Knee joint personalized rehabilitation scheme generation method and system based on digital twinning

Abstract

The invention provides a knee joint personalized rehabilitation scheme generation method and system based on digital twinning, and the method comprises the steps: obtaining the multi-dimensional feature data of a patient to construct a patient rehabilitation data set, constructing a knee joint injury knowledge graph based on the patient rehabilitation data set, constructing a knee joint minimal injury model based on the knee joint injury knowledge graph, and generating a knee joint personalized rehabilitation scheme. Functional weak links and injury degrees of the patient during rehabilitation training are accurately recognized, a data basis is provided for subsequently formulating a personalized scheme for the condition of the patient, emotion extraction is performed on voice feedback data in a patient rehabilitation data set, the patient emotion index is obtained, rehabilitation efficiency diagnosis is performed according to the patient emotion index, and the rehabilitation efficiency of the patient is improved. According to the method, the initial rehabilitation scheme of the knee joint is generated, the initial rehabilitation scheme of the knee joint is subjected to efficiency evaluation and scheme optimization, and the rehabilitation scheme of the knee joint is generated, so that the rehabilitation scheme is dynamically adjusted according to the personalized requirements of the patient while the personalized requirements of the patient are met, and the accuracy and efficiency of the generated rehabilitation scheme are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method and system for generating personalized knee joint rehabilitation programs based on digital twins. Background Technology

[0002] Developing knee rehabilitation programs is a crucial research area in the medical and health field, directly impacting patients' quality of life and the recovery of motor function. With an aging population and increasing sports injuries, the demand for personalized rehabilitation programs is growing rapidly. Effective rehabilitation programs can help patients quickly regain joint mobility and reduce the risk of secondary injuries. However, existing knee rehabilitation methods still have limitations in meeting patients' individual needs and dynamically adjusting rehabilitation programs.

[0003] On the one hand, existing knee rehabilitation methods mostly rely on standardized training plans, which are usually based on doctors' experience or general templates. These plans lack accurate capture of the patient's real-time condition and fail to fully consider individual differences and dynamic changes. When faced with complex rehabilitation needs, these methods often cannot address specific weaknesses in the patient's knee joint function. On the other hand, patients' acceptance of training programs and their willingness to implement them vary greatly. Existing methods often lack an effective mechanism to incorporate the patient's subjective feedback while evaluating the rehabilitation plan, resulting in poor implementation of the rehabilitation plan. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating personalized knee joint rehabilitation programs based on digital twins, the method comprising: Acquire multi-dimensional feature data of patients and construct a patient rehabilitation dataset; A knowledge graph of knee joint injury was constructed based on the aforementioned patient rehabilitation dataset. A model of minor knee injuries was constructed based on the aforementioned knee injury knowledge graph. Emotion extraction was performed on the voice feedback data within the patient rehabilitation dataset to obtain the patient's emotion index; Based on the patient's emotional index, rehabilitation efficiency is diagnosed, and an initial rehabilitation plan for the knee joint is generated. The effectiveness of the initial knee rehabilitation program is evaluated and the program is optimized to generate a new knee rehabilitation plan.

[0005] As a further aspect of the present invention, a knee joint micro-injury model is constructed based on the knee joint injury knowledge graph, including: The data corresponding to the knee joint biomechanical nodes and knee joint image nodes contained in the knee joint injury knowledge graph are input into a deep learning neural network to generate knee joint injury simulation data. Based on the temporal and causal relationships between nodes in the knee injury knowledge graph, a preliminary mapping rule for knee injury is constructed. The preliminary mapping rule for knee injury is iteratively corrected based on the data corresponding to the rehabilitation tag nodes in the knee injury knowledge graph until the similarity between the mapping result corresponding to the preliminary mapping rule for knee injury and the data corresponding to the rehabilitation tag nodes is greater than a preset similarity threshold, thus generating a knee injury mapping rule. Based on the knee joint injury simulation data and knee joint injury mapping rules, an original knee joint micro-injury model is constructed, and the finite element algorithm is used for simulation verification. Based on the simulation verification results, the original knee joint micro-injury model is optimized and corrected to generate a new knee joint micro-injury model.

[0006] As a further aspect of the present invention, emotion extraction is performed on the voice feedback data within the patient rehabilitation dataset to obtain the patient's emotion index, including: Obtain the text-based feedback data corresponding to the voice feedback nodes contained in the knee joint injury knowledge graph, as well as the corresponding raw voice audio data. Based on the BERT model, keywords are extracted from textual feedback data to generate a set of patient emotion keywords. Each emotion keyword in the patient emotion keyword set is classified into two emotion categories: positive emotion and negative emotion. The probability distribution of the emotion category corresponding to the emotion keyword is output based on the Softmax activation function, and the probability value of positive emotion is output as the emotion polarity score. Feature extraction is performed on the raw audio data to obtain an audio feature set, and combined with the corresponding text-based feedback data to obtain patient emotional tendency data; Obtain the corresponding data of the biomechanical nodes of the knee joint in the knowledge graph of knee joint injury, and compare them with the preset fatigue degree threshold and load state threshold. Generate the patient's physiological load data based on the comparison results. A multimodal fusion network is constructed based on the attention mechanism. The emotional polarity score, patient emotional tendency data, and patient physiological load data are adaptively weighted using a dynamic weight allocation strategy to generate a patient emotional index.

[0007] As a further aspect of the present invention, rehabilitation efficiency is diagnosed based on the patient's emotional index, and an initial rehabilitation plan for the knee joint is generated, including: Based on the patient rehabilitation dataset, the best historical rehabilitation plan is matched with the preset best rehabilitation plan library, and the best historical rehabilitation plan is sent to the user terminal. The patient's emotional index when the patient executes the best historical rehabilitation plan is monitored in real time, and the patient rehabilitation dataset is updated synchronously. If the patient's emotional index is within the preset emotional index threshold range, the patient's emotional index will continue to be monitored and the patient's rehabilitation dataset will be updated synchronously. If the patient's emotional index is not within the preset emotional index threshold range, the updated patient rehabilitation dataset is input into the knee joint micro-injury model, the patient rehabilitation dataset is converted into knee joint injury simulation data corresponding to the knee joint micro-injury model, and the weak areas of the patient's rehabilitation training are located according to the knee joint injury mapping rules contained in the knee joint micro-injury model, and rehabilitation training optimization data is generated. Among them, the data for optimizing rehabilitation training should include at least the location of the weak areas in rehabilitation training, the upper limit of stress tolerance, and the current degree of injury; The rehabilitation training optimization data is analyzed based on the proximal strategy optimization algorithm to obtain the decision on adjusting the rehabilitation training plan; Based on the optimized rehabilitation training data and the rehabilitation training program adjustment decisions, an initial rehabilitation plan for the knee joint is generated.

[0008] As a further aspect of the present invention, the rehabilitation training optimization data is analyzed based on the proximal strategy optimization algorithm to obtain a rehabilitation training program adjustment decision, including: Based on rehabilitation training optimization data, and combined with the corresponding data of rehabilitation tag nodes and knee joint biomechanical nodes in the knee joint injury knowledge graph, a state space is constructed. With the goal of reducing stress load in weak areas of rehabilitation training, a reward function is set, and a proximal strategy optimization algorithm is used for iterative training to generate rehabilitation training program adjustment decisions.

[0009] As a further aspect of the present invention, the effectiveness of the initial knee joint rehabilitation program is evaluated and the program is optimized to generate a knee joint rehabilitation program, including: The initial rehabilitation plan for the knee joint is input into the knee joint micro-injury model, and the model prediction control algorithm is used to simulate and predict, and the simulation prediction results are obtained. The simulation prediction results include at least the predicted rehabilitation effect data and the predicted safety risk data. The predicted rehabilitation effect data includes at least the stress reduction magnitude of the injury area, the improvement rate of the knee joint range of motion, and the improvement curve of the knee joint range of motion. The simulation prediction results are validated by combining the causal relationships between nodes in the set of knee joint injury nodes contained in the knee joint injury knowledge graph. If the simulation prediction results fail the verification, the parameters of the initial knee joint rehabilitation plan will be adjusted until the verification is passed. If the simulation prediction results pass the verification, the corresponding initial knee joint rehabilitation plan will be sent to the user's terminal as a knee joint rehabilitation plan.

[0010] As a further aspect of the present invention, the simulation prediction results are validated by combining the causal relationships between nodes within the knee joint injury node set contained in the knee joint injury knowledge graph, including: Obtain the causal relationship between knee joint biomechanical nodes and rehabilitation tag nodes in the knee joint injury knowledge graph, and define it as a rehabilitation association rule; The extent of stress reduction in the damaged area was used as the verification index, and the damage repair efficiency was verified by combining rehabilitation correlation rules. The rate of improvement in knee joint range of motion was used as the verification index, and the stability of rehabilitation effect was verified by combining rehabilitation association rules. The improvement curve of knee joint range of motion was used as the verification index, and the sustainability of rehabilitation effect was verified by combining rehabilitation correlation rules. Obtain the causal relationship between the knee joint biomechanical nodes and the voice feedback nodes in the knee joint injury knowledge graph, and verify the tolerance of rehabilitation training.

[0011] As a further aspect of the present invention, multi-dimensional feature data of patients are obtained to construct a patient rehabilitation dataset, including: Collect knee joint status data of patients during rehabilitation exercises and generate a continuous knee joint status sequence; Feature extraction is performed on the knee joint state sequence to obtain knee joint biomechanical data and knee joint activity data; Simultaneously collect the patient's voice information during rehabilitation exercises to generate the patient's voice feedback data, which includes at least the patient's subjective feeling data, pain description data, and compliance feedback data. Feature extraction is performed on the patient's knee joint pathology report to obtain knee joint imaging data and knee joint diagnostic data. The knee joint imaging data includes at least knee joint structural data, and the knee joint diagnostic data includes at least injury degree data and injury rehabilitation stage data. A patient rehabilitation dataset was constructed based on the aforementioned knee joint biomechanical data, knee joint activity data, voice feedback data, knee joint imaging data, and knee joint diagnostic data.

[0012] As a further aspect of the present invention, a knowledge graph of knee joint injury is constructed based on the patient rehabilitation dataset, including: The knee joint biomechanical data and knee joint activity data in the patient rehabilitation dataset are defined as knee joint biomechanical nodes, the voice feedback data in the patient rehabilitation dataset are defined as voice feedback nodes, the knee joint imaging data in the patient rehabilitation dataset are defined as knee joint imaging nodes, and the knee joint diagnostic data in the patient rehabilitation dataset are defined as rehabilitation label nodes, thereby generating a set of knee joint injury nodes. The data corresponding to each node in the set of knee joint injury nodes are timestamped to obtain the temporal correlation between each node in the set of knee joint injury nodes. By adjusting the corresponding data of each node in the knee joint injury node set and monitoring the data changes of nodes in the knee joint injury node set that have not undergone data adjustment, the causal relationship between each node in the knee joint injury node set can be obtained. A knowledge graph of knee joint injury is constructed based on the set of knee joint injury nodes and the temporal and causal relationships between the nodes in the set.

[0013] Furthermore, embodiments of the present invention also provide a system for generating personalized knee joint rehabilitation plans based on digital twins, including: The data acquisition module is used to acquire multi-dimensional feature data of patients with knee joint injuries and construct a patient rehabilitation dataset. A knowledge graph construction module is used to construct a knowledge graph of knee joint injury based on the patient rehabilitation dataset. A model building module is used to build a model of minor knee injuries based on a knowledge graph of knee injuries. An emotion extraction module is used to extract emotions from the voice feedback data in the patient rehabilitation dataset to obtain the patient's emotion index. The plan generation module is used to diagnose rehabilitation efficiency based on the patient's emotional index and generate an initial rehabilitation plan for the knee joint. The program optimization module is used to evaluate the effectiveness of the initial knee joint rehabilitation program and optimize the program to generate a knee joint rehabilitation program.

[0014] Compared with the prior art, the present invention has the following beneficial effects: Multidimensional feature data of patients are acquired to construct a patient rehabilitation dataset. Based on the patient rehabilitation dataset, a knee joint injury knowledge graph is constructed. By collecting biomechanical parameters and range of motion data of the injured knee joint in motion state, as well as knee joint imaging data, and combining speech recognition technology to synchronize patient feedback in real time, a patient rehabilitation dataset is constructed, providing data support for subsequent simulation and solution generation. At the same time, by generating a knee joint injury knowledge graph, the discrete and complex data in the patient rehabilitation dataset are integrated into a data aggregate with correlation between data, improving the efficiency of subsequent data analysis. Based on the knee joint injury knowledge graph, a knee joint micro-injury model is constructed. Digital twin technology is used to construct the knee joint micro-injury model. Based on the knee joint micro-injury model, the data contained in the knee joint injury knowledge graph is mapped to the patient's simulation virtual model, thereby accurately identifying the patient's functional weaknesses and the degree of injury during rehabilitation training, providing a data foundation for the subsequent development of personalized plans for the patient's condition. Emotional extraction is performed on the voice feedback data within the patient rehabilitation dataset to obtain the patient's emotional index. Based on the patient's emotional index, rehabilitation efficiency is diagnosed, and an initial rehabilitation plan for the knee joint is generated. The effectiveness of the initial rehabilitation plan for the knee joint is evaluated and optimized to generate a new knee joint rehabilitation plan. This allows the rehabilitation plan to be dynamically adjusted according to the patient's individual needs while meeting the individual needs of the patient, thereby improving the accuracy and efficiency of the generated rehabilitation plan. Attached Figure Description

[0015] Figure 1 This is a flowchart of the steps in the method for generating a personalized knee joint rehabilitation plan based on digital twins according to the present invention. Figure 2 This is a flowchart of step S1 in the method for generating a personalized knee rehabilitation plan based on digital twins in this invention. Figure 3 This is a schematic diagram of the knee joint personalized rehabilitation program generation system based on digital twins of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating the steps of the method for generating personalized knee joint rehabilitation plans based on digital twins, as described in this invention. Figure 2 This is a flowchart of step S1 in the method for generating a personalized knee rehabilitation plan based on digital twins according to the present invention. The following is a detailed introduction to the method for generating a personalized knee rehabilitation plan based on digital twins.

[0017] Step S1: Obtain multi-dimensional feature data of patients and construct a patient rehabilitation dataset.

[0018] In this embodiment, step S1 includes: Step S11: Obtain knee joint biomechanical data and knee joint activity data.

[0019] Specifically, knee joint status data is collected during the patient's rehabilitation exercises to generate a continuous knee joint status sequence. Feature extraction is performed on the knee joint status sequence to obtain knee joint biomechanical data and knee joint activity data. The knee joint biomechanical data includes at least six-degree-of-freedom kinematic parameters, joint torque, and plantar pressure distribution map of the knee joint in flexion, extension, and rotation. The knee joint activity data includes at least the range of motion of the knee joint.

[0020] In one possible embodiment, a six-axis inertial measurement unit wearable sensor is selected to collect six degrees of freedom kinematic parameters, joint torques, and knee joint range of motion during flexion, extension, and rotation, such as knee flexion angle [0°, 150°] and knee internal or external rotation angle [0°, 30°].

[0021] Plantar pressure sensors are used to collect plantar pressure distribution data during patients' rehabilitation training. Based on this data, a plantar pressure distribution map is constructed. The plantar pressure distribution data in the map can quantify the peak pressure and pressure center trajectory parameters during a gait cycle, thus helping to determine whether the pressure on the inner and outer sides of the patient's knee joint is balanced. For example, if the plantar pressure distribution map shows that the pressure is concentrated in the patient's forefoot or hindfoot during knee joint rehabilitation training, it may lead to abnormal load on the patient's knee joint, thereby affecting the rehabilitation process. By developing targeted balance training and gait correction training, the plantar pressure distribution can be improved, thereby reducing abnormal load on the knee joint and promoting better rehabilitation results.

[0022] Understandably, wavelet transform algorithm is used to denoise the collected raw data, outlier removal is performed based on the 3σ principle, and timestamp alignment is performed to generate a continuous knee joint state sequence. By extracting features from the preprocessed data, the corresponding knee joint biomechanical data and knee joint activity data can be obtained.

[0023] Step S12: Obtain voice feedback data.

[0024] Specifically, the system collects the patient's voice information during rehabilitation exercises to generate voice feedback data, which includes at least the patient's subjective feelings, pain description data, and compliance feedback data.

[0025] In one possible embodiment, a lapel microphone is used in conjunction with a pre-installed speech recognition module to recognize the patient's speech during rehabilitation training. The recognized content includes the patient's feedback on training intensity, such as "This movement is so tiring," pain descriptions, such as "Why is my inner knee aching a little?", willingness to perform, such as "Although it's tiring, I can persevere and complete the rest of the training to recover faster," and evaluation of movement difficulty, such as "This training movement is too difficult for me; I can hardly complete a standard movement." Based on the speech recognition module, the recognized content of the patient's speech is converted into text data, and natural language processing is used to extract keywords, label sentiment (positive, negative, neutral), and record feedback timestamps. Simultaneously, the timestamps are aligned with the collected knee joint biomechanical data and knee joint activity data.

[0026] Step S13: Obtain knee joint imaging data and knee joint diagnostic data.

[0027] Specifically, feature extraction is performed on the patient's knee joint pathology report to obtain knee joint imaging data and knee joint diagnostic data. The knee joint imaging data includes at least knee joint structural data, and the knee joint diagnostic data includes at least damage degree data and damage rehabilitation stage data.

[0028] In one possible embodiment, feature extraction is performed on the knee joint medical images in the patient's pathology report to obtain knee joint structural data. The knee joint structural data is then extracted from the patient's MRI and CT images using an image segmentation tool to obtain the patient's ligament length data, ligament diameter data, cartilage thickness data, and bone marrow edema area volume data. For example, a patient's anterior cruciate ligament length is 31 mm, ligament diameter is 2.4 mm, femoral trochlear cartilage thickness is 2.3 mm, and bone marrow edema area volume is 1.3 cm³. Combined with clinical diagnostic records, injury-related diagnostic information is obtained.

[0029] Based on the patient's clinical medical records, the following information is obtained: the type of knee injury, such as a minor tear of the anterior cruciate ligament; the degree of injury, such as mild, moderate, or severe, and the degree of injury is quantified and scored according to preset evaluation rules, such as a score of 85 points corresponding to mild knee injury; the rehabilitation stage, such as acute phase, subacute phase, or recovery phase, such as setting the first two weeks after the patient's knee injury as the acute phase; and past medical history, such as whether there is a history of knee surgery.

[0030] Step S14: Construct a patient rehabilitation dataset.

[0031] Specifically, a patient rehabilitation dataset is constructed based on knee biomechanical data, knee activity data, voice feedback data, knee imaging data, and knee imaging data and knee diagnostic data.

[0032] For example, part of the content of a record in the patient rehabilitation dataset is: {Knee joint biomechanical data collection timestamp: 10:15:10.000, Joint torque: 32 N·m, Knee joint range of motion: 80°, Voice feedback: Mild pain on the outside of the knee, Emotional tendency: Neutral, Anterior cruciate ligament length: 31 mm, Diagnostic label: [Minor tear of anterior cruciate ligament, mild injury, subacute phase]}.

[0033] Step S2: Construct a knowledge graph of knee joint injury based on the patient rehabilitation dataset.

[0034] In this embodiment, step S2 includes: Step S21: Construct a set of knee joint injury nodes.

[0035] Specifically, the knee joint biomechanical data and knee joint activity data in the patient rehabilitation dataset are defined as knee joint biomechanical nodes, the voice feedback data in the patient rehabilitation dataset are defined as voice feedback nodes, the knee joint imaging data in the patient rehabilitation dataset are defined as knee joint imaging nodes, and the knee joint diagnostic data in the patient rehabilitation dataset are defined as rehabilitation label nodes, thereby generating a set of knee joint injury nodes, wherein each node in the set of knee joint injury nodes retains the original data corresponding to the node attributes.

[0036] In one possible embodiment, the biomechanical node attributes of the knee joint include parameter name, parameter value, unit, acquisition timestamp, and normal range threshold. For example, the biomechanical node attributes of a certain knee joint are {joint torque, 32, N·m, 10:15:10.000, [0, 40]}.

[0037] The attributes of a voice feedback node include text content, sentiment, keywords, and collection timestamp. For example, the attributes of a voice feedback node are {slight pain on the outside of the knee, neutral, [pain, outside], 10:15:10.200}.

[0038] The attributes of a knee joint image node include the structure name, parameter value, unit, image acquisition timestamp, and reference interval. For example, the attributes of a knee joint image node are {anterior cruciate ligament, 31, mm, 10:35:20, [31, 35]}.

[0039] The attributes of a rehabilitation tag node include the injury type, injury severity, injury severity score, rehabilitation stage, and tag effective time. For example, a rehabilitation tag node attribute is {anterior cruciate ligament micro-tear, mild, 82, subacute phase, xx25.10.13}.

[0040] Step S22: Obtain the relationships between nodes within the set of knee joint injury nodes.

[0041] Specifically, the data corresponding to each node in the set of knee joint injury nodes are timestamped to obtain the temporal correlation between each node in the set of knee joint injury nodes; by adjusting the data corresponding to each node in the set of knee joint injury nodes and monitoring the data changes of nodes in the set of knee joint injury nodes that have not been adjusted, the causal correlation between each node in the set of knee joint injury nodes is obtained.

[0042] Understandably, temporal correlation represents the synchronous or sequential relationship of different nodes in the knee joint injury node set in the time dimension. The establishment of temporal correlation is determined by the difference in timestamps between nodes. If the difference in timestamps between two nodes is less than 1 second, it is determined that there is a temporal correlation between the two nodes. The strength of the temporal correlation is defined according to the difference in timestamps. The smaller the difference in timestamps between two nodes, the higher the corresponding temporal correlation strength between the two nodes.

[0043] For example, the timestamp of the knee biomechanical node is 10:00:15.000, and the timestamp of the voice feedback node is 10:00:15.200. The timestamp difference between the two nodes is 0.2s. Since the timestamp difference between the two nodes is lower than the preset timestamp difference threshold of 1s, it is determined that there is a temporal correlation between the knee biomechanical node and the voice feedback node. Assuming that the temporal correlation strength is calculated as "preset timestamp difference threshold - timestamp difference between the two nodes", the temporal correlation strength between the knee biomechanical node and the voice feedback node is 0.8.

[0044] The establishment of causal relationships is represented by determining the causal relationships between nodes within a set of knee joint injury nodes based on a pre-defined medical diagnostic database. The attributes of causal relationships include causal direction and association confidence.

[0045] For example, based on a pre-defined medical diagnostic database, the following causal relationship exists between the knee joint biomechanical node and the rehabilitation tag node: when the medial joint torque exceeds a threshold, the patient's ligaments will face a high risk of overload. The causal direction of this relationship is that the medial joint torque exceeding the threshold points to the ligaments facing a high risk of overload. Assuming that the confidence level of this causal relationship is obtained based on the percentage of similar cases in the pre-defined medical diagnostic database, the confidence level of this causal relationship is 0.95. Similarly, the following causal relationship exists between the voice feedback node and the biomechanical node: the reasons for frequent pain reports from patients include excessive training intensity and training movements that are not suitable for the patient's condition. The causal direction of this relationship is that frequent pain reports from patients simultaneously point to excessive training intensity and training movements that are not suitable for the patient's condition. Furthermore, if the percentage of similar cases in the pre-defined medical diagnostic database is 85%, the confidence level of this causal relationship is 0.85.

[0046] Step S23: Construct a knowledge graph of knee joint injuries.

[0047] Specifically, a knowledge graph of knee joint injury is constructed based on the set of knee joint injury nodes and the temporal and causal relationships between the nodes in the set of knee joint injury nodes.

[0048] In one possible embodiment, a graph database structure is adopted, with the nodes included in the set of knee joint injury nodes as the entities of the graph, and the temporal and causal relationships between the nodes in the set of knee joint injury nodes as the edges of the graph. The attributes of the edges include two types of association: temporal and causal association, as well as the temporal association strength or association confidence corresponding to the association type.

[0049] Step S3: Construct a knee joint micro-injury model based on the knee joint injury knowledge graph.

[0050] In this embodiment, step S3 includes: Step S31: Convert the data corresponding to the nodes in the knee injury knowledge graph into knee injury simulation data.

[0051] Specifically, the data corresponding to the knee joint biomechanical nodes and knee joint image nodes contained in the knee joint injury knowledge graph are input into a deep learning neural network to generate knee joint injury simulation data.

[0052] In one possible embodiment, the corresponding data of knee joint biomechanical nodes and knee joint imaging nodes in the knee joint injury knowledge graph are input into a pre-trained deep learning neural network model. The pre-training process of the model is based on a preset medical diagnostic database, which automatically learns the mapping relationship between the data features of the nodes in the knee joint injury knowledge graph and the injury simulation parameters, thereby outputting the corresponding knee joint injury simulation data.

[0053] For example, the range of motion of the knee joint in the biomechanical node is 80°, the joint torque is 32 N·m, the length of the anterior cruciate ligament in the knee joint imaging node is 31 mm, and the cartilage thickness is 2.1 mm. These values ​​are then input into a deep learning neural network model. The neural network model performs nonlinear operations through multiple convolutional and fully connected layers, outputting a knee flexion angle of 80°, a knee flexion torque of 32 N·m, a simulated ligament modeling length of 31 mm, and a ligament elastic modulus that decreases from 200 MPa in a healthy state to 180 MPa.

[0054] Step S32: Convert the relationships between nodes in the knee injury knowledge graph into knee injury mapping rules.

[0055] Specifically, a preliminary mapping rule for knee injury is constructed based on the temporal and causal relationships between nodes in the knee injury knowledge graph. The preliminary mapping rule is then iteratively revised using the data corresponding to the rehabilitation tag nodes in the knee injury knowledge graph as a benchmark, until the similarity between the mapping result corresponding to the preliminary mapping rule and the data corresponding to the rehabilitation tag nodes is greater than a preset similarity threshold, thus generating a knee injury mapping rule.

[0056] In one possible embodiment, by performing correlation analysis on the knee joint biomechanical data and changes in knee joint range of motion in the knee joint biomechanical node and the patient's knee joint diagnostic data in the rehabilitation tag node, an original knee joint injury temporal mapping rule is established. For example, when a patient's joint torque in the knee joint biomechanical node exceeds the predetermined normal range by 30% due to improper exercise, a temporary node for joint swelling symptoms will be added to the patient's corresponding knee joint injury knowledge graph. If intervention and treatment are not carried out in time, in subsequent periods, this temporary node will be converted into a rehabilitation tag node with the injury type of meniscus injury. This process presents a clear temporal development chain from biomechanical abnormality to symptom manifestation, symptom manifestation, and finally diagnosis. By conducting in-depth analysis of the temporal correlation between nodes in the knee joint injury knowledge graph, an original knee joint injury temporal mapping rule is established.

[0057] By conducting in-depth analysis of the causal relationships between nodes in the knowledge graph of knee joint injury using Bayesian networks, original causal mapping rules for knee joint injury are established. Taking the correlation analysis between knee joint biomechanical nodes and rehabilitation indicators as an example, a quantitative relationship between knee joint biomechanical nodes and rehabilitation indicators is established based on a pre-set medical diagnostic database. It is assumed that for every 10% increase in the patient's plantar pressure peak value beyond the preset plantar pressure peak value threshold, the patient's corresponding injury severity score increases by an average of 8 points. Thus, it can be inferred that when the patient's plantar pressure peak value exceeds the preset plantar pressure peak value threshold, there is a causal correlation with the injury severity score in the rehabilitation indicator node. It is also assumed that for every 10° increase in the knee flexion-extension angle beyond the upper limit of the normal range, the injury severity score increases by an average of 9 points, and for every 80 N·m of joint torque under load, the injury severity score increases by an average of 15 points. Similarly, it can be inferred that excessive knee flexion-extension angle and excessive joint torque under load are causally correlated with the injury severity score in the rehabilitation indicator node.

[0058] The original knee injury temporal mapping rule and the original knee injury causal mapping rule are encapsulated into a preliminary knee injury mapping rule. Based on digital twin, the preliminary knee injury mapping rule is used to assess the injury development process and risk, and the simulation assessment results are obtained. The simulation assessment results are quantitatively compared with the data corresponding to the rehabilitation tag nodes contained in the knee injury knowledge graph. In addition, the voice feedback node, knee biomechanical node, and knee imaging node data corresponding to the rehabilitation tag node data are combined to obtain the cosine similarity between the output results of the simulated injury development process and risk assessment and the data corresponding to the rehabilitation tag nodes. When the cosine similarity does not reach the preset threshold, the parameters in the preliminary knee injury mapping rule are dynamically adjusted, and the injury development process and risk assessment are re-performed until the cosine similarity reaches the preset threshold, thus generating the knee injury mapping rule.

[0059] Step S33: Construct a model of minimal knee joint injury.

[0060] Specifically, an original knee joint micro-injury model is constructed based on the knee joint injury simulation data and the knee joint injury mapping rules, and the finite element algorithm is used for simulation verification. Based on the simulation verification results, the original knee joint micro-injury model is optimized and corrected to generate a knee joint micro-injury model.

[0061] In one possible embodiment, based on the raw data corresponding to the knee joint image nodes, the geometric parameters of the actual knee joint bones are obtained. A three-dimensional solid model of the knee joint is constructed using tetrahedral units of 5 to 8 mm according to the geometric parameters to simulate the knee joint's flexion, extension, rotation and other movements. The motion trajectory of the three-dimensional solid model is constrained by the knee joint biomechanical node data as a reference. For example, the motion trajectory of the three-dimensional solid model cannot violate the normal range defined by the range of motion of the knee joint in the knee joint biomechanical nodes.

[0062] The minute structures in the three-dimensional solid model of the knee joint are modeled using 1 to 2 mm hexahedral elements. For example, ligaments and cartilage are modeled using 1.5 mm and 1 mm hexahedral elements, respectively, and an interface for simulating minute injuries is reserved.

[0063] The knee joint injury mapping rules are used as the simulation rules for the three-dimensional entity model to generate an original knee joint micro-injury model. Node data from the knee joint injury knowledge graph are input into the original knee joint micro-injury model for simulation. The simulation results are obtained, and the residual values ​​between the simulation results and the node data in the knee joint injury knowledge graph are calculated. The residual values ​​are compared with a preset residual threshold. If the residual value exceeds the preset residual threshold, a gradient descent algorithm is used to iteratively train the model by adjusting the model parameters, such as ligament elastic modulus and cartilage thickness, with the goal of minimizing the error, until the residual value is lower than the preset residual threshold. During the iterative training process, each time the residual value is calculated, node data from the knee joint injury knowledge graph that has not been input into the original knee joint micro-injury model is used to calculate the residual value, thereby ensuring the correlation between the simulation results and the nodes in the knee joint injury knowledge graph, and finally generating a knee joint micro-injury model.

[0064] Step S4: Extract emotions from the voice feedback data in the patient rehabilitation dataset to obtain the patient's emotion index.

[0065] Specifically, the text feedback data corresponding to the voice feedback nodes contained in the knee joint injury knowledge graph, as well as the corresponding raw voice audio data, are obtained.

[0066] Based on the BERT model, keywords are extracted from textual feedback data to generate a set of patient emotion keywords. Each emotion keyword in the patient emotion keyword set is classified into two emotion categories: positive emotion and negative emotion. The probability distribution of the emotion category corresponding to the emotion keyword is output based on the Softmax activation function, and the probability value of positive emotion is output as the emotional polarity score.

[0067] Feature extraction is performed on the raw audio data to obtain an audio feature set, and combined with the corresponding text-based feedback data to obtain patient emotional tendency data.

[0068] Data corresponding to biomechanical nodes of the knee joint within the knowledge graph of knee joint injury is obtained and compared with preset fatigue and load state thresholds. Based on the comparison results, patient physiological load data is generated.

[0069] A multimodal fusion network is constructed based on the attention mechanism. The emotional polarity score, patient emotional tendency data, and patient physiological load data are adaptively weighted using a dynamic weight allocation strategy to generate a patient emotional index.

[0070] For example, the sentiment of a patient's voice feedback, "I always feel a slight pain on the outside of my knee when doing the last 5 knee flexion movements, but fortunately I can still finish them each time," was extracted. A BERT-based sentiment analysis model was used to analyze the patient's voice feedback. "Slight pain" indicates that the patient's pain level is relatively mild, while "can still persevere" reflects the user's subjective initiative. The analysis shows that the sentiment of the patient's voice feedback is positive. Based on the preset mapping rules, the text data corresponding to the patient's voice feedback was mapped to the interval [-1, 1], and the sentiment polarity score was obtained as 0.6.

[0071] Using the Praat speech analysis tool, acoustic features were extracted from the raw audio data of the patient's speech feedback. The average pitch was 250Hz, and the speech rate was 1.5 words / second, which is a steady narrative rhythm. Combined with features such as speech intensity and formants, the patient's emotional tendency data was obtained as 0.5 based on a pre-trained SVM classification model. This emotional tendency data reflects that the patient's emotions were relatively calm when speaking, and there was no anxiety or irritability caused by pain.

[0072] The patient's knee joint biomechanical data were extracted from the knee injury knowledge graph and analyzed. The analysis showed that during knee flexion, the peak joint stress reached 2.8 times the patient's body weight. Although this did not exceed the preset load threshold of 3 times body weight, it was close to the critical value. The quadriceps activation remained high, with a cumulative muscle contraction duration of 45 minutes, which was lower than the preset fatigue threshold of 60 minutes. Based on the preset evaluation rules, the patient's current physiological load was determined to be mild fatigue, close to the risk of overload. The patient's physiological load data was 0.6. This physiological load data reflects that the patient's knee joint is currently under great pressure, and close monitoring and appropriate adjustment of rehabilitation training intensity are required.

[0073] Assuming the weight ratio of the network emotional polarity score, patient emotional tendency data, and patient physiological load data output by the multimodal fusion is 4:3:3, the patient emotional index is obtained as 0.6*0.4+0.5*0.3+0.6*0.3=0.57.

[0074] Step S5: Diagnose rehabilitation efficiency based on the patient's emotional index and generate an initial rehabilitation plan for the knee joint.

[0075] In this embodiment, step S5 includes: Step S51: Diagnose rehabilitation efficiency based on the patient's emotional index and generate rehabilitation training optimization data.

[0076] Specifically, based on the patient rehabilitation dataset, the system matches the historical best rehabilitation plan in the preset optimal rehabilitation plan library, sends the historical best rehabilitation plan to the user terminal, monitors the patient's emotional index when the patient executes the historical best rehabilitation plan in real time, and updates the patient rehabilitation dataset synchronously.

[0077] If the patient's emotional index is within the preset emotional index threshold range, the patient's emotional index will continue to be monitored, and the patient's rehabilitation dataset will be updated synchronously.

[0078] If the patient's emotional index is not within the preset emotional index threshold range, the updated patient rehabilitation dataset is input into the knee joint micro-injury model, the patient rehabilitation dataset is converted into knee joint injury simulation data corresponding to the knee joint micro-injury model, and the weak areas of the patient's rehabilitation training are located according to the knee joint injury mapping rules contained in the knee joint micro-injury model, generating rehabilitation training optimization data.

[0079] Among them, the data for optimizing rehabilitation training should include at least the location of the weak areas in rehabilitation training, the upper limit of stress tolerance, and the current degree of injury.

[0080] In one possible embodiment, assuming the preset emotional index threshold range is [0.6, 0.8], a value within this range indicates that the current rehabilitation program is meeting expectations and no adjustment is needed; a value within the range of (0, 0.3) indicates that the current rehabilitation program is ineffective and needs adjustment to improve patient compliance; a value within the range of (0.3, 0.6) indicates that the current rehabilitation program is less effective than expected and needs adjustment to balance patient compliance with the training intensity of the rehabilitation program; a value within the range of (0.6, 1) indicates that the current rehabilitation program is more effective than expected and needs adjustment to increase training intensity and improve patient recovery speed.

[0081] For example, if a patient's emotional index is 0.58 at a certain stage of rehabilitation, it is determined that the effectiveness of the current rehabilitation program is lower than expected, and the program needs to be adjusted to balance patient compliance and the training intensity of the rehabilitation program. The patient's rehabilitation dataset at the current stage is input into the knee joint micro-injury model for weak area analysis. It is found that the stress in the mid-segment of the patient's ligament is 26MPa and the maximum range of motion of the knee flexion is 90°. The generated rehabilitation training optimization data is to reduce the training intensity of the rehabilitation training movements in the original program to 60% to 70% of the original intensity, replace high-load training movements with low-load training movements, and modify the duration of a single set of rehabilitation training movements to 25 minutes.

[0082] Step S52: Analyze the rehabilitation training optimization data to obtain decisions on adjusting the rehabilitation training plan.

[0083] Specifically, based on rehabilitation training optimization data and combined with the corresponding data of rehabilitation tag nodes and knee joint biomechanical nodes in the knee joint injury knowledge graph, a state space is constructed; with the optimization objective of reducing the stress load in the weak areas of rehabilitation training, a reward function is set, and iterative training is performed through a proximal strategy optimization algorithm to generate rehabilitation training program adjustment decisions.

[0084] Using the example from step S52, reduce the intensity of the rehabilitation exercises in the original plan to 65% of the original intensity. If the patient's mood index improves by more than 0.05 per week, increase the intensity of the exercises by 3% in the second week.

[0085] Prioritize replacing the high-intensity exercise "open-chain straight leg raise" in the original training program with low-intensity exercises such as "closed-chain wall squat" and "seated knee bend".

[0086] Adjust the training frequency in the rehabilitation training program, reducing it from 4 times a week to 3 times a week, with each session lasting 25 minutes, and adding a 1-minute rest period at the 10th and 20th minute.

[0087] Furthermore, an initial rehabilitation plan for the knee joint is generated based on the rehabilitation training optimization data and the rehabilitation training plan adjustment decision.

[0088] Step S6: Evaluate the effectiveness of the initial knee rehabilitation plan and optimize the plan to generate a knee rehabilitation plan.

[0089] In this embodiment, step S6 includes: Step S61: Evaluate the effectiveness of the initial knee rehabilitation program and obtain simulation prediction results.

[0090] Specifically, the initial knee joint rehabilitation plan is input into the knee joint micro-injury model, and the model prediction control algorithm is used to simulate and predict, and the simulation prediction results are obtained. The simulation prediction results include at least the predicted rehabilitation effect data and the predicted safety risk data. The predicted rehabilitation effect data includes at least the stress reduction magnitude of the injury area, the improvement rate of the knee joint range of motion, and the improvement curve of the knee joint range of motion.

[0091] For example, if the knee joint micro-injury model outputs an absolute improvement of 16° in the range of motion of the knee joint after 4 weeks of initial knee joint rehabilitation, then the average improvement rate of the range of motion of the knee joint is 4°, the stress in the injured area decreases from 32MPa before training to 20MPa, and the weekly improvement rate of the range of motion of the knee joint shows a gradual upward trend, with an improvement of 2.5° in the first week, 3.5° in the second week, 4.5° in the third week, and 5.5° in the fourth week.

[0092] The knee joint minimal injury model only experienced ligament stress exceeding the threshold twice during the entire simulation training process, and the probability of cartilage pressure exceeding the standard was 10%. Combining the patient's weight, injury level, and parameters corresponding to the initial knee joint rehabilitation plan input into the knee joint minimal injury model, assuming that the calculation method for the predicted safety risk data is "number of times ligament stress exceeds the threshold * 10% + probability of cartilage pressure exceeding the standard", then the predicted safety risk data for this initial knee joint rehabilitation plan is 30%.

[0093] Step S62: Verify the simulation prediction results.

[0094] Specifically, the simulation prediction results are validated by combining the causal relationships between nodes in the set of knee injury nodes contained in the knee injury knowledge graph.

[0095] If the simulation prediction results fail the verification, the parameters of the initial knee joint rehabilitation plan will be adjusted until the verification is passed.

[0096] If the simulation prediction results pass the verification, the corresponding initial knee joint rehabilitation plan will be sent to the user's terminal as a knee joint rehabilitation plan.

[0097] Understandably, verifying the simulation prediction results means obtaining the causal relationship between the knee joint biomechanical nodes and rehabilitation tag nodes within the knee joint injury knowledge graph, and defining it as a rehabilitation association rule.

[0098] The reduction in stress in the injured area is used as a verification index, and the damage repair efficiency is verified in conjunction with rehabilitation association rules. The damage repair efficiency verification means that after the patient implements the initial knee joint rehabilitation program, the reduction in stress in the injured area should not show a negative increasing trend.

[0099] Using the rate of improvement in knee joint range of motion as a verification indicator, and combining it with rehabilitation association rules, the stability of rehabilitation effect is verified. The stability of rehabilitation effect is defined as the training intensity of rehabilitation exercises in the initial knee joint rehabilitation program not fluctuating frequently. For example, assuming the weekly fluctuation rate of training intensity is less than 5%, and the fluctuation range of the rate of improvement in knee joint range of motion in the simulation prediction results is less than 10% per week, it can be inferred that "if the training intensity of rehabilitation exercises in the initial knee joint rehabilitation program does not fluctuate drastically, then the rate of improvement in knee joint range of motion in the simulation prediction results should not show any sudden drop or rise." Based on this causal relationship, the stability of the rehabilitation effect of the initial knee joint rehabilitation program is verified.

[0100] The improvement curve of knee joint range of motion is used as the verification index, and rehabilitation association rules are combined to verify the persistence of rehabilitation effect. The verification of the persistence of rehabilitation effect means that the initial knee joint rehabilitation program should have a positive effect on the user's condition and should not aggravate the patient's condition.

[0101] The causal relationship between the knee joint biomechanical nodes and the voice feedback nodes in the knee joint injury knowledge graph is obtained, and the rehabilitation training tolerance is verified. The rehabilitation training tolerance verification means that after the patient implements the initial knee joint rehabilitation program, the patient should not frequently report pain. For example, if the patient reports pain every time rehabilitation training is performed.

[0102] Figure 3 This is a schematic diagram of the knee joint personalized rehabilitation plan generation system based on digital twins of the present invention. The following is a detailed introduction to the knee joint personalized rehabilitation plan generation system based on digital twins.

[0103] Specifically, the personalized knee rehabilitation plan generation system based on digital twins includes: The data acquisition module is used to acquire multi-dimensional feature data of patients with knee joint injuries and construct a patient rehabilitation dataset.

[0104] The graph construction module is used to construct a knowledge graph of knee joint injury based on the patient rehabilitation dataset.

[0105] The model building module is used to build a model of minor knee injuries based on a knowledge graph of knee injuries.

[0106] The emotion extraction module is used to extract emotions from the voice feedback data in the patient rehabilitation dataset to obtain the patient's emotion index.

[0107] The plan generation module is used to diagnose rehabilitation efficiency based on the patient's emotional index and generate an initial rehabilitation plan for the knee joint.

[0108] The program optimization module is used to evaluate the effectiveness of the initial knee joint rehabilitation program and optimize the program to generate a knee joint rehabilitation program.

[0109] The specific usage and function of this embodiment are explained below: First, multi-dimensional feature data of patients was acquired to construct a patient rehabilitation dataset. This was achieved by collecting biomechanical parameters and range of motion data of the injured knee joint during movement, as well as knee joint imaging data, and combining this with real-time patient feedback using speech recognition technology. This data provided data support for subsequent simulation and solution generation. Next, a knowledge graph of knee joint injury is constructed based on the patient rehabilitation dataset. By generating the knowledge graph of knee joint injury, a large amount of data in the patient rehabilitation dataset is associated and aggregated, while clarifying the relationship between different types of data, thus providing data support for subsequent data analysis. Secondly, a model of minor knee injuries was constructed by combining digital twin technology with a knowledge graph of knee injuries. Based on the model of minor knee injuries, the data contained in the knowledge graph of knee injuries was mapped to the simulation virtual model of the patient. At the same time, the relationships between the data nodes in the knowledge graph of knee injuries were combined to accurately identify the weak points of the patient's function and the degree of injury during rehabilitation training, which provided a data foundation for the subsequent development of personalized plans for the patient's condition. Finally, emotion extraction was performed on the voice feedback data within the patient rehabilitation dataset to obtain the patient's emotion index. Based on the patient's emotion index, rehabilitation efficiency was diagnosed, and an initial rehabilitation plan for the knee joint was generated. The effectiveness of the initial rehabilitation plan for the knee joint was evaluated and optimized to generate a new knee joint rehabilitation plan. This approach not only meets the individual needs of patients but also dynamically adjusts the rehabilitation plan according to their individual needs, thereby improving the accuracy and efficiency of the generated rehabilitation plan.

[0110] The present invention also provides an electronic device, comprising: At least one processor; and at least one memory communicatively connected to the processor; wherein the memory stores instructions executable by at least one processor, the instructions being executed by at least one processor to enable at least one processor to perform the method proposed in Embodiment 1 of the present invention.

[0111] The following is a detailed introduction to the various components of the electronic device: In this context, the processor is the control center of the electronic device. It can be a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement Embodiment 1 of this invention, such as one or more digital signal processors (DSPs) or one or more field-programmable gate arrays (FPGAs).

[0112] The processor can perform various functions of an electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0113] The memory is used to store the software program that executes the solution of the present invention, and the execution is controlled by the processor. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0114] The memory can be a real-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only (CD-ROM), or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory can be integrated with the processor or exist independently and coupled to the processor through an interface circuit of an electronic device; this embodiment of the invention does not specifically limit this.

[0115] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via limited means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0116] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0117] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for generating personalized knee joint rehabilitation plans based on digital twins, characterized in that, It includes the following steps: Acquire multi-dimensional feature data of patients and construct a patient rehabilitation dataset; A knowledge graph of knee joint injury was constructed based on the aforementioned patient rehabilitation dataset. A model of minor knee injuries was constructed based on the aforementioned knee injury knowledge graph. Emotion extraction was performed on the voice feedback data within the patient rehabilitation dataset to obtain the patient's emotion index; Based on the patient's emotional index, rehabilitation efficiency is diagnosed, and an initial rehabilitation plan for the knee joint is generated. The effectiveness of the initial knee rehabilitation program is evaluated and the program is optimized to generate a new knee rehabilitation plan.

2. The method for generating personalized knee joint rehabilitation plans based on digital twins according to claim 1, characterized in that, A model of minor knee injuries is constructed based on the aforementioned knee injury knowledge graph, including: The data corresponding to the knee joint biomechanical nodes and knee joint image nodes contained in the knee joint injury knowledge graph are input into a deep learning neural network to generate knee joint injury simulation data. Based on the temporal and causal relationships between nodes in the knee injury knowledge graph, a preliminary mapping rule for knee injury is constructed. The preliminary mapping rule for knee injury is iteratively corrected based on the data corresponding to the rehabilitation tag nodes in the knee injury knowledge graph until the similarity between the mapping result corresponding to the preliminary mapping rule for knee injury and the data corresponding to the rehabilitation tag nodes is greater than a preset similarity threshold, thus generating a knee injury mapping rule. Based on the knee joint injury simulation data and knee joint injury mapping rules, an original knee joint micro-injury model is constructed, and the finite element algorithm is used for simulation verification. Based on the simulation verification results, the original knee joint micro-injury model is optimized and corrected to generate a new knee joint micro-injury model.

3. The method for generating personalized knee joint rehabilitation plans based on digital twins according to claim 1, characterized in that, Emotion extraction was performed on the voice feedback data within the patient rehabilitation dataset to obtain the patient's emotion index, including: Obtain the text-based feedback data corresponding to the voice feedback nodes contained in the knee joint injury knowledge graph, as well as the corresponding raw voice audio data. Based on the BERT model, keywords are extracted from textual feedback data to generate a set of patient emotion keywords. Each emotion keyword in the patient emotion keyword set is classified into two emotion categories: positive emotion and negative emotion. The probability distribution of the emotion category corresponding to the emotion keyword is output based on the Softmax activation function, and the probability value of positive emotion is output as the emotion polarity score. Feature extraction is performed on the raw audio data to obtain an audio feature set, and combined with the corresponding text-based feedback data to obtain patient emotional tendency data; Obtain the corresponding data of the biomechanical nodes of the knee joint in the knowledge graph of knee joint injury, and compare them with the preset fatigue degree threshold and load state threshold. Generate the patient's physiological load data based on the comparison results. A multimodal fusion network is constructed based on the attention mechanism. The emotional polarity score, patient emotional tendency data, and patient physiological load data are adaptively weighted using a dynamic weight allocation strategy to generate a patient emotional index.

4. The method for generating personalized knee joint rehabilitation plans based on digital twins according to claim 1, characterized in that, Based on the patient's emotional index, rehabilitation efficiency is diagnosed, and an initial knee joint rehabilitation plan is generated, including: Based on the patient rehabilitation dataset, the best historical rehabilitation plan is matched with the preset best rehabilitation plan library, and the best historical rehabilitation plan is sent to the user terminal. The patient's emotional index when the patient executes the best historical rehabilitation plan is monitored in real time, and the patient rehabilitation dataset is updated synchronously. If the patient's emotional index is within the preset emotional index threshold range, the patient's emotional index will continue to be monitored and the patient's rehabilitation dataset will be updated synchronously. If the patient's emotional index is not within the preset emotional index threshold range, the updated patient rehabilitation dataset is input into the knee joint micro-injury model, the patient rehabilitation dataset is converted into knee joint injury simulation data corresponding to the knee joint micro-injury model, and the weak areas of the patient's rehabilitation training are located according to the knee joint injury mapping rules contained in the knee joint micro-injury model, and rehabilitation training optimization data is generated. Among them, the data for optimizing rehabilitation training should include at least the location of the weak areas in rehabilitation training, the upper limit of stress tolerance, and the current degree of injury; The rehabilitation training optimization data is analyzed based on the proximal strategy optimization algorithm to obtain the decision on adjusting the rehabilitation training plan; Based on the optimized rehabilitation training data and the rehabilitation training program adjustment decisions, an initial rehabilitation plan for the knee joint is generated.

5. The method for generating a personalized knee joint rehabilitation plan based on digital twins according to claim 4, characterized in that, The rehabilitation training optimization data is analyzed based on the proximal strategy optimization algorithm to obtain decisions on adjusting the rehabilitation training plan, including: Based on rehabilitation training optimization data, and combined with the corresponding data of rehabilitation tag nodes and knee joint biomechanical nodes in the knee joint injury knowledge graph, a state space is constructed. With the goal of reducing stress load in weak areas of rehabilitation training, a reward function is set, and a proximal strategy optimization algorithm is used for iterative training to generate rehabilitation training program adjustment decisions.

6. The method for generating a personalized knee joint rehabilitation plan based on digital twins according to claim 1, characterized in that, The effectiveness of the initial knee rehabilitation program is evaluated and optimized to generate a knee rehabilitation plan, including: The initial rehabilitation plan for the knee joint is input into the knee joint micro-injury model, and the model prediction control algorithm is used to simulate and predict, and the simulation prediction results are obtained. The simulation prediction results include at least the predicted rehabilitation effect data and the predicted safety risk data. The predicted rehabilitation effect data includes at least the stress reduction magnitude of the injury area, the improvement rate of the knee joint range of motion, and the improvement curve of the knee joint range of motion. The simulation prediction results are validated by combining the causal relationships between nodes in the set of knee joint injury nodes contained in the knee joint injury knowledge graph. If the simulation prediction results fail the verification, the parameters of the initial knee joint rehabilitation plan will be adjusted until the verification is passed. If the simulation prediction results pass the verification, the corresponding initial knee joint rehabilitation plan will be sent to the user's terminal as a knee joint rehabilitation plan.

7. The method for generating a personalized knee joint rehabilitation plan based on digital twins according to claim 6, characterized in that, By combining the causal relationships between nodes within the knee injury knowledge graph, the simulation prediction results are validated, including: Obtain the causal relationship between knee joint biomechanical nodes and rehabilitation tag nodes in the knee joint injury knowledge graph, and define it as a rehabilitation association rule; The extent of stress reduction in the damaged area was used as the verification index, and the damage repair efficiency was verified by combining rehabilitation correlation rules. The rate of improvement in knee joint range of motion was used as the verification index, and the stability of rehabilitation effect was verified by combining rehabilitation association rules. The improvement curve of knee joint range of motion was used as the verification index, and the sustainability of rehabilitation effect was verified by combining rehabilitation correlation rules. Obtain the causal relationship between the knee joint biomechanical nodes and the voice feedback nodes in the knee joint injury knowledge graph, and verify the tolerance of rehabilitation training.

8. The method for generating a personalized knee joint rehabilitation plan based on digital twins according to claim 1, characterized in that, Obtain multi-dimensional feature data of patients and construct a patient rehabilitation dataset, including: Collect knee joint status data of patients during rehabilitation exercises and generate a continuous knee joint status sequence; Feature extraction is performed on the knee joint state sequence to obtain knee joint biomechanical data and knee joint activity data; Simultaneously collect the patient's voice information during rehabilitation exercises to generate the patient's voice feedback data, which includes at least the patient's subjective feeling data, pain description data, and compliance feedback data. Feature extraction is performed on the patient's knee joint pathology report to obtain knee joint imaging data and knee joint diagnostic data. The knee joint imaging data includes at least knee joint structural data, and the knee joint diagnostic data includes at least injury degree data and injury rehabilitation stage data. A patient rehabilitation dataset was constructed based on the aforementioned knee joint biomechanical data, knee joint activity data, voice feedback data, knee joint imaging data, and knee joint diagnostic data.

9. The method for generating personalized knee joint rehabilitation plans based on digital twins according to claim 1, characterized in that, A knowledge graph of knee joint injury was constructed based on the aforementioned patient rehabilitation dataset, including: The knee joint biomechanical data and knee joint activity data in the patient rehabilitation dataset are defined as knee joint biomechanical nodes, the voice feedback data in the patient rehabilitation dataset are defined as voice feedback nodes, the knee joint imaging data in the patient rehabilitation dataset are defined as knee joint imaging nodes, and the knee joint diagnostic data in the patient rehabilitation dataset are defined as rehabilitation label nodes, thereby generating a set of knee joint injury nodes. The data corresponding to each node in the set of knee joint injury nodes are timestamped to obtain the temporal correlation between each node in the set of knee joint injury nodes. By adjusting the corresponding data of each node in the knee joint injury node set and monitoring the data changes of nodes in the knee joint injury node set that have not undergone data adjustment, the causal relationship between each node in the knee joint injury node set can be obtained. A knowledge graph of knee joint injury is constructed based on the set of knee joint injury nodes and the temporal and causal relationships between the nodes in the set.

10. A personalized knee joint rehabilitation plan generation system based on digital twins, characterized in that, include: The data acquisition module is used to acquire multi-dimensional feature data of patients with knee joint injuries and construct a patient rehabilitation dataset. A knowledge graph construction module is used to construct a knowledge graph of knee joint injury based on the patient rehabilitation dataset. A model building module is used to build a model of minor knee injuries based on a knowledge graph of knee injuries. An emotion extraction module is used to extract emotions from the voice feedback data in the patient rehabilitation dataset to obtain the patient's emotion index. The plan generation module is used to diagnose rehabilitation efficiency based on the patient's emotional index and generate an initial rehabilitation plan for the knee joint. The program optimization module is used to evaluate the effectiveness of the initial knee joint rehabilitation program and optimize the program to generate a knee joint rehabilitation program.

Images