Intelligent recommendation system for personalized nursing scheme of patient in department of cardiology and optimization method of intelligent recommendation system

Abstract

The invention relates to the technical field of nursing scheme intelligent recommendation systems, in particular to a cardiology patient personalized nursing scheme intelligent recommendation system optimization method, which comprises a dynamic rare case recognition engine for accurately capturing and recognizing cardiology rare cases by using medical data and text data, and comprises a data preparation module, a data retrieval module, a data processing module and a data processing module, the candidate sample classification and recognition module is used for recognizing whether rare cases exist or not, and the problems existing in the prior art are effectively solved by introducing innovation modules such as a dynamic rare case recognition engine, a knowledge graph builder and a rare case simulation generator. According to the invention, continuous learning and updating can be realized, rare cases can be accurately identified, and customized nursing suggestions can be provided based on individual features of patients. Meanwhile, the decision-making process of the system has good interpretability, decision-making basis can be provided for doctors, and usability and credibility of the system in clinical practice are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recommendation systems for nursing plans, and in particular to an optimization method for intelligent recommendation systems for personalized nursing plans for patients in cardiology departments. Background Art

[0002] With the continuous advancement of medical technology, the diagnosis and treatment of cardiology diseases have been continuously improved. However, in the face of the increasingly complex spectrum of cardiovascular diseases and individual differences among patients, traditional diagnosis and treatment methods still face huge challenges. In particular, in the formulation of personalized care plans, existing technologies still have many shortcomings.

[0003] Currently, the field of cardiology relies mainly on experienced specialists to make diagnoses and develop treatment plans. Although this approach works well for common cases, it often falls short when faced with complex or rare cases. Doctors' personal experience and knowledge limitations may lead to missed or misdiagnoses, especially when dealing with rare cardiovascular diseases. In addition, traditional methods are difficult to fully consider individual differences in patients, and often use a one-size-fits-all standardized treatment plan, which cannot truly achieve personalized medicine.

[0004] In recent years, with the development of artificial intelligence technology, some auxiliary diagnosis systems based on machine learning have begun to be applied in the field of cardiology. These systems have improved the accuracy and efficiency of diagnosis to a certain extent. However, there are still some key problems with existing intelligent diagnosis systems. First, most of them are trained based on static data, lack the ability of dynamic learning and updating, and it is difficult to absorb the latest medical discoveries and clinical experience in a timely manner. Secondly, these systems perform poorly when dealing with rare cases because the sample size of rare cases in the training data is insufficient. Furthermore, existing systems often focus only on disease diagnosis and have limited functions in the formulation of personalized care plans.

[0005] In addition, existing intelligent diagnosis systems generally lack interpretability. They are usually black-box and cannot provide doctors with a basis for decision-making, which greatly limits their application in clinical practice. At the same time, these systems often ignore the structured expression of medical knowledge and find it difficult to fully utilize the existing medical knowledge system.

[0006] In view of the above problems, there is an urgent need for an intelligent recommendation system that can dynamically learn, accurately identify rare cases, provide personalized care plans, and have good explainability. Summary of the invention

[0007] The present invention is proposed to solve these technical problems. The present invention proposes an intelligent recommendation system for personalized nursing plans for cardiology patients and an optimization method thereof, which can not only accurately diagnose common and rare cardiology diseases, but also provide patients with highly personalized nursing plans.

[0008] The present invention provides an intelligent recommendation system for personalized nursing plans for cardiology patients, which includes: a dynamic rare case identification engine, which uses medical data and text data to accurately capture and identify rare cardiology cases, and includes: a data preparation module, which obtains internal hospital data and performs data preprocessing for classification of cardiology diseases according to specified cardiology diagnosis and treatment guidelines, the preprocessing including multi-domain union of data of multiple modalities, unifying data formats, and constructing a globally unique identifier for each patient; a data retrieval module, which first screens an initial data set, then matches based on a dynamic index, sets a similarity threshold and a candidate sample retrieval quantity to control the recognition accuracy, and outputs the retrieved candidate samples that meet the similarity threshold or the retrieved quantity; a candidate sample classification and recognition module, which is used to identify whether it is a rare case, outputs a retrieval result if it is identified as a rare case, and feeds the recognition result back to the data preparation module as a training sample set, so that the data preparation module preprocesses more data, and at the same time, the similarity threshold of the retrieval module and the number of samples to be retrieved are adaptively adjusted.

[0009] Specifically, the dynamic rare case identification engine includes: a data preprocessing preparation module, which preprocesses the cardiology disease classification data and text data within the hospital; a data retrieval module, which is given a data set consisting of a group of patient data, and matches the data based on a dynamic index to find similar case groups and store them in the memory; based on the similarity calculation method, find the candidate patients with a high probability of belonging to rare cases. If a similarity threshold t is given, there are K data of patient data p that satisfy the similarity sim(p,q)>t, then the probability P=p / K that p is a rare case; a candidate sample classification and identification module, if the patient is judged according to the data retrieval, If the patient data is a high-probability patient of a rare case, it will be regarded as rare case data and added to the training data set for training the classifier; at the same time, the similarity between the retrieved patient p and the high-probability patient q regarded as a rare case is calculated in the range of (0, t), then p is the confidence R = p / K of the rare case, when R is greater than a certain threshold λ, λ is determined by the training data, then p should be regarded as a rare case, if the proportion of patients regarded as rare cases has reached the threshold, then the similarity threshold t should be adjusted, after the adjustment, the amount of data retrieved is reduced, and when the number of rare cases increases, the data retrieval module should retrieve larger scale data.

[0010] Specifically, the medical data includes patient imaging data, and the data content is represented in DICOM format. The DI COM information includes text content, basic patient information, identifiers and pixel information; the text data content is represented in PDF format, and the text content is extracted by image conversion. The data content includes basic patient information and doctor's diagnosis opinions. The collected text content data is carried by imaging reports of X-ray, CT and MRI images, and the PDF format text content is extracted by optical character recognition.

[0011] Specifically, a rare case knowledge graph builder based on the dynamic rare case identification engine is used to construct a rare case knowledge graph and multi-example discovery based on knowledge reasoning. The rare case knowledge graph builder performs multi-domain union, format conversion and information extraction on the input cardiology data, and uses the existing standard medical record knowledge graph to convert the obtained cardiology data into the form in the knowledge graph, thereby constructing a cardiology data knowledge graph with added value, including: an information mining module, which includes using a data preprocessing preparation module to obtain clinical data information and text information, using natural language processing and data analysis methods to mine Mining the cardiology classification data and text description information; A fusion module of cardiology classification data and text data based on the knowledge graph, which incorporates the classification data and text data into the same knowledge graph for subsequent processing and application; A rare case simulation generator based on the rare case knowledge graph builder: By constructing a knowledge graph, matching rare cases are found for clinical control and guidance, and the clinical diagnosis and treatment data are annotated with additional knowledge graph information to construct guiding medical data, a simulation module, which combines the guiding medical data constructed by matching rare cases with a simulated virtual human body to achieve simulation realization.

[0012] Specifically, the adaptive learning optimizer based on the rare case simulation generator includes an interactive module, which is used to optimize the simulation module, and optimizes it according to the case data and the given diagnostic data provided by the user, and is also used for the multi-expert learning module; the multi-expert learning module learns multiple experts through machines, obtains the processing methods of different experts to optimize the matching and simulation capabilities of the simulation module, and the learning module generates optimized matching and simulation samples for the input case data based on the simulation module and the multi-expert module, and judges whether it is a potential rare case based on the case learning data and the corresponding diagnostic data, so as to optimize the simulation module.

[0013] Specifically, the multi-expert integrated decision system based on the adaptive learning optimizer learns different expert information to improve the simulation capability, so that the system can simulate the expert system of the doctor's decision-making ability. The decision system converges after the simulation capability reaches a certain amount of training, and the final output case matching information and diagnosis information are provided to the user. The multi-expert integrated system makes auxiliary judgments based on the existing knowledge graph and outputs nursing plans, which include: a multi-expert decision fusion module for the multi-expert module integration process, which constructs guiding medical data based on the processing methods from different experts and provides it to the simulation module.

[0014] The optimization method for realizing the intelligent recommendation system for personalized nursing plans for cardiology patients is implemented in the following specific steps:

[0015] S1, inputting the patient's cardiology data and diagnosis data, processing the data by the dynamic rare case identification engine, and inputting the processed data into the case library and knowledge base;

[0016] S2, output the existing cases through the case library, and the user confirms the analysis results;

[0017] S3, data is input into the knowledge base to obtain additional information, and then passes through the simulation generator to generate optimized case results and case similarity feedback;

[0018] S4, adjusts the output results of the simulation generator according to the real case results and the simulation case results, determines whether it is a potential rare case, outputs expert opinions, provides them to users, and inputs them into the knowledge base at the same time.

[0019] Specifically, the functional modules of the dynamic rare case identification engine, the rare case simulation generator based on the rare case knowledge graph builder, the adaptive learning optimizer based on the rare case simulation generator, and the multi-expert integrated decision system based on the adaptive learning optimizer are all described by a knowledge graph, which includes multiple relationships and multiple entities.

[0020] Specifically, the simulation learning model based on the rare case simulation generator and the adaptive learning optimizer is based on Transformer. The model needs to optimize the number of data samples, that is, the quality and quantity of samples recognized in advance. A large amount of learning data is required to reach a convergence state, and a higher efficiency output is required to achieve a better state. The loss function used in the learning process is the cross entropy loss:

[0021]

[0022] y is the true result vector, Y is the model prediction value vector, i is the sample, N is the number of samples, C is the number of categories, y ij represents the true probability that the i-th sample belongs to the j-th class, Yij It represents the probability that the model predicts that the i-th sample belongs to the j-th class.

[0023] Specifically, the adaptive learning process is as follows: a. input diagnostic data T and disease classification data y, and input them into the matching module to obtain candidate results L and candidate samples X; b. output patient treatment method D; c. determine the inconsistent information between D and the knowledge graph in T, and generate a new auxiliary knowledge graph and input it into the simulation module; d. perform multi-field expert training on the diagnostic result D. If the expert result D reaches a certain training amount and meets the convergence condition, the training is terminated and the auxiliary information is output; e. The simulation output result is auxiliary information, and the similarity L is given, and the real result Y of the case is given as the loss, and the process is repeated until convergence; f. Output case D, and the case output similarity L; g. Perform knowledge graph learning on candidate sample X and case D, extract medical information assisted by the knowledge graph and input it into the case knowledge base; h. Determine whether sample X is marked as a rare disease. If so, randomly select other cases and add them to the training set; i. Adjust the sample size of the matching module and output the result.

[0024] Advantages and beneficial effects of the present invention:

[0025] The present invention effectively solves the problems existing in the prior art by introducing innovative modules such as a dynamic rare case identification engine, a knowledge graph builder, and a rare case simulation generator. It can continuously learn and update, accurately identify rare cases, and provide customized nursing advice based on individual patient characteristics. At the same time, the decision-making process of this system has good interpretability, which can provide doctors with a basis for decision-making, greatly improving the usability and credibility of the system in clinical practice.

[0026] In general, the present invention has brought revolutionary improvements to the diagnosis and care of cardiology patients. It can not only significantly improve the accuracy of diagnosis, especially the ability to identify rare cases, but also provide the most suitable personalized care plan for each patient. This system is expected to become a powerful assistant for cardiologists, greatly improve the diagnosis and treatment level and quality of life of cardiology patients, and make an important contribution to promoting the development of cardiology medical care in a more accurate and personalized direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is the overall logic block diagram of the system of the present invention.

[0028] Figure 2 This is a logic block diagram of the dynamic rare case identification engine of the present invention.

[0029] Figure 3 This is a logical block diagram of the rare case knowledge graph builder of the present invention.

[0030] Figure 4It is a logic block diagram of the rare case simulation generator of the present invention.

[0031] Figure 5 It is a logical block diagram of the adaptive learning optimizer of the present invention.

[0032] Figure 6 It is a logic block diagram of the multi-expert integrated decision-making system of the present invention.

[0033] Figure 7 It is a logic block diagram of the optimization method of the present invention. DETAILED DESCRIPTION

[0034] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0035] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0036] See also Figure 1-7 , the present invention discloses an intelligent recommendation system for personalized nursing plans for cardiology patients, including a dynamic rare case identification engine 100, which uses medical data and text data to accurately capture and identify rare cardiology cases, and includes: a data preparation module 101, which obtains internal hospital data and performs data preprocessing for classification of cardiology diseases according to designated cardiology diagnosis and treatment guidelines, the preprocessing includes multi-domain union of data of multiple modalities, unifying data formats, and constructing a globally unique identifier for each patient;

[0037] The data retrieval module 102 first screens the initial data set, then performs matching based on the dynamic index, sets a similarity threshold and a number of candidate sample retrievals to control the recognition accuracy, and outputs the retrieved candidate samples that meet the similarity threshold or the number of retrieved samples;

[0038] The candidate sample classification and identification module 103 is used to identify whether it is a rare case. If it is identified as a rare case, the search result is output and the recognition result is fed back to the data preparation module 101 as a training sample set, so that the data preparation module 101 can pre-process more data and adaptively adjust the similarity threshold and the number of samples to be retrieved of the search module 102.

[0039] First, the core of the present invention is the dynamic rare case identification engine 100. The engine analyzes medical data and text data through intelligent algorithms to accurately identify rare cardiology cases. This function is crucial for personalized care of cardiology patients, because rare cases often require special care plans.

[0040] Secondly, the data preparation module 101 plays a key role in the entire system. It not only obtains internal hospital data, but also pre-processes the data according to the specified cardiology diagnosis and treatment guidelines. For example, for a patient with myocarditis, the module may integrate multimodal data such as blood test results, electrocardiogram data, and echocardiogram results, and unify them into a standard format. The advantage of this is that it provides a standardized data basis for subsequent data analysis, greatly improving the accuracy and efficiency of the analysis.

[0041] Then, the data retrieval module 102 comes into play. It first screens the initial data set, for example, it may select cardiology case data within the last five years. Then, matching is performed based on the dynamic index. In this process, the system sets the similarity threshold and the number of candidate sample retrievals. For example, we may set the similarity threshold to 0.85 and the number of candidate sample retrievals to 200. This means that the system will find the first 200 cases with a similarity greater than 0.85 with the current patient from the database as candidate samples. This method not only ensures the accuracy of the matching, but also controls the amount of calculation, thereby improving the overall efficiency of the system.

[0042] Finally, the candidate sample classification and identification module 103 is responsible for identifying rare cases. If a rare case is identified, such as a rare type of myocardial lesion, the system will immediately output the search results and feed back the case as a new training sample to the data preparation module 101. This feedback mechanism enables the system to continuously learn and optimize, improving its ability to identify rare cases. At the same time, the system will automatically adjust the similarity threshold and the number of samples to be retrieved of the data retrieval module 102 according to the newly added rare cases. For example, if the recognition rate of a certain type of rare case is found to be low, the system may lower the similarity threshold or increase the number of samples to be retrieved to improve the capture rate.

[0043] The dynamic rare case identification engine 100 includes: a data preprocessing preparation module 104, which preprocesses the cardiology disease classification data and text data within the hospital;

[0044] The data retrieval module 102 is given a data set consisting of a group of patient data, and matches the data based on a dynamic index to find a similar case group and store it in a memory; based on a similarity calculation method, it is found that the candidate patient is a patient with a high probability of being a rare case. If a similarity threshold t is given, there are K data of patient data p that satisfy the similarity sim(p,q)>t, then the probability P=p / K that p is a rare case;

[0045] The candidate sample classification identification module 103, if the patient data is judged to be a high-probability patient of a rare case according to the data retrieval, then it is regarded as rare case data and added to the training data set for training the classifier (SVM classifier); at the same time, the similarity between the retrieved patient p and the high-probability patient q regarded as a rare case is calculated in the range of (0, t), then p is the confidence R = p / K of the rare case, when R is greater than a certain threshold λ (λ can be determined by the training data), then p should be regarded as a rare case, if the proportion of patients regarded as rare cases has reached the threshold, then the similarity threshold t should be adjusted, after the adjustment, the amount of data retrieved is reduced, and when the number of rare cases increases, the data retrieval module 102 should retrieve larger scale data.

[0046] In a preferred embodiment of the present invention, the workflow of the dynamic rare case identification engine 100 is more sophisticated. First, the data preprocessing preparation module 104 preprocesses the cardiology disease classification data and text data within the hospital. For example, for a patient suspected of myocardial infarction, the module may integrate the electrocardiogram data, blood biochemical indicators, symptom description and other information, and perform standardized processing.

[0047] Secondly, the data retrieval module 102 matches similar cases based on the preprocessed data. In this process, the system uses a similarity calculation method to identify potential rare cases. Specifically, if the similarity threshold t is set to 0.9, for a patient data p, if there are K data that satisfy the similarity sim(p,q)>0.9, then the probability P that p is considered a rare case is equal to p / K. This method allows us to quickly identify potential rare cases and improve the efficiency of the system.

[0048] Then, the candidate sample classification and identification module 103 starts working. If a patient's data is judged to be a high-probability rare case, the system will add it to the training data set for training the SVM classifier. The SVM classifier is selected here because it performs well in processing high-dimensional data and small sample problems, and is very suitable for the identification task of rare cases.

[0049] Finally, the system also introduces a confidence calculation mechanism. For a retrieved patient p and a high-probability patient q that is considered a rare case, if their similarity is in the range (0, t), then the confidence R that p is a rare case is equal to p / K. When R is greater than a certain threshold λ (for example, λ = 0.8), p is considered a rare case. This mechanism further improves the accuracy of the system in identifying rare cases.

[0050] It is worth noting that the system also has adaptive capabilities. If the proportion of patients considered rare cases reaches a preset threshold (such as 5%), the system will automatically adjust the similarity threshold t. The advantage of this is that the computational complexity can be controlled while maintaining recognition accuracy. When the number of rare cases increases, the data retrieval module 102 will automatically expand the retrieval scale to ensure that potential rare cases are not missed.

[0051] This design enables the system to continuously learn and optimize. As the amount of data increases, the ability to identify rare cases will continue to improve, thereby providing more accurate personalized care plans for cardiology patients.

[0052] Medical data includes patient imaging data, which is represented in DICOM format. DICOM information includes text content, basic patient information, identifiers, and pixel information. Text data is represented in PDF format. Text content is extracted by image conversion. The data includes basic patient information and doctor's diagnosis. The collected text content data is carried in the form of X-ray, CT, and MRI image reports.

[0053] In another embodiment of the present invention, the system processes more diverse medical data. First, the patient's imaging data is stored in the DICOM (Digital Imaging and Communication sin Medi cine) format. This format not only contains the pixel information of the image itself, but also important metadata such as the patient's basic information, examination date, and equipment parameters. For example, for a coronary artery CT scan result of a patient with myocardial infarction, the DICOM file contains not only an image of vascular stenosis, but also information such as the patient's age, gender, and scanning parameters.

[0054] Secondly, the system also processes text data stored in PDF format. These PDF files are usually reports of imaging examinations such as X-rays, CT scans, and MRIs. The system extracts the text content in these PDFs through image conversion, including the patient's basic information and the doctor's diagnosis. For example, for a cardiac MRI report, the system can extract key information such as reduced left ventricular ejection fraction and myocardial fibrosis.

[0055] This comprehensive data processing method enables the system to obtain the patient's complete medical information, thereby providing a more comprehensive basis for formulating personalized care plans. For example, the system can combine the degree of coronary artery stenosis in CT images and the myocardial function assessment in MRI reports to recommend more accurate care plans for patients. The PDF format text content is extracted through optical character recognition (OCR).

[0056] In a preferred embodiment of the present invention, the system uses optical character recognition (OCR) technology to extract text content in PDF format. The advantage of this method is that it can process PDF files in various formats, including scanned paper reports.

[0057] For example, for a handwritten electrocardiogram report, OCR technology can identify key information such as ST segment elevation and T wave inversion. This information is crucial for diagnosing heart diseases such as myocardial infarction. Through OCR technology, the system can digitize this important information for subsequent analysis and processing.

[0058] In addition, OCR technology can also recognize data in tables. For example, in a cardiac ultrasound report, important parameters such as left ventricular ejection fraction and right ventricular systolic pressure are usually presented in table form. OCR technology can accurately extract these values ​​and provide the system with more accurate quantitative analysis data.

[0059] This method greatly improves the efficiency and accuracy of the system in processing text data, laying a solid data foundation for subsequent rare case identification and personalized care plan formulation.

[0060] The rare case knowledge graph builder 200 based on the dynamic rare case identification engine 100 is used to build a rare case knowledge graph and multi-instance discovery based on knowledge reasoning. The rare case knowledge graph builder 200 based on the dynamic rare case identification engine 100 performs multi-domain union, format conversion and information extraction on the input cardiology data, and uses the existing standard medical record knowledge graph to convert the obtained cardiology data into the form in the knowledge graph, thereby building a cardiology data knowledge graph with added value, including:

[0061] The information mining module 201 includes obtaining clinical data information and text information by using the data preprocessing preparation module 104, and mining the cardiology classification data and text description information therein by using natural language processing and data analysis methods;

[0062] The cardiology classification data and text data fusion module 202 based on the knowledge graph incorporates the classification data and text data into the same knowledge graph for subsequent processing and application.

[0063] In another embodiment of the present invention, we introduce a rare case knowledge graph builder 200. The main task of this module is to build a rare case knowledge graph and perform multi-instance discovery based on this knowledge graph.

[0064] First, the information mining module 201 uses the data preprocessing preparation module 104 to obtain clinical data information and text information. For example, for a patient suspected of myocarditis, the module may extract key information such as troponin levels in blood test results, ST segment changes in electrocardiograms, and left ventricular ejection fraction in echocardiograms. Then, through natural language processing and data analysis methods, cardiology classification data and text description information are mined. For example, the system may extract classification information such as diffuse myocarditis and viral myocarditis from the doctor's diagnosis report.

[0065] Secondly, the cardiology classification data and text data fusion module 202 based on the knowledge graph comes into play. This module integrates the classification data and text data into the same knowledge graph. For example, for the above-mentioned myocarditis patient, the knowledge graph may contain the following relations:

[0066] Patient A->diagnosed with->viral myocarditis;

[0067] Viral myocarditis -> symptoms include -> chest pain;

[0068] Viral myocarditis -> laboratory tests -> elevated troponin;

[0069] Viral myocarditis -> treatment -> antiviral drugs;

[0070] This way of constructing a knowledge graph has several significant advantages. First, it can intuitively display the relationship between diseases, symptoms, test results, and treatments, helping doctors quickly understand the patient's overall condition. Second, by integrating data from different patients into the same knowledge graph, the system can more easily discover disease patterns and correlations, especially for rare cases. Finally, this structured data representation provides a good foundation for subsequent machine learning and intelligent reasoning.

[0071] It is worth noting that in the process of constructing the knowledge graph, the system will use the existing standard medical record knowledge graph as a reference. The advantage of this is that it can ensure that the newly constructed knowledge graph is consistent with medical standards and improve its reliability and versatility. For example, the system may refer to the disease classification system in the "Guidelines for the Diagnosis and Treatment of Cardiovascular Diseases" to ensure that the disease classification in the knowledge graph is consistent with the standard.

[0072] The cardiology data knowledge graph constructed in this way has high added value. It not only contains the information in the original data, but also presents the complex relationships between various entities in the form of a knowledge graph. This provides strong support for the subsequent identification of rare cases and the formulation of personalized care plans.

[0073] A rare case simulation generator 300 based on a rare case knowledge graph builder 200: by constructing a knowledge graph, matching rare cases are found for clinical comparison and guidance, and the clinical diagnosis and treatment data are annotated with additional knowledge graph information to construct guiding medical data. A simulation module 301 combines the guiding medical data constructed from matching rare cases with a simulated virtual human body to achieve simulation realization.

[0074] In yet another embodiment of the present invention, we introduce a rare case simulation generator 300 based on the rare case knowledge graph builder 200. The main function of this module is to find matching rare cases through the knowledge graph and use these cases for clinical control and guidance.

[0075] First, the rare case simulation generator 300 searches for rare cases similar to the current patient based on the previously constructed knowledge graph. For example, for a patient with unexplained myocardial injury, the system may find a similar rare cardiomyopathy case in the knowledge graph.

[0076] Secondly, the system will conduct an in-depth analysis of these matching rare cases and attach relevant information in the knowledge graph to the clinical diagnosis and treatment data. For example, the system may mark key information such as the typical symptoms, diagnostic criteria, and treatment methods of the rare cardiomyopathy. The advantage of this is that doctors can quickly understand the characteristics of this rare disease and provide a reference for diagnosis and treatment.

[0077] Then, the simulation module 301 comes into play. It combines the matched rare case data with the simulated virtual human body to simulate the disease progression. For example, for rare cardiomyopathy, the system may simulate the heart function indicators that change over time, such as the gradual decrease in left ventricular ejection fraction, the change in the degree of myocardial fibrosis, etc. This simulation can not only help doctors understand the development of the disease, but also be used to predict the progression of the patient's condition.

[0078] The design of the rare case simulation generator 300 has several significant advantages: By comparing with known rare cases, doctors can more easily identify similar rare diseases. Based on the simulation results, doctors can foresee possible complications and treatment responses, thereby formulating more accurate treatment plans. This simulation system can be used in medical education to help doctors learn to identify and deal with rare cases. At the same time, it also provides valuable virtual data for the study of rare diseases. Through virtual simulation, doctors can try different diagnosis and treatment plans without affecting actual patients, reducing the risks of actual treatment.

[0079] In general, the Rare Case Simulation Generator 300 greatly enhances the system's capabilities to not only identify rare cases, but also provide in-depth analysis and simulation for these cases, which is of great significance for improving the diagnosis and treatment of rare diseases in cardiology.

[0080] An adaptive learning optimizer 400 based on a rare case simulation generator 300 includes an interactive module 401 for optimizing the simulation module 301 according to case data and given diagnostic data provided by a user, and also for a multi-expert learning module 402;

[0081] The multi-expert learning module 402 learns from multiple experts through machines and obtains the processing methods of different experts to optimize the matching and simulation capabilities of the simulation module 301.

[0082] The learning module 403 generates optimized matching and simulation samples for the input case data based on the simulation module 301 and the multi-expert module 402, and determines whether it is a potential rare case based on the case learning data and the corresponding diagnostic data, so as to optimize the simulation module 301.

[0083] In a preferred embodiment of the present invention, we introduce an adaptive learning optimizer 400 based on the rare case simulation generator 300. The purpose of this optimizer is to continuously improve the simulation ability and rare case identification ability of the system.

[0084] First, the interactive module 401 plays an important role. It allows users to input new case data and diagnostic data, which are used to optimize the simulation module 301. For example, when a doctor handles a new rare cardiomyopathy case, he can enter the detailed information of this case into the system. The interactive module 401 will use this new data to optimize the simulation module 301 so that it can simulate this type of rare case more accurately.

[0085] Secondly, the multi-expert learning module 402 comes into play. This module uses machine learning to learn how different cardiology experts handle rare cases. For example, it may learn that one expert prefers to use a specific diagnostic method, or that another expert has unique insights on the choice of treatment options. By learning these different processing methods, the system can optimize the matching and simulation capabilities of the simulation module 301, making its simulation results more diverse and comprehensive.

[0086] Finally, the learning module 403 conducts in-depth analysis of the input case data based on the simulation module 301 and the multi-expert module 402. It generates optimized matching and simulation samples and determines whether there are potential rare cases based on these samples. For example, if the system finds that a case has a high degree of similarity with a known rare cardiomyopathy, but has some unique features, it may mark this case as a potential new rare case.

[0087] The design of this adaptive learning optimizer 400 has several significant advantages: the system can continuously learn from new cases and expert opinions, so that its performance continues to improve over time. By learning the processing methods of multiple experts, the system can provide more comprehensive and diverse diagnostic suggestions. By analyzing and comparing a large number of cases, the system has the potential to discover new and rare cases that were previously unrecognized. Based on the different expert processing methods learned, the system can recommend the most suitable diagnosis and treatment plan based on the patient's specific situation.

[0088] In general, the Adaptive Learning Optimizer 400 enables the entire system to have the ability to self-optimize and continuously learn, which is of great significance for handling complex and changeable rare cardiology cases. It not only improves the accuracy of the system, but also helps doctors discover new disease patterns and promote the development of cardiology.

[0089] The multi-expert integrated decision system 500 based on the adaptive learning optimizer 400 learns different expert information to improve the simulation capability, so that the system can simulate the expert system of the doctor's decision-making ability. The decision system converges after the simulation capability reaches a certain amount of training, and the final output case matching information and diagnosis information are provided to the user. The multi-expert integrated system performs auxiliary judgment based on the existing knowledge graph and outputs a nursing plan, which includes: a multi-expert decision fusion module 501 for the multi-expert module integration process, which constructs guiding medical data based on the processing methods from different experts and provides it to the simulation module 301.

[0090] In another embodiment of the present invention, we introduce a multi-expert integrated decision system 500 based on the adaptive learning optimizer 400. The goal of this system is to build an expert system that can simulate the decision-making ability of doctors by learning the decision-making methods of different experts.

[0091] First, the multi-expert integrated decision-making system 500 conducts in-depth learning of the diagnosis and treatment information of different cardiology experts. For example, it may learn that one expert pays special attention to virological examinations when dealing with myocarditis, while another expert pays more attention to cardiac magnetic resonance imaging results. By learning these different diagnosis and treatment focuses, the system can fully simulate the decision-making process of doctors.

[0092] Second, the system will gradually converge after reaching a certain amount of simulation capacity. This means that the system's performance will stabilize and it will be able to continuously output high-quality diagnostic recommendations. For example, after processing a large number of cardiomyopathy cases, the system's ability to identify and classify cardiomyopathy may reach a level comparable to that of a senior cardiologist.

[0093] Then, the system will output case matching information and diagnostic information to the user. For example, for a complex heart failure case, the system may provide the following information: The similarity with the previously recorded dilated cardiomyopathy case is 85%. Considering dilated cardiomyopathy, genetic testing is recommended to rule out hereditary cardiomyopathy. ACE inhibitors and beta-blockers are recommended, and implantable cardioverter defibrillators (ICDs) are considered to prevent sudden death.

[0094] The multi-expert integrated system will also make auxiliary judgments based on the existing knowledge graph and output a care plan. For example, based on the information in the knowledge graph, the system may recommend the following care plan for the above heart failure patients: daily monitoring of weight, blood pressure and heart rate. Low-salt diet and controlled fluid intake. Moderate aerobic exercise and avoid strenuous exercise. Take medications on time and pay attention to possible side effects.

[0095] Finally, the multi-expert decision fusion module 501 plays a key role. It integrates the treatment methods from different experts, constructs guiding medical data, and provides these data to the simulation module 301. The advantage of this is that the simulation module 301 can simulate the disease based on more comprehensive and diverse expert opinions, thereby improving the accuracy and comprehensiveness of the simulation. For example, when simulating a complex case of myocarditis, the system may combine the opinions of multiple experts, while considering multiple factors such as virological test results, cardiac magnetic resonance imaging, and myocardial biopsy, to give a more comprehensive prediction of disease progression.

[0096] The design of this multi-expert integrated decision-making system 500 has the following significant advantages: By integrating the knowledge and experience of multiple experts, the system can provide more comprehensive and balanced diagnostic recommendations and care plans. Based on the treatment methods of different experts, the system can recommend the most suitable diagnosis, treatment and care strategies according to the specific situation of the patient. As the system continues to learn new cases and expert opinions, its decision-making ability will continue to improve, gradually approaching or even surpassing the level of human experts. This system can effectively capture and preserve the diagnosis and treatment experience of senior experts, providing valuable resources for medical education and training of new doctors. In the face of complex or rare cases, this system can provide clinicians with strong decision-making support and improve diagnostic accuracy and treatment effectiveness.

[0097] In general, the multi-expert integrated decision system 500 is a core innovation of the present invention. It can not only provide high-quality diagnosis and nursing advice, but also continuously learn and optimize, providing powerful intelligent support for personalized nursing plan recommendations for cardiology patients.

[0098] In another embodiment of the present invention, an optimization method for realizing an intelligent recommendation system for personalized nursing plans for cardiology patients is disclosed, and the specific operation steps are as follows:

[0099] S1, inputting the cardiology data and diagnosis data of the patient, processing the data by the dynamic rare case identification engine 100, and inputting the processed data into the case library and knowledge base;

[0100] S2, output the existing cases through the case library, and the user confirms the analysis results;

[0101] S3, data is input into the knowledge base to obtain additional information, and then passes through the simulation generator to generate optimized case results and case similarity feedback;

[0102] S4, adjusts the output results of the simulation generator according to the real case results and the simulation case results, determines whether it is a potential rare case, outputs expert opinions, provides them to users, and inputs them into the knowledge base at the same time.

[0103] In a preferred embodiment of the present invention, we propose an optimization method for realizing an intelligent recommendation system for personalized nursing plans for cardiology patients. This method includes four main steps, each of which makes an important contribution to the optimization of the system and the formulation of personalized nursing plans.

[0104] First, in step S1, the system receives the patient's cardiology data and diagnostic data. These data may include the patient's basic information, symptom description, examination results, etc. For example, for a patient suspected of myocardial infarction, the input data may include chest pain symptom description, electrocardiogram results, myocardial enzyme spectrum, etc. These data are then processed by the dynamic rare case identification engine 100. The processed data will be input into the case library and knowledge base to provide a basis for subsequent analysis and decision-making.

[0105] Secondly, in step S2, the system retrieves and outputs existing similar cases from the case library. These cases may include past cases with similar symptoms to the current patient, or related typical cases. For example, the system may find several typical cases of myocardial infarction, including their symptom evolution, treatment process, and prognosis. Users (usually doctors) can review these cases and confirm the accuracy of the analysis results. This step helps doctors quickly understand the treatment methods and results of similar cases, and provide reference for the diagnosis and treatment of current patients.

[0106] Then, in step S3, the system inputs the data into the knowledge base to obtain additional information. This additional information may include the latest diagnosis and treatment guidelines, drug interaction information, rare complications, etc. For example, for a patient with myocardial infarction, the system may provide the latest thrombolytic therapy guidelines and antiplatelet drug use recommendations. Subsequently, this information is input into the simulation generator. The simulation generator generates optimized case results based on this information and gives case similarity feedback. This simulation can help doctors foresee possible disease progression and treatment effects, and provide important reference for formulating personalized treatment plans.

[0107] Finally, in step S4, the system compares the real case results with the simulated case results and adjusts the output results of the simulation generator accordingly. The purpose of this step is to continuously improve the accuracy of the simulation. At the same time, the system will determine whether the current case is a potential rare case. If a possible rare case is found, the system will immediately remind the user and give corresponding expert opinions. For example, if the system finds that the symptoms and examination results of a patient with myocardial infarction are significantly different from those of a typical case, it may indicate that this may be a rare variant of myocardial infarction, and a more in-depth examination is recommended. Finally, these new findings and conclusions will be entered into the knowledge base to further enrich the system's knowledge reserves.

[0108] The main advantages of this optimization method include: By constantly comparing real cases and simulation results, the system can continuously optimize its simulation capabilities. The system can promptly identify potential rare cases and improve the accuracy of diagnosis. By continuously inputting new cases and findings into the knowledge base, the system's knowledge reserve is continuously updated and expanded. Based on detailed case analysis and simulation results, the system can provide highly personalized care plan recommendations for each patient.

[0109] In general, this optimization method enables the intelligent recommendation system for personalized care plans for cardiology patients to continuously improve itself and provide patients with increasingly accurate diagnosis and care recommendations.

[0110] The functional modules of the dynamic rare case identification engine 100, the rare case simulation generator 300 based on the rare case knowledge graph builder 200, the adaptive learning optimizer 400 based on the rare case simulation generator 300, and the multi-expert integrated decision system 500 based on the adaptive learning optimizer 400 are all described by a knowledge graph, which includes multiple relationships and multiple entities.

[0111] In another embodiment of the present invention, we use knowledge graph to describe each functional module of the system. The advantage of this method is that it can clearly show the relationship between each module, as well as the entities and relationships involved in each module.

[0112] First, let’s take the dynamic rare case identification engine 100 as an example. It may be described in the knowledge graph as:

[0113] Dynamic rare case identification engine 100->comprising->data preparation module 101;

[0114] Dynamic rare case identification engine 100->comprising->data retrieval module 102;

[0115] Dynamic rare case identification engine 100->including->candidate sample classification and identification module 103;

[0116] Data preparation module 101->processing->medical data;

[0117] Data retrieval module 102->use->dynamic index;

[0118] Candidate sample classification and identification module 103->output->rare cases;

[0119] This representation clearly demonstrates the components of the dynamic rare case identification engine 100 and their functions.

[0120] Secondly, for the rare case knowledge graph builder 200, its knowledge graph representation may be as follows:

[0121] Rare case knowledge graph builder 200->build->rare case knowledge graph;

[0122] Rare case knowledge graph builder 200->includes->information mining module 201;

[0123] Rare case knowledge graph builder 200->includes->data fusion module 202;

[0124] Information mining module 201->use->natural language processing;

[0125] Data fusion module 202->integration->typing data and text data;

[0126] Again, the representation of the rare case simulation generator 300 in the knowledge graph may be:

[0127] Rare Case Simulation Generator 300->Use->Rare Case Knowledge Graph;

[0128] Rare case simulation generator 300->contains->simulation module 301;

[0129] Simulation module 301->Generate->Virtual case data;

[0130] Then, the knowledge graph representation of the adaptive learning optimizer 400 may be as follows:

[0131] Adaptive learning optimizer 400->optimization->simulation module 301;

[0132] Adaptive learning optimizer 400->includes->interaction module 401;

[0133] Adaptive learning optimizer 400->includes->multi-expert learning module 402;

[0134] Adaptive learning optimizer 400->includes->learning module 403;

[0135] Multi-expert learning module 402->learning->expert processing method;

[0136] Learning module 403->Generate->Optimized matching and simulation samples;

[0137] Finally, the multi-expert integrated decision system 500 may be described in the knowledge graph as:

[0138] Multi-expert integrated decision-making system 500->output->personalized care plan;

[0139] The multi-expert integrated decision system 500->includes->a multi-expert decision fusion module 501;

[0140] Multi-expert decision fusion module 501->integration->different expert processing methods.

[0141] This description method based on knowledge graph has the following significant advantages: Through the knowledge graph, we can intuitively see the relationship between the various components of the system and the flow of data. The knowledge graph can easily add new entities and relationships, making the system easy to expand and update. Based on the knowledge graph, the system can perform complex logical reasoning to improve the accuracy of diagnosis and decision-making. The knowledge graph can effectively integrate knowledge from different sources, including medical literature, clinical experience, and patient data. Through the knowledge graph, the system's decision-making process becomes more transparent and explainable, which helps doctors understand and verify the system's recommendations.

[0142] In general, the use of knowledge graphs to describe the various functional modules of the system makes the entire intelligent recommendation system for personalized nursing plans for cardiology patients clearer, more flexible and more intelligent. This approach not only improves the performance of the system, but also enhances its interpretability and credibility, which is of great significance for clinical applications.

[0143] The simulation learning model based on the rare case simulation generator 300 and the adaptive learning optimizer 400 is based on Transformer. The model needs to optimize the number of data samples, that is, the quality and quantity of samples recognized in advance. A large amount of learning data is required to reach a convergence state, and a higher efficiency output is required to achieve a better state. The loss function used in the learning process is the cross entropy loss:

[0144]

[0145] Among them, y is the true result vector, Y is the model prediction value vector, i is the sample, N is the number of samples, C is the number of categories, and y ij represents the true probability that the i-th sample belongs to the j-th class, Y ij It represents the probability that the model predicts that the i-th sample belongs to the j-th class.

[0146] In a preferred embodiment of the present invention, we use a Transformer-based simulation learning model to implement the core functions of the rare case simulation generator 300 and the adaptive learning optimizer 400. The Transformer model is selected for its excellent performance in processing sequence data, and is particularly suitable for processing time series medical data.

[0147] First, we need to optimize the quantity and quality of data samples. In the field of cardiology, high-quality data samples are often scarce, especially for rare cases. Therefore, we adopted the following strategies: Increase the number of samples by slightly transforming existing samples (such as adding noise, time axis scaling, etc.). Use models pre-trained on common cardiology diseases and then fine-tune them on rare cases. The system actively selects the most informative samples for annotation to improve learning efficiency.

[0148] For example, for a rare cardiomyopathy, we may only have 50 real cases. Through data augmentation, we can expand the sample size to 500. We then use a model pre-trained on common cardiomyopathies and fine-tune it on these 500 samples.

[0149] Secondly, the cross entropy loss function is used in the training process of the model. The choice of this loss function is based on the following considerations: The cross entropy loss function is particularly suitable for multi-classification problems, and the diagnosis of cardiology diseases is usually a multi-classification problem. Compared with other loss functions, cross entropy is more sensitive to low-probability events (such as rare cases), which is exactly the characteristic required by our system. The derivative form of the cross entropy loss function is simple, which is conducive to the rapid convergence of the model.

[0150] In practical applications, we may encounter such a situation: suppose we are dealing with a rare case of myocarditis. The model needs to accurately identify this rare myocarditis from a large number of cardiology diseases. We may have 10,000 training samples, of which only 50 are cases of this rare myocarditis. In this case, the calculation process of the cross entropy loss function is as follows:

[0151] For each sample i, if it is indeed rare myocarditis (y ij =1), while the model predicts that it is a rare myocarditis with a probability of 0.7 (Y ij =0.7), then the contribution of this sample to the loss function is:

[0152] -1*log(0.7)≈0.357

[0153] On the contrary, if the model mistakenly judges a sample that is not rare myocarditis as rare myocarditis, such as y ij =0, but Y ij =0.3, then the contribution of this sample to the loss function is:

[0154] -0*log(0.3)-1*log(0.7)≈0.357

[0155] It can be seen that whether it is a missed diagnosis or a misdiagnosis, the model will suffer the same penalty, which ensures the accuracy of the model when dealing with rare cases.

[0156] In addition, to further improve the performance of the model, we also adopted the following optimization strategy: a dynamic learning rate strategy was used. A larger learning rate was used in the early stage of training to quickly approach the optimal solution, and a smaller learning rate was used in the later stage of training to fine-tune the parameters. For example, we might start with a learning rate of 0.001 and halve the learning rate every 50 epochs. To prevent overfitting, especially when dealing with small sample problems such as rare cases, we adopted the L2 regularization technique. Specifically, we added a regularization term to the loss function:

[0157] L new =L+λ·∑(w 2 )

[0158] Where λ is the regularization coefficient, which is usually set to a small positive number such as 0.0001. 2 ) is the sum of squares of all model weights. We set up an early stopping mechanism. If the performance on the validation set does not improve for 10 consecutive epochs, we stop training. This not only prevents overfitting but also saves computational resources. We train multiple models (e.g. 5) and then average their outputs. This approach can reduce the variance of a single model and improve the stability of the overall prediction.

[0159] Through these optimization strategies, our model can achieve efficient learning on limited data samples, especially when dealing with rare cardiology cases. For example, in an internal test, our model's accuracy in identifying a rare cardiomyopathy increased from 75% to 92%, far exceeding the average level of 80% for human experts.

[0160] This Transformer-based simulation learning model, combined with a carefully designed optimization strategy, enables our intelligent recommendation system for personalized care plans for cardiology patients to accurately identify rare cases and provide highly personalized care recommendations for patients. It not only improves the accuracy of diagnosis, but also provides doctors with valuable decision-making support, which is of great significance to improving the diagnosis and treatment level of cardiology patients.

[0161] In the last embodiment of the present invention, the specific process of adaptive learning is described in detail. This process is very cleverly designed to continuously optimize the performance of the system, especially in handling rare cardiology cases. Let's analyze this process step by step:

[0162] Step a: First, the system receives diagnostic data T and disease classification data y. For example, T may contain the patient's electrocardiogram data, blood test results, etc., while y may be a preliminary disease classification (such as suspected myocarditis). These data are input into the matching module to obtain candidate results L and candidate samples X. L may be a list of known cases similar to the current case, while X is the detailed information of these candidate cases.

[0163] Step b: Based on the results of the previous step, the system outputs a recommended patient management approach D. This may include further examination recommendations, preliminary diagnosis, and treatment options.

[0164] Step c: The system compares the treatment D with the knowledge graph information in the original diagnosis data T to find inconsistencies. For example, if D recommends a myocardial biopsy and there is no relevant information in T, the system will mark this discrepancy. Based on these differences, the system generates a new auxiliary knowledge graph and inputs it into the simulation module. This step helps the system continuously improve its knowledge base.

[0165] Step d: The system performs multi-domain expert training on the diagnosis result D. This means that the system simulates the views of multiple experts in different fields (such as arrhythmia experts, heart failure experts, etc.) on the case. When the training reaches a certain amount and converges, the system outputs auxiliary information. This multi-expert training method can provide a more comprehensive diagnostic perspective.

[0166] Step e: The simulation module outputs the result as auxiliary information and gives the similarity L. At the same time, the system will use the real result Y of the case as part of the loss function. This process will continue to cycle until the model converges. This ensures that the model can continue to learn and improve.

[0167] Step f: The system outputs the final case diagnosis D and similarity L. This provides doctors with detailed diagnostic suggestions and references to similar cases.

[0168] Step g: The system conducts knowledge graph learning on candidate samples X and case D, extracts valuable medical information from them and inputs it into the case knowledge base. This step continuously enriches the system's knowledge reserve.

[0169] Step h: The system determines whether sample X is a rare disease. If so, the system randomly selects other cases to add to the training set. This strategy helps balance the dataset and improve the model's ability to identify rare cases.

[0170] Step i: Finally, the system will adjust the sample size of the matching module according to the results of the previous steps and output the final result.

[0171] The main advantages of this adaptive learning process include: by constantly comparing predicted results with actual results, the system can continuously optimize its performance; by generating and updating knowledge graphs, the system can effectively integrate new medical knowledge. By simulating the diagnosis process of experts in multiple fields, the system can provide more comprehensive diagnostic suggestions. Special attention is paid to rare cases, and the recognition ability of rare diseases is improved through special data processing strategies. The parameters of the matching module can be dynamically adjusted according to new data and results to maintain the adaptability of the system.

[0172] Through this carefully designed adaptive learning process, our intelligent recommendation system for personalized care plans for cardiology patients can continuously improve itself and provide patients with increasingly accurate diagnosis and care recommendations. Especially when dealing with complex and rare cardiology cases, this system can provide valuable decision support and effectively improve the diagnosis and treatment level of cardiology patients.

[0173] In the practical application of the present invention, we conducted a series of rigorous experiments to verify the superiority of our intelligent recommendation system for personalized nursing plans for cardiology patients. We selected an optimal embodiment and compared it in detail with two comparative examples. These experiments not only reflect the innovation of our system, but also highlight its significant advantages in practical applications.

[0174] First, let's look at the best implementation. In this implementation, we used a complete system architecture, including a dynamic rare case identification engine, a knowledge graph builder, a rare case simulation generator, an adaptive learning optimizer, and a multi-expert ensemble decision system. We used approximately 50,000 cardiology cases as the initial training set, including 500 rare cases. The system uses a Transformer-based model and is trained using a cross-entropy loss function.

[0175] Comparative Example 1 uses a traditional rule-based expert system. Although this system can handle common cardiology diseases, it is often unable to cope with complex and rare cases.

[0176] Comparative Example 2 is a simple machine learning model that uses the random forest algorithm for classification. Although this method is more advanced than the traditional expert system, it lacks key features such as dynamic learning and knowledge graph in our system.

[0177] We selected the following key indicators to evaluate system performance:

[0178] 1. Diagnostic accuracy: measures the proportion of cases correctly diagnosed by the system.

[0179] 2. Rare case identification rate: evaluates the system's ability to identify rare cardiology cases.

[0180] 3. Accuracy of personalized plan recommendation: measures the degree of consistency between the care plan recommended by the system and the expert's advice.

[0181] 4. System response time: Evaluate the time it takes for the system to provide diagnosis and recommendations.

[0182] 5. Knowledge updating efficiency: measures the speed at which the system absorbs new knowledge and updates the decision model.

[0183] The testing methods for these indicators are as follows:

[0184] Diagnostic Accuracy and Identification of Rare Cases: We used a test set of 10,000 cases, including 100 rare cases that were confirmed by three senior cardiologists.

[0185] Accuracy of personalized plan recommendations: We randomly selected 1,000 cases, compared the care plans recommended by the system with the recommendations of three cardiology experts, and calculated the consistency ratio.

[0186] System response time: We measured the average time it took the system to process 100 typical cases.

[0187] Knowledge update efficiency: We simulated 10 medical knowledge updates, each containing 100 new medical findings or guideline changes, and measured the time required for the system to fully absorb this new knowledge and update the decision model.

[0188] Now, let's look at the test results:

[0189]

[0190]

[0191] In terms of diagnostic accuracy, our system achieved 95.8%, far exceeding the 82.3% of traditional expert systems and the 88.5% of simple machine learning models. This means that out of 100 cases, our system can correctly diagnose nearly 96, while traditional methods may miss or misdiagnose 13 to 18 of them.

[0192] Of particular note is the rare case identification rate. Our system performed exceptionally well on this metric, achieving an 89.2% recognition rate, compared to 45.6% for traditional expert systems and 62.1% for simple machine learning models. This result fully demonstrates the effectiveness of our dynamic rare case identification engine, which greatly improves physicians’ ability to identify and treat rare cardiology diseases.

[0193] Our system also performed well in terms of personalized plan recommendations, achieving an accuracy rate of 93.5%. This means that our system is able to provide high-quality personalized care plans that are consistent with expert recommendations for the vast majority of patients. This is due to the multi-expert integrated decision-making system and adaptive learning optimizer in our system.

[0194] Although our system is slightly inferior to simple rule-based expert systems in terms of system response time, considering the amount and complexity of data processed by our system, a response time of 1.2 seconds is still very excellent and can fully meet clinical needs.

[0195] Finally, in terms of knowledge updating efficiency, our system has shown great advantages. It only takes 2.5 hours to fully absorb and integrate new medical knowledge, while traditional expert systems take a full 24 hours and simple machine learning models take 12 hours. This rapid learning and adaptability ensures that our system can always stay at the forefront of medicine and provide patients with the latest and best diagnosis and treatment recommendations.

[0196] In general, these test results strongly demonstrate the significant advantages of our intelligent recommendation system for personalized care plans for cardiology patients in key aspects such as diagnostic accuracy, rare case identification, personalized plan recommendation, and knowledge update. It can not only improve the diagnostic accuracy of cardiology diseases, but also greatly improve the ability to identify rare cases, while ensuring the personalization and advancement of care plans. This system is expected to become a powerful assistant for cardiologists and greatly improve the diagnosis and treatment level and quality of life of cardiology patients.

[0197] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Intelligent recommendation system for personalized nursing plans for cardiology patients, characterized by: include: The dynamic rare case identification engine uses medical data and text data to accurately capture and identify rare cardiology cases. It includes: a data preparation module that obtains internal hospital data and performs data preprocessing for classification of cardiology diseases according to the specified cardiology diagnosis and treatment guidelines. The preprocessing includes combining multiple modal data into multiple domains, unifying the data format, and constructing a globally unique identifier for each patient; The data retrieval module first screens the initial data set, then matches based on the dynamic index, sets the similarity threshold and the number of candidate samples to control the recognition accuracy, and outputs the retrieved candidate samples that meet the similarity threshold or the number of retrieved samples; the candidate sample classification and recognition module is used to identify whether it is a rare case. If it is identified as a rare case, the retrieval result is output and the recognition result is fed back to the data preparation module as a training sample set, allowing the data preparation module to pre-process more data and adaptively adjust the similarity threshold and the number of samples to be retrieved of the retrieval module.

2. The intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 1 is characterized by: The dynamic rare case identification engine includes: a data preprocessing preparation module, which preprocesses the cardiology disease classification data and text data within the hospital; a data retrieval module, which is given a data set consisting of a group of patient data, and matches the data based on a dynamic index to find similar case groups and store them in memory; based on the similarity calculation method, find the candidate patients with a high probability of belonging to rare cases. If a similarity threshold t is given, there are K data of patient data p that satisfy the similarity sim(p,q)>t, then the probability P=p / K that p is a rare case; a candidate sample classification identification module, if the patient data is judged according to the data retrieval, If it is a high-probability patient who is a rare case, it will be regarded as a rare case data and added to the training data set for training the classifier; at the same time, the similarity between the retrieved patient p and the high-probability patient q regarded as a rare case is calculated in the range of (0, t), then p is the confidence R = p / K of the rare case, when R is greater than a certain threshold λ, λ is determined by the training data, then p should be regarded as a rare case, if the proportion of patients regarded as rare cases has reached the threshold, then the similarity threshold t should be adjusted, after adjustment, the amount of data retrieved is reduced, and when the number of rare cases increases, the data retrieval module should retrieve larger scale data.

3. The intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 1 is characterized by: Medical data includes patient imaging data, and the data content is represented in DICOM format. DICOM information includes text content, basic patient information, identifiers and pixel information; text data content is represented in PDF format, and the text content is extracted by image conversion. The data content includes basic patient information and doctor's diagnosis opinions. The collected text content data is carried by imaging reports of X-ray, CT and MRI images, and the PDF format text content is extracted by optical character recognition.

4. The intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 1 is characterized by: A rare case knowledge graph builder based on the dynamic rare case identification engine is used to construct a rare case knowledge graph and multi-example discovery based on knowledge reasoning. The rare case knowledge graph builder performs multi-domain union, format conversion and information extraction on the input cardiology data, and uses the existing standard medical record knowledge graph to convert the obtained cardiology data into the form in the knowledge graph, thereby constructing a cardiology data knowledge graph with added value, including: an information mining module, which includes using a data preprocessing preparation module to obtain clinical data information and text information, using natural language processing and data analysis methods to mine its The cardiology classification data and text description information in the cardiology classification data and text data; the fusion module of cardiology classification data and text data based on the knowledge graph, which incorporates the classification data and text data into the same knowledge graph for subsequent processing and application; the rare case simulation generator based on the rare case knowledge graph builder: by constructing a knowledge graph, finding matching rare cases for clinical control and guidance, and annotating the clinical diagnosis and treatment data with additional knowledge graph information, constructing guiding medical data, simulation module, which constructs guiding medical data from matching rare cases and combines it with the simulated virtual human body to achieve simulation realization.

5. The intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 4 is characterized by: An adaptive learning optimizer based on a rare case simulation generator, including an interactive module for optimizing a simulation module according to case data and given diagnosis data provided by a user, and also for a multi-expert learning module; The multi-expert learning module uses machine learning to learn from multiple experts and obtain the processing methods of different experts to optimize the matching and simulation capabilities of the simulation module. The learning module generates optimized matching and simulation samples for the input case data based on the simulation module and the multi-expert module. Based on the case learning data and the corresponding diagnostic data, it determines whether it is a potential rare case in order to optimize the simulation module.

6. The intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 5 is characterized by: The multi-expert integrated decision-making system based on the adaptive learning optimizer learns different expert information to improve the simulation capability, so that the system can simulate the expert system of the doctor's decision-making ability. The decision-making system converges after the simulation capability reaches a certain amount of training, and the final output case matching information and diagnosis information are provided to the user. The multi-expert integrated system makes auxiliary judgments based on the existing knowledge graph and outputs nursing plans, which include: a multi-expert decision fusion module for the multi-expert module integration process, which constructs guiding medical data based on the processing methods from different experts and provides it to the simulation module.

7. A method for optimizing the intelligent recommendation system for personalized nursing plans for cardiology patients according to any one of claims 1 to 6, characterized in that: The specific steps are: S1, inputting the patient's cardiology data and diagnosis data, processing the data by the dynamic rare case identification engine, and inputting the processed data into the case library and knowledge base; S2, output the existing cases through the case library, and the user confirms the analysis results; S3, data is input into the knowledge base to obtain additional information, and then passes through the simulation generator to generate optimized case results and case similarity feedback; S4, adjusts the output results of the simulation generator according to the real case results and the simulation case results, determines whether it is a potential rare case, outputs expert opinions, provides them to users, and inputs them into the knowledge base at the same time.

8. The optimization method for realizing an intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 7, characterized in that: The functional modules of the dynamic rare case identification engine, the rare case simulation generator based on the rare case knowledge graph builder, the adaptive learning optimizer based on the rare case simulation generator, and the multi-expert integrated decision system based on the adaptive learning optimizer are all described by a knowledge graph, which includes multiple relationships and multiple entities.

9. The optimization method for realizing an intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 8, characterized in that: The simulation learning model based on the rare case simulation generator and the adaptive learning optimizer is based on Transformer. The model needs to optimize the number of data samples, that is, the quality and quantity of the samples identified in advance. A large amount of learning data is required to reach a convergence state, and a higher efficiency output is required to achieve a better state. The loss function used in the learning process is the cross entropy loss: y is the true result vector, Y is the model prediction value vector, i is the sample, N is the number of samples, C is the number of categories, y ij represents the true probability that the i-th sample belongs to the j-th class, Y ij It represents the probability that the model predicts that the i-th sample belongs to the j-th class.

10. The optimization method for realizing an intelligent recommendation system for personalized nursing plans for cardiology patients according to claim 9, characterized in that: The adaptive learning process is specifically as follows: a. input diagnostic data T and disease typing data y, and input them into the matching module to obtain candidate results L and candidate samples X; b. output patient treatment method D; c. determine the inconsistency information between the knowledge graphs in D and T, and generate a new auxiliary knowledge graph and input it into the simulation module; d. Perform multi-domain expert training on the diagnosis result D. If the expert result D reaches a certain amount of training and meets the convergence condition, the training is terminated and auxiliary information is output; e. The simulation output result is auxiliary information, and the similarity L is given. The real case result Y is given as the loss, and the process is repeated until convergence; f. Output case D, and the case output similarity L; g. Perform knowledge graph learning on candidate samples X and case D, extract medical information assisted by the knowledge graph and input it into the case knowledge base; h. Determine whether sample X is marked as a rare disease. If so, randomly select other cases and add them to the training set; i. Adjust the sample size of the matching module and output the results.

Images