Prediction and rehabilitation guidance system for joint function recovery after knee arthrodesis osteotomy

Abstract

The application discloses a knee-preserving osteotomy postoperative joint function recovery prediction and rehabilitation guidance system and relates to the technical field of rehabilitation medicine, which comprises the following modules: a data assimilation module, which acquires postoperative rehabilitation data of a patient in a whole cycle and extracts a high-dimensional feature set; a state matching module, which searches and matches a historical rehabilitation trajectory cluster in a pre-constructed rehabilitation state evolution map; a blueprint generation module, which generates an individualized intervention blueprint according to the trajectory cluster, guides an intelligent device to execute and collect feedback; a function deduction module, which inputs the feedback into a dynamics model coupled with the biomechanical response of a bone-cartilage-ligament complex and the muscle recruitment strategy adaptive adjustment, calculates a future expected joint function score; and a path adjustment module, which corrects the blueprint online according to the score deviation. The model shows that the coupling of biomechanics and muscle adjustment improves the prediction accuracy, and the combination of state matching and dynamic blueprint enhances the real-time feedback of the rehabilitation scheme, thereby improving the accuracy and adaptability of the rehabilitation scheme.

Description

Technical Field

[0001] This invention belongs to the field of rehabilitation medicine technology, specifically a system for predicting and guiding the recovery of joint function after knee osteotomy and orthopedic surgery. Background Technology

[0002] Current technologies for predicting and guiding joint function recovery after knee osteotomy and orthopedic surgery primarily rely on static statistical models or clinical experience frameworks. While attempting to optimize the rehabilitation process through limited data collection, they fail to effectively address the dynamic complexity of the postoperative biomechanical environment and individual recovery differences. The shortcomings of existing technologies lie in the fact that the prediction phase often uses simplified models, failing to deeply integrate the biomechanical response of the bone-cartilage-ligament complex and the adaptive adjustment of muscle recruitment strategies. This leads to deviations in future joint function score projections from the actual recovery trajectory. Furthermore, rehabilitation guidance often employs generic, phased approaches, lacking high-dimensional feature extraction from individual full-cycle rehabilitation data. It fails to construct a patient rehabilitation state evolution map covering multi-dimensional state transitions and cannot obtain intervention logic adapted to the current state from historical rehabilitation trajectories through similarity retrieval, resulting in a disconnect between intervention measures and the patient's actual progress.

[0003] The core problems to be solved by this invention include: how to construct a dynamic inference mechanism that can predict future joint function scores based on real-time execution feedback data by explicitly coupling the biomechanical response of the bone-cartilage-ligament complex with the adaptive adjustment of muscle recruitment strategies; and how to achieve the retrieval and matching of similar historical trajectory clusters in a pre-constructed rehabilitation state evolution map based on the high-dimensional feature set extracted from full-cycle rehabilitation data, thereby dynamically generating an individualized rehabilitation intervention blueprint, and correcting the intensity, frequency or action combination of the intervention path online through real-time feedback deviation. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] Therefore, this invention proposes a prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery, comprising:

[0006] The data assimilation module acquires the patient's full-cycle postoperative rehabilitation data and sends it to the data assimilation engine for processing, in order to extract a high-dimensional rehabilitation feature set from the assimilated data.

[0007] The state matching module, based on the high-dimensional rehabilitation feature set, performs similarity retrieval in the pre-constructed patient rehabilitation state evolution map and matches the historical rehabilitation trajectory clusters that are closest to the current patient state.

[0008] The blueprint generation module dynamically generates an individualized rehabilitation intervention blueprint based on the matched historical rehabilitation trajectory clusters, executes the individualized rehabilitation intervention blueprint, guides the patient to use intelligent rehabilitation equipment to complete the specified actions, and simultaneously collects execution feedback data.

[0009] The functional deduction module inputs the execution feedback data into the joint function dynamics deduction model in real time. The joint function dynamics deduction model couples the biomechanical response of the bone-cartilage-ligament complex with the adaptive adjustment process of muscle recruitment strategy. Through the calculation of the joint function dynamics deduction model, the expected joint function score of the patient at future time points is deduced.

[0010] The path adjustment module activates a rehabilitation path dynamic adjuster based on the deviation between the expected joint function score and the current actual score, and makes online corrections to the intensity, frequency or movement combination in the individualized rehabilitation intervention blueprint being implemented.

[0011] Furthermore, the full-cycle postoperative rehabilitation data is fed into a data assimilation engine for processing, including:

[0012] The full-cycle postoperative rehabilitation data includes multimodal physiological signals, imaging sequences, and patient self-assessment records;

[0013] The data assimilation engine calibrates data from different sources and time phases to a unified time-space reference framework;

[0014] The multimodal physiological signals were time-stamped and denoised to extract stable segments of electromyography, angle and pressure signals.

[0015] The imaging sequences were reconstructed and registered in three dimensions to quantify the bony healing area, force line angle changes, and dynamic characteristics of the joint space in the osteotomy region.

[0016] The unstructured descriptions in the patient self-assessment records were converted into standardized functional limitation levels and pain intensity indices;

[0017] A unified timeline is established, and the processed multimodal physiological signal features, quantified imaging features, and standardized self-assessment record features are mapped onto the unified timeline according to the number of days after surgery, forming a time series feature vector.

[0018] Furthermore, a high-dimensional rehabilitation feature set is extracted from the assimilated data, including:

[0019] From the time series feature vector, the rate of change of features between adjacent time points is calculated to form a first-order dynamic feature;

[0020] Identify periodic patterns of specific feature combinations in the time series feature vector and extract pattern parameters as periodic features;

[0021] Analyze the time lag correlation between different features, construct an influence relationship network between features, and extract network topology features from the influence relationship network;

[0022] The original time series feature vector, first-order dynamic features, periodic features, and network topology features are combined to form the high-dimensional rehabilitation feature set.

[0023] The high-dimensional rehabilitation feature set covers joint biomechanical state, soft tissue healing process, and neuromuscular control patterns.

[0024] Furthermore, the construction process of the pre-constructed patient rehabilitation state evolution map includes:

[0025] It compiles a large amount of complete rehabilitation data from historical patients, which includes assimilation data from the entire process from postoperative surgery to functional recovery;

[0026] The assimilation data of each historical patient throughout the entire process is divided into sliding segments according to time windows to obtain a series of snapshots of the recovery status arranged in chronological order;

[0027] Using a nonlinear manifold learning method, the high-dimensional rehabilitation feature set corresponding to each rehabilitation state snapshot is projected into a low-dimensional latent state space, and the coordinate points obtained by the projection are the rehabilitation state points.

[0028] Connect the consecutive recovery status points of the same historical patient in chronological order to form a historical recovery trajectory;

[0029] The historical rehabilitation trajectory is associated with its final joint function recovery level and stored as an evolution map of the patient's rehabilitation status.

[0030] Furthermore, based on the high-dimensional rehabilitation feature set, a similarity search is performed in the pre-constructed patient rehabilitation state evolution map to match the historical rehabilitation trajectory clusters that are closest to the current patient state, including:

[0031] Projecting the current patient's high-dimensional rehabilitation feature set onto the latent state space yields the current rehabilitation state point;

[0032] Calculate the multi-dimensional distance between the current rehabilitation status point and all rehabilitation status points in the patient's rehabilitation status evolution map;

[0033] Select the nearest several rehabilitation status points and backtrack to find the historical rehabilitation trajectory to which the several rehabilitation status points belong;

[0034] Based on the final recovery level, early feature similarity, and trajectory morphology consistency of the historical recovery trajectories, a set of trajectories most similar to the current patient's potential recovery pattern is selected, namely the historical recovery trajectory cluster.

[0035] Furthermore, the individualized rehabilitation intervention blueprint is implemented, guiding the patient to use intelligent rehabilitation equipment to complete designated actions, and execution feedback data is collected simultaneously, including:

[0036] The individualized rehabilitation intervention blueprint details the phased goals, training content, and intensity prescriptions for the future rehabilitation cycle;

[0037] The training content in the individualized rehabilitation intervention blueprint is broken down into a series of specific joint movement commands and resistance settings;

[0038] The intelligent rehabilitation equipment controls the patient's limbs to complete movements according to a predetermined range of motion, speed, and resistance.

[0039] By integrating a multi-axis force sensor and a high frame rate optical capture system into the intelligent rehabilitation device, the joint torque curve, motion trajectory deviation and muscle activation sequence are recorded during the execution of the movement.

[0040] The recorded joint torque curves, motion trajectory deviations, and muscle activation timings are integrated into the time-synchronized execution feedback data.

[0041] Furthermore, the construction and operation process of the joint function dynamics deduction model includes:

[0042] A personalized lower limb skeletal muscle multibody dynamics model was established for each patient. The parameters of the lower limb skeletal muscle multibody dynamics model were determined by the patient's preoperative imaging data and anatomical landmarks.

[0043] The lower limb skeletal muscle multibody dynamics model integrates a time-varying stiffness model that reflects the changes in biomechanical strength of bone healing after osteotomy, and a contact mechanics model that simulates the adaptive changes in cartilage load.

[0044] The motion trajectory and torque information in the execution feedback data are used as boundary conditions and input into the lower limb skeletal muscle multibody dynamics model to inversely calculate the activation level of the main muscle groups and the joint contact force.

[0045] The calculated joint contact force and muscle activation level are input into the time-varying stiffness model and contact mechanics model to positively extrapolate the micromechanical response and functional adaptation of bone tissue, articular cartilage and surrounding ligaments in the osteotomy area under the current load.

[0046] The process of iteratively performing reverse calculations and forward deductions simulates the tissue state and neural control patterns accumulated after multiple rehabilitation training sessions in the future, and predicts the expected joint function score based on this.

[0047] Furthermore, the parameter calibration process of the joint function dynamics deduction model includes:

[0048] Baseline feedback data were collected in the early postoperative period using a low-load passive activity mode.

[0049] The baseline feedback data is input into the uncalibrated joint function dynamics extrapolation model to obtain the initial extrapolation results;

[0050] The initial projection results are compared with the actual clinical assessment results to calculate the model prediction error;

[0051] By utilizing the principle of backpropagation of errors, the muscle lever arm parameters, tissue stiffness parameters, and neural control gain parameters in the model are adjusted to make the model's projection results approximate the actual clinical assessment results, thus completing the model calibration.

[0052] Furthermore, based on the deviation between the expected joint function score and the current actual score, a rehabilitation pathway dynamic adjuster is activated to make online corrections to the intensity, frequency, or movement combinations in the currently implemented individualized rehabilitation intervention blueprint, including:

[0053] The difference between the expected joint function score and the immediate function score calculated based on the current actual performance is used as the rehabilitation progress deviation.

[0054] Analyze the composition of the aforementioned rehabilitation progress deviation to identify whether it stems from insufficient strength, limited range of motion, or abnormal motor control.

[0055] The system queries a predefined rehabilitation strategy adjustment rule base and matches targeted adjustment suggestions based on the type and degree of deviation. These suggestions include increasing isometric contraction training for specific muscle groups, introducing additional joint mobilization techniques, or adjusting the proportion of auxiliary resistance in the movement.

[0056] The proposed adjustments will be integrated into the individualized rehabilitation intervention blueprint for subsequent cycles, covering the original training plan portion.

[0057] Furthermore, it also includes: a closed-loop control module, which converts the online correction instructions into specific device control parameters and patient guidance prompts, forming a closed-loop rehabilitation guidance process, including:

[0058] The training content in the adjusted individualized rehabilitation intervention blueprint is analyzed into motor angle curves, resistance torque curves, and safety range thresholds that can be executed by intelligent rehabilitation equipment.

[0059] Generate synchronized audiovisual guidance prompts, which include verbal commands for action key points, visual indications of target locations, and real-time feedback on completion quality;

[0060] When the next training cycle starts, the motor angle curve, resistance torque curve, and safety range threshold are loaded into the intelligent rehabilitation equipment controller, and the audiovisual guidance prompts are played simultaneously to implement the revised rehabilitation plan.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] A joint function dynamics model was constructed that adaptively adjusts the biomechanical response of the bone-cartilage-ligament complex and the muscle recruitment strategy. This model explicitly couples biomechanical behaviors such as stress redistribution in bone structure after osteotomy, changes in cartilage load transfer characteristics, and adaptive adjustments in ligament tension with the dynamic optimization process of agonist / antagonist muscle synergistic contraction patterns under muscle atrophy. It uses execution feedback data such as joint range of motion, muscle force output, and pain threshold collected by intelligent rehabilitation equipment as input to directly calculate the expected joint function score at future time points. This overcomes the limitations of traditional models that simplify biomechanical interactions and ignore muscle compensation mechanisms, enabling predictions to more realistically reflect the evolution of joint function after surgery with changes in the biomechanical environment, and improving the consistency between the predicted score and actual recovery.

[0063] By using a data assimilation engine to process postoperative rehabilitation data throughout the patient's lifecycle and extract a high-dimensional rehabilitation feature set, matching historical rehabilitation trajectory clusters are retrieved from a pre-constructed patient rehabilitation state evolution map. This map uses time as an axis to set nodes that associate different stages of state and transition probabilities. Each node stores historical patient rehabilitation trajectory data. Cosine similarity or Mahalanobis distance algorithms are used to locate the trajectory cluster closest to the current state based on the high-dimensional feature set. Based on this, an individualized rehabilitation intervention blueprint is dynamically generated, guiding intelligent devices to perform specified actions and simultaneously collecting feedback. A dynamic adjuster is activated based on the deviation between the expected and actual scores, correcting the blueprint's intensity, frequency, or action combinations online. This overcomes the limitations of universal rehabilitation programs, relying on the experience of similar trajectory cluster groups to achieve source adaptation between intervention and patient state, avoiding the blindness of experience-based programs. It shifts the rehabilitation path from pre-set to dynamic generation and real-time optimization based on similar cases, enhancing the accuracy and adaptability of the program. Attached Figure Description

[0064] Figure 1 This is a timing diagram of the prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in this invention.

[0065] Figure 2 Flowchart for high-dimensional rehabilitation feature set extraction;

[0066] Figure 3 A flowchart for matching historical recovery trajectory clusters;

[0067] Figure 4 A dual-indicator trend graph showing the relationship between the number of rehabilitation training sessions and joint biomechanical response;

[0068] Figure 5 Trend chart of full-cycle rehabilitation monitoring after knee osteotomy and orthopedic surgery. Detailed Implementation

[0069] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0070] See Figure 1 The present invention provides a predictive and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery, comprising a data assimilation module, a state matching module, a blueprint generation module, a functional deduction module, and a path adjustment module. The overall implementation scheme of the system is as follows: The data assimilation module acquires full-cycle postoperative rehabilitation data containing multi-stage and multi-type information of the patient, and sends this data to the data assimilation engine for processing. The data assimilation engine calibrates, aligns, extracts features, and fuses the data, extracting a high-dimensional rehabilitation feature set from the assimilated data that comprehensively reflects the joint's mechanical state, soft tissue healing, and neural control. The state matching module, based on this high-dimensional rehabilitation feature set, searches and matches it within a pre-constructed patient rehabilitation state evolution map. This map records the entire state change process of numerous historical patients from surgery to rehabilitation completion; the matching process aims to find a set of historical rehabilitation trajectory clusters most similar to the current patient's state. The blueprint generation module dynamically generates an individualized rehabilitation intervention blueprint customized for the current patient based on the matched historical rehabilitation trajectory clusters. This blueprint details the stages, content, and goals of rehabilitation training. The system executes this blueprint, guiding the patient to operate intelligent rehabilitation equipment to complete designated training movements, and simultaneously collecting execution feedback data such as joint torque, movement trajectory, and muscle activation during the training process. The functional deduction module inputs this execution feedback data in real time into a joint function dynamics deduction model. This model couples the biomechanical response process of the skeleton-cartilage-ligament complex under mechanical load with the process by which the nervous system adjusts muscle recruitment strategies to adapt to changes. Through dynamic calculations by the model, the expected joint function score of the patient at a specific future time point is deduced. The path adjustment module then activates a rehabilitation path dynamic adjuster based on the deviation between this expected score and the patient's current actual function score. This adjuster makes online corrections to the training intensity, frequency, or movement combinations in the ongoing individualized rehabilitation intervention blueprint, thereby achieving adaptive optimization of the rehabilitation path.

[0071] See Figure 2 In one embodiment of the present invention, a data assimilation engine processes full-cycle postoperative rehabilitation data. This full-cycle postoperative rehabilitation data includes multimodal physiological signals, imaging sequences, and patient self-assessment records. The data assimilation engine calibrates data from different sources and at different acquisition phases to a unified temporal-spatial reference framework. Multimodal physiological signals undergo timestamp alignment and denoising, extracting stable signal segments from signals such as electromyography, joint angles, and plantar pressure. Imaging sequences undergo three-dimensional reconstruction and multi-temporal registration, quantifying dynamic features such as the bony healing area of ​​the osteotomy region, changes in lower limb force line angles, and joint space width. Unstructured textual descriptions in patient self-assessment records are converted into standardized functional limitation levels and pain intensity indices using natural language processing technology. The system establishes a unified timeline, precisely mapping the processed multimodal physiological signal features, quantified imaging features, and standardized self-assessment record features to this unified timeline according to postoperative days, forming a time-series feature vector of the patient's rehabilitation process.

[0072] A high-dimensional rehabilitation feature set is extracted from the assimilated time-series feature vector. The rate of change of each feature between adjacent time points is calculated from the time-series feature vector to form first-order dynamic features describing the trend of change. Periodic patterns of specific feature combinations in the time-series feature vector are identified, such as repetitive patterns of gait or specific movements, and parameters such as frequency and amplitude of these patterns are extracted as periodic features. The temporal lag correlation between different features is analyzed, constructing an influence relationship network between features, and extracting network topology features such as node centrality and clustering coefficients from this network. The original time-series feature vector, first-order dynamic features, periodic features, and network topology features are merged to form a high-dimensional rehabilitation feature set, which comprehensively covers joint biomechanical state, soft tissue healing process, and neuromuscular control patterns.

[0073] In practice, a system for predicting and guiding joint function recovery after knee-preserving osteotomy was implemented for a patient who had undergone medial open wedge high tibial osteotomy. During the first week post-surgery, the system acquired the patient's full-cycle postoperative rehabilitation data, including multimodal physiological signals, imaging sequences, and patient self-assessment records. The data assimilation engine initiated its processing flow, sending multimodal physiological signals from surface electromyography sensors, articular angle meters, and pressure plates, along with knee X-rays and CT imaging sequences from immediate postoperative period, two weeks, and six weeks post-surgery, and the patient's daily self-assessment records submitted via tablet computer. The data assimilation engine calibrated the data from these different sources and acquisition phases to a unified temporal-spatial reference framework, with the time of surgery completion as zero and the patient's anatomical coordinate system as the baseline. Strict timestamp alignment and wavelet transform-based denoising were performed on multimodal physiological signals to extract stable segments of quadriceps electromyography, knee flexion-extension angle, and plantar pressure center trajectory during straight leg raises. Three-dimensional reconstruction and registration based on bony landmarks were performed on the imaging sequences. The percentage of bony healing area in the osteotomy region, the change in lower limb mechanical axis angle, and the dynamic characteristics of the medial joint space width were quantified on the registered model. The unstructured description of "needs to hold the handrail when going up and down stairs" in the patient's self-report was converted into a standardized functional limitation level of "moderate limitation" using a pre-defined semantic mapping table. The description of "dull pain, tolerable" was converted into a digitized pain intensity index of "3" (range 0-10). The system establishes a unified timeline with "postoperative days" as the scale. It maps the processed electromyography amplitude, joint range of motion, pressure center swing characteristics, quantified bone healing area, force line angle, joint space width, and standardized functional level and pain index onto the unified timeline according to their respective collection time, forming a series of time series feature vectors. Each dimension of the vector corresponds to a specific feature value of a specific date.

[0074] In practice, the process begins with extracting a high-dimensional rehabilitation feature set from the assimilated time-series feature vector. From the time-series feature vector, the rate of change of each feature value between adjacent time points is calculated; for example, the rate of increase in bone healing area between postoperative day 7 and day 14 is calculated, forming a first-order dynamic feature describing the trend of healing and functional changes. Periodic patterns of gait-related feature combinations in the time-series feature vector are identified, and pattern parameters such as gait cycle duration and standing phase ratio are extracted from continuous plantar pressure sequences as periodic features. The temporal lag correlation between the two features, "pain index" and "rectus femoris integral electromyography value," is analyzed. It is found that an increase in the pain index usually lags behind a decrease in electromyography value by 1 to 2 days. An influence relationship network between features is constructed, and the network's clustering coefficient and betweenness centrality of feature nodes are extracted as network topological features. Finally, the time-series feature vector containing original electromyography, angle, and imaging measurements, the first-order dynamic feature describing the rate of change, the periodic feature describing the gait cycle, and the network topological feature describing the feature association structure are vertically merged to form a high-dimensional rehabilitation feature set with significantly expanded dimensions. The high-dimensional rehabilitation feature set simultaneously encompasses the joint biomechanics represented by the lower limb force line, the soft tissue healing process represented by the bone healing area, and the neuromuscular control pattern represented by the muscle activation sequence.

[0075] In some embodiments, the calculation of first-order dynamic features can be achieved through a well-defined difference formula. For a specific feature component in a time series feature vector... At the point of time The value of , and its corresponding first-order dynamic eigencomponent. This can be expressed as the rate of change of characteristic values ​​at adjacent time points:

[0076]

[0077] in: Indicates at a point in time The calculated first-order dynamic eigenvalues, Indicates the current time point eigenvalues, Indicates the previous time point eigenvalues, This represents a fixed time interval (in days) between two consecutive time points. This formula is applied to each quantifiable feature component in the time series feature vector to generate the corresponding dynamically changing sequence.

[0078] It is understandable that the extraction of periodic features is not limited to gait. In some embodiments, the system also monitors the physiological signals of patients performing periodic rehabilitation exercises, such as continuous knee flexion and extension exercises, and extracts the periodicity and symmetry parameters of the movements from the joint angle signals as periodic features. Optionally, the construction of network topology features is not limited to pairwise feature relationships. The system can use multivariate time series analysis methods to simultaneously analyze the dynamic correlation between electromyography, angle, and pain index, construct a directed weighted influence relationship network, and then extract more complex network features.

[0079] In practical implementation, multimodal data alignment is a fundamental step. The data assimilation engine maintains a master clock, and the timestamps of all physiological signal acquisition devices, imaging equipment, and patient self-assessment terminals are synchronized with this master clock during initialization. Optionally, for historical imaging sequences that cannot be synchronized in real time, the data assimilation engine reads the acquisition date and time from the metadata information of the image files and combines it with the exact time in the surgical record to convert it into a unified postoperative time point, achieving temporal alignment. Spatial calibration is completed by having the patient perform a joint multimodal data acquisition in a specific posture when using the system for the first time. The data acquired in this acquisition provides a transformation matrix for spatial registration of all subsequent 3D imaging models and real-time sensor data. It can be understood that the unstructured text conversion of patient self-assessment records relies on a continuously optimized medical natural language processing model. This model maps common pain descriptors and functional limitation descriptive phrases to predefined standardized scale levels, ensuring the objectivity and quantification of subjective evaluation.

[0080] See Figure 3 In one embodiment of the present invention, a pre-constructed patient rehabilitation state evolution map is constructed. A large amount of complete rehabilitation data from historical patients is collected, including assimilation data throughout the entire process from post-surgery to full functional recovery or reaching a plateau. For each historical patient's complete assimilation data, a sliding segment is performed according to a fixed-length time window, resulting in a series of rehabilitation state snapshots arranged chronologically. Using a nonlinear manifold learning method, the high-dimensional rehabilitation feature set corresponding to each rehabilitation state snapshot is projected into a low-dimensional latent state space, with each projection point representing a rehabilitation state point. Continuous rehabilitation state points from the same historical patient are connected chronologically to form a historical rehabilitation trajectory representing the patient's complete rehabilitation process. Each historical rehabilitation trajectory is associated with its final joint function recovery level, and all trajectories are stored to collectively constitute the patient rehabilitation state evolution map.

[0081] Based on a high-dimensional rehabilitation feature set, similarity retrieval is performed in a pre-constructed patient rehabilitation state evolution map to match the historical rehabilitation trajectory clusters closest to the current patient's state. The current patient's high-dimensional rehabilitation feature set is projected onto the same latent state space to obtain a rehabilitation state point representing their current state. The multi-dimensional distance between this current rehabilitation state point and all historical rehabilitation state points in the patient's rehabilitation state evolution map is calculated; the distance metric can combine Euclidean distance and feature weights. Several closest rehabilitation state points are selected, and their historical rehabilitation trajectories are traced back to find their corresponding historical rehabilitation trajectories. Based on the similarity between the final recovery level, early stage features, and the current patient's state of these historical rehabilitation trajectories, as well as the morphological consistency of the trajectory direction, the set of trajectories most similar to the current patient's potential recovery pattern is selected; this is the historical rehabilitation trajectory cluster.

[0082] In practice, the process of pre-constructing the patient rehabilitation status evolution map begins with the collection of historical data. The system collects complete rehabilitation data from 100 historical patients who have completed all rehabilitation courses. The complete rehabilitation data includes assimilation data from the first day after surgery to the functional plateau period at the twelfth month after surgery. For the assimilation data of each historical patient, a sliding segment is performed with a fixed-length time window of two weeks and a step size of one week to obtain a series of rehabilitation status snapshots arranged in chronological order. For example, the rehabilitation status snapshot sequence of a certain patient is "days 1-14 after surgery", "days 8-21 after surgery", and up to "days 330-343 after surgery". Using the t-distributed random neighborhood embedding algorithm in the nonlinear manifold learning method, the high-dimensional rehabilitation feature set corresponding to each rehabilitation status snapshot is projected into a two-dimensional latent state space. The two-dimensional coordinate points obtained by projection are the rehabilitation status points representing the rehabilitation status at that moment. Connecting the continuous rehabilitation status points of the same historical patient in chronological order forms a historical rehabilitation trajectory with a clear direction, extending from the early postoperative period to the end of rehabilitation. Each historical rehabilitation trajectory is associated with and labeled with the knee joint association score level used at the final follow-up, such as "excellent", "good", "average". All historical rehabilitation trajectories with recovery level labels are stored to form a patient rehabilitation status evolution map containing multiple evolution paths.

[0083] In practice, based on the high-dimensional rehabilitation feature set, similarity retrieval is performed in the pre-constructed patient rehabilitation state evolution map to match historical rehabilitation trajectory clusters. Once the data of a new patient in the fourth week post-surgery is processed and a high-dimensional rehabilitation feature set is generated, the state matching module is activated. The system projects the patient's current high-dimensional rehabilitation feature set into the same two-dimensional latent state space using the same t-distributed random neighborhood embedding algorithm used when constructing the map, obtaining a rehabilitation state point representing the current state, with coordinates assumed to be (0.15, -0.02). The state matching module calculates the multi-dimensional Euclidean distance between this current rehabilitation state point and all historical rehabilitation state points stored in the patient's rehabilitation state evolution map. The system selects the 50 historical rehabilitation state points with the smallest Euclidean distance and backtracks through the map database to find the 30 different historical rehabilitation trajectories to which each of these 50 points belongs. Based on the final recovery level of these 30 historical rehabilitation trajectories, the similarity between the trajectories' shape in the first four weeks and the distribution of the current patient's condition points in the first four weeks, and the consistency of the trajectories' subsequent trends, a comprehensive screening was conducted. Finally, 5 historical rehabilitation trajectories that are most similar to the current patient's potential recovery pattern were selected, and the set of these 5 trajectories was determined as the matched historical rehabilitation trajectory cluster.

[0084] In some embodiments, the nonlinear manifold learning method is not limited to the t-distributed random neighborhood embedding algorithm. Optionally, the system may employ a uniform manifold approximation and projection algorithm to project the high-dimensional rehabilitation feature set onto the latent state space. This method has different characteristics in maintaining the balance between local and global data structures. When calculating the multi-dimensional distance between the current rehabilitation state point and historical rehabilitation state points, the distance metric can be Mahalanobis distance to consider the covariance relationship between different feature dimensions. Multi-dimensional distance The weighted Euclidean distance can be calculated using the following formula:

[0085]

[0086] in: Indicates the current recovery status point and the first The distance between historical recovery status points This represents the dimension of the potential state space. Indicates the first Preset weighting factors for each dimension, This indicates the current recovery status point is at the [number]th [location]. Coordinate values ​​in each dimension Indicates the first The historical recovery status point at the first Coordinate values ​​in each dimension. Weighting factor. The settings can be based on the importance of the feature in predicting the recovery level.

[0087] It is understandable that the number of nearest historical rehabilitation status points selected is a configurable parameter. In some embodiments, the system dynamically adjusts the number of points selected based on the distribution density of points in the potential state space, selecting more points in dense areas and fewer points in sparse areas to ensure that the backtracked trajectory is statistically representative. Optionally, the evaluation of trajectory morphological consistency relies not only on visual judgment, but the system also calculates the dynamic time-normalized distance between the current patient status point sequence and each candidate historical rehabilitation trajectory in the corresponding time period to numerically measure morphological similarity. During the matching process, the system prioritizes historical rehabilitation trajectories with a final recovery level of "excellent" and high early similarity as the main reference for generating the rehabilitation intervention blueprint.

[0088] In one embodiment of the invention, an individualized rehabilitation intervention blueprint is executed, guiding the patient to complete specified movements using intelligent rehabilitation equipment, and execution feedback data is collected simultaneously. The individualized rehabilitation intervention blueprint details the phased goals, training content, and intensity prescriptions for a future rehabilitation cycle. The system decomposes the training content in the individualized rehabilitation intervention blueprint into a series of specific joint movement commands and resistance setting parameters. The system controls the intelligent rehabilitation equipment to drive the patient's limbs to complete training movements according to the predetermined range of motion, movement speed, and resistance settings. Through a multi-axis force sensor and a high-frame-rate optical capture system integrated into the intelligent rehabilitation equipment, the joint torque curves, the deviation between the actual movement trajectory and the preset trajectory, and the activation sequence of major muscle groups are recorded in real time during the movement execution process. The recorded joint torque curves, movement trajectory deviations, and muscle activation sequences are synchronized and integrated in time to form a structured execution feedback data stream.

[0089] In practice, the blueprint generation module generates an individualized rehabilitation intervention blueprint for a patient in the fourth week post-surgery, based on the matched historical rehabilitation trajectory clusters. This individualized intervention blueprint details the phased goals, training content, and intensity prescriptions for the next two-week rehabilitation cycle. For example, the phased goal is "to achieve an active knee flexion angle of 120 degrees and restore quadriceps strength to 60% of the healthy side." Training content includes "seated active assisted knee flexion" and "short-arc quadriceps extension," while the intensity prescription specifies "10 repetitions per set, 3 sets per day, with resistance set at 30% of the maximum assist force." The system breaks down the "seated active assisted knee flexion" training content in the individualized intervention blueprint into a series of specific joint movement commands and resistance settings. The joint movement commands include "starting angle 0 degrees (fully extended), target angle 100 degrees, movement speed 15 degrees per second," and the resistance settings include "providing a 5 Nm assist torque in the 0-30 degree flexion range and a 3 Nm assist torque in the 30-100 degree range." The system controls the intelligent rehabilitation device connected to the patient's lower limbs. The motor of the intelligent rehabilitation device drives the patient's lower leg to complete the knee flexion movement according to the predetermined range of motion from 0 to 100 degrees, the movement speed of 15 degrees per second, and the segmented assist torque.

[0090] In practice, a multi-axis force sensor and a high-frame-rate optical capture system integrated into the intelligent rehabilitation device record the joint torque curve, motion trajectory deviation, and muscle activation sequence during the movement execution process. The multi-axis force sensor, installed at the end of the lever arm of the intelligent rehabilitation device, measures the torque experienced by the lower leg in the sagittal plane in real time, forming a joint torque curve that varies over time. This curve reflects the combined effect of the patient's active force and the device's assisted force. Four high-frame-rate optical capture cameras deployed around the training area track reflective markers pasted on the patient's thigh and lower leg, recording the three-dimensional coordinates of the markers at a frequency of 200 frames per second. By calculating the angle between the lower leg marker and the thigh marker, the actual motion trajectory is obtained and compared with the preset ideal trajectory of "uniform motion from 0 to 100 degrees," calculating the motion trajectory deviation. The motion trajectory deviation is quantified as the root mean square difference between the actual angle and the ideal angle at each time point. The system simultaneously records electromyographic signals from the rectus femoris, vastus medialis, and hamstring muscles using surface electromyography electrodes. After rectification and filtering, the activation initiation time and activation intensity sequence of each muscle during knee flexion are determined. The system rigorously synchronizes the recorded joint torque curves, motion trajectory deviations, and muscle activation sequence data, using the action start command issued by the intelligent rehabilitation device controller as a unified zero point, integrating them into a single time-labeled, multi-channel parallel execution feedback data set.

[0091] In some embodiments, the calculation of motion trajectory deviation can be quantized using a continuous error function. For each sampling time... Movement trajectory deviation This can be expressed as the instantaneous difference between the actual motion trajectory and the ideal motion trajectory in the joint angle space:

[0092]

[0093] in: This represents the average trajectory deviation over a complete training cycle. This represents the total number of sampling points within a complete action cycle. Indicates the first The actual knee joint angle value obtained by the optical capture system at each sampling time. Indicates the first The ideal knee joint angle value specified in the individualized rehabilitation intervention blueprint at each sampling moment. This calculation process is executed automatically after each training movement to quantify the quality of the movement execution.

[0094] It is understandable that intelligent rehabilitation devices have diverse control modes. In some embodiments, the resistance setting parameter is not a constant assist torque. The system can estimate the patient's active torque component in real time based on the joint torque curve and dynamically adjust the assist torque using an adaptive control algorithm, ensuring that the patient always bears the target percentage of the activity load. Optionally, the guidance of movement execution relies not only on the mechanical drive of the device. The system can present a virtual target arc of the knee joint angle on the display screen of the intelligent rehabilitation device. The patient can actively attempt to follow the arc by observing the positional relationship between their real-time angle cursor and the target arc. The intelligent rehabilitation device only provides differential assistance when the patient's force is insufficient. This mode emphasizes the patient's active participation. The acquisition of muscle activation timing is not only used for feedback. The system can calculate the co-contraction ratio of the quadriceps and hamstrings in real time. If the co-contraction ratio exceeds a safe threshold, the intelligent rehabilitation device will pause the movement and issue a prompt to avoid inappropriate muscle compensation patterns.

[0095] In one embodiment of the present invention, the construction and operation of a joint function dynamics extrapolation model are described. A personalized lower limb musculoskeletal multibody dynamics model is established for the patient. The skeletal geometric parameters and muscle-bone attachment points of this model are obtained from the patient's preoperative CT or MRI imaging data through three-dimensional reconstruction and anatomical landmark calibration. This lower limb musculoskeletal multibody dynamics model integrates a time-varying stiffness model reflecting the change in biomechanical strength of bone healing over time after osteotomy, and a contact mechanics model simulating adaptive changes in cartilage under load. The actual motion trajectory and measured joint torque information from the execution feedback data are used as boundary conditions and input into the lower limb musculoskeletal multibody dynamics model. Through inverse dynamics calculation, the activation level of the major muscle groups and the joint contact force during this movement are solved. The calculated joint contact force and muscle group activation level are input into the integrated time-varying stiffness model and contact mechanics model for forward biomechanical extrapolation, simulating the micromechanical response and functional adaptive changes of bone tissue, articular cartilage, and surrounding ligaments in the osteotomy area under the current load. The process of iteratively performing reverse calculations and forward deductions simulates the adaptation results of bone and cartilage tissue state, ligament mechanical properties, and nervous system control patterns after multiple rehabilitation training sessions, and comprehensively predicts the expected joint function score based on this.

[0096] Parameter calibration of the joint function dynamics extrapolation model. In the early postoperative period, baseline feedback data was collected using a low-load passive activity mode. This baseline data was input into the joint function dynamics extrapolation model, which had not yet undergone individualized calibration, to obtain initial model extrapolation results. The initial extrapolation results output by the model were compared with actual clinical assessment results, such as muscle strength tests and joint range of motion measurements, to calculate the model's prediction error. Using the principle of error backpropagation, the muscle lever arm parameters, tissue stiffness parameters, and neural control gain parameters in the model were adjusted to continuously approximate the actual clinical assessment results, thus completing the calibration of the individualized model for this patient.

[0097] In practice, a personalized lower limb musculoskeletal multibody dynamic model is established as the core of the joint function dynamics derivation model. The parameters of the lower limb musculoskeletal multibody dynamic model are obtained from the patient's preoperative imaging data and anatomical landmarks. The system imports the patient's preoperative high-resolution knee CT scan data and generates a skeletal geometric model containing the distal femur, proximal tibia, patella, and major ligament attachment points through 3D reconstruction software. On the geometric model, the origin and insertion points and muscle force lines of major muscle groups such as the vastus medialis, vastus lateralis, rectus femoris, and hamstrings are accurately calibrated on the skeleton, thereby determining personalized muscle lever arm parameters. In the constructed multibody dynamics model of the lower limb skeletal muscles, a time-varying stiffness model is integrated to reflect the change of biomechanical strength of bone healing over time after osteotomy. The time-varying stiffness model uses postoperative time as a variable to describe the gradual increase of the equivalent stiffness of the osteotomy area. At the same time, a contact mechanics model is integrated to simulate the deformation, fluid exudation and matrix remodeling of articular cartilage under mechanical load. The contact mechanics model defines the mechanical response of cartilage based on nonlinear elasticity and porous media theory.

[0098] In practice, the joint function dynamics model receives execution feedback data and performs calculations. The actual flexion-extension angle trajectory of the knee joint and the measured joint torque curve from the execution feedback data recorded during a "seated active assisted knee flexion" training movement are used as dynamic boundary conditions and input into the lower limb skeletal muscle multibody dynamics model. Through inverse dynamics calculations, the model solves for the activation level time-series curves of major muscle groups such as the quadriceps and hamstrings during this movement, as well as the peak and distribution of joint contact forces in the tibiofemoral and patellofemoral joints, under known movement trajectories and external torques. The calculated joint contact forces and muscle activation levels are then input into the integrated time-varying stiffness model and contact mechanics model for forward biomechanical simulation. This simulates the microscopic strain and adaptive bone remodeling of the callus in the osteotomy area, the contact stress distribution of the articular cartilage and the mechanical response of the subchondral bone, and the stress state of the surrounding ligament structures under this training load. The system iteratively executes the above reverse calculation and forward deduction process to simulate the adaptive changes in bone healing stiffness, cartilage mechanical properties, and neuromuscular control patterns due to cumulative effects after 36 similar rehabilitation training sessions within the next two weeks. Based on the deduced final tissue state and neural control efficiency, the system comprehensively calculates and predicts the patient's expected joint function score two weeks later.

[0099] In practice, parameter calibration of the joint function dynamics model is initiated early postoperatively. On the third postoperative day, the system-controlled intelligent rehabilitation device drives the patient's knee joint to perform low-load passive movements in a weightless state, collecting baseline feedback data on angles, torques, and surface electromyography. This baseline feedback data is input into the joint function dynamics model, which has not yet undergone individualized calibration. Initial projection results are obtained through model calculations; for example, the projected quadriceps activation level should be 15% of maximum voluntary contraction, and the peak joint contact force should be 0.2 times the body weight. The initial projection results are compared with actual clinical assessment results obtained by rehabilitation therapists through manual muscle strength testing and instrument measurements. The comparison reveals that the model-projected quadriceps activation level is higher than the measured value, and the projected joint contact force is lower than the measured value. The model prediction error vector is then calculated. Utilizing the principle of error backpropagation, the system automatically adjusts the muscle lever arm parameters in the lower limb musculoskeletal multibody dynamics model, the tissue stiffness parameters in the time-varying stiffness model and contact mechanics model, and the neural control gain parameters in the inverse calculation module. Through multiple iterations, the model's output of muscle activation levels and joint contact forces under the same input is made to approximate the actual measured values, thus completing the calibration of the patient's personalized joint function dynamics model. See Table 1 for key parameter adjustments before and after calibration.

[0100] Table 1: Calibration Table of Key Parameters for Joint Functional Dynamics Deduction Model

[0101]

[0102] In some embodiments, the time-varying stiffness model can be described by an exponential asymptotic function to describe the stiffness of the bone healing region. Changes over time:

[0103]

[0104] in: Indicates the first postoperative day Equivalent mechanical stiffness of the osteotomy area This indicates the final expected stiffness of the bone tissue after complete healing. Indicates the initial stiffness immediately after surgery. This represents the healing rate constant determined by factors such as the patient's age and nutritional status. It is a natural constant. This function defines the relationship between stiffness and its increasing time from the initial value to the final value.

[0105] It is understandable that model parameter calibration is an ongoing process. In some embodiments, the system performs a one-time calibration not only in the early postoperative period but also recalibrates the model parameters using new assessment data at key rehabilitation milestones (such as week 4 and week 12 postoperatively) to dynamically track changes in patient tissue healing and neural adaptation. Optionally, the cartilage contact mechanics model in forward modeling can employ biphasic or multiphasic porous media theory to calculate not only stress distribution but also the impact of synovial fluid flow in the cartilage matrix on load transfer and energy dissipation. Error backpropagation calibration can be based on gradient descent or more advanced optimization algorithms to automatically find a set of optimal patient-specific model parameters by minimizing the loss function between model predictions and clinically measured values.

[0106] See Figure 4 This is a dual-indicator trend chart showing the relationship between the number of rehabilitation training sessions and joint biomechanical response. Focusing on the rehabilitation scenario after knee osteotomy and orthopedic surgery, it clearly presents the cumulative changes of two core biomechanical indicators with training, representing a visualization of the positive analysis results of the joint function dynamics model. The tibiofemoral joint contact force increases non-linearly, with the rate of increase gradually slowing down, reflecting the recovery of the joint's load-bearing capacity. The slower rate of increase suggests improved joint adaptability to load, indicating no risk of overload. The cartilage contact stress increases non-linearly, with an overall stable rate of increase. The cartilage stress is within the physiological tolerance range, and the stable increase indicates that the cartilage matrix gradually completes biomechanical adaptation and remodeling under rehabilitation load. Based on this trend, a biomechanical assessment can be added after 18 training sessions. If the stress rate of increase is abnormal, the training intensity can be reduced through the path adjustment module.

[0107] In one embodiment of the invention, a dynamic adjustment mechanism for the rehabilitation pathway is activated based on the deviation between the expected joint function score and the current actual score, allowing for online correction of the ongoing individualized rehabilitation intervention blueprint. The difference between the projected expected joint function score and the immediate function score calculated based on the patient's current performance is calculated; this difference serves as the rehabilitation progress deviation. The components of the rehabilitation progress deviation are analyzed to identify whether the deviation primarily stems from insufficient strength, limited joint mobility, or abnormal motor control. A predefined rehabilitation strategy adjustment rule base is queried, and targeted adjustment suggestions are matched based on the identified deviation type and degree. These suggestions include increasing isometric contraction training for specific muscle groups, introducing additional joint mobilization techniques, or adjusting the auxiliary resistance ratio of training movements. The generated adjustment suggestions are integrated into the individualized rehabilitation intervention blueprint for subsequent cycles, overriding or modifying the corresponding training plan sections.

[0108] The system also includes a closed-loop control module, which translates online correction instructions into specific equipment control parameters and patient guidance prompts, forming a closed-loop rehabilitation guidance process. The training content in the adjusted individualized rehabilitation intervention blueprint is parsed into motor angle curves, resistance torque curves, and safe range threshold parameters executable by the intelligent rehabilitation equipment. Audiovisual guidance prompts synchronized with the training movements are generated, including verbal explanations of the key points of the movements, visual indicators of the target locations, and real-time feedback sound effects indicating the quality of completion. At the start of the next training cycle, the system loads the parsed motor angle curves, resistance torque curves, and safe range thresholds into the controller of the intelligent rehabilitation equipment and simultaneously plays the generated audiovisual guidance prompts, thereby automatically implementing the corrected rehabilitation plan.

[0109] In practice, the path adjustment module, based on the deviation between the expected joint function score derived from the joint function dynamics model and the patient's current actual score, activates the rehabilitation path dynamic adjuster. The dynamic adjuster then makes online corrections to the ongoing individualized rehabilitation intervention blueprint. The joint function dynamics model predicts a joint function score of 85 points (out of 100) for the patient two weeks after implementing the current rehabilitation plan. The system's immediate function score, calculated based on the patient's most recent training performance, is 78 points. The path adjustment module calculates the difference between the expected and immediate function scores, resulting in a rehabilitation progress deviation of -7 points. The module analyzes the composition of this deviation. By analyzing various sub-scores under the immediate function score (such as muscle strength score, range of motion score, and motor control score), it identifies that the deviation primarily originates from the strength dimension represented by "insufficient peak torque of the quadriceps," and secondarily from the range of motion dimension represented by "lag at the end of active knee flexion." No significant abnormalities were found in the motor control dimension. The Rehabilitation Path Dynamic Adjuster queries a predefined rehabilitation strategy adjustment rule base. Based on the deviation type (primarily insufficient strength, secondarily limited range of motion) and the deviation level of -7 points, it matches targeted adjustment suggestions. These suggestions include: "Add isometric contraction training of the quadriceps before the original training, holding each set for 6 seconds, for a total of 3 sets," "Introduce passive joint mobilization techniques at the end of knee flexion to improve joint gliding," and "Increase the proportion of auxiliary resistance of the equipment in active assisted knee flexion training from 30% to 40%." The Rehabilitation Path Dynamic Adjuster integrates the generated adjustment suggestions into the individualized rehabilitation intervention blueprint for the following week, covering the intensity and warm-up activities of the "seated active assisted knee flexion" section of the original training plan.

[0110] In practical implementation, the closed-loop control module included in the system translates online correction instructions into specific device control parameters and patient guidance prompts. The closed-loop control module receives the adjusted individualized rehabilitation intervention blueprint and parses the training content of "increasing isometric contraction training of the quadriceps" and "adjusting the auxiliary resistance ratio to 40%" in the adjusted individualized rehabilitation intervention blueprint into motor angle curves, resistance torque curves, and safety range thresholds that can be executed by the intelligent rehabilitation device. The motor angle curve is defined as "keeping the knee joint in a 30-degree flexion position", the resistance torque curve is defined as "providing a constant resistance of 20 Newton-meters in the opposite direction of the patient's force within 0-6 seconds", and the safety range threshold is set to "joint angle fluctuation not exceeding ±2 degrees". The closed-loop control module generates synchronized audiovisual guidance prompts, including the following commands: "Prepare for quadriceps isometric contraction, press down forcefully, hold, 6, 5, 4, 3, 2, 1, relax." A progress bar gradually changes from green to red on the screen as a visual indicator of the target position. Encouraging sound effects are played when the patient reaches more than 80% of the target torque, providing real-time feedback on the completion quality. At the start of the next training cycle, the system automatically loads the analyzed motor angle curve, resistance torque curve, and safety range threshold to the intelligent rehabilitation equipment controller. The intelligent rehabilitation equipment fixes the patient's knee joint in a 30-degree flexion position and applies resistance, while simultaneously playing the generated audiovisual guidance prompts, thus automatically and in a closed loop implementing the modified rehabilitation program.

[0111] In some embodiments, the deviation in recovery progress This can be quantified as a difference function of a comprehensive score:

[0112]

[0113] in: This indicates the deviation in recovery progress. This represents the future target time point predicted by the joint function dynamics model. Expected joint function score, Indicates based on the current time point The instantaneous joint function score is calculated from the actual performance data. A positive deviation indicates that the progress is ahead of schedule or in line with expectations, while a negative deviation indicates that the progress is behind schedule.

[0114] It is understandable that the rehabilitation strategy adjustment rule base is a set of if-then rules built upon clinical knowledge and historical data. In some embodiments, the rule base considers not only the type and degree of deviation but also the patient's current postoperative time and pain index. For example, in the early postoperative period, even if strength deviation occurs, high-load isometric contractions are avoided first, and alternative strategies such as neuromuscular electrical stimulation are used instead. Optionally, the generation of audiovisual guidance prompts can be highly personalized. The system can select different styles of guiding voice or visual themes based on the patient's historical response preferences to improve patient compliance and training experience. The closed-loop control module performs a safety check before loading new parameters, such as comparing the newly set resistance torque with the patient's historical maximum tolerable torque. If it exceeds the safe range, the system will refuse to execute and prompt the therapist for manual review.

[0115] See Figure 5 This is a trend chart of full-cycle rehabilitation monitoring after knee osteotomy and orthopedic surgery, spanning from March 1st to May 20th, 2026. It visually presents the dynamic changes of three core rehabilitation indicators and is a typical output visualization of the data assimilation module and functional deduction module. Joint range of motion shows a continuous linear increase with a stable rate of increase, reflecting the core process of joint function recovery and conforming to the expected trajectory of postoperative rehabilitation, indicating effective improvement in joint adhesions and contractures. Quadriceps muscle strength initially increased slowly but then rapidly, with a significant increase in the rate of increase after 4 months. Muscle strength recovery lags behind joint range of motion, consistent with the rehabilitation logic of "recovering range of motion first, then strengthening muscle strength," indicating the entry into the active muscle strength training phase after 4 months. Pain scores continued to decline and stabilized at a low level after 4 months, indicating good pain control and suggesting inflammation resolution and tissue healing, providing a safe foundation for high-intensity rehabilitation training.

[0116] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A system for predicting and guiding the recovery of joint function after knee osteotomy and orthopedic surgery, characterized in that, include: The data assimilation module acquires the patient's full-cycle postoperative rehabilitation data and sends it to the data assimilation engine for processing, in order to extract a high-dimensional rehabilitation feature set from the assimilated data. The state matching module, based on the high-dimensional rehabilitation feature set, performs similarity retrieval in the pre-constructed patient rehabilitation state evolution map and matches the historical rehabilitation trajectory clusters that are closest to the current patient state. The blueprint generation module dynamically generates an individualized rehabilitation intervention blueprint based on the matched historical rehabilitation trajectory clusters, executes the individualized rehabilitation intervention blueprint, guides the patient to use intelligent rehabilitation equipment to complete the specified actions, and simultaneously collects execution feedback data. The functional deduction module inputs the execution feedback data into the joint function dynamics deduction model in real time. The joint function dynamics deduction model couples the biomechanical response of the bone-cartilage-ligament complex with the adaptive adjustment process of muscle recruitment strategy. Through the calculation of the joint function dynamics deduction model, the expected joint function score of the patient at future time points is deduced. The path adjustment module activates a rehabilitation path dynamic adjuster based on the deviation between the expected joint function score and the current actual score, and makes online corrections to the intensity, frequency or action combination in the individualized rehabilitation intervention blueprint being implemented. The construction and operation process of the joint function dynamics deduction model includes: A personalized lower limb skeletal muscle multibody dynamics model was established for each patient. The parameters of the lower limb skeletal muscle multibody dynamics model were determined by the patient's preoperative imaging data and anatomical landmarks. The lower limb skeletal muscle multibody dynamics model integrates a time-varying stiffness model that reflects the changes in biomechanical strength of bone healing after osteotomy, and a contact mechanics model that simulates the adaptive changes in cartilage load. The motion trajectory and torque information in the execution feedback data are used as boundary conditions and input into the lower limb skeletal muscle multibody dynamics model to inversely calculate the activation level of the main muscle groups and the joint contact force. The calculated joint contact force and muscle activation level are input into the time-varying stiffness model and contact mechanics model to positively extrapolate the micromechanical response and functional adaptation of bone tissue, articular cartilage and surrounding ligaments in the osteotomy area under the current load. The process of iteratively performing reverse calculations and forward deductions simulates the tissue state and neural control patterns accumulated after multiple rehabilitation training sessions in the future, and predicts the expected joint function score based on this.

2. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 1, characterized in that, The full-cycle postoperative rehabilitation data is fed into a data assimilation engine for processing, including: The full-cycle postoperative rehabilitation data includes multimodal physiological signals, imaging sequences, and patient self-assessment records; The data assimilation engine calibrates data from different sources and time phases to a unified time-space reference framework; The multimodal physiological signals were time-stamped and denoised to extract stable segments of electromyography, angle and pressure signals. The imaging sequences were reconstructed and registered in three dimensions to quantify the bony healing area, force line angle changes, and dynamic characteristics of the joint space in the osteotomy region. The unstructured descriptions in the patient self-assessment records were converted into standardized functional limitation levels and pain intensity indices; A unified timeline is established, and the processed multimodal physiological signal features, quantified imaging features, and standardized self-assessment record features are mapped onto the unified timeline according to the number of days after surgery, forming a time series feature vector.

3. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 2, characterized in that, Extract a high-dimensional rehabilitation feature set from the assimilated data, including: From the time series feature vector, the rate of change of features between adjacent time points is calculated to form a first-order dynamic feature; Identify periodic patterns of specific feature combinations in the time series feature vector and extract pattern parameters as periodic features; Analyze the time lag correlation between different features, construct an influence relationship network between features, and extract network topology features from the influence relationship network; The original time series feature vector, first-order dynamic features, periodic features, and network topology features are combined to form the high-dimensional rehabilitation feature set. The high-dimensional rehabilitation feature set covers joint biomechanical state, soft tissue healing process, and neuromuscular control patterns.

4. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 3, characterized in that, The construction process of the pre-constructed patient rehabilitation state evolution map includes: It compiles a large amount of complete rehabilitation data from historical patients, which includes assimilation data from the entire process from postoperative surgery to functional recovery; The assimilation data of each historical patient throughout the entire process is divided into sliding segments according to time windows to obtain a series of snapshots of the recovery status arranged in chronological order; Using a nonlinear manifold learning method, the high-dimensional rehabilitation feature set corresponding to each rehabilitation state snapshot is projected into a low-dimensional latent state space, and the coordinate points obtained by the projection are the rehabilitation state points. Connect the consecutive recovery status points of the same historical patient in chronological order to form a historical recovery trajectory; The historical rehabilitation trajectory is associated with its final joint function recovery level and stored as an evolution map of the patient's rehabilitation status.

5. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 4, characterized in that, Based on the high-dimensional rehabilitation feature set, a similarity search is performed in the pre-constructed patient rehabilitation state evolution map to match the historical rehabilitation trajectory clusters that are closest to the current patient state, including: Projecting the current patient's high-dimensional rehabilitation feature set onto the latent state space yields the current rehabilitation state point; Calculate the multi-dimensional distance between the current rehabilitation status point and all rehabilitation status points in the patient's rehabilitation status evolution map; Select the nearest several rehabilitation status points and backtrack to find the historical rehabilitation trajectory to which the several rehabilitation status points belong; Based on the final recovery level, early feature similarity, and trajectory morphology consistency of the historical recovery trajectories, a set of trajectories most similar to the current patient's potential recovery pattern is selected, namely the historical recovery trajectory cluster.

6. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 5, characterized in that, The individualized rehabilitation intervention blueprint is executed, guiding the patient to use intelligent rehabilitation equipment to complete designated actions, and execution feedback data is collected simultaneously, including: The individualized rehabilitation intervention blueprint details the phased goals, training content, and intensity prescriptions for the future rehabilitation cycle; The training content in the individualized rehabilitation intervention blueprint is broken down into a series of specific joint movement commands and resistance settings; The intelligent rehabilitation equipment controls the patient's limbs to complete movements according to a predetermined range of motion, speed, and resistance. By integrating a multi-axis force sensor and a high frame rate optical capture system into the intelligent rehabilitation device, the joint torque curve, motion trajectory deviation and muscle activation sequence are recorded during the execution of the movement. The recorded joint torque curves, motion trajectory deviations, and muscle activation timings are integrated into the time-synchronized execution feedback data.

7. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 6, characterized in that, The parameter calibration process of the joint function dynamics deduction model includes: Baseline feedback data were collected in the early postoperative period using a low-load passive activity mode. The baseline feedback data is input into the uncalibrated joint function dynamics extrapolation model to obtain the initial extrapolation results; The initial projection results are compared with the actual clinical assessment results to calculate the model prediction error; By utilizing the principle of backpropagation of errors, the muscle lever arm parameters, tissue stiffness parameters, and neural control gain parameters in the model are adjusted to make the model's projection results approximate the actual clinical assessment results, thus completing the model calibration.

8. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 7, characterized in that, Based on the deviation between the expected joint function score and the current actual score, a rehabilitation pathway dynamic adjuster is activated to make online corrections to the intensity, frequency, or movement combinations in the currently implemented individualized rehabilitation intervention blueprint, including: The difference between the expected joint function score and the immediate function score calculated based on the current actual performance is used as the rehabilitation progress deviation. Analyze the composition of the aforementioned rehabilitation progress deviation to identify whether it stems from insufficient strength, limited range of motion, or abnormal motor control. The system queries a predefined rehabilitation strategy adjustment rule base and matches targeted adjustment suggestions based on the type and degree of deviation. These suggestions include increasing isometric contraction training for specific muscle groups, introducing additional joint mobilization techniques, or adjusting the proportion of auxiliary resistance in the movement. The proposed adjustments will be integrated into the individualized rehabilitation intervention blueprint for subsequent cycles, covering the original training plan portion.

9. The prediction and rehabilitation guidance system for joint function recovery after knee osteotomy and orthopedic surgery as described in claim 8, characterized in that, Also includes: Closed-loop control module, The online correction instructions are translated into specific device control parameters and patient guidance prompts to form a closed-loop rehabilitation guidance process, including: The training content in the adjusted individualized rehabilitation intervention blueprint is analyzed into motor angle curves, resistance torque curves, and safety range thresholds that can be executed by intelligent rehabilitation equipment. Generate synchronized audiovisual guidance prompts, which include verbal commands for action key points, visual indications of target locations, and real-time feedback on completion quality; When the next training cycle starts, the motor angle curve, resistance torque curve, and safety range threshold are loaded into the intelligent rehabilitation equipment controller, and the audiovisual guidance prompts are played simultaneously to implement the revised rehabilitation plan.

Images