Movement rehabilitation method, device and equipment for patients with cardiovascular diseases combined with weakness and fear of movement

Abstract

The invention provides a motion rehabilitation method, device and equipment for movement fear of a cardiovascular disease combined weak patient. The method comprises the following steps: S101, establishing different potential category influence factor models of movement fear of the cardiovascular disease combined weak patient on the basis of a potential category analysis method; s102, based on a knowledge graph technology, aggregating related evidence-based evidences of health education and exercise rehabilitation of the movement fear of the senile cardiovascular disease and weak patient, and constructing to obtain a knowledge graph-based exercise rehabilitation knowledge base of the movement fear of the senile cardiovascular disease and weak patient; s103, in combination with the potential category influence factor model and the multi-modal data of the patient, constructing a multi-modal data-based digital twinborn body of the senile cardiovascular disease combined with the dynamic terrorism of the weak patient; and S104, according to the dynamically updated digital twinborn model, performing matching from the exercise rehabilitation knowledge base according to the health target, the body condition, the real-time environment and the preference characteristics of the patient to obtain personalized health education and exercise prescription recommendation.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and particularly to a method, device and equipment for kinesiophobia exercise rehabilitation of patients with cardiovascular diseases complicated with frailty. Background Art

[0002] Exercise-based cardiac rehabilitation (CR) is a key strategy for the treatment of CVD. However, due to limitations, most eligible patients, especially the elderly population, do not participate in CR, resulting in a common phenomenon of kinesiophobia in the elderly population.

[0003] Digital health technologies, integrating cutting-edge technologies such as the Internet and wearable devices, enable instant access to health data and personalized education, with effects comparable to those of traditional outpatient CR, and have become a new approach for CVD rehabilitation. They not only achieve comprehensive data recording, real-time monitoring, and personalized consultation, but also through the integration of AI and machine learning technologies, in-depth analysis can be obtained based on the multimodal data collected by wearable devices, and then a personalized exercise plan can be customized for patients. In addition, the application of gamification design has also significantly improved the treatment compliance of patients. At the same time, knowledge graph and digital twin technologies have also gradually been applied, providing strong technical support for the formulation of precision medicine and personalized rehabilitation plans. Although the application of digital health technologies in the field of CR is increasing, there are still deficiencies in scientificity, safety, personalization, and popularity. Summary of the Invention

[0004] The present invention provides a method, device and equipment for kinesiophobia exercise rehabilitation of patients with cardiovascular diseases complicated with frailty to improve the above problems.

[0005] S101, establishing a model of influencing factors of different latent classes of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on the latent class analysis method;

[0006] S102, aggregating the evidence-based evidence related to health education and exercise rehabilitation of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on knowledge graph technology, and constructing an exercise rehabilitation knowledge base of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on the knowledge graph;

[0007] S103, combining the latent class influencing factor model and the multimodal data of the patient to construct a digital twin of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on multimodal data;

[0008] S104, according to the dynamically updated digital twin model, matching personalized health education and exercise prescription recommendations from the exercise rehabilitation knowledge base for the patient's health goals, physical conditions, real-time environment, and preference characteristics.

[0009] Preferably, step S101 specifically includes:

[0010] A phobia classification method using prior knowledge and information theory enhancement technology, a technique for determining the optimal number of classes and boundaries based on the Bayesian information criterion and the expectation-maximization algorithm, and identifying potential subgroups through latent class analysis and hierarchical clustering, to obtain a model of influencing factors for different latent classes of phobia in elderly patients with cardiovascular disease complicated by frailty; among them, for different models of influencing factors for latent classes:

[0011] Based on BIC = -2·ln(L)+p·ln(n), by calculating the model complexity and goodness of fit under different numbers of classes, select the number of classes K that minimizes BIC as the optimal number of classes; L is the maximum likelihood estimate of the model; p is the number of model parameters; n is the number of samples;

[0012] Use the EM algorithm to estimate the parameters of the class distribution, and achieve the maximum likelihood estimation of the parameters through iterative update steps:

[0013] In the E step: Calculate the posterior probability P(Zi; = k|Xi,θ), where Zi is the class to which sample i belongs, Xi is the sample feature, and θ is the current model parameter; in the M step: Maximize the posterior probability and update the parameter θ;

[0014] Use the latent class model to stratify the samples, assuming that there is a dependence relationship between the observed variable X and the latent class variable C, and maximize the likelihood function of the joint distribution P(X,C);

[0015] Use the hierarchical clustering algorithm to further analyze the fine-grained structure of the latent classes, adopt the Ward method to minimize the within-class sum of squares error, calculate the distance d(xi, xj) between sample points xi and xj, and merge according to the following criteria:

[0016]

[0017] Where: A and B are two classes; μA and μB are the means of the classes; |A| and |B| are the number of class samples.

[0018] Preferably, step S102 specifically includes:

[0019] Retrieve evidence-based medicine literature and clinical data, perform standardized processing to obtain standard data;

[0020] Based on natural language processing technology, combined with the named entity recognition method, extract key entities in the standard data;

[0021] Use a relationship classification model to identify the semantic relationships between key entities, and the relationship types include "treatment", "influence", "indications", "contraindications";

[0022] Based on the semantic relationships between key entities, construct a sports rehabilitation knowledge base for phobia in elderly patients with cardiovascular disease complicated by frailty.

[0023] Preferably, it further includes:

[0024] Derive the deep associations between key entities based on ontology reasoning and vector embedding techniques to improve the motion rehabilitation knowledge base.

[0025] Preferably, deriving the deep associations between key entities based on ontology reasoning and vector embedding techniques to improve the motion rehabilitation knowledge base specifically includes:

[0026] Construct a domain ontology for elderly cardiovascular diseases, define the core concepts and their hierarchical structures, attributes, and rules, and use description logic for reasoning; or

[0027] Adopt a knowledge graph embedding model to map entities and relationships into a high-dimensional vector space, and the embedding representations of entities and relationships are: h + r ≈ t; where: h and t are the vector representations of the head entity and the tail entity respectively, and r is the vector representation of the relationship; or

[0028] Calculate the relevance between entities based on cosine similarity or Euclidean distance.

[0029] Preferably, for the digital twin:

[0030] Represent the patient's health status as a hidden variable and construct a state space model;

[0031] Construct a causal network including health status, intervention measures, and multi-modal observation variables to model the dynamic changes of the patient's health status;

[0032] For a linear Gaussian system, use a Kalman filter to dynamically update the health status;

[0033] For a non-linear non-Gaussian system, use a particle filter for health status prediction.

[0034] The embodiment of the present invention also provides a motion rehabilitation device for patients with phobia of movement due to cardiovascular diseases combined with frailty, including:

[0035] A model establishment unit for establishing a model of influencing factors of different latent classes of phobia of movement in elderly patients with cardiovascular diseases combined with frailty based on the latent class analysis method;

[0036] A motion rehabilitation knowledge base construction unit for aggregating the evidence-based evidence of health education and motion rehabilitation of phobia of movement in elderly patients with cardiovascular diseases combined with frailty based on knowledge graph technology to construct a motion rehabilitation knowledge base of phobia of movement in elderly patients with cardiovascular diseases combined with frailty based on the knowledge graph;

[0037] A digital twin construction unit, which is used to combine a latent class influence factor model and multimodal data of a patient to construct a digital twin of fear of movement disorder in elderly patients with cardiovascular disease complicated with frailty based on the multimodal data;

[0038] A recommendation unit, which is used to match personalized health education and exercise prescription recommendations from the exercise rehabilitation knowledge base according to the dynamically updated digital twin model for the patient's health goals, physical conditions, real-time environment, and preference characteristics.

[0039] An embodiment of the present invention also provides a fear of movement disorder exercise rehabilitation device for patients with cardiovascular disease complicated with frailty, which includes a memory and a processor. A computer program is stored in the memory, and the computer program can be executed by the processor to implement the fear of movement disorder exercise rehabilitation method for patients with cardiovascular disease complicated with frailty as described above.

[0040] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0041] By acquiring the patient's multimodal data and constructing a digital twin, and combining the domain ontology knowledge base and exercise rehabilitation knowledge graph constructed based on evidence-based medical literature, the present invention realizes the real-time dynamic assessment of the patient's health status and the intelligent recommendation of personalized exercise rehabilitation programs. The present invention can dynamically adjust and optimize the rehabilitation program, improve its aging adaptation level and patient compliance, and thus effectively improve the rehabilitation effect and quality of life of elderly patients with cardiovascular disease complicated with frailty. Description of the Drawings

[0042] Figure 1 It is a flowchart of the fear of movement disorder exercise rehabilitation method for patients with cardiovascular disease complicated with frailty provided by the first embodiment of the present invention.

[0043] Figure 2 It is a structural diagram of the fear of movement disorder exercise rehabilitation device for patients with cardiovascular disease complicated with frailty provided by the second embodiment of the present invention. Detailed Embodiments

[0044] Next, the technical solution of the present invention will be described clearly and completely in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0045] As Figure 1 , the fear of movement disorder exercise rehabilitation method for patients with cardiovascular disease complicated with frailty provided in this embodiment may specifically include:

[0046] S101, establishing different latent class influence factor models of fear of movement disorder in elderly patients with cardiovascular disease complicated with frailty based on the latent class analysis method.

[0047] In this embodiment, under the condition of limited samples, a phobia classification method that combines prior knowledge and information theory enhancement techniques can be adopted. The optimal number of classes and boundaries are determined based on the Bayesian information criterion and the expectation-maximization algorithm, and techniques for identifying potential subgroups through latent class analysis and hierarchical clustering are used.

[0048] Among them, for the determination of the optimal number of classes and boundaries, based on the BIC, by calculating the model complexity and goodness of fit under different numbers of classes, the number of classes K that minimizes the BIC is selected.

[0049] BIC = -2·ln(L) + p·ln(n)

[0050] Among them: L is the maximum likelihood estimate of the model; p is the number of model parameters; n is the number of samples.

[0051] For parameter estimation, the EM algorithm can be used to estimate the parameters of the class distribution, and the maximum likelihood estimation of the parameters is achieved through iterative update steps:

[0052] E step (Expectation): Calculate the posterior probability P(Zi; = k|Xi,θ), where Zi is the class to which sample i belongs, Xi is the sample feature, and θ is the current model parameter.

[0053] M step (Maximization): Maximize the posterior probability and update the parameter θ.

[0054] Latent class analysis can use the Latent Class Analysis (LCA) model to stratify the samples. Assuming that there is a dependence relationship between the observed variable X and the latent class variable C, the likelihood function of the joint distribution P(X,C) is maximized.

[0055] Hierarchical clustering: Use the hierarchical clustering algorithm to further analyze the fine-grained structure of the latent classes, and adopt the Ward method to minimize the within-class sum of squares error. Calculate the distance d(xi, xj) between sample points xi and xj, and merge them according to the following criteria:

[0056]

[0057] Among them: A and B are two classes; μA and μB are the means of the classes; |A| and |B| are the number of class samples.

[0058] S102. Based on knowledge graph technology, aggregate the evidence-based evidence on the health education and exercise rehabilitation of phobia in elderly patients with cardiovascular disease complicated with frailty, and construct a knowledge base for exercise rehabilitation of phobia in elderly patients with cardiovascular disease complicated with frailty based on the knowledge graph.

[0059] Specifically, it includes:

[0060] Retrieve evidence-based medical literature and clinical data, perform standardized processing, and obtain standard data.

[0061] Based on natural language processing technology, combined with named entity recognition methods, extract key entities from the standard data.

[0062] Use a relation classification model to identify the semantic relationships between key entities. The relation types include "treatment", "influence", "indication", and "contraindication".

[0063] Among them, based on natural language processing (NLP) technology, combined with named entity recognition (NER) methods, extract key entities from the literature and clinical data. The entity types include diseases, symptoms, drugs, intervention measures, exercise programs, etc. The specific implementation steps are as follows:

[0064] First, preprocess the text: including word segmentation, stop word removal, and word stemming.

[0065] Then, perform entity recognition.

[0066] In this embodiment, an NER model can be fine-tuned in combination with domain annotation data to achieve high-precision entity recognition.

[0067] P(e|x) = softmax(W·h x + b)

[0068] Where: e is the entity category label; x is the input text; h is the context embedding vector output by BERT.

[0069] Among them, the NER model is pre-trained using BERT (Bidirectional Encoder Representations from Transformers).

[0070] Next, perform relation extraction:

[0071] In this embodiment, a relation classification model can be used to identify the semantic relationships between entities. The relation types include "treatment", "influence", "indication", "contraindication", etc. The method steps are as follows:

[0072] Feature representation: Encode the identified entities and their contexts, and use bidirectional LSTM to capture sequence dependencies.

[0073] h t = LSTM(x t , h t-1 )

[0074] Classifier: Use a multi-layer perceptron (MLP) for relationship classification:

[0075] P(r|e1,e2,c) = softmax(W r ·[h e1 ; h e2 ; c] + b r )

[0076] where: r is the relationship category; e1, e2 are entities; c is the context representation.

[0077] Finally, based on the semantic relationships between key entities, a sports rehabilitation knowledge base for fear of movement disorder in elderly patients with cardiovascular disease and frailty is constructed.

[0078] Specifically, it also includes:

[0079] Derive the deep associations between key entities based on ontology reasoning and vector embedding technology to improve the sports rehabilitation knowledge base.

[0080] It includes:

[0081] 1. Association relationship derivation method based on ontology reasoning

[0082] Specifically, construct a domain ontology for elderly cardiovascular disease, define core concepts (such as diseases, intervention measures) and their hierarchical structures, attributes, and rules.

[0083] Ontology rule example: If X has disease D and the intervention measure for D is I, then X is suitable for I:

[0084]

[0085] The reasoning algorithm uses description logic (DL) for reasoning, for example, an ontology reasoner based on ALCHQ (such as Fact++, Pellet) executes rule calculations.

[0086] 2. Association relationship derivation method based on vector calculation

[0087] Among them, use a knowledge graph embedding model (such as TransE) to learn the embedding representations of entities and relationships by mapping entities and relationships to a high-dimensional vector space: h + r ≈ t

[0088] where: h and t are the vector representations of the head entity and the tail entity respectively; r is the vector representation of the relationship.

[0089] 3. Relevance calculation: Calculate the relevance between entities based on cosine similarity or Euclidean distance:

[0090]

[0091] Among them, the inference process can use an optimization algorithm based on gradient descent (such as Adam) to minimize the loss function of the embedding:

[0092]

[0093] where: S is the positive sample set, (h', r, t') is the negative sample; d is the distance function; γ is the margin value.

[0094] S103. Combine the latent class influence factor model and the multimodal data of the patient to construct a digital twin of fear of movement in elderly patients with cardiovascular disease complicated with frailty based on multimodal data.

[0095] Specifically, in this implementation, before constructing the digital twin, it is first necessary to achieve multimodal data fusion.

[0096] Among them, multimodal data of elderly patients with cardiovascular disease complicated with frailty are obtained through various sources such as intelligent wearable devices and sensors, including physiological indicators such as heart rate and blood pressure, behavioral data such as steps and sleep duration, and laboratory test indicators such as blood glucose and blood lipids. The obtained multimodal data are preprocessed, and feature extraction and standardization techniques are used to eliminate the differences between different modal data, and the standardized multimodal patient data are obtained.

[0097] For example:

[0098]

[0099] where x is the original feature value, and μ and σ are the mean and standard deviation respectively.

[0100] Then, represent the health status of the patient as a hidden variable and construct a state space model:

[0101]

[0102] where: Zt is the hidden state (health status); u t is the control input (such as an intervention measure); x t is the observed data (such as multimodal input); w t and v t are the state transition noise and the observation noise respectively.

[0103] Then, construct a causal network including health status, intervention measures and multimodal observation variables to model the dynamic changes of the patient's health status:

[0104] P(Z t |Z t-1 ,X t ,U t ) = P(Zt |Z t-1 )·P(X t |Z t )·P(U t |Z t )

[0105] Optimize the network parameters through maximum likelihood estimation or Bayesian estimation.

[0106] In this embodiment, real-time monitoring and prediction of the patient's health status can adopt Kalman filtering or particle filtering.

[0107] For a linear Gaussian system, use a Kalman filter to dynamically update the health status. The principle is as follows:

[0108] State prediction:

[0109]

[0110] Observation update:

[0111] Update covariance: P t|t =(I - K t ·H)·P t|t-1

[0112] For a non-linear non-Gaussian system, use a particle filter. The principle is as follows:

[0113] · Initialization: Generate N particles according to the initial distribution

[0114] · State prediction:

[0115]

[0116] · Weight update:

[0117]

[0118] · Resampling: Update the particles according to the weights Perform particle update.

[0119] S104. According to the dynamically updated digital twin model, match personalized health education and exercise prescription recommendations from the exercise rehabilitation knowledge base for the patient's health goals, physical conditions, real-time environment, and preference characteristics.

[0120] In this embodiment, the digital twin model can be updated based on a dynamic update method of Bayesian inference, which includes:

[0121] Prior update of health status:

[0122] Updating the health status using the dynamic Bayesian formula:

[0123] P(Z t |X 1：t , U 1：t ) ∝ P(X t |Z t ) · ∫ P(Z t |Z t-1 , U t ) P(Z t-1 |X 1:t-1 , U 1:t-1 ) dZ

[0124] Real-time prediction:

[0125] Predicting the future health status based on the observed data and the prior state:

[0126] P(Z t+1 |X 1：t , U 1:t ) = ∫ P(Z t+1 |Z t , U t+1 ) P(Z t |X 1：t , U 1:t ) dZ t

[0127] In this embodiment, according to the dynamically updated digital twin model, access the pre-constructed knowledge base for fear of movement disorder exercise rehabilitation of elderly patients with cardiovascular diseases and frailty. According to the patient's category, match a suitable exercise rehabilitation plan. Compare and simulate the key parameters in the exercise rehabilitation plan with the patient's digital twin. By evaluating the impact of exercise load on cardiovascular function, personalize and optimize the exercise prescription. Feed back the personalized exercise rehabilitation prescription to the patient in the form of voice / text, etc., to guide them to carry out exercise rehabilitation training. During the process, continuously collect multi-modal data and update the digital twin. According to the physiological indicators and behavioral data that change dynamically during the patient's exercise rehabilitation process, use the prediction model to evaluate the severity and prognosis of their fear of movement disorder, obtain personalized rehabilitation effect feedback, and provide a reference for subsequent adjustment of the exercise prescription.

[0128] For example, for patients with severe kinesiophobia, safety thresholds for heart rate, respiratory rate, and blood pressure are set. Once the monitored values exceed the standards, warning messages are pushed through the App. According to the clustering results, a matching exercise plan is retrieved from the knowledge base. For example, for a 60-year-old male patient with moderate kinesiophobia, a moderate-intensity aerobic exercise of 30 minutes each time, three times a week is recommended. The exercise plan parameters are input into the digital twin model of the patient's corresponding cardiovascular system. Through simulation analysis, a personalized exercise prescription is obtained. For example, the intensity of aerobic exercise is reduced by 20%, and the frequency is adjusted to twice a week. The optimized prescription is conveyed to the patient in the form of voice synthesis to guide them to complete the exercise training, and the training data is collected in real time and transmitted to the cloud. The long short-term memory neural network is used to predict the severity of the patient's kinesiophobia symptoms in the next week. If there is no obvious improvement in the kinesiophobia symptoms, the exercise prescription is further adjusted and optimized until the ideal rehabilitation effect is achieved.

[0129] In summary, in this embodiment, by obtaining the multi-modal data of the patient and constructing the digital twin, combined with the domain ontology knowledge base and the exercise rehabilitation knowledge graph constructed based on evidence-based medical literature, the real-time dynamic assessment of the patient's health status and the intelligent recommendation of personalized exercise rehabilitation programs are realized. The present invention can dynamically adjust and optimize the rehabilitation program, improve its aging adaptation level and patient compliance, thereby effectively improving the rehabilitation effect and quality of life of elderly patients with cardiovascular diseases complicated with frailty.

[0130] Please refer to Figure 2 , the second embodiment of the present invention also provides a kinesiophobia exercise rehabilitation device for elderly patients with cardiovascular diseases complicated with frailty, including:

[0131] A model establishment unit 210, configured to establish a model of influencing factors of different latent classes of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on the latent class analysis method;

[0132] An exercise rehabilitation knowledge base construction unit 220, configured to aggregate the evidence-based evidence of health education and exercise rehabilitation of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on the knowledge graph technology, and construct an exercise rehabilitation knowledge base of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on the knowledge graph;

[0133] A digital twin construction unit 230, configured to construct a digital twin of kinesiophobia in elderly patients with cardiovascular diseases complicated with frailty based on multi-modal data by combining the latent class influencing factor model and the multi-modal data of the patient;

[0134] A recommendation unit 240, configured to match and obtain personalized health education and exercise prescription recommendations from the exercise rehabilitation knowledge base according to the dynamically updated digital twin model for the patient's health goals, physical conditions, real-time environment, and preference characteristics.

[0135] The third embodiment of the present invention also provides a fear-of-movement exercise rehabilitation device for patients with cardiovascular diseases complicated by frailty, which includes a memory and a processor. A computer program is stored in the memory and can be executed by the processor to implement the fear-of-movement exercise rehabilitation method for patients with cardiovascular diseases complicated by frailty as described above.

[0136] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the present application that have similar functions.

Claims

1. A method for the rehabilitation of cardiovascular disease combined with frail patients with akinesia, characterized in that: include: S101, based on latent class analysis method, establish different latent class influencing factor models of aphrodisiac syndrome in elderly patients with cardiovascular disease and frailty; S102, based on knowledge graph technology, aggregate the relevant evidence-based evidence of health education and exercise rehabilitation for elderly patients with cardiovascular disease and frailty with akinesia, and construct a knowledge base of exercise rehabilitation for elderly patients with cardiovascular disease and frailty with akinesia based on knowledge graph; S103, combining the latent class influencing factor model and the multimodal data of patients, to construct a digital twin of akinesia in elderly patients with cardiovascular disease and frailty based on multimodal data; S104, according to the dynamically updated digital twin model, personalized health education and exercise prescription recommendations are obtained from the sports rehabilitation knowledge base based on the patient's health goals, physical conditions, real-time environment, and preference characteristics.

2. The method for exercise rehabilitation of akinesia in patients with cardiovascular disease and frailty according to claim 1, characterized in that: Step S101 specifically includes: The classification method of phobia using prior knowledge and information theory enhancement technology was used to determine the optimal number of categories and boundaries based on the Bayesian information criterion and the maximum expectation algorithm. The latent class analysis and hierarchical clustering technology were used to identify potential subgroups, and different latent category influencing factor models of phobia in elderly patients with cardiovascular disease and frailty were obtained. Among them, for different latent category influencing factor models: Based on BIC = -2·ln(L) + p·ln(n), by calculating the model complexity and goodness of fit under different numbers of categories, the number of categories K that minimizes BIC is selected as the optimal number of categories; L is the maximum likelihood estimate of the model; p is the number of model parameters; n is the number of samples; The EM algorithm is used to estimate the parameters of the category distribution, and the maximum likelihood estimation of the parameters is achieved through iterative update steps: In step E: calculate the posterior probability P(Zi; = k|Xi,θ), where Zi is the category to which sample i belongs, Xi is the sample feature, and θ is the current model parameter; in step M: maximize the posterior probability and update the parameter θ; Use the latent class model to stratify the samples, assuming that there is a dependency relationship between the observed variable X and the latent class variable C, and maximize the likelihood function of the joint distribution P(X,C); The hierarchical clustering algorithm is used to further analyze the fine-grained structure of the latent classes, and the Ward method is used to minimize the intra-class squared error. The distance d(xi, xj) between sample points xi, xj is calculated and merged according to the following criteria: Where: A, B are two categories; μA, μB are the means of the categories; |A|, |B| are the number of category samples.

3. The method for exercise rehabilitation of akinesia in patients with cardiovascular disease and frailty according to claim 1, characterized in that: Step S102 specifically includes: Retrieve evidence-based medical literature and clinical data, perform standard processing, and obtain standard data; Based on natural language processing technology and named entity recognition method, key entities in standard data are extracted; A relational classification model is used to identify semantic relations between key entities. The relation types include "treatment", "effect", "indications", and "contraindications"; Based on the semantic relationship between key entities, a knowledge base of exercise rehabilitation for akinesia in elderly patients with cardiovascular disease and frailty was constructed.

4. The method for exercise rehabilitation of akinesia in patients with cardiovascular disease and frailty according to claim 3, characterized in that: Also includes: Based on ontology reasoning and vector embedding technology, deep associations between key entities are derived to improve the sports rehabilitation knowledge base.

5. The method for exercise rehabilitation of akinesia in patients with cardiovascular disease and frailty according to claim 3, characterized in that: Based on ontology reasoning and vector embedding technology, the deep associations between key entities are derived to improve the sports rehabilitation knowledge base, including: Construct a domain ontology for cardiovascular diseases in the elderly, define core concepts and their hierarchical structures, attributes, and rules, and use description logic for reasoning; or The knowledge graph embedding model is used to map entities and relations into a high-dimensional vector space. The embedding representation of entities and relations is: h+r≈t; where: h, t are the vector representations of the head entity and the tail entity respectively, and r is the vector representation of the relation; or Calculate the relatedness between entities based on cosine similarity or Euclidean distance.

6. The method for exercise rehabilitation of akinesia in patients with cardiovascular disease and frailty according to claim 1, characterized in that: For digital twins: Represent the patient's health status as a hidden variable and construct a state space model; Construct a causal network that includes health status, intervention measures, and multimodal observation variables to model the dynamic changes of patients' health status; For linear Gaussian systems, the health status is dynamically updated using a Kalman filter; For nonlinear non-Gaussian systems, particle filters are used for health state prediction.

7. A motor rehabilitation device for patients with cardiovascular disease and frailty who suffer from akinesia, characterized in that: include: A model building unit, used to build different latent class influencing factor models of aphobia in elderly patients with cardiovascular disease and frailty based on latent class analysis method; The exercise rehabilitation knowledge base construction unit is used to aggregate the health education and evidence-based evidence of exercise rehabilitation for elderly patients with cardiovascular disease and frailty with akinesia based on knowledge graph technology, and to construct an exercise rehabilitation knowledge base for elderly patients with cardiovascular disease and frailty with akinesia based on knowledge graph; A digital twin construction unit is used to combine the latent class influencing factor model and the multimodal data of patients to construct a digital twin of akinesia in elderly patients with cardiovascular disease and frailty based on multimodal data; The recommendation unit is used to match the patient's health goals, physical conditions, real-time environment, and preference characteristics from the sports rehabilitation knowledge base to obtain personalized health education and exercise prescription recommendations based on the dynamically updated digital twin model.

8. A sports rehabilitation device for patients with cardiovascular disease and frailty who suffer from akinesia, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement the exercise rehabilitation method for akinesia in patients with cardiovascular disease and frailty as described in any one of claims 1 to 6.

Images