Patient personalized nursing management system

Abstract

The invention provides a patient personalized nursing management system, and relates to the field of nursing management, which comprises the following steps: carrying out structured embedded coding on preprocessed patient nursing information and multiple pieces of candidate nursing path information to generate respective semantic embedded coding representations; a semantic feature search network based on nursing path matching is adopted to perform feature search matching on patient nursing information embedded semantics and each candidate nursing path information embedded semantics so as to explore the matching degree between each nursing path and the patient. Therefore, simple text or rule matching can be exceeded, and the multi-dimensional information of the patient and the structured requirement of the nursing path are placed in the same high-dimensional semantic space to be compared, so that the nursing path query response semantics capable of more accurately reflecting the matching relationship can be output; and a solid foundation is laid for calculating the adaptation degree and selecting the optimal personalized nursing path.

Description

Technical Field

[0001] The present application relates to the field of nursing management, and more particularly, in an embodiment of the present application, to a patient personalized nursing management system. Background Art

[0002] With the development of modern medicine and people's increasing demand for health, providing high-quality, personalized medical care services has become an important goal pursued in the medical field. Patients' conditions are often complex and diverse. Even the same disease may show different characteristics in different individuals, affected by multiple factors such as age, gender, underlying diseases, complications, living habits, differences in test results, and medication history. Traditional, standardized nursing models are often difficult to fully meet the unique needs of each patient, which may lead to poor nursing effects, waste of resources, and even delayed disease. Therefore, building a personalized nursing management plan that can be formulated according to the specific circumstances of each patient is of great significance to improving nursing quality, optimizing treatment effects, and improving patient satisfaction and the efficiency of medical resource utilization.

[0003] At present, the methods that try to realize personalized patient care in practice include relying on the clinical experience of senior nurses for judgment, or using some auxiliary systems based on simple rules (such as based on the main diagnosis or specific keywords) to recommend nursing pathways. However, these existing methods have obvious limitations. For example, manual judgment is highly dependent on the experience level and knowledge reserve of nursing staff, and it is difficult to ensure standardization and consistency. In addition, when faced with massive and complex patient information, processing efficiency and accuracy are challenged. Although systems based on simple rules or keyword matching can provide certain automation support, they can often only perform relatively superficial information matching, and it is difficult to deeply understand the complex internal connections between various patient data indicators (such as diagnostic details, subtle changes in test results, the mutual influence of multiple complications, the complexity of medication records, etc.) and the overall semantics of nursing needs. This matching method is prone to overlooking key subtle differences and cannot accurately capture the full picture of the patient's status, which may recommend non-optimal nursing pathways and affect the accuracy of personalized care.

[0004] Therefore, an optimized patient personalized care management system is desired. Summary of the invention

[0005] To solve the above technical problems, the present application is proposed. An embodiment of the present application provides a patient personalized care management system, which, after performing structured embedding encoding on the preprocessed patient care information and multiple candidate care path information to generate their respective semantic embedding encoding representations, uses a semantic feature search network based on care path matching to perform feature search and matching on the semantic embedding of the patient care information and the semantic embedding of each candidate care path information to explore the matching degree between each care path and the patient. In this way, it can go beyond simple text or rule matching, place the multi-dimensional information of the patient and the structured requirements of the care path in the same high-dimensional semantic space for comparison, so as to output a care path query response semantics that can more accurately reflect the matching relationship, laying a solid foundation for calculating the fitness and selecting the optimal personalized care path.

[0006] According to one aspect of the present application, there is provided a patient personalized care management system, which includes: A care information data collection module for obtaining care information data of a target patient object, where the care information data includes basic data, diagnosis data, test results, complications, and medication records; A care information data preprocessing module for performing data cleaning and format conversion on the care information data to obtain preprocessed care information data; A care path coarse-grained matching module for matching multiple care path information from a care path information library based on the preprocessed care information data; A care path fine-grained semantic matching module for performing care path search and matching processing based on structured embedding semantics on the preprocessed care information data and the multiple care path information to obtain multiple care path query response semantic coding representations; A care plan generation module for performing care path matching degree analysis based on the multiple care path query response semantic coding representations to determine the fitness of multiple care paths and generate a care plan.

[0007] Compared with the prior art, the patient personalized care management system provided by the present application, after performing structured embedding encoding on the preprocessed patient care information and multiple candidate care path information to generate their respective semantic embedding encoding representations, uses a semantic feature search network based on care path matching to perform feature search and matching on the semantic embedding of the patient care information and the semantic embedding of each candidate care path information to explore the matching degree between each care path and the patient. In this way, it can go beyond simple text or rule matching, place the multi-dimensional information of the patient and the structured requirements of the care path in the same high-dimensional semantic space for comparison, so as to output a care path query response semantics that can more accurately reflect the matching relationship, laying a solid foundation for calculating the fitness and selecting the optimal personalized care path. Brief Description of the Drawings

[0008] The above and other objects, features, and advantages of the present application will become more apparent by describing the embodiments of the present application in more detail with reference to the accompanying drawings. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 It is a system block diagram of a patient personalized care management system according to an embodiment of the present application.

[0010] Figure 2 It is a schematic diagram of data flow of a patient personalized care management system according to an embodiment of the present application.

[0011] Figure 3 It is a block diagram of a nursing path fine-grained semantic matching module in a patient personalized care management system according to an embodiment of the present application.

[0012] Figure 4 It is a block diagram of a nursing path query and matching unit in a patient personalized care management system according to an embodiment of the present application.

[0013] Figure 5 It is a block diagram of a nursing information semantic weight calculation sub-unit in a patient personalized care management system according to an embodiment of the present application. Detailed Description of the Embodiments

[0014] The following will detail various exemplary embodiments, features, and aspects of the present application with reference to the accompanying drawings. The same reference numerals in the drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0015] The special term "exemplary" here means "serving as an example, embodiment, or illustrative". Any embodiment described as "exemplary" here does not have to be construed as superior to or better than other embodiments.

[0016] In addition, to better illustrate the present application, numerous specific details are given in the following detailed description. Those skilled in the art should understand that the present application can be implemented without some of these specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0017] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, "a plurality of" means two or more unless otherwise specifically defined.

[0018] Currently, methods for attempting to achieve personalized patient care in practice include relying on the clinical experience of senior nurses for judgment or using some auxiliary systems based on simple rules (such as based on the primary diagnosis or specific keywords) to recommend care paths. However, these existing methods have obvious limitations, easily overlook key nuances, and cannot accurately capture the full picture of the patient's condition, thus possibly recommending non-optimal care paths and affecting the accuracy of personalized care.

[0019] To overcome the deficiencies of the above existing methods and achieve a deeper and more accurate matching between patients and care paths for the screening of care paths, in the technical solution of this application, a patient personalized care management system is proposed. Figure 1 FIG. is a system block diagram of a patient personalized care management system according to an embodiment of this application. Figure 2 FIG. is a schematic diagram of data flow of a patient personalized care management system according to an embodiment of this application. As Figure 1 and Figure 2 shown, the patient personalized care management system 100 according to an embodiment of this application includes: a nursing information data acquisition module 110 for obtaining nursing information data of a target patient object, where the nursing information data includes basic data, diagnostic data, test results, complications, and medication records; a nursing information data preprocessing module 120 for performing data cleaning and format conversion on the nursing information data to obtain preprocessed nursing information data; a coarse-grained care path matching module 130 for matching multiple care path information from a care path information library based on the preprocessed nursing information data; a fine-grained semantic matching module 140 for performing a care path search and matching process based on structured embedded semantics on the preprocessed nursing information data and the multiple care path information to obtain multiple care path query response semantic coding representations; and a care plan generation module 150 for performing a care path matching degree analysis based on the multiple care path query response semantic coding representations to determine the adaptability of multiple care paths and generate a care plan.

[0020] It should be understood that in the process of patient care path management, relying solely on preliminary rule matching to screen candidate care paths is not enough. The key lies in how to accurately find the most suitable care path that matches the complex situation of the current patient from these candidate paths. This requires a technology that can understand and compare the deep semantic relationships between the comprehensive patient information and the detailed content of the care path. Therefore, in the technical solution of this application, after performing structured embedding encoding on the preprocessed patient care information and multiple candidate care path information to generate their respective semantic embedding encoding representations, further, a semantic feature search network based on care path matching is used to perform feature search and matching on the patient care information embedded semantics and each candidate care path information embedded semantics to explore the matching degree between each care path and the patient. In this way, it is possible to go beyond simple text or rule matching and compare the multi-dimensional information of the patient and the structured requirements of the care path in the same high-dimensional semantic space. This semantic feature search network can capture and compare the complex and non-linear feature associations between the two, deeply mine the semantic similarity and matching degree between the patient's state and the care path requirements, and thus output a care path query response semantics that can more accurately reflect the matching relationship, laying a solid foundation for subsequent accurate calculation of the adaptation degree and selection of the optimal personalized care path.

[0021] In the above patient personalized care management system 100, the care information data collection module 110 is used to obtain the care information data of the target patient object, and the care information data includes basic data, diagnosis data, test results, complications, and medication records. It should be understood that since the patient's condition and needs are complex and highly individualized, they are comprehensively affected by multiple factors such as basic conditions such as age and gender, specific disease diagnoses, test results reflecting physiological states, possible complications, and past and current medication situations. Relying solely on a single dimension or partial information cannot comprehensively and accurately depict the patient's health status and potential care needs, easily leading to a disconnection between the care plan and the actual situation of the patient and failing to achieve the expected effect of personalized care. Therefore, systematically collecting multi-dimensional information covering basic data, diagnosis, tests, complications, and medication records is a necessary prerequisite for constructing a complete portrait of the patient, understanding their unique care needs, and providing sufficient basis for subsequent accurate matching of care paths.

[0022] In particular, in a specific example of the present application, data requests can be initiated to these source systems using standardized data interfaces (such as HL7, FHIR or specific API interfaces) under the premise of complying with data security and privacy protection regulations, or patient information pushed by the source system can be received. For example, basic data and diagnostic information are mainly obtained from the core database of HIS or EMR; test results are obtained by docking with the LIS system; complication records may come from diagnostic records or specific problem lists in the EMR; medication records are retrieved from the doctor's order system or drug management system. The acquired data is usually structured, such as fields stored in database tables, but may also contain semi-structured or unstructured text information that needs further processing (for example, partial diagnosis descriptions or medical records). The entire process aims to collect core data that is strongly related to patient care and is scattered in different systems in an automated and standardized manner to form a preliminary and comprehensive patient information data set, laying a solid data foundation for subsequent data cleaning, format conversion, and intelligent matching of personalized care pathways.

[0023] In the above-mentioned patient personalized nursing management system 100, the nursing information data preprocessing module 120 is used to perform data cleaning and format conversion on the nursing information data to obtain preprocessed nursing information data. It should be understood that since the originally acquired nursing information data, including basic data, diagnostic data, test results, complications and medication records, etc., often come from different medical information systems or recording methods, their formats may not be unified and the data quality is also uneven. Patient information itself has complex and diverse characteristics and is easily affected by multiple factors, which means that the original data may have missing values, abnormal values, duplicate records, inconsistent units, and non-standardized terms (such as differences in diagnostic descriptions and drug names). The "noise" and non-standardization of these original data will seriously interfere with subsequent analysis and processing, and cannot be directly used for accurate matching and modeling. Therefore, the nursing information data is further cleaned and format converted to obtain preprocessed nursing information data. The purpose of performing data cleaning and format conversion is to solve the above-mentioned data quality problems and convert the original, heterogeneous, and possibly erroneous nursing information data into a unified, standardized, and clean data set, namely "preprocessed nursing information data". This includes removing duplicate or invalid records, filling or processing missing values, unifying data units and codes (such as disease codes, drug codes), standardizing terminology, converting data types (such as digitizing text-based test results), etc. Its core goal is to ensure that the data input into subsequent analysis links (such as preliminary pathway matching, semantic embedding coding) is high-quality, consistent and computable, laying the foundation for understanding and comparing the deep semantic relationship between comprehensive patient information and detailed content of the care pathway.

[0024] Specifically, in an embodiment of the present application, data cleaning can detect and correct (or remove) outliers through statistical methods or setting thresholds; use imputation techniques (such as mean, median filling, or more complex model prediction) to handle missing key information; identify and delete duplicate patient records or data entries; standardize the mapping of diagnosis names, test item names, and drug names according to medical terminology standard libraries (such as ICD coding, SNOMED CT, LOINC, RxNorm, etc.) to unify the term expression; verify and unify the measurement units of data from different sources. Immediately following is format conversion, whose purpose is to convert the cleaned data into a unified structured format that can be accepted and processed by downstream algorithms or models. This can be achieved by converting text-based diagnosis descriptions into standard disease codes, unifying various formats of dates and times into a standard format, normalizing or standardizing numerical data such as test results to eliminate the influence of dimensions, and encoding categorical information (such as gender, presence or absence of specific complications) (such as one-hot encoding or label encoding), etc. Through this series of rigorous data cleaning and format conversion operations, the final result is a high-quality, highly consistent, and neatly structured "preprocessed nursing information data", which provides a solid and reliable data foundation for subsequent accurate nursing path matching and personalized plan generation.

[0025] In the above patient personalized care management system 100, the care path coarse-grained matching module 130 is used to match multiple care path information from the care path information library based on the preprocessed care information data. It should be understood that the care path information library usually contains a vast amount of standardized or semi-standardized care processes for different diseases, stages, and patient groups. If all paths are directly incorporated into the subsequent complex deep semantic analysis, it will face a huge computational burden and low processing efficiency. Therefore, a preliminary screening must be carried out first. Based on the patient's current core condition, a part of the relatively highly relevant candidate paths can be quickly located from the vast library. That is to say, on the one hand, due to the complexity and diversity of patient information, and the care path information library may contain a large number of standardized or semi-standardized paths for different diseases, different stages, and different conditions, directly performing the subsequent complex semantic analysis has a huge computational amount and low efficiency; on the other hand, although simply relying on simple rules or keyword matching has the defects of "surface information matching" and "difficult to deeply understand", as a rapid screening method, it can initially identify the care paths related to the patient's core situation (such as keywords like main diagnosis, key complications, etc.) based on the relatively standardized data after preprocessing (which solves the noise problem of the original data). Therefore, further based on the preprocessed care information data, multiple care path information is matched from the care path information library. In particular, it is worth mentioning that the purpose of executing this step is not to directly select the final path, but to perform a preliminary and coarse-grained screening. Its core goal is to quickly filter out the obviously irrelevant care paths from the vast care path information library based on the key features (captured by rules or keyword matching) in the preprocessed care information, and narrow down the search scope to a candidate path set containing multiple potentially suitable options (i.e., "multiple care path information"). This provides optimized input for the subsequent "deeper and more accurate" matching, improving the processing efficiency and pertinence of the entire system.

[0026] Specifically, in a specific embodiment of the present application, first, the system extracts key features from the preprocessed patient care information data. These features usually include but are not limited to the standardized coding of the main diagnosis (such as ICD coding), clearly recorded key complication labels, patient age group, gender, and sometimes may also include certain clinically significant test result ranges (such as blood glucose level, renal function indicators, etc.) or specific core medication information. These extracted features constitute a summary description of the patient's current state and needs.

[0027] Next, the system uses these extracted key features as query conditions to perform matching retrieval in the nursing path information database. Each path in the nursing path information database usually also comes with metadata or tags describing its scope of application, such as applicable disease codes, target population characteristics (age, specific conditions), nursing stages, etc. Common matching techniques include keyword-based text retrieval, that is, checking whether the path description contains the patient's key diagnosis or complication terms; or rule-based logical matching, such as setting "if the patient's diagnosis is X and the age is greater than Y, then match the path marked as Z"; and more precise matching based on standardized coding, directly comparing whether the codes in the patient information are the same as or belong to a specific classification associated with the path.

[0028] Subsequently, to ensure the quality of the screening results and retain a certain degree of diversity for subsequent selection, the matching process usually sets certain thresholds or uses combined rules, aiming to screen out "multiple" rather than a single most matching path. Some basic sorting mechanisms may also be applied, such as roughly sorting the initially screened paths according to the number or importance of the matched key features, but the main purpose is still to form a candidate set.

[0029] Finally, the system aggregates all the nursing path information that meets the matching conditions (which may include path ID, name, key description, etc.) to form a result list containing "multiple nursing path information". This list will be used as input and passed to the subsequent structured embedding coding and fine-grained matching link based on the semantic feature search network in the technical solution, so as to achieve initial focus from massive information to precise personalized recommendation under in-depth semantic understanding. This phased processing method takes into account both efficiency and accuracy and is a common strategy for implementing complex personalized decision support systems.

[0030] In the above patient personalized care management system 100, the nursing path fine-grained semantic matching module 140 is used to perform nursing path search matching processing based on structured embedding semantics on the preprocessed nursing information data and the multiple nursing path information to obtain multiple nursing path query response semantic coding representations. Figure 3 Block diagram of the nursing path fine-grained semantic matching module in the patient personalized care management system according to an embodiment of the present application. As Figure 3As shown in the figure, in the embodiment of the present application, the fine-grained semantic matching module 140 of the care path includes: a care information structured encoding unit 141, configured to perform structured embedding encoding on the preprocessed care information data to obtain a preprocessed care information semantic embedding encoding vector; a care path information structured encoding unit 142, configured to perform structured embedding encoding on the multiple care path information to obtain multiple care path semantic embedding encoding vectors; a care path query matching unit 143, configured to pass the preprocessed care information semantic embedding encoding vector and the multiple care path semantic embedding encoding vectors through a semantic feature search network based on care path matching to obtain multiple care path query response semantic encoding vectors as the multiple care path query response semantic encoding representations.

[0031] In the above patient personalized care management system 100, the care information structured encoding unit 141 and the care path information structured encoding unit 142 are configured to perform structured embedding encoding on the preprocessed care information data to obtain a preprocessed care information semantic embedding encoding vector, and perform structured embedding encoding on the multiple care path information to obtain multiple care path semantic embedding encoding vectors. It should be understood that considering the traditional rule-based or keyword-based matching methods, only "surface information matching" can be performed, and it is impossible to "deeply understand the complex internal relationships between the patient's various data indicators and the semantic meaning of the overall care needs contained therein". To overcome this defect, relying solely on the preprocessed clean data and the initially screened candidate paths is not enough. Therefore, in the technical solution of the present application, the preprocessed care information data is further subjected to structured embedding encoding to obtain a preprocessed care information semantic embedding encoding vector, and the multiple care path information is subjected to structured embedding encoding to obtain multiple care path semantic embedding encoding vectors. The preprocessed care information containing the patient's multi-dimensional features and the detailed content of each initially screened candidate care path are mapped into a unified high-dimensional vector space. The core goal is to transform these heterogeneous and complex information into quantitative "care path semantic embedding encoding vectors" that can reflect their semantic content.

[0032] In the above-mentioned patient personalized care management system 100, the care path query matching unit 143 is used to pass the pre-processed care information semantic embedding coding vector and the multiple care path semantic embedding coding vectors through a semantic feature search network based on care path matching to obtain multiple care path query response semantic coding vectors as the multiple care path query response semantic coding representations. It should be understood that only obtaining the semantic embedding vectors of the patient and the care path, and performing a simple similarity calculation (such as the limitations of the traditional method mentioned in the technical background), is still difficult to completely overcome the drawbacks of "surface information matching", and it is impossible to fully and deeply understand the complex internal connections between the various data indicators of the patient and the overall semantics of the care needs implied. Therefore, in the technical solution of the present application, the pre-processed care information semantic embedding coding vector and the multiple care path semantic embedding coding vectors are further passed through a semantic feature search network based on care path matching to obtain multiple care path query response semantic coding vectors. That is to say, although the preprocessed semantic embedding encoding vector of nursing information provides a basis, in order to capture the subtle and complex interactions between information such as diagnostic details, subtle changes in test results, the mutual influence of multiple complications, and the complexity of medication records, it is necessary to capture the deep semantic associations between features, rather than just the surface feature similarity comparison. The processing of the semantic feature search network based on nursing path matching aims to use the capabilities of similar deep learning models to mine the complex matching patterns hidden behind the vectors. Specifically, the semantic feature search network based on nursing path matching can deeply interact and match a single vector representing the patient's status with a set of vectors representing multiple candidate nursing paths. Its core goal is to go beyond simple distance or similarity calculations, and through complex processing mechanisms within the network, such as single semantic query, feature enhancement, context-aware adjustment, relationship gating, and self-attention aggregation, dynamically and context-awarely evaluate the deep semantic fit between the patient information vector and each nursing path vector, thereby significantly improving the accuracy and reliability of personalized recommendations for the entire system.

[0033] Figure 4 FIG. 1 is a block diagram of a nursing path query matching unit in a patient personalized nursing management system according to an embodiment of the present application. Figure 4As shown, in the embodiment of the present application, the nursing path query matching unit 143 includes: a monomer semantic query encoding subunit 1431, configured to perform feature enhancement on the preprocessed nursing information semantic embedding encoding vector, and then perform monomer semantic query encoding on each of the nursing path semantic embedding encoding vectors in the multiple nursing path semantic embedding encoding vectors respectively to obtain a set of preprocessed nursing information monomer semantic query score encoding vectors; a nursing information semantic weight calculation subunit 1432, configured to calculate the monomer semantic weight of each of the preprocessed nursing information monomer semantic query score encoding vectors in the set of preprocessed nursing information monomer semantic query score encoding vectors respectively to obtain a set of preprocessed nursing information monomer query semantic self-attention weights; a nursing path query response encoding subunit 1433, configured to weight the set of preprocessed nursing information monomer semantic query score encoding vectors based on the set of preprocessed nursing information monomer query semantic self-attention weights to obtain the multiple nursing path query response semantic encoding vectors.

[0034] Specifically, the monomer semantic query encoding subunit 1431 is configured to perform feature enhancement on the preprocessed nursing information semantic embedding encoding vector, and then perform monomer semantic query encoding on each of the nursing path semantic embedding encoding vectors in the multiple nursing path semantic embedding encoding vectors respectively to obtain a set of preprocessed nursing information monomer semantic query score encoding vectors. Correspondingly, in the embodiment of the present application, the monomer semantic query encoding subunit 1431 is configured to: perform feature enhancement on the preprocessed nursing information semantic embedding encoding vector based on deconvolution encoding to obtain an enhanced preprocessed nursing information semantic embedding encoding vector, and the enhanced preprocessed nursing information semantic embedding encoding vector has the same feature scale as each of the nursing path semantic embedding encoding vectors in the multiple nursing path semantic embedding encoding vectors; perform monomer semantic query encoding on the enhanced preprocessed nursing information semantic embedding encoding vector and each of the nursing path semantic embedding encoding vectors in the multiple nursing path semantic embedding encoding vectors respectively to obtain a set of preprocessed nursing information monomer semantic query score encoding vectors.

[0035] Specifically, in the monomer semantic query encoding subunit 1431, performing feature enhancement on the preprocessed nursing information semantic embedding encoding vector based on deconvolution encoding to obtain an enhanced preprocessed nursing information semantic embedding encoding vector, and the enhanced preprocessed nursing information semantic embedding encoding vector has the same feature scale as each of the nursing path semantic embedding encoding vectors in the multiple nursing path semantic embedding encoding vectors, which is expressed by the formula: ; where is the semantic embedding coding vector of the post - processed care information, is the transposed convolution coding, is the transposed convolution weight matrix, is the L1 norm of the vector, is the enhanced semantic embedding coding vector of the post - processed care information. It should be understood that since both patient information and care path information are converted into semantic embedding coding vectors, there may be differences in the generation processes and the characteristics of the original data on which these two types of vectors are based. Directly comparing a single vector representing the patient's state with a set of vectors representing multiple candidate care paths may lead to inaccuracies in subsequent comparisons and matches due to potential "feature scale" inconsistencies. That is to say, if the feature vectors from different sources have different scales, the numerical stability of subsequent calculations and the fairness of comparisons will both be affected, and it is impossible to establish a reliable basis for comparison within the same semantic space. Therefore, in order to perform meaningful and fair deep semantic comparisons, this potential scale mismatch problem must be solved. Based on this, through feature enhancement based on transposed convolution coding, it aims to improve the feature expressiveness of the semantic embedding coding vector of the post - processed care information, enabling it to capture and present the patient's complex personalized state more richly and meticulously. Selecting transposed convolution coding instead of a simple method reflects the emphasis on "data - driven adaptive feature representation learning" and can learn the optimal enhancement method from the data. Secondly, and most importantly, the clear goal of this step is to ensure that the enhanced patient information vector (i.e., "the enhanced semantic embedding coding vector of the post - processed care information") is "aligned in feature scale" with "the semantic embedding coding vectors of multiple care paths". This achieves dimension alignment or scale alignment, creating a prerequisite for fair and just comparison and integration of the two in the semantic feature search network.

[0036] Specifically, in the single - entity semantic query coding sub - unit 1431, the enhanced semantic embedding coding vector of the post - processed care information and each care path semantic embedding coding vector in the multiple care path semantic embedding coding vectors are respectively subjected to single - entity semantic query coding to obtain a set of single - entity semantic query score coding vectors of the post - processed care information, which is expressed by the formula: ; ; where, is the multiple care path semantic embedding coding vectors, are respectively the 1st, 2nd, th, and th care path semantic embedding coding vectors in the multiple care path semantic embedding coding vectors, and are respectively the trainable weight matrix and the trainable bias vector, is a vector concatenation operation, is a function, the th semantic query score encoding vector of the post - processing care information monomer. It should be understood that although the enhanced semantic embedding encoding vector of the post - processing care information and the semantic embedding encoding vector of the care path have been aligned in terms of feature scale, to truly achieve an understanding of the "deep semantic relationship" beyond "surface information matching", simple vector - to - vector distance or similarity calculations are still insufficient. These simple metrics are difficult to capture the subtle and multi - dimensional fit between the complex patient states (such as diagnostic details, subtle changes in test results, interactions between complications, medication complexity) and the multi - faceted requirements of the care path. As described in the provided corpus, the monomer semantic query encoding method is the processing of an implicit representor for semantic similarity or matching based on a deep neural network architecture, and its original design purpose is to measure the deep "semantic correlation" between vectors. Therefore, the introduction of monomer semantic query encoding is to utilize this more powerful deep - learning ability to conduct a refined semantic association assessment for each pair of "patient - path" vectors, rather than staying at the macroscopic similarity level. The purpose of this step is very specific, that is, for each candidate care path, use the monomer semantic query unit to independently and deeply explore the semantic matching degree between it and the enhanced patient information vector. The core goal is not simply to give a matching score, but to generate a "monomer semantic query score encoding vector". This vector "is not a simple scalar score, but contains richer semantic association information", and its different dimensions may represent the semantic similarity between the two at "different levels". Therefore, for each candidate path, an information - rich vectorized representation is generated that can reflect its matching situation with the patient in multiple semantic dimensions (such as disease feature coincidence degree, specific care need response degree, risk factor matching degree, etc.).

[0037] Figure 5 is a block diagram of the care information semantic weight calculation sub - unit in the patient personalized care management system according to an embodiment of the present application. As Figure 5As shown, in the embodiments of the present application, the nursing information semantic weight calculation subunit 1432 includes: a monomer semantic matching secondary subunit 14321, configured to determine the monomer semantic matching degree of each preprocessed nursing information monomer semantic query score encoding vector in the set of preprocessed nursing information monomer semantic query score encoding vectors based on the self-distribution characteristics of the feature set of the set of preprocessed nursing information monomer semantic query score encoding vectors, so as to obtain a set of preprocessed nursing information monomer semantic matching degrees; a gating weight calculation secondary subunit 14322, configured to input the set of preprocessed nursing information monomer semantic matching degrees into a relational gating proxy module to obtain a set of self-attention weights of the preprocessed nursing information monomer query semantics.

[0038] Specifically, the monomer semantic matching secondary subunit 14321 is configured to determine the monomer semantic matching degree of each preprocessed nursing information monomer semantic query score encoding vector in the set of preprocessed nursing information monomer semantic query score encoding vectors based on the self-distribution characteristics of the feature set of the set of preprocessed nursing information monomer semantic query score encoding vectors, so as to obtain a set of preprocessed nursing information monomer semantic matching degrees. Correspondingly, in the embodiments of the present application, the monomer semantic matching secondary subunit 14321 is configured to: perform inherent alignment calibration on each preprocessed nursing information monomer semantic query score encoding vector in the set of preprocessed nursing information monomer semantic query score encoding vectors to obtain a set of calibrated preprocessed nursing information monomer semantic query score encoding vectors; calculate the monomer semantic matching degree of each calibrated preprocessed nursing information monomer semantic query score encoding vector based on the distribution characteristics of the set of calibrated preprocessed nursing information monomer semantic query score encoding vectors to obtain the set of preprocessed nursing information monomer semantic matching degrees.

[0039] Specifically, performing inherent alignment calibration on each preprocessed nursing information monomer semantic query score encoding vector in the set of preprocessed nursing information monomer semantic query score encoding vectors to obtain a set of calibrated preprocessed nursing information monomer semantic query score encoding vectors is expressed by the formula: ; ; ; ; ; Wherein, is the preprocessed nursing information query potential vector, is the projection weight matrix, is the preprocessed nursing information query interaction encoding vector, is the coupling constant, which is calculated in the same way as when deconvolution is enhanced to maintain symmetry. is the covariance matrix of the post - treatment care information query after pre - processing. is the encoded vector of the semantic query score of the post - treatment care information monomer after calibration. It should be understood that although the encoded vector of the semantic query score of the post - treatment care information monomer generated in the previous step is intended to capture the semantic correlation between the patient information and each candidate care path, its generation process (such as feature splicing) may lead to inaccuracies in representation. Specifically, during the process of mapping from the original feature space to the semantic query encoding space, there may be problems such as unsatisfactory representation of dynamic characteristics or insufficient inherent alignment characteristics between the two spaces. Therefore, it is necessary to improve the representation accuracy of the semantic query score of the care information monomer by optimizing this mapping process. Without calibration, these subtle deviations in representation or spatial misalignments may accumulate and affect the accuracy of subsequent matching degree quantification, weight assignment, and final decision - making based on these vectors, and cannot meet the required high - precision personalized matching needs. Therefore, the purpose of this step is to refine and correct the original encoded vector of the semantic query score of the care information monomer by implementing inherent alignment calibration. Specifically, it uses the calculated interaction potential vector as the inherent alignment generator, and by imposing constraints on the covariant path under the condition of gauge symmetry (i.e., the "mapping gauge fixing guarantee" in the corpus), it systematically improves the inherent alignment between the feature space and the semantic query encoding space. This is not only to optimize the representation of the dynamic characteristics of each encoded vector of the semantic query score of the post - treatment care information monomer itself, but also to ensure that these vectors are expressed in a unified and aligned semantic space, thereby improving the accuracy and reliability of their representation.

[0040] Specifically, calculate the monomer semantic matching degree of each encoded vector of the semantic query score of the post - treatment care information monomer after calibration based on the distribution characteristics of the set of the encoded vectors of the semantic query score of the post - treatment care information monomer after calibration to obtain the set of the monomer semantic matching degrees of the post - treatment care information, which is expressed by the formula: ; where, is the natural exponential function with base e, is function, is the monomer semantic matching degree of the post - treatment care information, is the number of in the set of the encoded vectors of the semantic query score of the post - treatment care information monomer after calibration, and A calibrated pre - processed post - care information monomer semantic query score encoding vector. It should be understood that although obtaining only the "calibrated pre - processed post - care information monomer semantic query score encoding vector" between each candidate care path and patient information already contains rich semantic association information, looking at these score vectors in isolation may not be sufficient to comprehensively and accurately evaluate the relative merits of each path. Since understanding the complex internal connections and overall care needs of patients is particularly important, and for the same patient, the absolute values of the matching scores of different candidate paths may be affected by various factors, direct comparison may ignore the relative position and significance of these scores in the current candidate set. As pointed out by the provided corpus, a context - aware method is needed, where each monomer semantic query score encoding vector should not be viewed in isolation but considered in the context of the overall score set. If the overall distribution characteristics of the score vector set are not considered, it may not be possible to distinguish the truly prominent and more confident matches among many options, thus affecting the accuracy of the final recommendation. Therefore, the purpose of this step is to adaptively adjust the matching degree represented by each monomer score vector by analyzing the overall distribution characteristics of the set of calibrated pre - processed post - care information monomer semantic query score encoding vectors. Its core goal is to determine the pre - processed post - care information monomer semantic matching degree based on the relative performance of each score vector in the entire set (e.g., comparison with statistical indicators such as mean and variance). If a calibrated pre - processed post - care information monomer semantic query score encoding vector is significantly prominent in the set (e.g., far above the average level), its corresponding semantic relevance is considered stronger and should be given a higher matching degree; conversely, if the score is close to or even lower than the overall level, it may represent a weaker association or noise, and its matching degree should be correspondingly reduced. The purpose of this is to extract a more reliable and quantifiable indicator that can better reflect the true matching strength.

[0041] Specifically, the gating weight calculation secondary subunit 14322 is used to input the set of pre - processed post - care information monomer semantic matching degrees into the relational gating proxy module to obtain the set of pre - processed post - care information monomer query semantic self - attention weights, which can be expressed by the formula: ; where is a masking function, is a preset threshold, Query the semantic self-attention weights for the preprocessed nursing information monomers. It should be understood that although the monomer semantic matching degrees between each candidate nursing path and the patient have been calculated in the previous steps, representing a preliminary matching quantitative assessment, there may be limitations when directly applying these matching degrees to subsequent information aggregation (such as based on self-attention). Specifically, directly using these matching degrees may seem crude and easily overlook the diversity and complexity of semantic relationships. That is, due to the complexity of patient information and the drawback that traditional methods are difficult to capture deep connections, this means that simple matching scores may not fully reflect the subtle differences and interrelationships of different paths in meeting the specific and multi-dimensional needs of patients. Therefore, a more complex mechanism is needed to process these matching degrees to ensure that the weights generated for subsequent information integration can more accurately reflect these nuances and complex relationships. Therefore, the purpose of this step is to introduce a relational gating proxy module as a precise intermediate layer or regulatory mechanism. Its core goal is not simply to transmit the matching degree information, but to selectively enhance, suppress, or reshape these matching degrees, so as to achieve fine-grained control of modeling complex semantic relationships. This module plays a role similar to a "switch" in the biological nervous system. By controlling the intensity and direction of the information flow, it transforms the set of original preprocessed nursing information monomer semantic matching degrees into a weight form more suitable for the self-attention mechanism. In short, its purpose is to optimize and refine the matching degree information so that it is transformed into self-attention weights that can more accurately guide the subsequent information aggregation process.

[0042] Specifically, the nursing path query response encoding subunit 1433 is used to weight the set of preprocessed nursing information monomer semantic query score encoding vectors based on the set of semantic self-attention weights queried for the preprocessed nursing information monomers to obtain the multiple nursing path query response semantic encoding vectors, which can be expressed by the formula: ; Where is the A semantic encoding vector for the query response of a care path. It should be understood that since the system has generated a monomer semantic query score encoding vector reflecting the multi-dimensional semantic matching details between each candidate care path and the patient information through the previous steps, and calculated the monomer query semantic self-attention weights that can reflect the relative importance of these matches in the current candidate set. However, these two sets of information need to be effectively combined to form the final and comprehensive semantic representation of the fitness of each care path to the patient. Therefore, the purpose of this step is to use the monomer query semantic self-attention weights generated in the previous step to perform weighted aggregation on the corresponding monomer semantic query score encoding vectors. The core goal is to execute a dynamic weight allocation process, enabling the model to flexibly focus on important semantic information and ignore secondary or noisy information according to the adaptively learned weights. By giving greater influence to the score encoding vectors with higher self-attention weights and reducing the contribution of vectors with lower weights, this weighted aggregation process simulates an information integration and decision-making process. It performs weighted combination (such as weighted average or summation) based on the matching information represented by each score vector and its relative importance, and finally generates a comprehensive judgment or conclusive vector representation for each candidate care path.

[0043] In the above patient personalized care management system 100, the care plan generation module 150 is used to perform care path matching degree analysis based on the above-mentioned multiple care path query response semantic encoding representations to determine the fitness of multiple care paths, and generate a care plan. The care plan generation module 150 includes: a care path fitness calculation unit for determining the fitness of multiple care paths based on the above-mentioned multiple care path query response semantic encoding vectors; a care plan generation unit for selecting the care path corresponding to the maximum value among the fitnesses of the multiple care paths as the care path for the target patient object, and generating a care plan. In particular, in the embodiments of the present application, the care plan generation module includes: passing the above-mentioned multiple care path query response semantic encoding vectors through a care path matching degree analyzer based on a classifier to obtain the fitnesses of the multiple care paths.

[0044] It should be understood that although the semantic encoding vectors of the nursing path query responses contain deep and complex semantic matching information between patient information and each candidate path, these high-dimensional vectors themselves are not intuitive bases for judging the quality of the match. However, personalized nursing needs to precisely fit the unique and multi-factor-influenced condition of the patient. Therefore, the system requires a clear mechanism to quantify and compare the "goodness" of these deep matching results in order to make a final recommendation decision. Directly using the semantic vectors for selection is not clear, and it is necessary to convert them into scalar values that are easy to compare, that is, "fitness". Therefore, in the technical solution of this application, the multiple semantic encoding vectors of the nursing path query responses are further passed through a classifier-based nursing path matching degree analyzer to obtain the fitness of multiple nursing paths. The role of the classifier-based nursing path matching degree analyzer is to learn and judge the matching degree represented by each semantic encoding vector of the nursing path query response, and give a quantified fitness evaluation result. Immediately afterwards, the path with the highest fitness score is selected as the nursing path finally recommended to the target patient. The purpose is to make an optimized, data-driven decision based on the results of the previous deep semantic analysis and the current quantitative evaluation, ensuring that the selected path is the most suitable for the specific situation of the current patient among all candidates. Finally, generating a nursing plan is to convert the selected path into an actually executable nursing plan, complete the transformation from data analysis to clinical application, and achieve the ultimate goal of the entire system - providing personalized nursing management.

[0045] Specifically, in a specific example of the application, the semantic encoding vectors of the query responses of all candidate paths are input into a pre-trained deep neural network classifier (such as a multi-layer perceptron or a Transformer-based classification head). This classifier uses these vectors as input features, and its output layer is designed to predict a fitness score ranging from 0 to 1. This classifier learns how to interpret the pattern of path fitness from the semantic encoding vectors through training on a large amount of labeled data (including historical patient cases and their corresponding best nursing paths or effect evaluations). For a new patient, after the system calculates the semantic encoding vectors of the nursing path query responses for each path, they are sent into this classifier one by one to obtain the fitness scores. Subsequently, the path with the highest score is selected, and its detailed content (such as nursing measures, time nodes, expected goals, etc.) is integrated into the final personalized nursing plan document and presented to medical staff.

[0046] In summary, the patient personalized care management system 100 according to the embodiments of the present application is elucidated. After structuring and embedding the preprocessed patient care information and multiple candidate care path information to generate their respective semantic embedding coding representations, a semantic feature search network based on care path matching is used to search and match the semantic features of the patient care information embedding semantics and each candidate care path information embedding semantics to explore the matching degree between each care path and the patient. In this way, it is possible to go beyond simple text or rule matching, place the multi-dimensional information of the patient and the structured requirements of the care path in the same high-dimensional semantic space for comparison, so as to output a care path query response semantics that can more accurately reflect the matching relationship, laying a solid foundation for calculating the fitness and selecting the optimal personalized care path.

[0047] As described above, the patient personalized care management system 100 according to the embodiments of the present application can be implemented in various terminal devices. In one example, the patient personalized care management system 100 can be integrated into the terminal device as a software module and / or a hardware module. For example, the patient personalized care management system 100 can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the patient personalized care management system 100 can also be one of the many hardware modules of the terminal device.

[0048] Alternatively, in another example, the patient personalized care management system 100 and the terminal device can also be separate devices, and the patient personalized care management system 100 can be connected to the terminal device through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0049] In several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation.

[0050] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0051] In addition, in each of the embodiments of the present invention, the functional modules can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware, or in the form of a combination of hardware and software functional modules.

[0052] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0053] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be encompassed by the present invention. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0054] In addition, it is obvious that the term "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The terms such as "second" are used to denote names and do not represent any specific order.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit of the technical solutions of the present invention.

Claims

1. A patient personalized care management system, characterized in that, Including: A nursing information data acquisition module, configured to obtain nursing information data of a target patient object, where the nursing information data includes basic data, diagnosis data, test results, complications, and medication records; A nursing information data preprocessing module, configured to perform data cleaning and format conversion on the nursing information data to obtain preprocessed nursing information data; A coarse-grained nursing path matching module, configured to match multiple nursing path information from a nursing path information library based on the preprocessed nursing information data; A fine-grained semantic matching module for nursing paths, configured to perform a nursing path search and matching process based on structured embedded semantics on the preprocessed nursing information data and the multiple nursing path information to obtain multiple nursing path query response semantic coding representations; A nursing plan generation module, configured to perform a nursing path matching degree analysis based on the multiple nursing path query response semantic coding representations to determine the fitness of multiple nursing paths, and generate a nursing plan.

2. The patient personalized care management system according to claim 1, characterized in that The fine-grained semantic matching module for nursing paths includes: A nursing information structured coding unit, configured to perform structured embedded coding on the preprocessed nursing information data to obtain a preprocessed nursing information semantic embedded coding vector; A nursing path information structured coding unit, configured to perform structured embedded coding on the multiple nursing path information to obtain multiple nursing path semantic embedded coding vectors; A nursing path query matching unit, configured to pass the preprocessed nursing information semantic embedded coding vector and the multiple nursing path semantic embedded coding vectors through a semantic feature search network based on nursing path matching to obtain multiple nursing path query response semantic coding vectors as the multiple nursing path query response semantic coding representations.

3. The patient personalized care management system according to claim 2, characterized in that, The nursing path query matching unit includes: A single entity semantic query coding subunit, configured to enhance the features of the preprocessed nursing information semantic embedded coding vector, and then perform single entity semantic query coding on each of the nursing path semantic embedded coding vectors in the multiple nursing path semantic embedded coding vectors to obtain a set of preprocessed nursing information single entity semantic query score coding vectors; A nursing information semantic weight calculation subunit, configured to calculate the single entity semantic weight of each preprocessed nursing information single entity semantic query score coding vector in the set of preprocessed nursing information single entity semantic query score coding vectors to obtain a set of preprocessed nursing information single query semantic self-attention weights; A nursing path query response coding subunit, configured to weight the set of preprocessed nursing information single entity semantic query score coding vectors based on the set of preprocessed nursing information single query semantic self-attention weights to obtain the multiple nursing path query response semantic coding vectors.

4. The patient personalized care management system according to claim 3, wherein The single entity semantic query coding subunit is configured to: Perform feature enhancement based on deconvolution coding on the semantic embedding coding vector of the preprocessed nursing information to obtain an enhanced semantic embedding coding vector of the preprocessed nursing information, and the enhanced semantic embedding coding vector of the preprocessed nursing information has the same feature scale as each semantic embedding coding vector of the multiple nursing path semantic embedding coding vectors; Perform monomer semantic query coding on the enhanced semantic embedding coding vector of the preprocessed nursing information and each semantic embedding coding vector of the multiple nursing path semantic embedding coding vectors respectively to obtain a set of monomer semantic query score coding vectors of the preprocessed nursing information.

5. The patient personalized care management system according to claim 4, wherein The nursing information semantic weight calculation subunit includes: A monomer semantic matching secondary subunit, configured to determine the monomer semantic matching degree of each monomer semantic query score coding vector in the set of monomer semantic query score coding vectors of the preprocessed nursing information based on the self-distribution characteristics of the feature set of the set of monomer semantic query score coding vectors of the preprocessed nursing information, so as to obtain a set of monomer semantic matching degrees of the preprocessed nursing information; A gating weight calculation secondary subunit, configured to input the set of monomer semantic matching degrees of the preprocessed nursing information into a relational gating proxy module to obtain a set of self-attention weights of the monomer query semantics of the preprocessed nursing information.

6. The patient personalized care management system according to claim 5, characterized in that, The monomer semantic matching secondary subunit is used for: Performing inherent alignment calibration on each monomer semantic query score coding vector in the set of monomer semantic query score coding vectors of the preprocessed nursing information to obtain a set of calibrated monomer semantic query score coding vectors of the preprocessed nursing information; Calculating the monomer semantic matching degree of each calibrated monomer semantic query score coding vector of the preprocessed nursing information based on the distribution characteristics of the set of calibrated monomer semantic query score coding vectors of the preprocessed nursing information to obtain the set of monomer semantic matching degrees of the preprocessed nursing information.

7. The patient personalized care management system according to claim 6, characterized in that The nursing plan generation module includes: A nursing path adaptability calculation unit, configured to determine the adaptability of multiple nursing paths based on the multiple nursing path query response semantic coding vectors; A nursing plan generation unit, configured to select the nursing path corresponding to the maximum value among the adaptabilities of the multiple nursing paths as the nursing path of the target patient object and generate a nursing plan.

8. The patient personalized care management system according to claim 7, wherein The nursing path adaptability calculation unit is used for: passing the multiple nursing path query response semantic coding vectors through a nursing path matching degree analyzer based on a classifier to obtain the adaptabilities of the multiple nursing paths.

Images