Blood transfusion treatment scheme recommendation system and method based on historical information

Abstract

The invention provides a blood transfusion treatment scheme recommendation system and method based on historical information, and relates to the field of medical information.The blood transfusion treatment scheme recommendation method comprises the steps that firstly, basic information and diagnosis information of a patient, real-time blood test information and historical blood transfusion medical record information are extracted from HIS, LIS, EMR and TMIS systems; performing joint semantic analysis on the basic information, the diagnosis information and the real-time blood test information, and converting the historical blood transfusion medical record information into a dynamic vector sequence; then, the current blood index and the historical blood transfusion time sequence are dynamically matched in a frequency domain space through nonlinear spectrum analysis, and a personalized recommendation scheme is generated in combination with the implicit reasoning ability of a pre-trained medical large model, so that the limitation of a traditional system in the aspects of static parameter setting, individual adaptability and semantic mining is broken through; and the accuracy and timeliness of blood transfusion treatment decision making are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of medical information, and more particularly, in an embodiment of the present application, to a blood transfusion therapy plan recommendation system and method based on historical information. Background Art

[0002] As an important treatment method in clinical medicine, blood transfusion therapy is widely used in scenarios such as surgery, trauma, blood diseases, and acute and chronic anemia. Its core goal is to quickly restore the patient's circulating blood volume, improve oxygen carrying capacity, or correct coagulation dysfunction through precise blood component transfusion.

[0003] Patent CN111564211B proposes a blood transfusion intelligent decision-making system, device and method, which includes a data layer, an interaction layer, an application layer and peripheral system applications. The data layer integrates HIS, LIS, EMR, TMIS and IDT systems to record patient information, diagnosis, blood type testing, transfusion history and intelligent decision-making data. The interaction layer is responsible for data exchange and business logic processing. The application layer formulates personalized blood transfusion treatment plans based on data analysis results and serves as a reference for blood application. Peripheral system applications submit actual clinical blood application requests to the blood bank based on these references.

[0004] Although the patent solution constructs the transfusion recommendation logic through keyword analysis and hedging algorithm, its technology has significant limitations: First, the hedging algorithm only relies on the fixed features of historical data (such as changes in indications after a single transfusion), and does not model the dynamic time series association of multiple transfusion records (such as the law of the evolution of patient antibody levels with the frequency of transfusions), resulting in delayed response to scenarios such as fluctuations in demand during the postoperative recovery period and repeated attacks of chronic anemia. Second, the preset formula uses standardized parameters (such as the default value of blood volume), ignoring the abnormal fluid distribution of obese and edematous patients or plasma protein variables of patients with liver disease, which can easily lead to deviations in blood volume calculation. Third, keyword extraction can only identify explicit symptom terms (such as "low hemoglobin"), and cannot parse the nonlinear associations between symptoms (such as the compound impact of "thrombocytopenia with coagulation abnormalities" on transfusion varieties). It lacks the ability of deep learning to capture implicit pathological patterns and may miss the core indications of complex conditions.

[0005] Therefore, an optimized transfusion therapy recommendation scheme based on historical information is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a blood transfusion therapy plan recommendation system and method based on historical information, which first extracts the patient's basic information and diagnostic information, real-time blood test information and historical blood transfusion medical record information from the HIS, LIS, EMR and TMIS systems, and then performs joint semantic analysis on the basic information and diagnostic information and real-time blood test information, and converts the historical blood transfusion medical record information into a dynamic vector sequence. Then, the current blood indicators and the historical blood transfusion time series are dynamically matched in the frequency domain space through nonlinear spectral analysis, and combined with the implicit reasoning ability of the pre-trained medical large model, a personalized recommendation plan is generated to break through the limitations of traditional systems in static parameter setting, individual adaptability and semantic mining, and significantly improve the accuracy and timeliness of blood transfusion therapy decisions.

[0007] According to one aspect of the present application, a blood transfusion therapy plan recommendation system based on historical information is provided, which includes: Information collection module, used to obtain basic information and diagnosis information of patients from HIS system, obtain real-time blood test information from LIS system, and obtain historical blood transfusion medical record information from EMR system and TMIS system; A semantic fusion module, used for performing semantic fusion of the basic information and diagnostic information of the patient object with the real-time blood test information based on feature coding to obtain a fusion semantic representation between the patient information and the blood test information; A blood transfusion information encoding module, used for embedding and encoding each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of blood transfusion information embedding encoding features; A dynamic matching analysis module is used to perform dynamic matching analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion treatment plan recommendation report, wherein the dynamic matching analysis is used to: first perform dynamic matching of the blood transfusion information based on nonlinear spectral analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information, and then perform generation processing to obtain the blood transfusion treatment plan recommendation report.

[0008] According to another aspect of the present application, a method for recommending a blood transfusion therapy regimen based on historical information is provided, comprising: Obtain basic information and diagnosis information of patients from the HIS system, obtain real-time blood test information from the LIS system, and obtain historical blood transfusion medical record information from the EMR system and TMIS system; Performing semantic fusion based on feature coding on the basic information and diagnostic information of the patient object and the real-time blood test information to obtain a fusion semantic representation between the patient information and the blood test information; Embedding and coding each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of embedded coding features of the blood transfusion information; A dynamic matching analysis is performed on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion treatment plan recommendation report, wherein the dynamic matching analysis is used to: first perform dynamic matching of the blood transfusion information based on nonlinear spectral analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information, and then perform generation processing to obtain the blood transfusion treatment plan recommendation report.

[0009] Compared with the prior art, the present application provides a blood transfusion therapy recommendation system and method based on historical information, which first extracts basic patient information and diagnostic information, real-time blood test information and historical blood transfusion medical record information from HIS, LIS, EMR and TMIS systems, then performs joint semantic analysis on the basic information and diagnostic information and real-time blood test information, and converts the historical blood transfusion medical record information into a dynamic vector sequence. Then, the current blood indicators are dynamically matched with the historical blood transfusion time series in the frequency domain space through nonlinear spectral analysis, and combined with the implicit reasoning ability of the pre-trained medical large model, a personalized recommendation plan is generated to break through the limitations of traditional systems in static parameter setting, individual adaptability and semantic mining, and significantly improve the accuracy and timeliness of blood transfusion therapy decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying 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 of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 This is a system block diagram of a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application.

[0012] Figure 2 This is a data flow diagram of a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application.

[0013] Figure 3 It is a block diagram of a dynamic matching analysis module in a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application.

[0014] Figure 4 This is a block diagram of a blood transfusion information dynamic query response unit in a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application.

[0015] Figure 5 The present invention is a flowchart of a method for recommending a blood transfusion therapy plan based on historical information according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0017] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0018] In addition, in order to better illustrate the present application, numerous specific details are provided in the following specific embodiments. It should be understood by those skilled in the art that the present application can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present application.

[0019] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0020] Blood transfusion therapy is an important treatment method in clinical medicine and is widely used in various scenarios. Patent CN111564211B proposes an intelligent decision-making system for blood transfusion, which integrates patient information, diagnosis and other data, and formulates personalized blood transfusion treatment plans through analysis. However, the system fails to model the dynamic time series association of multiple blood transfusion records, ignores the abnormal fluid distribution or plasma protein variables of specific patients, and causes calculation deviations. It is also unable to analyze the nonlinear associations between symptoms, making it difficult to capture the core indications of complex diseases.

[0021] In response to the above technical problems, the technical concept of this application is to adopt an artificial intelligence-based data analysis and processing algorithm. Specifically, firstly, basic patient information and diagnostic information, real-time blood test information and historical transfusion medical record information are extracted from HIS, LIS, EMR and TMIS systems. Then, the basic information, diagnostic information and real-time blood test information are jointly semantically parsed to capture individual characteristics such as abnormal fluid distribution in obese patients and hypoproteinemia in cirrhosis. Subsequently, the historical transfusion medical record information is converted into a dynamic vector sequence to track long-term laws such as the evolution of the patient's antibody level and the attenuation of transfusion efficacy. Then, the current blood indicators are dynamically matched with the historical transfusion time series in the frequency domain space through nonlinear spectral analysis to identify the association between the current coagulation parameter fluctuations and the previous effective transfusion schemes. Finally, combined with the implicit reasoning ability of the pre-trained medical large model, a personalized recommendation scheme integrating individual metabolic characteristics, dynamic blood needs and complex symptom associations is generated. This solution achieves in-depth analysis of patients' individual characteristics, dynamic blood needs, and complex symptom associations, thus breaking through the limitations of traditional systems in static parameter setting, individual adaptability, and semantic mining, and significantly improving the accuracy and timeliness of transfusion therapy decisions.

[0022] This application proposes a blood transfusion therapy plan recommendation system based on historical information. Figure 1 This is a system block diagram of a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the blood transfusion treatment plan recommendation system 100 based on historical information includes: an information acquisition module 110, which is used to obtain basic information and diagnostic information of the patient object from the HIS system, obtain real-time blood test information from the LIS system, and obtain historical blood transfusion medical record information from the EMR system and the TMIS system; a semantic fusion module 120, which is used to perform semantic fusion based on feature coding on the basic information and diagnostic information of the patient object and the real-time blood test information to obtain a fused semantic representation between the patient information and the blood test information; a blood transfusion information encoding module 130, which is used to embed and encode each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of embedded coding features of the blood transfusion information; a dynamic matching analysis module 140, which is used to perform dynamic matching analysis on the time series of the fused semantic representation between the patient information and the blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion treatment plan recommendation report.

[0023] In the above-mentioned blood transfusion treatment plan recommendation system 100 based on historical information, the information acquisition module 110 is used to obtain the basic information and diagnosis information of the patient object from the HIS system, obtain the real-time blood test information from the LIS system, and obtain the historical blood transfusion medical record information from the EMR system and the TMIS system. It should be understood that the HIS system, the full name of which is the Hospital Information System, is an integrated information management system. The HIS system stores a large amount of information about patients, including but not limited to personal basic information (such as name, age, gender, contact information), health background (such as allergy history, previous medical history), clinical diagnosis results (ICD code), surgical records and other related medical data. Among them, the HIS system contains structured values ​​such as age, weight, height, and BMI, as well as category data such as gender, blood type, department, clinical diagnosis, and surgical type, and also contains text data such as symptom descriptions (such as "dizziness" and "chest pain") and diagnostic notes. By integrating these structured and unstructured data, a basic framework for a comprehensive understanding of the patient's health status can be constructed. Secondly, the laboratory information system (LIS) usually contains detailed laboratory test results, such as hemoglobin level, white blood cell count, coagulation function indicators, etc. Among them, the LIS system provides numerical time series data including HB (hemoglobin), PLT (platelets), PT (prothrombin time), APTT (activated partial thromboplastin time), INR (international normalized ratio), etc., and also includes static values ​​of blood type and DIC related indicators. These indicators reflect the patient's immediate physiological state. Finally, obtaining historical transfusion medical record information from the electronic medical record (EMR) system and transfusion management system (TMIS) is to track the patient's long-term health trends and treatment responses. Historical transfusion data in the EMR / TMIS system (such as transfusion time, variety, transfusion reaction, etc.) is usually stored in the form of discrete events. Among them, the historical transfusion medical record information provided by the EMR / TMIS system includes time series data of transfusion time, type (such as red blood cells, plasma), transfusion volume, transfusion reaction (such as allergy, fever), as well as event labels such as post-transfusion evaluation (such as "indication improvement") and organ transplant history. In this way, by extracting different types of information from HIS, LIS, EMR and TMIS systems, not only the efficiency of data utilization is improved, but also the understanding of individual differences of patients is enhanced, laying a solid foundation for the formulation of more accurate and personalized transfusion treatment plans, ensuring that the proposed treatment recommendations can be closely combined with the actual situation of the patient, thereby significantly improving the quality and effectiveness of medical decision-making.

[0024] In the above-mentioned blood transfusion therapy plan recommendation system 100 based on historical information, the semantic fusion module 120 is used to perform semantic fusion based on feature coding on the basic information and diagnostic information of the patient object and the real-time blood test information to obtain a fused semantic representation between patient information and blood test information, including: a patient object personal information embedding coding unit 121, used to embed the basic information and diagnostic information of the patient object to obtain a patient object personal information embedding coding vector; a real-time blood test information association coding unit 122, used to pass the real-time blood test information through a blood test information association encoder based on a fully connected layer to obtain a blood test information association coding vector; a patient information-blood test information feature domain fusion unit 123, used to perform feature domain fusion on the patient object personal information embedding coding vector and the blood test information association coding vector to obtain a patient information-blood test information fused semantic representation vector as the patient information-blood test information fused semantic representation.

[0025] Specifically, the patient object personal information embedding coding unit 121 is used to embed the basic information and diagnosis information of the patient object to obtain the patient object personal information embedding coding vector. In an embodiment of the present application, the patient object personal information embedding coding unit 121 is used to: embed the basic information and diagnosis information of the patient object using a structured-text dual-stream encoder based on the TabTransformer-BioBERT model to obtain the patient object personal information embedding coding vector. It should be understood that considering that the basic information and diagnosis information of the patient cover structured numerical values ​​(such as age, weight, etc.), category data (such as gender, blood type, etc.) and text data (such as symptom description, diagnosis notes, etc.). Different types of data are difficult to integrate and process directly. If keyword extraction or numerical threshold judgment is used, it is difficult to effectively model the dynamic association between structured fields, and it is also impossible to capture the pathological patterns implicit in the text, such as the intrinsic connection between disease diagnosis and related symptoms, and the association between different diagnoses. Based on this, the present application embeds the basic information and diagnosis information of the patient object to establish a deep mapping between individual patient features to obtain a patient object personal information embedding coding vector. In particular, in a specific example of the present application, a structured-text dual-stream encoder based on the TabTransformer-BioBERT model is used to embed the basic information and diagnostic information of the patient object to obtain the patient object personal information embedding encoding vector. It can be understood that the BioBERT model is pre-trained specifically for the biomedical field and can better understand the professional terms and semantic relationships in medical texts. For text content such as symptom descriptions and disease names in diagnostic information, BioBERT can capture more accurate semantic information. TabTransformer can effectively process structured data and mine potential connections between structured data. The combination of the two can achieve a deeper semantic understanding of patient information, which is impossible with simple data processing methods. Specifically, the TabTransformer architecture is used for the patient's numerical features (such as height, weight, laboratory indicators) and categorical features (such as gender, blood type, department). The model interactively learns categorical features through a self-attention mechanism, such as the correlation between "blood type" and "department" (a specific department pays more attention to a specific blood type). Dynamic encoding is performed as a dense vector, and the standardized numerical features (such as INR value and platelet count) are mapped to vector dimension vectors through a fully connected layer. For hierarchical classification features such as ICD codes, hierarchical embedding can be used to integrate the hierarchical relationship of the code (such as disease category-subcategory) into the vector representation to enhance the medical semantics of the code.Next, for unstructured texts such as clinical diagnosis texts and patient region descriptions, the pre-trained BioBERT model (a BERT variant fine-tuned in the medical field) is used to extract context-related semantic features. For example, the medical entity and status description in "DIC with reduced fibrinogen level" captures its clinical meaning through BioBERT's word vector and sentence vector, avoiding the problem of semantic loss in the traditional bag-of-words model. For short texts (such as diagnostic conclusions), the CLS-tagged vector can be extracted as the overall representation; for long texts (such as detailed medical history), key information is aggregated through attention pooling. Finally, the structured feature embedding (from TabTransformer) is cross-modally aligned with the text semantic vector (from BioBERT), and the gated fusion mechanism is used to automatically learn the weight ratio of the two types of features. For example, when laboratory indicators (such as extremely low platelet counts) are highly correlated with text diagnoses (such as acute leukemia), the model will enhance the contribution of structured features; when the diagnostic description contains complex complications, it focuses on the analysis of text semantics. The fused multimodal vector is compressed into a unified low-dimensional patient object personal information embedding coding vector through a multi-layer perceptron (MLP), retaining key discriminant information.

[0026] Specifically, the real-time blood test information association encoding unit 122 is used to pass the real-time blood test information through a blood test information association encoder based on a fully connected layer to obtain a blood test information association encoding vector. It should be understood that, considering that there are often complex associations between the various indicators in the real-time blood test information, such as HB (hemoglobin) is related to the degree of anemia, PT (prothrombin time), APTT (activated partial thromboplastin time), etc. are related to coagulation function. Based on this, in order to capture the nonlinear association between multiple parameters (such as the synergistic effect of increased INR value and decreased fibrinogen on coagulation function defects), in the technical solution of the present application, the real-time blood test information is passed through a blood test information association encoder based on a fully connected layer to obtain a blood test information association encoding vector. That is to say, the blood test information association encoder based on the fully connected layer constructs a high-order feature interaction network through multi-layer nonlinear transformation, and automatically mines the potential associations between these indicators by learning the statistical laws in the data, thereby more comprehensively reflecting the patient's current blood state. Specifically, the fully connected layer in the blood test information association encoder based on the fully connected layer is composed of multiple neurons, and each neuron is connected to all neurons in the previous layer through weights to form a dense connection structure to capture the complex relationship and potential association between different blood test indicators. Among them, in the input layer, various blood test data are converted into feature vectors and input into the network. As the network goes deeper, each fully connected layer will perform nonlinear transformation on the input data, and this process is achieved through activation functions. Commonly used activation functions such as ReLU, tanh or sigmoid can introduce nonlinear factors to the model, making the model more expressive. In the training process of the blood test information association encoder based on the fully connected layer, the connection weights between the layers are adjusted through the back propagation algorithm to minimize the error between the predicted value and the actual value. This process not only allows the model to learn the direct connection between blood test indicators, but also discovers those indirect relationships that are not easy to detect. For example, by analyzing a large amount of case data, the model may find that a decrease in hemoglobin level is accompanied by a decrease in platelet count, which may be due to a common pathological mechanism caused by a certain disease. In addition, the model can also identify some complex patterns. For example, in certain specific cases, even if the hemoglobin and platelet counts are within the normal range, the patient may still experience anemia symptoms. This is due to the influence of other factors that have not been taken into account (such as abnormal iron metabolism). In order to further improve the generalization ability and accuracy of the model, regularization techniques (such as L2 regularization, Dropout, etc.) are used to prevent overfitting. Finally, the output generated after multiple layers of nonlinear transformation is the blood test information association coding vector that integrates multiple blood test indicators and their interactions.This representation method can not only more accurately reflect the patient's current blood status, but also provide strong support for subsequent diagnostic decisions.

[0027] Specifically, the patient information-blood test information feature domain fusion unit 123 is used to perform feature domain fusion on the patient object personal information embedded coding vector and the blood test information associated coding vector to obtain the patient information-blood test information fusion semantic representation vector as the patient information-blood test information fusion semantic representation. It should be understood that considering that the patient object personal information embedded coding vector contains the basic characteristics of the patient (such as age, gender, weight, etc.), disease diagnosis related information and symptom description, etc., it reflects the individual attributes and overall health status of the patient. The blood test information associated coding vector focuses on the current patient's blood indicators, such as hemoglobin, coagulation related indicators, etc. These two types of information each have a unique focus. Therefore, in order to more comprehensively understand the patient's situation, the present application integrates multiple key information by performing feature domain fusion on the patient object personal information embedded coding vector and the blood test information associated coding vector to obtain the patient information-blood test information fusion semantic representation vector. In particular, in a specific example of the present application, the cross-attention network can be used to perform feature domain fusion on the patient object personal information embedding coding vector and the blood test information associated coding vector to achieve deep collaborative representation of individual patient characteristics and real-time physiological state, and obtain a fusion semantic representation vector between patient information and blood test information. The cross-attention network allows the model to dynamically evaluate the importance of information from different sources without losing the original information, and automatically adjust the weight distribution between each feature according to the current task requirements. Specifically, under this framework, the patient personal information embedding coding vector and the blood test information associated coding vector are input into the cross-attention module. The module constructs an attention distribution map by calculating the similarity or correlation between the two vectors. This distribution map reflects the importance of each type of information to the final fusion result, thereby ensuring that the information that is most critical to the prediction can obtain a higher weight. When calculating the attention score, the cross-attention network compares the two types of vectors and quantifies their mutual influence through a series of nonlinear transformations and dot product operations. For example, if a patient has a specific underlying disease (represented by the personal information embedding encoding vector) and his current hemoglobin level is lower than normal (represented by the blood test information association encoding vector), the cross-attention network can identify the possible connection between the two and give a corresponding high weight. This mechanism helps to reveal potential causal relationships or association patterns, beyond the simple feature superposition or averaging method. Finally, these features with different weights are combined through a weighted summation strategy to form a fusion semantic representation vector between patient information and blood test information. This vector not only integrates the individual characteristics and real-time physiological state of the patient, but also because of the introduction of the cross-attention mechanism, the model can more accurately capture the complex interaction between the two.

[0028] In the above-mentioned blood transfusion treatment plan recommendation system 100 based on historical information, the blood transfusion information encoding module 130 is used to embed and encode each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of blood transfusion information embedding encoding features, including: using an RNN-based embedding encoder to embed and encode each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of blood transfusion information embedding encoding vectors as the time series of blood transfusion information embedding encoding features. It should be understood that considering that the temporal dynamics of historical blood transfusion records contain key treatment response laws, the traditional processing of such time series information has essential defects. The historical blood transfusion data (such as transfusion time, type, transfusion reaction, etc.) in the EMR / TMIS system is usually stored in the form of discrete events. The traditional method only extracts the dominant features of a single blood transfusion (such as transfusion volume, blood product type), but ignores the dynamic association between multiple blood transfusion events (such as the nonlinear trend of antibody titer with increasing transfusion frequency). For example, the time series pattern of decreasing hemoglobin recovery after three consecutive red blood cell transfusions in the past six months in patients with chronic anemia may indicate that the production of alloantibodies in the body has shortened the survival of red blood cells, but traditional static analysis cannot capture such gradual evolution, resulting in subsequent transfusion plans still being recommended according to the initial dose, exacerbating the risk of ineffective transfusion. Therefore, in the technical solution of the present application, each transfusion information in the historical transfusion medical record information is embedded and encoded to obtain a time series of transfusion information embedded coding vectors. In particular, in a specific example of the present application, an RNN-based embedding encoder is used to embed and encode each transfusion information in the historical transfusion medical record information to obtain a time series of transfusion information embedded coding vectors. In particular, the RNN-based temporal embedding encoder can convert discrete transfusion events into a sequence of semantic vectors with temporal continuity. The characteristics of each transfusion event (such as transfusion time, type of blood product, hemoglobin increase 24 hours after transfusion, and allergic reaction grade) are mapped to high-dimensional vectors through the embedding layer. The recurrent structure of the RNN establishes long-range dependencies between events in this process: the gating mechanism (such as the update gate of the GRU) identifies the potential association between the abnormal decrease in platelet count after the third transfusion of the postoperative patient and the previous two plasma transfusions; the hidden state transfer mechanism tracks the absolute value change trend of lymphocytes after bone marrow transplant patients are transfused with irradiated red blood cells. In this way, the time series of the generated transfusion information embedding encoding vector can be used as a feature representation of the patient's transfusion history, recording the details and dynamic changes of the patient's past transfusion treatment.

[0029] In the above-mentioned blood transfusion therapy recommendation system 100 based on historical information, the dynamic matching analysis module 140 is used to perform dynamic matching analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion therapy recommendation report, wherein the dynamic matching analysis is used to: first perform dynamic matching of the blood transfusion information based on nonlinear spectral analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information, and then perform generation processing to obtain the blood transfusion therapy recommendation report.

[0030] Figure 3 FIG. 4 is a block diagram of a dynamic matching analysis module in a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application. Figure 3 As shown, in an embodiment of the present application, the dynamic matching analysis module 140 includes: a blood transfusion information dynamic query response unit 141, which is used to perform blood transfusion information dynamic matching based on nonlinear spectral analysis on the time series of the fusion semantic representation vector between the patient information and blood test information and the blood transfusion information embedded coding vector to obtain a blood transfusion information dynamic query response representation vector; a blood transfusion treatment plan recommendation unit 142, which is used to pass the blood transfusion information dynamic query response representation vector through a blood transfusion treatment plan recommender based on a large model to obtain the blood transfusion treatment plan recommendation report.

[0031] Figure 4 FIG. 4 is a block diagram of a dynamic query response unit for blood transfusion information in a blood transfusion therapy plan recommendation system based on historical information according to an embodiment of the present application. Figure 4As shown, in an embodiment of the present application, the dynamic query response unit 141 of blood transfusion information includes: a decision point state domain matrix degree matrix calculation subunit 1411, which is used to calculate the blood transfusion information decision point state class neighborhood matrix and the blood transfusion information decision point state class degree matrix between the fusion semantic representation vector between the patient information-blood test information and the time series of the blood transfusion information embedded coding vector respectively; a blood transfusion information decision point state Laplace matrix calculation subunit 1412, which is used to calculate the blood transfusion information decision point state Laplace matrix based on the blood transfusion information decision point state class neighborhood matrix and the blood transfusion information decision point state class degree matrix; a blood transfusion information characteristic spectrum decomposition adaptive fusion subunit 1413, which is used to perform spectrum decomposition and adaptive fusion on the blood transfusion information decision point state Laplace matrix to obtain the dynamic query response representation vector of blood transfusion information. It should be understood that considering that the fusion semantic representation vector between patient information and blood test information reflects the patient's current comprehensive condition, the time series of the blood transfusion information embedded coding vector records the historical blood transfusion situation and changes in physical response. However, the patient's condition is complex and changeable, and traditional linear analysis is difficult to capture the dynamic and nonlinear associations between these information. For example, in the treatment of chronic diseases, the efficacy of blood transfusion will produce complex fluctuations due to factors such as physical adaptability and the stage of disease progression. In traditional linear models, it is easy to simplify it into independent events, ignoring the dynamic balance mechanism behind it. Based on this, in the technical solution of the present application, the time series of the fusion semantic representation vector between the patient information and blood test information and the embedded coding vector of the blood transfusion information is dynamically matched based on nonlinear spectral analysis to obtain a dynamic query response representation vector of the blood transfusion information. Specifically, first, the nonlinear decision response embeds each transfusion information into a coding vector and interacts with the fused semantic representation vector between patient information and blood test information to generate an implicit code representing the "expected response of the t-th transfusion effect under the current pathological conditions". For example, the response pattern of INR dropping from 2.5 to 2.1 when 200 mL of plasma was transfused for the second time was implicitly represented under the current low albumin state. Subsequently, the decision point class neighborhood matrix and class degree matrix were constructed based on the implicitly coded time series to quantify the association strength between each historical event and the current state (such as the topological proximity between the third transfusion and the current coagulation abnormality is 0.87). The core component vector was extracted through Laplace matrix spectral decomposition to capture the cross-temporal and spatial correlation principal component of "the amount of plasma transfusion needs to increase as the albumin level decreases". Finally, the adaptive fusion mapped the core component vector into a dynamic query response vector, mapped the high-dimensional time series features into the low-dimensional spectral domain space, and revealed the potential evolution trajectory.

[0032] In an embodiment of the present application, the decision point state domain matrix degree matrix calculation subunit 1411 is used to: perform nonlinear decision response on each blood transfusion information embedded coding vector in the time series of the fusion semantic representation vector between the patient information and the blood test information and the blood transfusion information embedded coding vector to obtain a time series of blood transfusion information decision point state implicit coding vectors; calculate the blood transfusion information decision point state class neighborhood matrix based on the time series of blood transfusion information decision point state implicit coding vectors; calculate the blood transfusion information decision point state class degree matrix based on the time series of blood transfusion information decision point state implicit coding vectors.

[0033] Specifically, the semantic representation vector of the fusion between the patient information and the blood test information and each blood transfusion information embedding coding vector in the time series of the blood transfusion information embedding coding vector are subjected to nonlinear decision response to obtain a time series of implicit coding vectors of the blood transfusion information decision point states, which is expressed as the blood transfusion information nonlinear decision response formula:

[0034] in, is the time series of the transfusion information embedded coding vector, , , and are the first, second, and third time series of the blood transfusion information embedding coding vector. and The transfusion information is embedded in the coding vector, is the fusion semantic representation vector between patient information and blood test information, is matrix multiplication, and They are and The corresponding decision response weight matrix and decision response bias vector, is the activation function, , , , and are the first, second, and third in the time series of the implicit coding vector of the blood transfusion information decision point state. , and The implicit coding vector of the blood transfusion information decision point state, It is the time series of implicit coding vectors of the state of the decision point of blood transfusion information. It should be understandable that there are multi-dimensional and dynamically changing association patterns between the fusion semantic representation of patient information and blood test information and the temporal representation of embedded coding of blood transfusion information, such as the interactive influence of anemia degree and abnormal coagulation function, and the nonlinear trend of antibody level evolution after multiple transfusions. These patterns cannot be effectively represented by simple linear superposition or threshold judgment, which easily leads to deviations between the recommended scheme and the real needs of patients. Therefore, by performing nonlinear decision response, the complex medical data correlation that is difficult to capture by traditional methods is solved. Nonlinear decision response maps the heterogeneous features of the patient's real-time status and historical events to a high-dimensional space by constructing a latent space, using nonlinear mapping and interaction between features, and realizes the explicit modeling of deep interaction and potential associations between features in this space. For example, when a patient has hypoproteinemia caused by cirrhosis, the system can identify the nonlinear relationship between hemoglobin level and coagulation factor activity through interactions in the latent space, rather than viewing a single indicator in isolation. This processing method allows case characteristics that were originally difficult to distinguish in the original data space (such as thrombocytopenia caused by different causes) to form separable representation clusters in the latent space, while converting the complex evolution of the disease into a parseable geometric structure, thereby significantly improving the accuracy and robustness of subsequent matching analysis.

[0035] Specifically, based on the time series of the implicit coding vector of the blood transfusion information decision point state, the blood transfusion information decision point state class neighborhood matrix is ​​calculated, and the blood transfusion information decision point state class neighborhood matrix calculation formula is expressed as:

[0036] in, To calculate the Euclidean norm of a vector, is the inverse hyperbolic cosine function, , , , and are the eigenvalues ​​of each position in the state class neighborhood matrix of the blood transfusion information decision point, is the state-like neighborhood matrix of the blood transfusion information decision point. It should be understood that by constructing the state-like neighborhood matrix of the blood transfusion information decision point, the system maps each state implicit coding vector of the blood transfusion information decision point to a high-dimensional feature space, and uses the inverse hyperbolic cosine function to calculate the similarity between the state implicit coding vectors of the blood transfusion information decision point, thereby explicitly encoding the local connection relationship between the data points into a graph structure. This graph structure can capture the nonlinear association of multi-dimensional elements. In this way, the state semantic association is quantified by eigenvalue calculation. For example, when a patient has an allergic reaction after a blood transfusion event, the corresponding state implicit coding vector of the blood transfusion information decision point will form a high-similarity connection with the historical allergic event in the feature space, and the strength of this connection is accurately represented by the edge weight of the state-like neighborhood matrix of the blood transfusion information decision point. This representation capability enables the system to identify the pathological association hidden behind seemingly unrelated blood transfusion events. This semantic association modeling based on graph structure breaks through the limitations of traditional methods in temporal dynamic modeling and complex semantic mining, and provides a structural knowledge foundation for the subsequent generation of accurate and personalized blood transfusion plans. By converting discrete transfusion events into computable graph relationships, the system has made the leap from simple data matching to intelligent reasoning, significantly improving the scientific nature and timeliness of transfusion decisions.

[0037] Specifically, based on the time series of the implicit coding vector of the blood transfusion information decision point state, the blood transfusion information decision point state class matrix is ​​calculated, and the blood transfusion information decision point state class matrix calculation formula is expressed as:

[0038] in, To calculate the square of the vector-norm, yes The number of vectors in the equation is reduced by one. , and are the eigenvalues ​​of each position on the diagonal of the state class matrix of the blood transfusion information decision point, is the state class degree matrix of the blood transfusion information decision point. It should be understood that the calculation of the state class degree matrix of the blood transfusion information decision point can dynamically analyze the manifold characteristics of high-dimensional time-series blood transfusion data by constructing a structured measurement system in the form of a diagonal matrix, in which each diagonal element not only records the connection density of a specific blood transfusion event in the implicit coding space, but also quantifies the hub status of the blood transfusion information decision point in the global graph structure through a weighted stacking mechanism. For example, when a serious adverse reaction occurs after a blood transfusion, the degree value of the corresponding node will increase significantly. This dynamic adjustment mechanism enables the system to adaptively identify key decision nodes. The construction of the state class degree matrix of the blood transfusion information decision point transforms discrete time-series decision points into graph nodes with topological weights, and realizes the bidirectional mapping of local connection strength and global influence through degree encoding, and finally provides an adaptive parameter benchmark for the spectral decomposition of the state Laplace matrix of the blood transfusion information decision point. This design allows for the identification of high-frequency co-occurring clinical patterns (such as the correlation between platelet transfusion and DIC risk) through high-density connections when dealing with heterogeneous transfusion events, as well as the detection of low-probability but high-harm hidden associations (such as acute hemolytic reactions after transfusion of certain rare blood types) through sparse connections.

[0039] Specifically, the blood transfusion information decision point state Laplace matrix calculation subunit 1412 is used to calculate the blood transfusion information decision point state Laplace matrix based on the blood transfusion information decision point state class neighborhood matrix and the blood transfusion information decision point state class degree matrix, and the blood transfusion information decision point state Laplace matrix calculation formula is expressed as:

[0040] in, is the state Laplace matrix of the transfusion information decision point. It should be understood that by fusing the previously constructed state neighborhood matrix of the transfusion information decision point and the state degree matrix of the transfusion information decision point, a transition modeling from local connection relationship to global structural characteristics can be achieved, wherein the state neighborhood matrix of the transfusion information decision point encodes the state semantic association between the transfusion information decision points, while the state degree matrix of the transfusion information decision point quantifies the centrality weight of each node in the graph structure. The synergistic effect of the two enables the state Laplace matrix of the transfusion information decision point to accurately characterize the manifold topological characteristics of the transfusion event sequence. In clinical decision-making, this representation transformation not only extracts low-frequency patterns reflecting the periodicity of transfusion reactions (such as the hemoglobin recovery curve every 4 weeks in patients with chronic anemia) through eigenvalue decomposition, but also captures implicit associations across treatment courses (such as the delayed effect of changes in coagulation factor activity after plasma transfusion) through high-order eigenvectors.

[0041] In an embodiment of the present application, the transfusion information feature spectrum decomposition adaptive fusion subunit 1413 is used to: perform matrix optimization based on a topological closure operator on the transfusion information decision point state Laplace matrix to obtain an optimized transfusion information decision point state Laplace matrix; perform spectral decomposition on the optimized transfusion information decision point state Laplace matrix to obtain a set of transfusion information decision point core component coding vectors; and perform adaptive fusion on the set of transfusion information decision point core component coding vectors to obtain the transfusion information dynamic query response representation vector.

[0042] Specifically, the blood transfusion information decision point state Laplace matrix is ​​optimized based on a topological closure operator to obtain an optimized blood transfusion information decision point state Laplace matrix, which is expressed as:

[0043] in, is the blood transfusion information cut set space matrix, yes The inverse matrix of is the blood transfusion information circulation space matrix, yes The inverse matrix of The natural constant The exponential function value with base , It is added by location point. It is the optimized Laplace matrix of the transfusion information decision point state. It should be understood that in the transfusion treatment plan recommendation system based on historical information, the state of the transfusion information decision point can be represented by a cluster of poles, and its corresponding Laplace matrix of the transfusion information decision point state is composed of the union of all pole clusters in the topological space. In order to enhance the stability of the topological features, it is necessary to extract the connected domain features of the class neighborhood matrix and the class degree matrix of the transfusion information decision point state respectively through generalized inverse linear transformation, wherein the generalized inverse result of the class neighborhood matrix of the transfusion information decision point state is projected to the cut set space to capture the key connectivity features under the network disconnection threshold, and the generalized inverse result of the class degree matrix of the transfusion information decision point state is mapped to the circulation space to extract the flow conservation characteristics of the loop structure. Based on the algebraic operation fusion of the cut set closure and the circulation closure, the optimized Laplace matrix of the transfusion information decision point state is generated by matrix multiplication. The optimization process uses the synergistic effect of generalized inverse closure to effectively capture the global topological features (such as the distribution law of connected components and the topological features of the loop path) in the network topology architecture, significantly improving the structural representation stability of the Laplacian matrix in the spectral space, making it more resistant to noise interference and local structural distortion in clinical transfusion data. Specifically, the minimum cut set threshold between transfusion decision points can be quantified through the cut set closure operation, reflecting the critical conditions for different transfusion events to trigger pathological state transfer; while the circulation closure operation can characterize the cyclic dependency of the transfusion reaction chain, such as the feedback loop between abnormal iron metabolism and liver function damage after red blood cell transfusion. The optimized transfusion information decision point state Laplacian matrix generated by the two collaboratively retains the connectivity characteristics of the original graph structure and enhances the dynamic response characteristics of the loop path. This dual closure mechanism enables the system to identify explicit associations between transfusion schemes (such as the linear correlation between platelet transfusion and DIC score) and capture implicit associations (such as the trend pattern of non-monotonic changes in fibrinogen levels after cryoprecipitate transfusion) when processing heterogeneous transfusion data. Through the collaborative synthesis of cut set closure and circulation closure, the system constructs an optimized transfusion information decision point state Laplace matrix with both local connectivity robustness and global topological consistency to overcome the topological noise and dynamic coupling imbalance problems that are common in clinical transfusion data manifolds.

[0044] Specifically, the optimized blood transfusion information decision point state Laplace matrix is ​​spectrally decomposed to obtain a set of core component encoding vectors of the blood transfusion information decision point, which is expressed as follows using the blood transfusion information decision point spectral decomposition formula:

[0045] in, For Perform spectral decomposition operation, is the blood transfusion information feature vector matrix, i.e., the set of the core component encoding vectors of the blood transfusion information decision point. , , and are the first, second, and third core component encoding vectors of the blood transfusion information decision point. and The core component encoding vector of the blood transfusion information decision point, The diagonal elements of the matrix are , , and The diagonal matrix of eigenvalues ​​of transfusion information, , , and They are , , and The corresponding eigenvalues ​​are is the diagonal matrix of eigenvalues ​​of transfusion information. It should be understood that the spectral decomposition operation converts the complex graph structure semantics into solvable frequency domain features. Specifically, the spectral decomposition operation extracts the intrinsic structure of the data manifold through eigenvalue decomposition, in which the low-dimensional encoding set consisting of the eigenvectors corresponding to the smallest eigenvalues ​​not only captures the periodic law of the transfusion event sequence (such as the hemoglobin fluctuation baseline after weekly red blood cell transfusion in patients with chronic kidney disease), but also reveals the implicit correlation pattern across treatment courses (such as the time lag effect of changes in coagulation factor activity after plasma transfusion). This frequency domain representation breaks through the window limitation of traditional time series analysis methods, enabling the system to separate the core features directly related to transfusion reactions from mixed clinical signals. Spectral decomposition can map the topological features of the graph structure (such as node centrality and path connectivity) into quantifiable frequency domain components. For example, the second smallest eigenvector can be used to identify the high-risk patient subgroup with acute hemolytic reaction within 3 hours after transfusion, and a dynamic weight model can be constructed based on the eigenvalue decay characteristics to give higher frequency domain response weights to recent key transfusion events. This conversion mechanism enables the system to process multi-source heterogeneous data streams in real time in emergency blood transfusion scenarios.

[0046] Specifically, the set of the core component encoding vectors of the blood transfusion information decision point is adaptively fused to obtain the blood transfusion information dynamic query response representation vector, which is expressed as follows using the blood transfusion information adaptive fusion formula:

[0047] in, For Perform adaptive fusion operations. and They are The corresponding fusion weight matrix and fusion bias vector, yes function, yes The corresponding blood transfusion information scoring weight vector, yes The corresponding blood transfusion information weight value, is a masking operation, is the preset threshold, yes The corresponding transfusion information mask weight value, is the dynamic query response representation vector of transfusion information. It should be understood that adaptive fusion realizes the mapping transition from discrete graph features to continuous clinical status through the synergy of attention mechanism and gated architecture, in which each core component encoding vector of transfusion information decision point contains both the time series pattern of specific transfusion events (such as the change curve of prothrombin time after platelet transfusion) and cross-dimensional pathological associations (such as the coupling relationship between red blood cell transfusion and ferritin metabolism). The adaptive fusion mechanism enables the system to automatically reconstruct feature combination strategies for different clinical scenarios (such as acute blood loss and chronic anemia) through dynamic weight allocation. For example, when the patient's hemoglobin concentration drops sharply, the low-frequency encoding vector reflecting the relationship between red blood cell transfusion frequency and tissue oxygen supply will be strengthened first; when sepsis patients have a tendency to DIC, the high-frequency feature subset representing the consumption of coagulation factors and the timing of platelet transfusion will be activated. The core component encoding vector of the transfusion information decision point constructs a dynamic representation vector with both global structural features and local temporal characteristics through a context-aware fusion strategy. This hybrid encoding approach enables the system to break through the static limitations of traditional methods.

[0048] Specifically, the transfusion therapy plan recommendation unit 142 is used to pass the transfusion information dynamic query response representation vector through the transfusion therapy plan recommender based on the large model to obtain the transfusion therapy plan recommendation report. That is, although the transfusion information dynamic query response representation vector integrates the patient's current status and historical transfusion related information, the complex medical knowledge and potential treatment strategies contained therein require strong reasoning capabilities to mine. During the pre-training process, the large model learned a large amount of medical literature and case data, covering a rich transfusion therapy knowledge system. It can associate the features in the vector with its own knowledge reserves, perform deep reasoning, and integrate information scattered in different data fragments, so as to give comprehensive and reasonable suggestions for transfusion therapy plans. For example, in the face of complex conditions (such as transfusion needs of patients with multiple complications), the large model can comprehensively consider various factors and give accurate solutions that are difficult to derive by conventional methods. In this way, with the transfusion information dynamic query response representation vector as input, the large model uses its own knowledge and reasoning ability to fully consider the individual characteristics of the patient, the current condition and the historical transfusion situation, and generate a highly personalized transfusion therapy plan recommendation report. The report covers the recommended blood components (specific types and proportions of red blood cells, plasma, platelets, etc.), transfusion doses, transfusion schedules, and possible complication prevention measures. For example, for an elderly patient with cirrhosis and coagulation disorders, the large model can provide a detailed plan accurate to the components and doses of each transfusion based on his unique physical condition and transfusion history. Among them, the transfusion treatment recommendation report takes into account the individual characteristics of the patient, the current condition, and the historical transfusion situation, and can provide clinicians with detailed and accurate guidance to develop the most suitable transfusion treatment plan for the patient. Specifically, the report will detail the recommended blood component types and their proportions. For example, when facing patients with anemia and coagulation disorders, it may be recommended to use red blood cell suspensions combined with fresh frozen plasma or specific types of concentrated coagulation factors to improve the anemic state and correct coagulation abnormalities. For patients with severe thrombocytopenia, specific recommendations for the use of platelet preparations may be included. Moreover, in terms of dosage, the report will accurately calculate the appropriate dosage required for each transfusion based on factors such as the patient's weight, age, gender, and specific health conditions, and propose a strategy for either fractional infusion or a single large-volume infusion. For example, for some patients with heart failure, in order to avoid excessive heart burden, the report may recommend the use of fractional small-volume infusions. In addition, considering the possible complications or special physical conditions (such as allergic reactions) of the patient, the report will also provide preventive measures and treatment plans to ensure the safety of the transfusion process. Regarding the timing of transfusions, the report also provides a detailed plan. It not only points out the best time for transfusions, but may also involve recommendations on the frequency of transfusions. For example, for patients with chronic diseases who require long-term supportive treatment, the report will design a continuous transfusion schedule based on the trend of their condition in order to maintain a stable treatment effect.At the same time, in order to respond to acute illness attacks or surgical needs, the report may also include a rapid response mechanism for emergency blood transfusions. It is worth noting that the report also covers the evaluation and management of potential risk factors. If the patient has a history of producing alloantibodies, the report will emphasize the importance of choosing blood products with a higher degree of matching and may recommend additional immunological testing. For patients who are at risk of iron overload due to repeated blood transfusions, the report will recommend the implementation of appropriate monitoring measures and timely adjustment of treatment plans, such as the introduction of drug-assisted therapies such as deferoxamine. The blood transfusion treatment plan recommendation report integrates multi-source data and uses advanced algorithm models and medical knowledge to develop a comprehensive and detailed treatment strategy for each patient.

[0049] In summary, the blood transfusion therapy recommendation system 100 based on historical information according to the embodiment of the present application is explained, which first extracts the patient's basic information and diagnostic information, real-time blood test information and historical blood transfusion medical record information from the HIS, LIS, EMR and TMIS systems, and then performs joint semantic analysis on the basic information and diagnostic information and real-time blood test information, and converts the historical blood transfusion medical record information into a dynamic vector sequence. Then, the current blood indicators and the historical blood transfusion time series are dynamically matched in the frequency domain space through nonlinear spectral analysis, and combined with the implicit reasoning ability of the pre-trained medical large model, a personalized recommendation plan is generated to break through the limitations of the traditional system in static parameter setting, individual adaptability and semantic mining, and significantly improve the accuracy and timeliness of blood transfusion therapy decisions.

[0050] As described above, the blood transfusion therapy scheme recommendation system 100 based on historical information according to the embodiment of the present application can be implemented in various terminal devices. In one example, the blood transfusion therapy scheme recommendation system 100 based on historical information can be integrated into the terminal device as a software module and / or a hardware module. For example, the blood transfusion therapy scheme recommendation system 100 based on historical information can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the blood transfusion therapy scheme recommendation system 100 based on historical information can also be one of the many hardware modules of the terminal device.

[0051] Alternatively, in another example, the blood transfusion therapy recommendation system 100 based on historical information and the terminal device may also be separate devices, and the blood transfusion therapy recommendation system 100 based on historical information may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0052] Figure 5 FIG. 1 is a flow chart of a method for recommending a blood transfusion treatment plan based on historical information according to an embodiment of the present application. Figure 5As shown, according to the method for recommending a blood transfusion treatment plan based on historical information in an embodiment of the present application, it includes: S110, obtaining basic information and diagnostic information of a patient object from an HIS system, obtaining real-time blood test information from an LIS system, and obtaining historical blood transfusion medical record information from an EMR system and a TMIS system; S120, performing semantic fusion based on feature coding on the basic information and diagnostic information of the patient object and the real-time blood test information to obtain a fused semantic representation between patient information and blood test information; S130, embedding coding each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of embedded coding features of the blood transfusion information; S140, performing dynamic matching analysis on the time series of the fused semantic representation between patient information and blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion treatment plan recommendation report.

[0053] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned method for recommending a blood transfusion treatment plan based on historical information have been described in detail above. Figures 1 to 4 The description of the blood transfusion therapy plan recommendation system based on historical information has been described in detail, and therefore, its repeated description will be omitted.

[0054] In summary, the method for recommending blood transfusion treatment plans based on historical information based on the embodiment of the present application is explained, which first extracts basic patient information and diagnostic information, real-time blood test information and historical blood transfusion medical record information from HIS, LIS, EMR and TMIS systems, then performs joint semantic analysis on the basic information and diagnostic information as well as real-time blood test information, and converts the historical blood transfusion medical record information into a dynamic vector sequence. Then, the current blood indicators are dynamically matched with the historical blood transfusion time series in the frequency domain space through nonlinear spectral analysis, and combined with the implicit reasoning ability of the pre-trained medical large model, a personalized recommendation plan is generated to break through the limitations of traditional systems in static parameter setting, individual adaptability and semantic mining, and significantly improve the accuracy and timeliness of blood transfusion treatment decisions.

[0055] In the 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 only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0056] 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 may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0057] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0058] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0059] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0060] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0061] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit of the technical solution of the present invention.

Claims

1. A blood transfusion therapy recommendation system based on historical information, characterized in that: include: Information collection module, used to obtain basic information and diagnosis information of patients from HIS system, obtain real-time blood test information from LIS system, and obtain historical blood transfusion medical record information from EMR system and TMIS system; A semantic fusion module, used for performing semantic fusion of the basic information and diagnostic information of the patient object with the real-time blood test information based on feature coding to obtain a fusion semantic representation between the patient information and the blood test information; A blood transfusion information encoding module, used for embedding and encoding each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of blood transfusion information embedding encoding features; A dynamic matching analysis module is used to perform dynamic matching analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion treatment plan recommendation report, wherein the dynamic matching analysis is used to: first perform dynamic matching of the blood transfusion information based on nonlinear spectral analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information, and then perform generation processing to obtain the blood transfusion treatment plan recommendation report.

2. The blood transfusion therapy recommendation system based on historical information according to claim 1, characterized in that: The semantic fusion module comprises: A patient object personal information embedding coding unit, used for embedding and coding the basic information and diagnosis information of the patient object to obtain a patient object personal information embedding coding vector; A real-time blood test information associated coding unit, used for passing the real-time blood test information through a blood test information associated encoder based on a fully connected layer to obtain a blood test information associated coding vector; The patient information-blood test information feature domain fusion unit is used to perform feature domain fusion on the patient object personal information embedding coding vector and the blood test information associated coding vector to obtain a fused semantic representation vector between the patient information and the blood test information as the fused semantic representation between the patient information and the blood test information.

3. The blood transfusion therapy recommendation system based on historical information according to claim 2, characterized in that: The patient object personal information embedding encoding unit is used to: use a structured-text dual-stream encoder based on the TabTransformer-BioBERT model to embed the basic information and diagnosis information of the patient object to obtain the patient object personal information embedding encoding vector.

4. The blood transfusion therapy recommendation system based on historical information according to claim 1, characterized in that: The blood transfusion information encoding module includes: using an RNN-based embedding encoder to embed code each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of blood transfusion information embedding coding vectors as the time series of blood transfusion information embedding coding features.

5. The blood transfusion therapy recommendation system based on historical information according to claim 4, characterized in that: The dynamic matching analysis module includes: A blood transfusion information dynamic query response unit, used for performing a blood transfusion information dynamic matching based on nonlinear spectrum analysis on the time series of the fusion semantic representation vector between the patient information and the blood test information and the blood transfusion information embedded coding vector to obtain a blood transfusion information dynamic query response representation vector; A blood transfusion therapy plan recommendation unit is used to pass the blood transfusion information dynamic query response characterization vector through a blood transfusion therapy plan recommender based on a large model to obtain the blood transfusion therapy plan recommendation report.

6. The blood transfusion therapy recommendation system based on historical information according to claim 5, characterized in that: The blood transfusion information dynamic query response unit comprises: A decision point state domain matrix degree matrix calculation subunit, used to respectively calculate a blood transfusion information decision point state class neighborhood matrix and a blood transfusion information decision point state class degree matrix between the patient information-blood test information fusion semantic representation vector and the time series of the blood transfusion information embedding coding vector; A blood transfusion information decision point state Laplace matrix calculation subunit, used to calculate the blood transfusion information decision point state Laplace matrix based on the blood transfusion information decision point state class neighborhood matrix and the blood transfusion information decision point state class degree matrix; The blood transfusion information feature spectrum decomposition and adaptive fusion subunit is used to perform spectrum decomposition and adaptive fusion on the blood transfusion information decision point state Laplace matrix to obtain the blood transfusion information dynamic query response representation vector.

7. The blood transfusion therapy recommendation system based on historical information according to claim 6, characterized in that: The decision point state domain matrix degree matrix calculation subunit is used to: Perform nonlinear decision response on each blood transfusion information embedding coding vector in the time series of the fusion semantic representation vector between the patient information and the blood test information and the blood transfusion information embedding coding vector to obtain a time series of implicit coding vectors of blood transfusion information decision point states; Calculating the blood transfusion information decision point state class neighborhood matrix based on the time series of the implicit coding vector of the blood transfusion information decision point state; Based on the time series of the implicit coding vector of the blood transfusion information decision point state, the class matrix of the blood transfusion information decision point state is calculated.

8. The blood transfusion therapy plan recommendation system based on historical information according to claim 7, characterized in that: The blood transfusion information characteristic spectrum decomposition adaptive fusion subunit is used to: Performing matrix optimization on the blood transfusion information decision point state Laplace matrix based on a topological closure operator to obtain an optimized blood transfusion information decision point state Laplace matrix; Performing spectral decomposition on the optimized blood transfusion information decision point state Laplace matrix to obtain a set of core component coding vectors of the blood transfusion information decision point; The set of the core component encoding vectors of the blood transfusion information decision point is adaptively fused to obtain the dynamic query response representation vector of the blood transfusion information.

9. A method for recommending a blood transfusion treatment plan based on historical information, characterized in that: include: Obtain basic information and diagnosis information of patients from the HIS system, obtain real-time blood test information from the LIS system, and obtain historical blood transfusion medical record information from the EMR system and TMIS system; Performing semantic fusion based on feature coding on the basic information and diagnostic information of the patient object and the real-time blood test information to obtain a fusion semantic representation between the patient information and the blood test information; Embedding and coding each blood transfusion information in the historical blood transfusion medical record information to obtain a time series of embedded coding features of the blood transfusion information; A dynamic matching analysis is performed on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information to obtain a blood transfusion treatment plan recommendation report, wherein the dynamic matching analysis is used to: first perform dynamic matching of the blood transfusion information based on nonlinear spectral analysis on the time series of the fused semantic representation between the patient information and blood test information and the embedded coding features of the blood transfusion information, and then perform generation processing to obtain the blood transfusion treatment plan recommendation report.

10. The method for recommending a blood transfusion treatment plan based on historical information according to claim 9, characterized in that: The basic information and diagnostic information of the patient object and the real-time blood test information are semantically fused based on feature coding to obtain a fused semantic representation between the patient information and the blood test information, including: Embedding and encoding the basic information and diagnosis information of the patient object to obtain an embedded encoding vector of the patient object personal information; The real-time blood test information is passed through a blood test information associated encoder based on a fully connected layer to obtain a blood test information associated encoding vector; The patient object personal information embedding coding vector and the blood test information associated coding vector are fused in feature domain to obtain a fused semantic representation vector between patient information and blood test information as the fused semantic representation between patient information and blood test information.

Images