A rehabilitation nursing management system based on artificial intelligence
By combining multi-source heterogeneous sensing and Riemannian manifold mapping techniques with federated learning and dynamic systems, a homogeneous feature matrix is generated, which solves the data processing and privacy protection problems in intelligent nursing systems and enables efficient rehabilitation pathway planning and resource scheduling.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Filing Date
- 2026-06-01
- Publication Date
- 2026-07-14
AI Technical Summary
Existing intelligent nursing systems suffer from limitations in data processing, insufficient privacy protection, low model training efficiency, lack of dynamism and predictability in rehabilitation path planning, and disconnect between decision-making and execution when facing complex rehabilitation scenarios.
Multi-source heterogeneous sensing modules are used to map multimodal data to a low-dimensional Riemannian manifold space. Combined with federated learning and dynamical systems theory, homogenized feature matrices are generated. Gradient residuals are uploaded through a federated distillation mechanism. Nursing tasks are generated using a large language model and assigned using game theory.
It achieves accurate representation of multimodal data, improves model training efficiency and prediction accuracy, dynamically generates safe and effective rehabilitation pathways, and improves the efficiency and response speed of nursing resource scheduling.
Smart Images
Figure CN122392786A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based rehabilitation and nursing management system. Background Technology
[0002] With the accelerating aging of the population, the demand for rehabilitation care is experiencing explosive growth, and utilizing artificial intelligence technology to assist rehabilitation care has become an important development direction for smart healthcare. Currently, most existing intelligent care systems rely on monitoring single or simple physiological parameters. While this has alleviated the burden on medical staff to some extent, significant technical bottlenecks still exist when facing complex rehabilitation scenarios.
[0003] Existing technologies have limitations in data processing. Rehabilitation and nursing data usually come from wearable devices, environmental sensors, and audio-visual equipment, and have the characteristics of being multi-source, heterogeneous, and high-dimensional.
[0004] Existing technologies mostly employ linear processing methods in Euclidean space, which makes it difficult to effectively preserve the nonlinear topological structure between cross-modal data, leading to information loss or semantic gaps during feature extraction.
[0005] In terms of privacy protection and model training, traditional centralized machine learning requires uploading sensitive patient data to the cloud, which poses a significant risk of privacy leakage; while conventional federated learning often faces problems of high communication overhead and slow model convergence when dealing with non-independent and identically distributed data.
[0006] Rehabilitation pathway planning lacks dynamism and foresight. Existing systems mostly set nursing plans based on static rules, which cannot treat patient behavior as a dynamic system for evolutionary analysis. It is difficult to capture chaotic inflection points in the behavioral trajectory in real time (such as pre-fall warnings), resulting in delayed risk warnings and a lack of optimality in pathway planning.
[0007] There is a disconnect between the decision-making and execution levels. The numerical warnings output by the underlying algorithms are difficult to translate directly into semantic instructions that medical staff can understand. Furthermore, the task allocation lacks a comprehensive game theory consideration of personnel skills, location, and workload, resulting in low efficiency in the scheduling of nursing resources. Summary of the Invention
[0008] The purpose of this invention is to provide an artificial intelligence-based rehabilitation and nursing management system, which aims to solve or improve at least one of the above-mentioned technical problems.
[0009] To achieve the above objectives, the present invention provides the following solution: An artificial intelligence-based rehabilitation and nursing management system includes: The multi-source heterogeneous sensing module is used to acquire multimodal data in the rehabilitation and nursing process, and to map the high-dimensional multimodal data to a low-dimensional Riemannian manifold space to generate a homogeneous feature matrix. The federated learning module predicts the physiological state of nursing targets locally based on the homogeneous feature matrix, and adopts a federated distillation mechanism, uploading only the gradient residuals. The nursing plan module analyzes the behavior of nursing goals based on multimodal data and uses dynamic systems theory to generate rehabilitation pathways with risk warnings. The multimodal large model decision module obtains rehabilitation pathways with risk warnings combined with nursing goal profiles, and generates nursing tasks based on a large language model; The task collaboration management module transforms nursing tasks into actual nursing actions after medical staff confirm them, and then assigns them to the optimal nursing staff using game theory principles.
[0010] Furthermore, the multi-source heterogeneous sensing module includes: The multimodal data acquisition unit acquires multimodal data of nursing goals through wearable devices, environmental sensors, and emotional interaction devices, and performs data cleaning to generate a structured feature matrix. The manifold space mapping unit adaptively adjusts the manifold curvature by the structured feature matrix, preserves the cross-modal topology, and generates a homogeneous feature matrix.
[0011] Furthermore, the federated learning module includes: The LSTM prediction unit predicts the current physiological state of the nursing target locally based on the homogenized feature matrix using the LSTM model. Riemannian covariant element, calculating gradient in the tangent space of the manifold; The residual subnet unit calculates the residual between the local model and the global model and uploads the gradient residual.
[0012] Furthermore, the Riemannian covariant unit includes: In the formula, Let be the Riemann covariant derivative, representing the loss function. In manifold Regarding parameters The gradient; For Christofel's symbol; The loss function; The first parameter of the local model One component; This represents a tiny change in the coordinates; Let be the partial derivative basis vectors on the manifold.
[0013] Furthermore, the nursing care protocol module includes: The dynamics analysis unit treats the behavior of nursing goals as a dynamic system, simulates the evolution process, and constructs evolution equations; The risk detection unit monitors the Lyapunov index of nursing goals in real time, detects behavioral inflection points of nursing goals, and marks them as risk points; The rehabilitation pathway planning unit, when generating risk points, solves the minimum cost function based on the Monge-Kandrovich optimal transport theory to generate a rehabilitation pathway from the current state to the rehabilitation goal.
[0014] Furthermore, the risk detection unit includes: The expression for the maximum Lyapunov exponent is: In the formula, The maximum Lyapunov index; End time; The start time; This represents the total number of time steps. It is an orthogonal matrix; The tangent space evolution matrix; This is the initial perturbation vector; If the maximum Lyapunov index is greater than 0, it is judged as a behavioral inflection point and marked as a risk point.
[0015] Furthermore, the cost function is expressed as: In the formula, To achieve the optimal nursing pathway; For the transmission plan, a mapping from the source distribution to the target distribution is defined; These are the initial state distribution and the target state distribution, respectively. Let $\mathbf{a}$ be the transition cost function, representing the transition from state $\mathbf{a} Transferred to The cost; This is the penalty coefficient for deviation from the inflection point, used to control the degree of avoidance of risk points; The set of detected risk points; This refers to the actual physiological state; The target physiological state; The range on the manifold measures the state difference.
[0016] Furthermore, the task collaboration management module includes: The meta-reinforcement learning unit adjusts the policy output by updating and adjusting the feature weighting coefficients; The task assignment unit comprehensively evaluates the multidimensional tags of nursing staff and generates assignment instructions.
[0017] Furthermore, the meta-reinforcement learning unit includes: In the formula, The weighting coefficient for the i-th feature at time t; For the feature gradient; For the policy sensitivity tensor; For risk temperature parameters; Let be the determinant of the matrix.
[0018] Furthermore, the task dispatch unit includes: In the formula, For comprehensive scoring; For skill matching; Location relevance indicates distance and route accessibility; Workload suitability indicates whether the current workload is saturated; Assign tasks to the overall score The highest level of nursing care.
[0019] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention discloses an AI-based rehabilitation and nursing management system. By constructing a multi-source heterogeneous perception and Riemannian manifold mapping mechanism, it effectively solves the problems of high-dimensional redundancy and cross-modal semantic gap in multimodal data in rehabilitation and nursing, achieving accurate representation of complex physiological characteristics. Combined with federated distillation technology, while ensuring patient data privacy and security, it significantly improves model training efficiency and prediction accuracy in a distributed environment by utilizing Riemannian covariant derivatives and gradient residual upload mechanisms. Introducing dynamical system theory and optimal transport planning, it can quantify nonlinear risk inflection points in the rehabilitation process in real time and dynamically generate rehabilitation paths that balance safety and efficacy. Furthermore, by integrating large-model decision-making and meta-reinforcement learning task collaboration mechanisms, it achieves intelligent transformation from numerical warnings to clinical semantic instructions and global game optimization of nursing resources, significantly improving the response speed, execution accuracy, and humanistic care level of rehabilitation and nursing, demonstrating significant clinical application value. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] The purpose of this invention is to provide an artificial intelligence-based rehabilitation and nursing management system, which aims to solve or improve at least one of the above-mentioned technical problems.
[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0025] like Figure 1 As shown, the present invention provides an artificial intelligence-based rehabilitation and nursing management system, comprising: A multi-source heterogeneous sensing module is used to acquire multimodal data during the rehabilitation and nursing process, and to map the high-dimensional multimodal data to a low-dimensional Riemannian manifold space to generate a homogeneous feature matrix, including: The multimodal data acquisition unit acquires multimodal data of nursing goals through wearable devices, environmental sensors, and emotional interaction devices, and performs data cleaning to generate a structured feature matrix. Data cleaning includes noise reduction and alignment.
[0026] Wearable devices include ECG and SpO2 sensors; environmental sensors include temperature and humidity sensors and fall detection sensors; and emotional interaction devices include voice devices and facial expression video stream acquisition devices.
[0027] The manifold space mapping unit adaptively adjusts the manifold curvature using the structured feature matrix, preserving the cross-modal topology, and generates a homogeneous feature matrix, expressed as: In the formula, A metric tensor describes distances and angles on a manifold and is used to define the geometric properties of space; This is the natural exponential function, used to map distance to similarity weights; This is the curvature adjustment factor, a hyperparameter used to control the degree of manifold bending; The gradient operator represents the partial derivative of the input feature x. This is a feature alignment loss function used to measure the differences between features of different modalities and to guide the construction of manifold structures.
[0028] The federated learning module predicts the physiological state of the nursing target locally based on the homogeneous feature matrix. It employs a federated distillation mechanism, uploading only the gradient residuals, including: The LSTM prediction unit predicts the current physiological state of the nursing target locally based on the homogenized feature matrix using the LSTM model. The expression is as follows: In the formula, for Predictable physiological state at any time; It is a long short-term memory network model used to process time-series data; These are the learnable parameters for the local model, including the weights and biases of the LSTM. The input feature sequence represents the sequence from... Time's up Multimodal data at any given time; The Riemannian covariant element calculates the gradient in the tangent space of the manifold, avoiding the dimensional distortion caused by traditional Euclidean space calculations. Its expression is: In the formula, Let be the Riemann covariant derivative, representing the loss function. In manifold Regarding parameters The gradient; The Christofel notation describes manifold connections and is used to correct for derivative deviations caused by coordinate system changes. This is the loss function, which measures the error between the predicted value and the actual value. The first parameter of the local model One component; This represents a tiny change in the coordinates; These are the partial derivative basis vectors on the manifold; The residual subnet unit calculates the residual between the local model and the global model, and uploads the gradient residual. The expression is: In the formula, This is the gradient residual vector, representing the difference between the local model and the global model; A logarithmic mapping maps points on a manifold to vectors in the tangent space. The projection operator projects the gradient of the global model onto the tangent space of the local manifold; This represents the gradient of the global model.
[0029] The nursing plan module, based on multimodal data, uses dynamic systems theory to analyze the behavior of nursing goals and generates a rehabilitation pathway with risk warnings, including: The dynamic analysis unit treats the behavior of nursing goals as a dynamic system, simulates the evolution process, and constructs an evolution equation, the expression of which is: In the formula, Let z be the rate of change of the hidden state z over time; The operator matrix, which is related to the model parameters, is used to describe the linear dynamics of the system; The Hamiltonian gradient is used to describe the conservative force component of the system. It is an antisymmetric matrix used to introduce nonlinear chaos and simulate the memory effect of behavior; The sign function represents the time delay. Behavioral direction in a given state; The risk detection unit monitors the Lyapunov index of nursing goals in real time, detects behavioral inflection points of nursing goals, and marks them as risk points, including: The expression for the maximum Lyapunov exponent is: In the formula, The maximum Lyapunov exponent measures the rate at which the system trajectory diverges or converges. End time; The start time; This represents the total number of time steps. It is an orthogonal matrix, obtained through QR decomposition, used to track the direction of disturbance; Let be the tangent space evolution matrix, which describes the evolution of the perturbation over time; This is the initial perturbation vector; If the maximum Lyapunov index is greater than 0, it is judged as a behavioral inflection point and marked as a risk point.
[0030] The rehabilitation pathway planning unit, when generating risk points, solves for minimizing the cost function based on the Monge-Kandrovich optimal transport theory to generate a rehabilitation path from the current state to the rehabilitation goal, including: The cost function is expressed as: In the formula, To achieve the optimal nursing pathway; For the transmission plan, a mapping from the source distribution to the target distribution is defined; These are the initial state distribution and the target state distribution, respectively. Let $\mathbf{a}$ be the transition cost function, representing the transition from state $\mathbf{a} Transferred to The cost; This is the penalty coefficient for deviation from the inflection point, used to control the degree of avoidance of risk points; The set of detected risk points; This refers to the actual physiological state; The target physiological state; The range on the manifold measures the state difference.
[0031] The multimodal large model decision module obtains rehabilitation pathways with risk warnings combined with nursing goal profiles, and generates nursing tasks based on a large language model.
[0032] The large language model is based on an external medical knowledge base as its training database.
[0033] The task collaboration management module, after medical staff confirm the nursing task, transforms the task into actual nursing actions and assigns them to the optimal nurse using game theory principles, including: The meta-reinforcement learning unit adjusts the policy output by updating and adjusting the feature weighting coefficients, as expressed in the following expression: In the formula, The weighting coefficient of the i-th feature at time t is used to determine the degree of influence of the current feature on the decision. For the feature gradient; For the policy sensitivity tensor; For risk temperature parameters; It is the determinant of the matrix, used to measure the volume change after transformation; The task assignment unit comprehensively evaluates the multidimensional labels of nursing staff and generates assignment instructions, expressed as: In the formula, For comprehensive scoring; For skill matching; Location relevance indicates distance and route accessibility; Workload suitability indicates whether the current workload is saturated; Assign tasks to the overall score The highest level of nursing care.
[0034] In one embodiment, a scenario is presented in the orthopedic ward of a tertiary hospital, where patient Zhang (68 years old) is in the early rehabilitation phase on the third day after undergoing joint replacement surgery for a left femoral neck fracture. The technical solution of this invention will be described below.
[0035] Patients wear smart bracelets (continuously collecting heart rate, ECG, and blood oxygen SpO2), and temperature and humidity sensors in the ward record environmental data. When nurses use tablets to talk to patients, emotional interaction devices capture the patient's micro-expressions (painful facial expressions) and tone of voice (trembling).
[0036] After cleaning the aforementioned multimodal data, dimensionality reduction was performed using a manifold space mapping unit. Due to the patient's pain state, feature alignment between speech and facial expressions was lost. If the curvature is too high, the system will automatically increase the curvature adjustment factor. This makes the manifold more closely fit the current "pain-anxiety" topology, generating a homogeneous feature matrix. .
[0037] The edge computing gateway in the ward runs a lightweight LSTM model, taking the feature sequence from the past 15 minutes as input. Predicting the patient's current cardiovascular stress state .
[0038] The gateway calculates the local gradient in the Riemann tangent space and compares it with the global teacher model distributed from the cloud to generate a gradient residual vector. The data was encrypted and uploaded to the cloud to participate in global model iteration, without leaking any original physiological data throughout the process.
[0039] The system detected a latent state when the patient attempted to stand. The maximum Lyapunov index was calculated after the dramatic fluctuations. The system immediately identified a "behavioral inflection point" and marked it as a high-risk point for a fall. .
[0040] Based on optimal transport theory, the system incorporates risk points into the cost function. High penalties ( (Weight increase).
[0041] Large models acquire structured data input, including: Risk point: Abnormal Lyapunov index of nursing goals .
[0042] Path constraints: The optimal transport plan has masked all "getting out of bed and walking" action nodes.
[0043] Nursing goal profile: 68-year-old female, 3 days post-operatively for femoral neck fracture, current emotional label: high anxiety / pain.
[0044] The large model generates task requirements based on medical knowledge: the original plan of "getting out of bed and walking 20 meters 3 times a day" is adjusted to "stop getting out of bed today and instead perform passive joint exercises in bed + psychological comfort".
[0045] Because the patient's emotional anxiety (risk temperature parameter) was detected (Upgrade), the meta-reinforcement learning unit rapidly increased the weight of the "communication skills" feature. .
[0046] The system iterates through the pool of nurses on duty and calculates that nurse A has the highest overall score. ), because she not only possesses orthopedic rehabilitation skills ( High), currently responsible for a small number of patients ( (High), and it's in the adjacent ward ( (High), and more importantly, she is skilled in geriatric psychological counseling. The system immediately pushed a personalized intervention task to Nurse A's handheld terminal: "Go to bed 3, perform bed joint exercises, and use the script template to soothe emotions."
[0047] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0048] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A rehabilitation and nursing management system based on artificial intelligence, characterized in that, include: The multi-source heterogeneous sensing module is used to acquire multimodal data in the rehabilitation and nursing process, and to map the high-dimensional multimodal data to a low-dimensional Riemannian manifold space to generate a homogeneous feature matrix. The federated learning module predicts the physiological state of nursing targets locally based on the homogeneous feature matrix, and adopts a federated distillation mechanism, uploading only the gradient residuals. The nursing plan module analyzes the behavior of nursing goals based on multimodal data and uses dynamic systems theory to generate rehabilitation pathways with risk warnings. The multimodal large model decision module obtains rehabilitation pathways with risk warnings combined with nursing goal profiles, and generates nursing tasks based on a large language model; The task collaboration management module transforms nursing tasks into actual nursing actions after medical staff confirm them, and then assigns them to the optimal nursing staff using game theory principles.
2. The rehabilitation and nursing management system based on artificial intelligence according to claim 1, characterized in that, The multi-source heterogeneous sensing module includes: The multimodal data acquisition unit acquires multimodal data of nursing goals through wearable devices, environmental sensors, and emotional interaction devices, and performs data cleaning to generate a structured feature matrix. The manifold space mapping unit adaptively adjusts the manifold curvature by the structured feature matrix, preserves the cross-modal topology, and generates a homogeneous feature matrix.
3. The rehabilitation and nursing management system based on artificial intelligence according to claim 1, characterized in that, The federated learning module includes: The LSTM prediction unit predicts the current physiological state of the nursing target locally based on the homogenized feature matrix using the LSTM model. Riemannian covariant element, calculating gradient in the tangent space of the manifold; The residual subnet unit calculates the residual between the local model and the global model and uploads the gradient residual.
4. The artificial intelligence-based rehabilitation and nursing management system according to claim 3, characterized in that, The Riemann covariant unit includes: In the formula, Let be the Riemann covariant derivative, representing the loss function. In manifold Regarding parameters The gradient; For Christofel's symbol; The loss function; The first parameter of the local model One component; This represents a tiny change in the coordinates; Let be the partial derivative basis vectors on the manifold.
5. The rehabilitation and nursing management system based on artificial intelligence according to claim 1, characterized in that, The nursing care plan module includes: The dynamics analysis unit treats the behavior of nursing goals as a dynamic system, simulates the evolution process, and constructs evolution equations; The risk detection unit monitors the Lyapunov index of nursing goals in real time, detects behavioral inflection points of nursing goals, and marks them as risk points; The rehabilitation pathway planning unit, when generating risk points, solves the minimum cost function based on the Monge-Kandrovich optimal transport theory to generate a rehabilitation pathway from the current state to the rehabilitation goal.
6. The rehabilitation and nursing management system based on artificial intelligence according to claim 5, characterized in that, The risk detection unit includes: The expression for the maximum Lyapunov exponent is: In the formula, The maximum Lyapunov index; End time; The start time; This represents the total number of time steps. It is an orthogonal matrix; The tangent space evolution matrix; This is the initial perturbation vector; If the maximum Lyapunov index is greater than 0, it is judged as a behavioral inflection point and marked as a risk point.
7. The rehabilitation and nursing management system based on artificial intelligence according to claim 5, characterized in that, The cost function is expressed as: In the formula, To achieve the optimal nursing pathway; For the transmission plan, a mapping from the source distribution to the target distribution is defined; These are the initial state distribution and the target state distribution, respectively. Let $\mathbf{a}$ be the transition cost function, representing the transition from state $\mathbf{a} Transferred to The cost; This is the penalty coefficient for deviation from the inflection point, used to control the degree of avoidance of risk points; The set of detected risk points; This refers to the actual physiological state; The target physiological state; The range on the manifold measures the state difference.
8. The rehabilitation and nursing management system based on artificial intelligence according to claim 1, characterized in that, The task collaboration management module includes: The meta-reinforcement learning unit adjusts the policy output by updating and adjusting the feature weighting coefficients; The task assignment unit comprehensively evaluates the multidimensional tags of nursing staff and generates assignment instructions.
9. The rehabilitation and nursing management system based on artificial intelligence according to claim 8, characterized in that, The meta-reinforcement learning unit includes: In the formula, The weighting coefficient for the i-th feature at time t; For the feature gradient; For the policy sensitivity tensor; For risk temperature parameters; Let be the determinant of the matrix.
10. A rehabilitation and nursing management system based on artificial intelligence according to claim 8, characterized in that, The task dispatching unit includes: In the formula, For comprehensive scoring; For skill matching; Location relevance indicates distance and route accessibility; Workload suitability indicates whether the current workload is saturated; Assign tasks to the overall score The highest level of nursing care.