180 mmHg, the blood pressure is determined as an abnormal physical indicator.

[0163] When the actual respiratory rate of an adult is greater than 24 times / minute or less than 12 times / minute, or the respiratory rate of a newborn is about 40-60 times / minute, a baby under 1 year old is about 30-40 times / minute, a toddler of 1-3 years old is about 25-30 times / minute, a child of 4-7 years old is about 20-25 times / minute, the respiratory rate is determined as an abnormal physical indicator.

[0164] A160, based on the preset feature indicator weight list, determines the feature weight of all abnormal physical indicators, and performs weighted summation on all abnormal physical indicators to obtain a target physical score.

[0165] Different physical indicators have different effects on the emergency degree of the patient. The preset feature indicator weight list gives each physical indicator a corresponding weight, and the target physical score is obtained by weighted summation, which can more scientifically consider the influence of each abnormal physical indicator on the data emergency degree.

[0166] For example, when the heart rate is an abnormal physical indicator, the heart rate weight is determined as 0.2; when the blood oxygen is an abnormal physical indicator, the blood oxygen weight is determined as 0.3; when the blood pressure is an abnormal physical indicator, the blood pressure weight is determined as 0.25; and when the respiratory rate is an abnormal physical indicator, the blood pressure weight is determined as 0.2.

[0167] A170, according to the target entity and the relevance information, determining a department emergency degree score and a corresponding department weight;

[0168] Multiplying the department emergency degree score by the department weight to obtain a department type score.

[0169] Different departments have different emergency treatment capabilities and different degrees of attention for different diseases. Considering the department emergency degree score and the department weight can further evaluate the emergency degree of the data from the department level, making the calculation of the emergency degree more comprehensive.

[0170] A180, obtaining an emergency degree score according to the electronic tag score, the target sign score, the department type score, and corresponding preset weights of the electronic tag, the target sign, and the department type.

[0171] The emergency degree score = the electronic tag score * the preset weight of the electronic tag + the target sign score * the preset weight of the target sign + the department type score * the preset weight of the department type.

[0172] By comprehensively considering the electronic tag score, the target sign score, and the department type score, and giving corresponding weights, the scene timeliness, the patient signs, and the department factors can be combined comprehensively and scientifically to obtain a comprehensive emergency degree score, which can more accurately reflect the emergency degree of the data.

[0173] A190, normalizing the emergency degree score to obtain a data emergency degree.

[0174] Specifically, the emergency degree score can be normalized to a value between 0 and 1, so that the emergency degrees of different patients are comparable, facilitating sorting and classification, and facilitating medical staff to quickly judge the emergency degree of the patient and take corresponding measures.

[0175] The data emergency degree acquisition method disclosed by A110-A190 comprehensively considers the scene timeliness, the patient sign index, and the department factor, and evaluates the data emergency degree from different angles, which can comprehensively and accurately reflect the emergency degree of the patient's condition; by using the preset weight list and the quantized score, the factors are scientifically weighted and calculated, avoiding the randomness of subjective judgment, so that the evaluation of the emergency degree is more objective and reasonable; the normalized emergency degree value has good comparability, facilitating quick loading and calling of related data, and improving the allocation efficiency of medical resources.

[0176] Referring to Figure 5 , the method for determining the semantic association degree of each association information with the target entity in S500 includes:

[0177] B110, searching for the shortest path of the target entity to the corresponding target data entity in each association information and the number of hops corresponding to the shortest path in the medical knowledge graph.

[0178] The shortest path can reflect the most direct and closest relationship between the target entity and the association information, and the number of hops as a quantitative indicator intuitively reflects the distance between them; the fewer the number of hops, the closer the relationship between them in the knowledge graph, which helps to more accurately evaluate the semantic association degree subsequently.

[0179] Assuming that the target entity is "coronary heart disease", there are "high blood lipids" and "chest pain" in the relevance information; in the medical knowledge graph, there may be multiple connection paths between "coronary heart disease" and "high blood lipids", and the shortest path found through searching is: coronary heart disease-atherosclerosis-high blood lipids, and the number of hops of this path is 2 (two connection relationships are passed); similarly, the shortest path of "coronary heart disease" and "chest pain" may be: coronary heart disease-myocardial ischemia-chest pain, and the number of hops is also 2.

[0180] B120, determining the maximum reference weight of the path in the knowledge graph.

[0181] The maximum reference weight is the maximum value of the weights on all possible paths in the knowledge graph, which is specifically obtained by traversing the knowledge graph, recording the weight of each path, and finding the maximum value among them.

[0182] B130, determining the path weight of the shortest path.

[0183] Specifically, the path weight is calculated according to the weights of each edge on the path through the addition rule or the multiplication rule.

[0184] The path weight comprehensively considers the relationship strength of each edge on the shortest path, and can more accurately reflect the semantic association strength between the target entity and the relevance information; different calculation rules (addition or multiplication) can be selected according to actual conditions to better adapt to the characteristics of the knowledge graph.

[0185] B140, obtaining the association degree score of each relevance information and the target entity based on a preset formula.

[0186] Wherein, the association degree score=(path weight / maximum reference weight) x e (-0.5×跳数) .

[0187] The association degree score formula comprehensively considers the path weight and the number of hops, the path weight reflects the strength of semantic association, and the number of hops reflects the closeness of semantic association, and the number of hops is processed through the exponential function, so that the influence of the number of hops on the association degree score presents exponential decay, which is more in line with the actual semantic association rule.

[0188] B150, normalizing the association degree score to obtain the semantic association degree between the relevance test information and the target entity.

[0189] Specifically, the association degree score is normalized to a value between 0 and 1, which can make the semantic association degrees between different relevance information and the target entity have a unified measurement standard, facilitating comparison and sorting, and being helpful for subsequent data analysis and decision-making.

[0190] The method for determining the semantic correlation degree of each correlation information with a target entity disclosed by B110-150 can more accurately evaluate the semantic correlation degree between the correlation information and the target entity by comprehensively considering multiple factors such as the shortest path, path weight, and hop count, and avoids the limitations of single-factor evaluation. The maximum reference weight and normalization processing enable the calculation of the semantic correlation degree to have a unified standard, so that the correlation degrees of different correlation information can be directly compared, improving the comparability and reliability of the evaluation results. This scheme can be applied to multiple fields such as medical information retrieval and auxiliary diagnosis, and accurate semantic correlation degrees help doctors quickly find important information related to diagnosis, improve diagnosis efficiency and accuracy, and also provide strong support for intelligent application of medical information systems.

[0191] With reference to Figure 6 The method for obtaining the cache capacity factor of each correlation information in S600 includes the following steps.

[0192] C110, periodically obtain the total storage space and used storage space of the edge node, and calculate the storage space usage rate.

[0193] The calculation formula is: usage rate = used storage space / total storage space; for different types of edge nodes, the total storage space and used storage space can be obtained through system calls provided by the operating system or related management tools, such as using the df command in the Linux system or the disk management tool in the Windows system.

[0194] Periodically obtaining the storage space usage of the edge node can help to real-time master the storage resource state of the edge node; the storage space usage rate is a key basis for subsequent dynamic adjustment of the cache capacity factor. Only by accurately understanding the current storage usage can the cache space be reasonably allocated to avoid excessive use or waste of storage resources.

[0195] C120, according to the calculated usage rate, dynamically adjust the cache capacity factor using a segmented function.

[0196] When the usage rate is less than or equal to 70%, a relaxed mode is adopted, and the cache capacity factor = 1-(usage rate)2.

[0197] When the usage rate is greater than 70%, a strict mode is adopted, and the cache capacity factor = (1-usage rate)3.

[0198] The segmented function is used to dynamically adjust the cache capacity factor, which can flexibly adjust the cache strategy according to the actual storage usage of the edge node. When the storage space usage is low (relaxed mode), a larger cache capacity factor can be given to allow more related information to be cached, thereby improving the access efficiency of data. When the storage space usage is high (strict mode), the cache capacity factor is reduced to limit the amount of cached information, thereby avoiding excessive occupation of storage resources and ensuring the stability of the system.

[0199] C130, if the storage space information acquisition fails, set the cache capacity factor to the default value 0.5, and record the error log.

[0200] In the case of failure to acquire storage space information, setting the default cache capacity factor can ensure the normal operation of the system and avoid the failure of the cache strategy due to the inability to acquire storage information. Recording error logs is helpful for subsequent troubleshooting and system maintenance, and administrators can discover problems in a timely manner and take appropriate measures to solve the problems according to the log information.

[0201] C140, if the calculated cache capacity factor is negative, correct it to 0.

[0202] A negative cache capacity factor has no meaning in practical applications, and correcting it to 0 can avoid unreasonable cache strategies, which ensures that the cache capacity factor is within a reasonable range, making the cache strategy more stable and reliable, and avoiding system errors or performance degradation due to abnormal calculation results.

[0203] The method for acquiring the cache capacity factor of each related information disclosed in C110-C140 can reasonably allocate cache space according to the storage usage of the edge node by dynamically adjusting the cache capacity factor, thereby avoiding waste and excessive use of storage resources and improving the utilization efficiency of storage resources. In the case of failure to acquire storage space information or abnormal calculation results, appropriate processing measures (setting a default value, correcting a negative number) are taken to ensure the normal operation of the system and improve the stability and reliability of the system. The use of a segmented function allows the cache strategy to be flexibly adjusted according to different storage usage, which adapts to the dynamic changes of the storage resources of the edge node and improves the adaptability of the system to different environments.

[0204] Referring to Figure 7 , the method for dynamic preloading of corresponding related information in S800 according to the ranking of cache level scores from high to low, includes:

[0205] D100, according to the ranking of cache level scores from high to low, dynamically preloads corresponding related information, and pre-stores it to the target edge node according to the first-in-first-out principle.

[0206] The target edge node includes a temporary queue and a protected queue, the protected queue occupies 70% of the cache space, and the temporary queue occupies 30% of the cache space.

[0207] The association information with a high cache level score is usually more important and more likely to be accessed. Preloading from high to low according to the score can preferentially store these data in the edge node, so that when a user needs to access these data, the data can be obtained from the edge node faster, reducing data transmission delay and improving data access efficiency. Dividing the cache space into a protected queue and a temporary queue and allocating them according to a certain proportion can ensure that important data have enough space for storage and provide a temporary storage place for some possibly useful data. At the same time, the first-in-first-out principle can ensure the effective use of cache space and avoid long-term occupation of cache space without updating.

[0208] D200 determines the management rules and data retention rules of the corresponding association information according to the cache level score.

[0209] The association information with different cache level scores has different importance and access frequency. Determining the management rules and data retention rules according to the score can finely manage different data and improve the efficiency and pertinence of cache management. Reasonable data retention rules can avoid unnecessary long-term occupation of cache space and save the storage resources of the edge node. At the same time, targeted management rules can reduce the workload of data maintenance and improve the overall performance of the system.

[0210] The target edge node refers to a computing node close to a data source or a user device, which is usually deployed at the network edge. It can store data cache and provide fast data access service for users, reducing data transmission delay. The edge node is like a small data warehouse that stores some frequently accessed data to improve the response speed of the system.

[0211] The temporary queue is a data storage area in the edge node cache, used for temporarily storing some data. These data may have been accessed recently, but the cache level score is low, and it is not yet determined whether they need to be retained for a long time. The data in the temporary queue is usually the easiest to be eliminated. When the edge node cache space is insufficient, data will be eliminated from the temporary queue first.

[0212] The protected queue is also a data storage area in the edge node cache, used for storing data with a high cache level score and importance. These data are usually frequently accessed or critical to the operation of the system, so they are protected to a certain extent. However, when the cache space is severely insufficient, even the data in the protected queue will be eliminated according to the cache level score, with the data with a lower score being eliminated first.

[0213] Referring to Figure 8 , the method for determining the management rule and the data retention rule of the corresponding relevance information according to the cache level score comprises:

[0214] D210, all relevance information with a cache level score greater than a preset threshold is obtained, denoted as first type target information.

[0215] Preliminary screening of the relevance information can distinguish relatively important information, because information with a cache level score greater than a preset threshold is usually more likely to be frequently accessed or has higher value data, and classifying them separately facilitates subsequent targeted management.

[0216] D220, the first type target information is stored in a protected queue in order according to the time of entering the queue.

[0217] Storing the first type target information in the protected queue in order of entry ensures that important information has a stable storage location and follows the first-in, first-out principle, making cache management more orderly and avoiding information storage chaos.

[0218] D230, the data retention duration of the first type target information is configured.

[0219] The data retention duration T is T = preset basic duration + weight of corresponding target information x 12.

[0220] Preferably, the preset basic duration is 24 hours.

[0221] This step considers the weight factor of information to determine the data retention duration. Higher weight data is usually more important, so it is given a longer retention time. This way, the retention duration can be dynamically adjusted according to the importance of information, ensuring that important information is retained in the cache for a longer period of time, improving data availability.

[0222] D240, all relevance information with a cache level score not greater than a preset threshold is obtained, denoted as second type target information.

[0223] Corresponding to the D210 step, further distinguishes relatively less important information. By dividing the relevance information into two categories, different management strategies can be adopted for information of different importance levels, improving the efficiency of cache management.

[0224] D250, the second type target information is stored in a temporary queue in order according to the time of entering the queue.

[0225] Storing the second type target information in the temporary queue in order of entry facilitates the unified management of this information, and when the cache space is tight, data can be removed from the temporary queue first, ensuring the rational use of cache space.

[0226] D260, configuring the data retention duration of the second type of target information as a preset duration.

[0227] wherein the preset duration is t, .

[0228] Further, it can also be set as a fixed duration, such as 2 hours, and automatically deleted after the duration.

[0229] Further, when the edge node cache space is insufficient, data is first eliminated from the temporary queue, and if it is still insufficient, data with a lower cache level score is eliminated from the protected queue.

[0230] D210-D260 discloses a method for determining the management rules of the corresponding relevance information according to the cache level score and the data retention rules. By dividing the relevance information into two categories and adopting different storage and retention strategies for different categories of information, the fine management of cache data is realized, and the cache resources can be reasonably allocated according to the importance of the information; the first type of target information is configured with a data retention duration related to the weight, which ensures that important information is retained in the cache for a long enough time, improves the availability of these information, and meets the frequent access demand of users for important data; the second type of target information is stored in the temporary queue and set with a shorter retention duration, which can timely clean up the data that is no longer needed, avoid the cache space being occupied by a large amount of low-value data, and optimize the use efficiency of the cache space.

[0231] The medical data preloading method disclosed in the present application optimizes and improves the traditional medical data caching strategy from multiple aspects. A multi-modal medical knowledge graph is constructed according to multi-dimensional data sources. This graph integrates the relationships between various medical data. In the medical field, data with business connections such as CT images and associated test reports will explicitly show their association in the knowledge graph. For example, the graph will show the corresponding relationship between the CT image of a patient and the corresponding blood routine test report, biochemical test report, etc. Based on the constructed medical knowledge graph, multiple relevance information associated with the target entity can be obtained, including associated test information, associated drug information, etc., which means that the system can automatically identify data with business association and consider them as a whole. When a doctor extracts a target entity (such as "pneumonia"), the system can simultaneously obtain the CT image, blood routine test report, and drug information commonly used for treating pneumonia, etc. associated with it, avoiding manual retrieval of multi-source data by the doctor, reducing the complexity and time cost of operation, and improving the diagnosis efficiency.

[0232] The application determines data urgency and data urgency weight according to target entities and relevance information. In actual diagnosis, real-time diagnosis results affect the urgency of data. For example, after preliminary diagnosis, it is suspected that the patient has acute myocardial infarction. At this time, the urgency of electrocardiogram examination results, myocardial enzyme spectrum test reports and other data will be greatly improved, and a higher urgency weight is correspondingly given. The cache level score is obtained by weighted summation of the semantic correlation degree, data urgency and cache capacity factor, and the dynamic preloading of the corresponding relevance information is performed in descending order of the score. When the real-time diagnosis result changes, the urgency of the data and the cache level score will also change accordingly. The system will preferentially pre-load the relevance information with a high score. In this way, the doctor can obtain the data most relevant to the current diagnosis in a timely manner, avoiding the problem of being unable to obtain the required data in a timely manner due to the inability of the traditional strategy to dynamically adjust the cache content, and improving the accuracy and timeliness of diagnosis.

[0233] The multi-modal medical knowledge graph in the application integrates various types of medical data such as images, texts, reports and the like, breaks the isolated state between different modal data, and can clearly show the semantic correlation between different modal data, such as the correlation between the lesion characteristics shown in the image data and the description of the disease in the text report and the diagnosis result. When determining the cache level score, the semantic correlation degree and the corresponding semantic correlation weight of the relevance information and the target entity are considered, which enables the data of different modalities to be sorted and loaded according to their semantic correlation degree with the target entity during the preloading process. For example, when the target entity is a brain tumor, the system will comprehensively evaluate the data of different modalities such as brain CT images, pathological reports and relevant expert diagnosis suggestions according to the semantic correlation degree, and then pre-load them according to the cache level score, so as to facilitate the doctor to quickly obtain and integrate these cross-modal data.

[0234] Through the medical data preloading method disclosed in the application, the traditional passive response type retrieval according to human experience is converted into diagnosis-driven preloading, the traditional multi-department cooperation to transfer paper medical records and repeated examination is converted into digital consultation package and real-time data synchronization, and the traditional average interruption of n times per case to call data of different departments is converted into seamless workflow data calling. The delay of the doctor or operator in calling related data is reduced from 320 ms to 45 ms, and manual operation is reduced by 87.5%.

[0235] In a second aspect, the application discloses a medical data preloading system for executing the medical data preloading method disclosed in the first aspect of the application, comprising:

[0236] The data acquisition module is configured to determine the type of the patient's medical record, call a target strategy according to the type, and extract a target entity from the patient's medical record based on the target strategy.

[0237] a graph construction module, configured to construct a multi-modal medical knowledge graph according to multi-dimensional data sources;

[0238] an association information acquisition module, configured to acquire a plurality of association information associated with the target entity based on the medical knowledge graph; the plurality of association information comprises association examination information, association drug information and association examination information;

[0239] a data urgency information acquisition module, configured to determine data urgency and data urgency weight according to the target entity and the association information;

[0240] a semantic association information acquisition module, configured to determine semantic association degrees of each association information with the target entity and corresponding semantic association weights thereof;

[0241] a cache capacity information acquisition module, configured to determine cache capacity factors of each association information and corresponding cache capacity weights thereof;

[0242] a calculation module, configured to perform weighted summation on the semantic association degrees, the data urgency and the cache capacity factors based on the semantic association weights, the data urgency weight and the cache capacity weight, to obtain cache level scores of the target entity and each association information;

[0243] a preloading module, configured to perform dynamic preloading on the corresponding association information in descending order of the cache level scores.

[0244] The computer device according to the embodiments of the present disclosure comprises a memory and a processor. The memory is configured to store non-transitory computer readable instructions. Specifically, the memory can comprise one or more computer program products, which can comprise various forms of computer readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, comprise random access memory (RAM) and / or cache memory, etc. The non-volatile memory may, for example, comprise read-only memory (ROM), hard disk, flash memory, etc.

[0245] The processor can be a central processing unit (CPU) or other forms of processing units with data processing capability and / or instruction execution capability, and can control other components in the computer device to perform desired functions. In an embodiment of the present disclosure, the processor is configured to run the computer readable instructions stored in the memory, so that the computer device performs all or part of the steps of the medical data preloading method of the embodiments of the present disclosure.

[0246] Those skilled in the art shall understand that, in order to solve the technical problem of how to obtain a good user experience effect, the embodiment can also include well-known structures such as a communication bus, an interface, and the like, which shall also be included in the protection scope of the present disclosure.

[0247] As Figure 9 A structural schematic diagram of a computer device provided by the embodiment of the present disclosure is shown. The structural schematic diagram is suitable for implementing the computer device in the embodiment of the present disclosure. Figure 9 The computer device shown is only an example and shall not bring any limitation to the functions and use range of the embodiment of the present disclosure.

[0248] As Figure 9 As shown, the computer device can include a processor (for example, a central processing unit, a graphics processing unit, and the like), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer device are also stored. The processor, the ROM, and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0249] Generally, the following devices can be connected to the I / O interface: an input device including, for example, a sensor or a visual information collection device; an output device including, for example, a display screen; a storage device including, for example, a magnetic tape, a hard disk, and the like; and a communication device. The communication device can allow the computer device to communicate with other devices (such as an edge computing device) to exchange data wirelessly or by wire. Although Figure 9 The computer device with various devices is shown, but it shall be understood that it is not required to implement or have all the devices shown. More or fewer devices can be alternatively implemented or provided.

[0250] In particular, according to the embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiment of the present disclosure includes a computer program product including a computer program carried on a non-transitory computer readable medium, the computer program containing program codes for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication device, or installed from the storage device, or installed from the ROM. When the computer program is executed by the processor, all or part of the steps of the medical data preloading method of the embodiment of the present disclosure are performed.

[0251] The detailed description of the embodiment can refer to the corresponding description in the foregoing embodiments, which will not be described here again.

[0252] A computer readable storage medium according to embodiments of the present disclosure has non-transitory computer readable instructions stored thereon. When the non-transitory computer readable instructions are run by a processor, all or part of the steps of the medical data preloading method of the embodiments of the present disclosure described above are performed.

[0253] The computer readable storage medium described above includes, but is not limited to, an optical storage medium (for example, a CD-ROM and a DVD), a magneto-optical storage medium (for example, an MO), a magnetic storage medium (for example, a magnetic tape or a mobile hard disk), a medium with a built-in rewritable nonvolatile memory (for example, a memory card), and a medium with a built-in ROM (for example, a ROM cartridge).

[0254] Detailed descriptions of the embodiments are described above with reference to the corresponding descriptions of the previous embodiments, and will not be repeated here.

[0255] The above describes the basic principles of the present disclosure in combination with specific embodiments, but it should be noted that the advantages, advantages, effects and the like mentioned in the present disclosure are only examples and are not limiting, and these advantages, advantages, effects and the like cannot be considered as the must-have of each embodiment of the present disclosure. In addition, the specific details of the above disclosure are only for the purpose of example and for the purpose of understanding, and are not limited to the above specific details, and the above specific details do not limit the present disclosure to be implemented with the above specific details.

[0256] In the present disclosure, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. The block diagrams of devices, apparatuses, equipment, systems involved in the present disclosure are only illustrative examples and are not intended to require or imply the connection, arrangement, configuration shown in the block diagram. As those skilled in the art will recognize, these devices, apparatuses, equipment, systems can be connected, arranged, configured in any manner. Words such as "include", "contain", "have" and the like are open-ended words, which mean "including but not limited to", and can be used interchangeably. The words "or" and "and" used herein mean the word "and / or", and can be used interchangeably unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably.

[0257] In addition, as used herein, "or" used in the list of items preceded by "at least one of" means a disjunctive list, such that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "example" does not mean that the described example is preferred or better than other examples.

[0258] It is also important to note that the systems and methods of the present disclosure can be embodied in a variety of contexts. For example, the systems and methods of the present disclosure can be implemented in the context of a computer system, a mobile device, a server, a network, a distributed computing environment, etc. It is also important to note that the various components and steps of the systems and methods of the present disclosure can be decomposed and / or recombined. Such decompositions and / or recombinations should be considered equivalents of the present disclosure.

[0259] Various changes, modifications, and alterations to the techniques described herein can be made without departing from the teachings of the technology defined by the appended claims. In addition, the scope of the claims of the present disclosure is not limited to the specific aspects described above. Rather, the scope of the claims of the present disclosure is intended to encompass all techniques, apparatuses, manufactures, compositions of matter, means, methods, and actions that would fall within the scope of a person of ordinary skill in the art, having the benefit of the benefit of this disclosure, and the equivalent thereof.

[0260] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other aspects without departing from the scope of the disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0261] The above description has been presented for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although various example aspects and embodiments have been discussed above, those of skill in the art will recognize certain modifications, permutations, alterations, additions and sub-combinations thereof.

Claims

1. A method of medical data preloading, characterized by, The method comprises the following steps: determining the type of the patient's medical record, calling a target strategy according to the type, and extracting target entities from the patient's medical record based on the target strategy; constructing a multi-modal medical knowledge graph according to multi-dimensional data sources; obtaining a plurality of association information associated with the target entity based on the medical knowledge graph; The plurality of association information includes association test index information, association drug information and association test information; determining data urgency and data urgency weight according to the target entity and the association information; determining the semantic association degree of each association information with the target entity and the corresponding semantic association weight; determining the cache capacity factor of each association information and the corresponding cache capacity weight; Based on the semantic association weight, the data urgency weight and the cache capacity weight, the semantic association degree, the data urgency and the cache capacity factor are weighted and summed to obtain the cache level score of the target entity and each association information; According to the cache level score from high to low, the corresponding association information is dynamically preloaded; The method for obtaining the data urgency comprises the following steps: determining the scene timeliness type according to the target entity; obtaining the timeliness weight of the scene timeliness type according to the preset scene timeliness weight list; obtaining the scene score of the scene timeliness type according to the preset label score list; multiplying the scene score and the timeliness weight to obtain the electronic label score; determining the abnormal sign index according to the association information; determining the feature weight of all abnormal sign indexes based on the preset feature index weight list, and weighting and summing all abnormal sign indexes to obtain the target sign score; determining the department emergency degree score and the corresponding department weight according to the target entity and the association information; multiplying the department emergency degree score and the department weight to obtain the department type score; obtaining the urgency degree score according to the electronic label score, the target sign score, the department type score and the corresponding electronic label preset weight, target sign preset weight and department type preset weight; normalizing the urgency degree score to obtain the data urgency; The determining of the semantic correlation degree of each of the correlation information and the target entity comprises: searching, in the medical knowledge graph, a shortest path of the target entity to a corresponding target data entity in each of the correlation information and a hop number corresponding to the shortest path; determining a maximum reference weight of a path in the knowledge graph; determining a path weight of the shortest path; obtaining a correlation degree score of each of the correlation information and the target entity based on a preset formula; correlation degree score=(path weight / maximum reference weight)×e (-0.5×跳数) ; and performing normalization processing on the correlation degree score to obtain a semantic correlation degree of the correlation verification information and the target entity. The method for obtaining the cache capacity factor of each association information comprises the following steps: periodically obtaining the total storage space and the used storage space of the edge node, calculating the storage space usage rate, and the calculation formula is: usage rate=used storage space / total storage space; according to the calculated usage rate, the cache capacity factor is dynamically adjusted using a segmented function: when the usage rate is less than or equal to 70%, the cache capacity factor=1-(usage rate^2); when the usage rate is greater than 70%, the cache capacity factor=(1-usage rate)^3.

2. The medical data preloading method of claim 1, wherein, The type of the patient's medical record is an image type; The target strategy is called according to the type, and the target entity is extracted from the patient's medical record based on the target strategy, which comprises calling an image processing large model to process the patient's medical record and identifying the target entity.

3. The medical data preloading method of claim 1, wherein, The type of the patient medical record is an unstructured text type; the unstructured text type is a medical record lacking a fixed format and containing doctor's handwritten records and disease descriptions; The target strategy is called according to the type, and target entities are extracted from the patient medical record based on the target strategy, including: The patient medical record is subjected to denoising processing, the patient medical record after denoising processing is split into independent words or phrases, and each word or phrase is labeled with a part of speech to obtain a first text data set; The BioBERT model is pre-trained using a large-scale medical corpus, and the first text data set is input into the trained BioBERT model to obtain target entities.

4. The medical data preloading method of claim 1, wherein, The type of the patient medical record is a semi-structured text type; the semi-structured text type is a medical record having a certain format and part of the content being a free description record; The target strategy is called according to the type, and target entities are extracted from the patient medical record based on the target strategy, including: The patient medical record is subjected to field analysis to determine fixed fields and non-fixed fields containing diagnosis information; Target entities are directly extracted from the fixed fields; The BioBERT model is pre-trained using a large-scale medical corpus, and the trained BioBERT model is used to identify the non-fixed fields to obtain target entities.

5. The medical data preloading method of claim 1, wherein, The type of the patient medical record is a structured text type; the structured text type is a medical record having a fixed format and no free description record; The target strategy is called according to the type, and target entities are extracted from the patient medical record based on the target strategy, including determining a target field storing diagnosis information according to the structure of the patient medical record, and extracting target entities from the target field.

6. The medical data preloading method of claim 1, wherein, The multi-modal medical knowledge graph is constructed based on entity name-association test-association drug, and combined with multi-dimensional data sources; The data sources include one or more of PubMed literature, NCCN clinical guidelines, hospital diagnosis and treatment rules, UMLS, and SNOMED CT, and the medical knowledge graph is a Neo4j knowledge graph.

7. The medical data preloading method of claim 1, wherein, The scene timeliness type includes any one of a first type, a second type, and a third type, wherein the scene emergency degree is sequentially sorted from large to small as the first type, the second type, and the third type; The vital sign indicators include heart rate, blood oxygen, blood pressure, and respiratory rate; The emergency degree score = the electronic tag score × electronic tag preset weight + the target vital sign score × target vital sign preset weight + the department type score × department type preset weight.

8. The medical data preloading method of claim 1, wherein, If the storage space information fails to be acquired, the cache capacity factor is set to a default value of 0.5, and an error log is recorded; if the calculated cache capacity factor is negative, it is corrected to 0.

9. The medical data preloading method of claim 1, wherein, The dynamic preloading of the corresponding relevance information according to the cache level score from high to low, including: The dynamic preloading of the corresponding relevance information according to the cache level score from high to low is pre-stored in the target edge node according to the first-in-first-out principle. According to the cache level score, management rules and data retention rules of corresponding relevance information are determined; The target edge node includes a temporary queue and a protected queue, the protected queue occupies 70% of the cache space, and the temporary queue occupies 30% of the cache space.

10. The medical data preloading method of claim 9, wherein, According to the cache level score, management rules and data retention rules of corresponding relevance information are determined, including: All relevance information with the cache level score greater than a preset threshold is obtained, denoted as first type target information; The first type target information is stored in the protected queue according to the time of entering the queue in sequence; The data retention time length of the first type target information is configured, and the data retention time length is T: T = preset basic time length + weight of corresponding target information × 12; All relevance information with the cache level score not greater than the preset threshold is obtained, denoted as second type target information; The second type target information is stored in the temporary queue according to the time of entering the queue in sequence; The data retention time length of the second type target information is configured as a preset time length.

11. A computer apparatus, comprising: The computer device includes: At least one processor; and The memory is in communication connection with the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the medical data preloading method of any one of claims 1-10.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing a computer to execute the medical data preloading method of any one of claims 1-10.

13. A computer program product comprising computer instructions, characterized in that, The computer instructions are executed by the processor to implement the steps of the method of any one of claims 1-10.

Images