Orthopedic complete-cycle rehabilitation management system based on artificial intelligence

Abstract

The invention relates to the technical field of medical rehabilitation, in particular to an artificial intelligence-based orthopedic full-cycle rehabilitation management system, which comprises a rehabilitation demand evaluation module, a three-dimensional pathological modeling module, a multi-modal training generation module, a rehabilitation effect prediction module, a dynamic parameter adjustment module and a full-cycle management module. Multi-dimensional data such as bone mineral density, joint motion range, gait mechanics and psychological state of a patient are collected, an artificial intelligence algorithm is combined for analysis, a personalized rehabilitation scheme is generated, and rehabilitation parameters are predicted and adjusted in real time. The system can dynamically correct the training intensity, the nutritional supplement dosage, the psychological intervention mode and the rehabilitation schedule, and the safety and reliability of data are guaranteed through the block chain technology; according to the system, biomechanics, nutriology, psychology and environmental factors are comprehensively fused, the system has strong adaptive ability and full-period management ability, real-time optimization can be performed according to the rehabilitation process of a patient, and the rehabilitation effect of the orthopedics department and the satisfaction degree of the patient are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical rehabilitation, and particularly to an orthopedic full-cycle rehabilitation management system based on artificial intelligence. Background Art

[0002] With the intensification of the global aging trend, the incidence of orthopedic diseases has been continuously rising. Orthopedic treatments such as fractures and joint replacement surgeries have gradually become important links in the rehabilitation treatment of patients. The core goal of orthopedic rehabilitation is to promote the functional recovery of bones and joints, reduce the occurrence of complications, and improve the quality of life of patients.

[0003] Although the research on intelligent orthopedic rehabilitation management systems has been continuously developing in recent years, there are still many deficiencies in the existing technologies. First of all, the rehabilitation programs of many existing systems lack customization for individual patients and cannot fully consider the physiological characteristics, psychological states, and environmental factor changes of patients. For example, traditional systems only rely on simple monitoring data and ignore the dynamic changes of patients during the rehabilitation process, resulting in the inability to provide the optimal personalized treatment plan. Secondly, there are also technical bottlenecks in the rehabilitation effect prediction and dynamic adjustment of existing intelligent rehabilitation systems. Although some systems use machine learning and deep learning algorithms for data analysis, they often cannot accurately predict physiological indicators such as callus growth and mechanical property recovery in real-time environments and lack sufficient adaptability to cope with the complex changes that occur during the rehabilitation process. In addition, another common problem in existing technologies is that the influence of environmental factors (such as temperature and humidity) is ignored, and the adaptability differences among different patients are not fully considered, resulting in the system not being able to achieve the best effect in practical applications. Summary of the Invention

[0004] Based on the above purposes, the present invention provides an orthopedic full-cycle rehabilitation management system based on artificial intelligence.

[0005] An orthopedic full-cycle rehabilitation management system based on artificial intelligence includes a rehabilitation requirement assessment module, a three-dimensional pathology modeling module, a multi-modal training generation module, a rehabilitation effect prediction module, a dynamic parameter adjustment module, and a full-cycle management module, wherein;

[0006] The rehabilitation requirement assessment module: collects the patient's bone density data, joint range of motion data, gait mechanics data, and psychological assessment data through multi-modal sensors, and outputs a multi-dimensional rehabilitation requirement parameter set;

[0007] The three-dimensional pathology modeling module: receives the multi-dimensional rehabilitation requirement parameters, and generates a three-dimensional dynamic bone model including a stress distribution heat map and a microstructural damage model based on the reverse deduction algorithm of the bone mechanics conduction chain;

[0008] Multimodal Training Generation Module: According to the three-dimensional dynamic bone model and real-time physiological monitoring data, generate a personalized training parameter set including a mechanical loading scheme, a nutritional supplement scheme, and a psychological intervention scheme through a transfer learning network;

[0009] Rehabilitation Effect Prediction Module: Input the personalized training parameter set into the LSTM-GAN hybrid network, and output a stage rehabilitation effect matrix including the predicted value of callus growth rate and the mechanical property recovery curve;

[0010] Dynamic Parameter Adjustment Module: Based on the temperature and humidity data collected by the environmental sensor and the stage rehabilitation effect matrix, dynamically correct the training intensity parameter and the nutritional supplement dose through an adaptive particle swarm algorithm;

[0011] Full-cycle Management Module: Integrate all the corrected parameters to generate a visual rehabilitation calendar, and store the data fingerprints of each stage through the blockchain to form an immutable rehabilitation evidence chain.

[0012] Optionally, the rehabilitation demand assessment module includes:

[0013] Bone Density Data Acquisition: Perform spiral tomography scanning on the affected bone of the patient through a dual-energy X-ray bone densitometer to obtain bone mineral content distribution data. Use an improved Hounsfield value calibration algorithm to eliminate soft tissue interference, generate a three-dimensional bone density distribution map including the interface between cortical bone and cancellous bone, and output the coordinate set of bone density abnormal areas and the intensity attenuation gradient parameters;

[0014] Joint Range of Motion Quantification: Deploy a millimeter-wave radar array around the patient's joint, capture the joint movement trajectory through the reflected signal of the 60GHz band electromagnetic wave, reconstruct a three-dimensional kinematic model using Doppler phase resolution technology, calculate the flexion angle, rotational angular velocity, and movement plane offset, and generate a joint range of motion feature vector including the maximum range of motion threshold and pain trigger points;

[0015] Gait Mechanics Analysis: Embed a piezoelectric six-dimensional force sensor array in the patient's sole, continuously collect the ground reaction force data during the gait cycle at a sampling frequency of 500Hz. After eliminating motion artifacts through a wavelet denoising algorithm, use an inverse dynamics model to calculate the lower limb joint torque distribution, and output a gait mechanics parameter set including the center of pressure trajectory, the proportion of the stance phase time, and the left-right symmetry index;

[0016] Psychological State Assessment: Continuously monitor the skin conductance response and heart rate variability through a smart bracelet, combine the electronic anxiety-depression scale filled in daily, calculate the psychological stress index using a fuzzy logic algorithm, and simultaneously detect the blood oxygen level of the prefrontal cortex using a micro near-infrared spectrometer to construct a psychological assessment matrix including the emotional stability score and the rehabilitation confidence index;

[0017] Multi-dimensional parameter fusion: Input the three-dimensional bone density distribution map, joint mobility feature vector, gait mechanics parameter set, and psychological assessment matrix into the feature-level fusion network for time-axis alignment and dimension normalization processing, and assign weight coefficients to each parameter through the attention mechanism, finally generating a multi-dimensional rehabilitation requirement parameter set including biomechanical demand level, psychological intervention priority, and rehabilitation stage identifier.

[0018] Optionally, the three-dimensional pathological modeling module includes:

[0019] Multi-dimensional data parsing: Receive the multi-dimensional rehabilitation requirement parameter set from the rehabilitation requirement assessment module, extract the coordinate set of abnormal bone density regions, joint mobility feature vector, gait mechanics parameter set, and psychological assessment matrix, perform three-dimensional spatial alignment of the bone density distribution map and the mechanical loading target area coordinates through the spatial coordinate system registration algorithm, and use the cubic spline interpolation method to fill the data acquisition blind area to generate a spatio-temporally unified multi-modal bone dataset;

[0020] Inverse construction of the mechanical conduction chain: Based on the lower limb biomechanical conduction principle, taking the flexion angle in the joint mobility feature vector as the boundary condition, combined with the center of pressure trajectory data in the gait mechanics parameter set, use the finite element inverse solution method to deduce the bone stress conduction path and establish a mechanical conduction chain topological network including the principal stress direction vector and energy dissipation coefficient, where the intensity attenuation gradient parameter of the bone density distribution map is used to correct the material anisotropy parameter;

[0021] Dynamic stress field reconstruction: Input the mechanical conduction chain topological network into the improved U-Net++ network, extract multi-scale stress features through cascaded dilation convolutional layers, and combine the electromyogram signals in the real-time physiological monitoring data to generate a dynamic stress distribution heat map including peak stress regions, stress gradient change rates, and fatigue accumulation effects;

[0022] Microdamage coupling modeling: Based on the cancellous bone porosity data of the three-dimensional bone density distribution map, use the fractal dimension algorithm to calculate the microstructural damage distribution, combine the peak stress region coordinates of the dynamic stress distribution heat map, and simulate the microcrack growth trajectory through the crack propagation finite element method to construct a three-dimensional microstructural damage model including pore connectivity index and trabecular bone fracture risk value;

[0023] Multi-physical field fusion: Perform spatial registration on the dynamic stress distribution heat map and the three-dimensional microstructural damage model, introduce the temperature and humidity data collected by the environmental sensor as boundary conditions, and calculate the influence factor of the temperature gradient on the bone remodeling rate through the thermal-mechanical coupling algorithm to generate a three-dimensional dynamic bone model integrating biomechanical characteristics, microstructural state, and environmental parameters;

[0024] Model dynamic update: According to the real-time rehabilitation data fed back by the full-cycle management module, perform incremental training on the three-dimensional dynamic bone model. When it is detected that the change in the coordinate set of the abnormal bone density area exceeds 5% or the offset of the joint range of motion feature vector exceeds the preset threshold, trigger the re-deduction of the mechanical conduction chain topology network and update the prediction parameters of the stress distribution heat map;

[0025] Pathological model verification: Register and compare the generated three-dimensional dynamic bone model with the postoperative CT images in the orthopedic PACS system. Use the Hausdorff distance algorithm to calculate the matching degree between the predicted contour of the model and the real bone morphology. When the error exceeds 1.2 mm, automatically start the parameter calibration process and output a verification report containing the confidence score to the multi-modal training generation module.

[0026] Optionally, the multi-modal training generation module includes:

[0027] Multi-source data fusion: Receive the three-dimensional dynamic bone model and real-time physiological monitoring data from the three-dimensional pathological modeling module. Extract the peak stress area coordinates of the dynamic stress distribution heat map, the trabecular fracture risk value of the microstructural damage model, the temperature and humidity data of the environmental sensor, and the patient's real-time blood oxygen saturation and electromyogram signal. Use the spatio-temporal alignment algorithm to spatially match the mechanical loading target area coordinates with the micro-damage area, and generate a fusion data set containing biomechanical constraint conditions and physiological state parameters;

[0028] Mechanical loading scheme generation: Based on the peak stress area coordinates of the fusion data set, use the inverse dynamics model to calculate the activation threshold of the target muscle group, and combine the stress gradient change rate data of the three-dimensional dynamic bone model to generate variable impedance electromagnetic damping parameters through the fuzzy PID control algorithm;

[0029] Nutritional supplement optimization: Analyze the trabecular bone porosity data of the three-dimensional bone density distribution map, combine the serum calcium and phosphorus concentrations and vitamin D levels in the real-time metabolism monitoring data, construct a differential equation of bone metabolism kinetics, and use the model predictive control algorithm to calculate the daily nutritional supplement dose;

[0030] Psychological intervention matching: Input the emotional stability score of the psychological assessment matrix into the pre-trained psychological state classifier, construct a virtual reality scene decision tree based on the reinforcement learning algorithm, and select the corresponding intervention mode according to the rehabilitation stage identifier.

[0031] Optionally, the multi-modal training generation module further includes:

[0032] Transfer learning parameter adjustment: Establish a pre-trained network containing 100,000 orthopedic rehabilitation cases. Input the mechanical loading scheme, nutritional supplement scheme, and psychological intervention parameter package into the feature mapping layer. Align the distribution differences between the target patient and the source domain data through the domain adaptation algorithm, and finally output a training parameter set containing personalized impedance curves, nutritional supplement gradient functions, and VR scene trigger conditions;

[0033] Multi-modal parameter collaborative optimization: Construct a three-objective optimization model containing mechanical loading intensity, nutritional supplement dosage, and psychological intervention effect. Use the NSGA-II algorithm to solve the Pareto optimal solution set. The constraint conditions include: the joint moment does not exceed 80% of the pain trigger point threshold, the blood calcium concentration is maintained in the range of 2.1 - 2.6 mmol / L, and the fluctuation range of the psychological stress index ≤ ±15%. Output the time collaboration matrix of each scheme parameter;

[0034] Dynamic parameter encapsulation: Synchronize the optimized mechanical loading electromagnetic pulse sequence, nutritional supplement gradient function, and virtual reality scene trigger conditions on the time axis to generate a personalized training parameter set containing the training intensity (0 - 100%) at each time period within a 24-hour cycle, nutritional agent types (I - IV), and psychological intervention modes (A - D). The parameter set is encapsulated in JSON-LD format and transmitted to the rehabilitation effect prediction module with an attached data verification code.

[0035] Optionally, the rehabilitation effect prediction module includes:

[0036] Spatio-temporal data preprocessing: Receive the personalized training parameter set from the multi-modal training generation module, parse the electromagnetic pulse sequence timestamps, nutritional supplement gradient functions, and VR scene trigger conditions contained therein. Convert the discrete parameter sequence into equally spaced time step data through the sliding window algorithm, and use the Z-score normalization method to perform feature encoding on the mechanical loading intensity, nutritional agent types, and psychological intervention modes to generate a 128-dimensional time series input vector with a unified dimension;

[0037] LSTM-GAN network construction: Build a hybrid neural network architecture with dual-channel input. The generator network consists of 4 layers of bidirectional LSTM, with each layer containing 256 memory units. Receive the encoded training parameter sequence and append the electromyogram signal spectral features in the real-time physiological monitoring data. The discriminator network uses a 3-layer temporal convolutional neural network to perform adversarial training on the input real rehabilitation effect data and the generated prediction data, and set a gated attention mechanism at the output layer;

[0038] Bone callus growth kinetics modeling: Embed the bone metabolism kinetics equation into the hidden state layer of the generator network, calculate the calcium and phosphorus deposition rate according to the nutrient supplementation gradient function, combine the electromagnetic pulse frequency and damping coefficient in the mechanical loading parameters, and simulate the three-dimensional growth process of the bone callus through a differential equation solver, and output time-varying prediction parameters including the density of new trabeculae, mineralization rate, and the proportion of bone callus volume;

[0039] Prediction of mechanical property recovery: Input the stress distribution heat map of the three-dimensional dynamic bone model into the auxiliary feature channel of the discriminator network, calculate the bone stiffness recovery coefficient by using the strain energy density integration method, combine the pore connectivity index decay curve of the microstructural damage model, and map the mechanical property recovery trajectory through a radial basis function neural network, and output a time series of mechanical recovery indicators including the maximum load capacity, elastic modulus, and fatigue life;

[0040] Multi-modal result fusion: Align the time-varying prediction parameters and the time series of mechanical recovery indicators in space and time, calculate the confidence weights of each indicator by using the evidence theory fusion algorithm, construct a third-order rehabilitation effect tensor including the time dimension, space dimension, and indicator dimension, and extract key features through tensor decomposition to generate a stage rehabilitation effect matrix;

[0041] Post-processing of prediction results: Conduct a medical rationality check on the generated stage rehabilitation effect matrix. When it is detected that the mutation of the proportion of bone callus volume exceeds 20% / week or the recovery rate of elastic modulus exceeds the anatomical limit, start the constraint satisfaction algorithm to correct abnormal data points, and finally output a visual prediction report including the predicted value of the bone callus growth rate per week within the next 8 weeks, the mechanical property recovery curve, and the compliance probability.

[0042] Optionally, the dynamic parameter adjustment module includes:

[0043] Multi-source data input: Receive the stage rehabilitation effect matrix from the rehabilitation effect prediction module and the temperature and humidity data collected by the environmental sensor, extract the predicted value of the bone callus growth rate, the mechanical property recovery curve, the environmental temperature compensation coefficient, and the humidity attenuation factor contained therein, and align the rehabilitation effect prediction data and the environmental parameters in time series through the time axis synchronization algorithm to generate a fusion input data set including the biomechanical response lag time and the environmental interference sensitivity index;

[0044] Environment-physiology coupling modeling: Based on the fusion input data set, construct a non-linear mapping model between the temperature and humidity parameters and the bone metabolism rate, calculate the temperature compensation coefficient and the humidity attenuation factor by using the bivariate polynomial regression algorithm, and generate a theoretical rehabilitation rate benchmark curve after correcting the environmental parameters;

[0045] Adaptive particle swarm initialization: The solution space dimension of the particle swarm optimization algorithm is defined as a two-dimensional vector of [training intensity parameter, nutritional supplement dosage], where the training intensity parameter includes the electromagnetic pulse frequency and the damping coefficient, and the nutritional supplement dosage includes the calcium-phosphorus supplement gradient and the vitamin D intake. The size of the initialized particle swarm is 200, and the position vector of each particle corresponds to a set of adjustable parameter combinations;

[0046] Construction of multi-objective fitness function: A three-objective evaluation system including rehabilitation progress deviation, physiological safety index, and environmental adaptability was established. The rehabilitation progress deviation was calculated as the root mean square error between the predicted callus volume ratio and the theoretical benchmark curve. The physiological safety index was determined by the joint torque safety threshold (not exceeding 80% of the pain trigger point) and the blood calcium concentration range (2.1-2.6mmol / L). The environmental adaptability calculated the dynamic safety margin based on the energy consumption rate after temperature and humidity compensation.

[0047] Particle swarm iterative optimization: A dynamic inertia weight adjustment strategy is used to update particle velocity and position. The inertia weight decreases linearly from 0.9 to 0.4. The individual learning factor and group learning factor are set to 1.8 and 2.2, respectively. The non-dominated sorting level and crowding distance of each particle are calculated in each iteration, and the feasible solutions that meet the joint torque constraints and blood calcium concentration constraints are retained in the Pareto front solution set.

[0048] Multi-objective decision-making: Select optimization strategies based on rehabilitation stage identifiers, prioritize minimizing physiological safety risks during the inflammatory period, and focus on optimizing the deviation of rehabilitation progress during the repair period. The entropy-weighted TOPSIS method is used to select the optimal parameter combination from the Pareto solution set, and output the electromagnetic pulse frequency adjustment amount (±0.3Hz), damping coefficient correction value (±0.25N·s / m), and calcium-phosphorus supplement gradient update amount (±50mg / day).

[0049] Dynamic parameter encapsulation: The optimized training intensity parameters and nutritional supplement dosages are divided into time windows to generate a dynamic training parameter package that includes the morning high-intensity training segment (06:00-08:00), the afternoon nutritional supplement window (12:00-14:00) and the evening rehabilitation maintenance period (18:00-20:00). The package is encapsulated using the Protobuf protocol and transmitted to the full-cycle management module.

[0050] Optionally, the full cycle management module includes:

[0051] Multi-source data integration: Receive the dynamic training parameter package from the dynamic parameter adjustment module, the stage rehabilitation effect matrix from the rehabilitation effect prediction module, and the microstructural damage model from the three-dimensional pathological modeling module. Extract the electromagnetic pulse frequency adjustment amount, calcium and phosphorus supplementation gradient update amount, predicted value of callus volume ratio, and pore connectivity index. Align the discrete parameters to the unified time axis through the timestamp alignment algorithm to generate a full-dimensional rehabilitation time series dataset including training intensity - nutritional supplementation - repair progress;

[0052] Visualized rehabilitation calendar generation: Map the full-dimensional rehabilitation time series dataset to the augmented reality interface and construct a four-dimensional visualized view based on the three-dimensional dynamic bone model;

[0053] Data fingerprint generation: Structurally process the multi-dimensional rehabilitation requirement parameters, personalized training parameter set, and stage rehabilitation effect matrix for each rehabilitation cycle. Use the SHA-3-256 hashing algorithm to calculate the data fingerprint layer by layer. First, generate the leaf node hash values of the original data of each module, and then gradually aggregate them through the Merkle tree structure to generate the global data fingerprint;

[0054] Blockchain evidence storage: Write the global data fingerprint into the medical alliance chain network and perform distributed storage using the Practical Byzantine Fault Tolerance (PBFT) consensus mechanism.

[0055] Advantages of the present invention:

[0056] In the present invention, through the generation of multi-modal data fusion and personalized training schemes, customized rehabilitation management services can be provided for each patient. Specifically, the system receives data from three-dimensional pathological modeling, real-time physiological monitoring, and environmental sensors, and combines information from multiple aspects such as mechanical loading, nutritional supplementation, and psychological intervention. Through the multi-source data fusion algorithm, a dynamic training scheme including biomechanical constraints and physiological states is generated. The fuzzy PID control algorithm, model predictive control algorithm, and reinforcement learning technology are used to further optimize and adjust the rehabilitation scheme to ensure that the training intensity, nutritional supplementation amount, and psychological intervention mode required by the patient at different rehabilitation stages can adapt to their physiological and psychological needs, greatly improving the rehabilitation effect and safety.

[0057] In the present invention, through the spatio-temporal data processing and prediction model based on deep learning, the patient's rehabilitation process is comprehensively monitored and predicted. By introducing the LSTM-GAN network and combining the bone metabolism kinetics and mechanical property recovery models, the callus growth, mechanical property recovery, and changes in related physiological indicators of the patient can be predicted in real time. This technology can provide accurate rehabilitation effect evaluation through multi-modal result fusion and the evidence theory algorithm, and further perform dynamic adjustment in aspects such as mechanical loading, nutritional supplementation, and psychological intervention.

[0058] The present invention utilizes the real-time collection of temperature and humidity sensors and environmental data, combines the influence of physiological characteristics on the patient's rehabilitation process, constructs an environment-physiological coupling model, and dynamically optimizes the training intensity and nutritional supplementation plan based on the particle swarm optimization algorithm. Through adaptive particle swarm initialization and multi-objective fitness functions, the system can customize the rehabilitation parameters according to different environments (temperature, humidity) and physiological states (blood calcium concentration, joint torque, etc.), thereby optimizing the patient's rehabilitation progress and physiological safety. This optimization process not only considers the individual differences of patients, but also adjusts the optimization objectives according to the rehabilitation stages (such as the inflammation stage, repair stage, etc.), provides targeted rehabilitation programs, effectively improves the rehabilitation efficiency and comfort of patients, and ensures that patients always maintain the best treatment plan during the dynamically changing rehabilitation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0060] Figure 1 It is a schematic diagram of the system flow of the embodiment of the present invention;

[0061] Figure 2 It is a schematic diagram of the three-dimensional pathological modeling module process of the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] The present invention will be described in detail below in conjunction with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0063] It should be pointed out that in the specification, it is mentioned that "an embodiment", "embodiment", "exemplary embodiment", "some embodiments", etc. indicate that the described embodiments may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. In addition, when combining embodiments to describe specific features, structures or characteristics, implementing such features, structures or characteristics in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0064] Generally, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or property in the singular sense, or can be used to describe a combination of features, structures, or properties in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but rather can alternatively, at least in part depending on the context, allow for the existence of other factors that are not necessarily explicitly described.

[0065] As Figure 1 - Figure 2 shown, an orthopedic full-cycle rehabilitation management system based on artificial intelligence includes a rehabilitation needs assessment module, a three-dimensional pathology modeling module, a multi-modal training generation module, a rehabilitation effect prediction module, a dynamic parameter adjustment module, and a full-cycle management module, where;

[0066] Rehabilitation needs assessment module: Collects patient bone density data, joint range of motion data, gait mechanics data, and psychological assessment data through multi-modal sensors, and outputs a multi-dimensional rehabilitation needs parameter set;

[0067] Three-dimensional pathology modeling module: Receives the multi-dimensional rehabilitation needs parameters, and generates a three-dimensional dynamic bone model including a stress distribution heat map and a microstructural damage model based on the reverse deduction algorithm of the bone mechanics conduction chain;

[0068] Multi-modal training generation module: According to the three-dimensional dynamic bone model and real-time physiological monitoring data, generates a personalized training parameter set including a mechanical loading plan, a nutritional supplement plan, and a psychological intervention plan through a transfer learning network;

[0069] Rehabilitation effect prediction module: Inputs the personalized training parameter set into an LSTM-GAN hybrid network, and outputs a stage rehabilitation effect matrix including predicted values of callus growth rate and mechanical property recovery curves;

[0070] Dynamic parameter adjustment module: Based on the temperature and humidity data collected by the environmental sensor and the stage rehabilitation effect matrix, dynamically corrects the training intensity parameter and the nutritional supplement dose through an adaptive particle swarm algorithm;

[0071] Full-cycle management module: Integrates all corrected parameters to generate a visual rehabilitation calendar, and stores the data fingerprints of each stage through the blockchain to form an immutable rehabilitation evidence chain.

[0072] The rehabilitation needs assessment module includes:

[0073] Bone density data acquisition: The affected bones of the patient are scanned in a spiral tomographic manner using a dual-energy X-ray absorptiometer to obtain bone mineral content distribution data. An improved Hounsfield value calibration algorithm is used to eliminate soft tissue interference, generating a three-dimensional bone density distribution map that includes the interface between cortical bone and cancellous bone, and outputting a coordinate set of bone density abnormal regions and intensity attenuation gradient parameters;

[0074] For the bone mineral content distribution data obtained by the dual-energy X-ray absorptiometer, an improved Hounsfield value calibration algorithm is used to eliminate soft tissue interference, generating a three-dimensional bone density distribution map, and setting the calibration coefficient as C Hounsfield , and its calculation formula is:

[0075] Bone Density(x,y,z)=C Hounsfield ·Raw Hounsfield Value(x,y,z);

[0076] where x, y, and z are three-dimensional space coordinates;

[0077] The output is a three-dimensional bone density distribution map that includes the interface between cortical bone and cancellous bone, a coordinate set of bone density abnormal regions {(x1,y1,z1),…,(x n ,y n ,z n )} and an intensity attenuation gradient parameter set G Decay ={g1,g2,…,g n}, where gi i represents the attenuation gradient of the i-th region;

[0078] Range of motion quantification: A millimeter-wave radar array is deployed around the patient's joint. The joint movement trajectory is captured through the reflected signal of the 60GHz band electromagnetic wave. The Doppler phase resolution technology is used to reconstruct the three-dimensional kinematic model, calculate the flexion angle, rotational angular velocity, and movement plane offset, and generate a range of motion feature vector that includes the maximum range of motion threshold and pain trigger point;

[0079] The joint movement trajectory is captured by the millimeter-wave radar array, and the three-dimensional kinematic modeling is performed using the Doppler phase resolution technology. The flexion angle θ flex , rotational angular velocity ω rot , and movement plane offset δ plane are calculated through the following formulas:

[0080]

[0081] where x(t), y(t) are the coordinates of the joint at time t, and x0, y0 are the reference points.

[0082] The output is the range of motion feature vector Ajoint = [θ flex , ω rot , δ plane , where the maximum range of motion threshold is θ max , and the pain trigger point is θ pain ;

[0083] Gait mechanics analysis: Embed a piezoelectric six - dimensional force sensor array in the patient's sole, continuously collect the ground reaction force data during the gait cycle at a sampling frequency of 500Hz. After removing the motion artifacts through the wavelet denoising algorithm, use the inverse dynamics model to calculate the lower limb joint moment distribution, and output a set of gait mechanics parameters including the center of pressure trajectory, the proportion of the stance phase time, and the left - right symmetry index;

[0084] Use the wavelet denoising algorithm to remove the artifact data in the gait cycle. The ground reaction force data is used to calculate the lower limb joint moment distribution through the inverse dynamics model. The calculation formula of the moment distribution τ(t) is:

[0085]

[0086] where F(t ′ ) is the ground reaction force, is the coordinate of the point of action;

[0087] The output is a set of gait mechanics parameters G gait = [P center , T support , S symmetry , which respectively represent the center of pressure trajectory P center , the proportion of the stance phase time T support , and the left - right symmetry index S symmetry ;

[0088] Mental state assessment: Continuously monitor the skin conductance response and heart rate variability through a smart bracelet, combine with the electronic anxiety - depression scale filled in daily, use the fuzzy logic algorithm to calculate the psychological stress index, and simultaneously use a micro - near - infrared spectrometer to detect the blood oxygen level of the prefrontal cortex to construct a psychological assessment matrix including the emotional stability score and the rehabilitation confidence index;

[0089] The skin conductance response and heart rate variability are used to calculate the psychological stress index I stress by the fuzzy logic algorithm,

[0090] I stress = f fuzzy (HRV, EDA);

[0091] where HRV is the heart rate variability and EDA is the skin conductance response.

[0092] Meanwhile, the blood oxygen level of the prefrontal cortex was monitored by a near-infrared spectrometer, and the emotional stability score S was calculated by combining the anxiety-depression scale. emotion , The formula is:

[0093] S emotion = g NIRS (O2, HRV);

[0094] Where O2 is the blood oxygen level, and the output is the psychological assessment matrix M psychology = [I stress , S emotion ;

[0095] Multi-dimensional parameter fusion: The three-dimensional bone density distribution map, joint mobility feature vector, gait mechanics parameter set, and psychological assessment matrix are input into the feature-level fusion network for time-axis alignment and dimension normalization processing, and the weight coefficients of each parameter are assigned through the attention mechanism. Finally, a multi-dimensional rehabilitation demand parameter set containing biomechanical demand levels, psychological intervention priorities, and rehabilitation stage identifiers is generated. Among them, the bone density abnormal area coordinate set is directly mapped to the mechanical loading target area coordinates, the joint mobility feature vector is used to calculate the initial training intensity benchmark value, and the gait symmetry index and psychological stress index jointly correct the predicted value of rehabilitation program compliance.

[0096] Input the bone density distribution map, joint mobility feature vector, gait mechanics parameter set, and psychological assessment matrix into the feature-level fusion network. After time-axis alignment and dimension normalization, calculate the weight coefficients W bio and W psych through the attention mechanism, and finally generate a multi-dimensional rehabilitation demand parameter set D rehab , expressed as:

[0097] D rehab = W bio · [Bone Density Features, Joint Mobility Features, Gait Mechanics] + W psych · [Psychological Stress, Emotion Stability];

[0098] Where W bio and W psych are the weight coefficients of biomechanical demand and psychological intervention priority respectively, and D rehab contains biomechanical demand levels, psychological intervention priorities, and rehabilitation stage identifiers.

[0099] The three-dimensional pathological modeling module includes:

[0100] Multi-dimensional data analysis: Receive the multi-dimensional rehabilitation requirement parameter set from the rehabilitation requirement assessment module, extract the coordinate set of abnormal bone density regions, the joint range of motion feature vector, the gait mechanics parameter set, and the psychological assessment matrix. Align the bone density distribution map with the mechanical loading target area coordinates in three-dimensional space through the spatial coordinate system registration algorithm. Use the cubic spline interpolation method to fill the data acquisition blind area and generate a spatio-temporally unified multi-modal bone data set;

[0101] After receiving the multi-dimensional rehabilitation requirement parameter set from the rehabilitation requirement assessment module, align the bone density distribution map with the mechanical loading target area coordinates in three-dimensional space through the spatial coordinate system registration algorithm. The specific registration formula is as follows:

[0102]

[0103] Among them, T bone is the registration matrix, which represents mapping the spatial coordinates (x bone , y bone , z bone ) of the bone density distribution map to the coordinate system (x target , y target , z target ) of the loading target area. a, b, c,... are the coefficients in the transformation matrix, and x0, y0, z0 are the translation parameters. Use the cubic spline interpolation method to fill the data acquisition blind area. The interpolation calculation formula is:

[0104]

[0105] Among them, f(x0) and f(x1) are the data point values, x0, x1 are adjacent data points, x is the target interpolation point, and the output is the spatio-temporally unified multi-modal bone data set D modal , which contains information such as the bone density distribution map, the mechanical loading target area coordinates, and the joint range of motion characteristics;

[0106] Reverse construction of the mechanical conduction chain: Based on the lower limb biomechanical conduction principle, taking the flexion angle in the joint range of motion feature vector as the boundary condition, combined with the center of pressure trajectory data in the gait mechanics parameter set, use the finite element reverse solution method to deduce the bone stress conduction path, and establish a mechanical conduction chain topological network including the principal stress direction vector and the energy dissipation coefficient. The intensity attenuation gradient parameter of the bone density distribution map is used to correct the material anisotropy parameter;

[0107] Based on the lower limb biomechanical conduction principle, combined with the flexion angle θ flex in the joint range of motion feature vector and the center of pressure trajectory data P center in the gait mechanics parameter set, deduce the bone stress conduction path through the finite element reverse solution method. The stress conduction formula of the mechanical conduction chain is:

[0108]

[0109] Among them, E(x) is the elastic modulus of the material, is the displacement gradient, Ω is the problem area, and σ stress is the stress field. The generated mechanical conduction chain topological network includes the principal stress direction vector S and the energy dissipation coefficient η;

[0110]

[0111] Among them, F node (i) is the load of the i-th node, and L(i) is the length of this node;

[0112] Dynamic stress field reconstruction: Input the mechanical conduction chain topological network into the improved U-Net++. Extract multi-scale stress features through cascaded dilated convolutional layers, and combine the electromyogram signals in the real-time physiological monitoring data to generate a dynamic stress distribution heat map that includes the peak stress area, the stress gradient change rate, and the fatigue accumulation effect. The time resolution of this heat map reaches 0.1 second / frame;

[0113] Stress distribution heat map: The dynamic stress field is reconstructed through the improved U-Net++. Use cascaded dilated convolutional layers to extract multi-scale stress features, and combine the real-time electromyogram signal data EMG(t). The generation formula of the dynamic stress distribution heat map H stress (t) is:

[0114] H stress (t) = U-Net++(σ stress , EMG(t));

[0115] Among them, H stress (t) is the stress distribution of the heat map at time t, and σ stress is the current stress field;

[0116] The output includes the peak stress area P peak 、the stress gradient change rate G stress (t), and the fatigue accumulation effect F fatigue (t);

[0117]

[0118] Microdamage coupling modeling: Based on the cancellous bone porosity data in the three-dimensional bone density distribution map, use the fractal dimension algorithm to calculate the microstructural damage distribution. Combine the peak stress area coordinates of the dynamic stress distribution heat map, and simulate the microcrack growth trajectory through the crack propagation finite element method to construct a three-dimensional microstructural damage model that includes the pore connectivity index and the trabecular bone fracture risk value;

[0119] Use the fractal dimension algorithm to calculate the cancellous bone porosity φtrabecular For the data, the calculation formula for the distribution of microstructural damage is as follows:

[0120]

[0121] The growth trajectory of microcracks is simulated by the finite element method of crack propagation, and the simulation formula is:

[0122]

[0123] Among them, K IC is the fracture toughness of the material, Y is the geometric factor, and a is the size of the crack;

[0124] Output the three-dimensional model M of microstructural damage damage , including the pore connectivity index C porosity and the trabecular bone fracture risk value R fracture ;

[0125] Multi-physics field fusion: Perform spatial registration on the thermogram of dynamic stress distribution and the three-dimensional model of microstructural damage, introduce the temperature and humidity data collected by environmental sensors as boundary conditions, and calculate the influence factor of temperature gradient on the bone remodeling rate through the thermal-mechanical coupling algorithm to generate a three-dimensional dynamic bone model that integrates biomechanical characteristics, microstructural state, and environmental parameters;

[0126] Thermal-mechanical coupling algorithm: After performing spatial registration on the thermogram of dynamic stress distribution and the three-dimensional model of microstructural damage, calculate the temperature gradient of the influence factor β on the bone remodeling rate remodeling , and its calculation formula is:

[0127]

[0128] Among them, α thermal is the thermal diffusivity, is the temperature gradient, and Ω is the bone region.

[0129] The output three-dimensional dynamic bone model M bone integrates biomechanical characteristics, microstructural state, and environmental parameters;

[0130] Model dynamic update: According to the real-time rehabilitation data feedback by the full-cycle management module, perform incremental training on the three-dimensional dynamic bone model. If it is detected that the change in the coordinate set of the bone density abnormal area exceeds 5% or the offset of the joint range of motion feature vector exceeds the preset threshold, trigger the re-deduction of the mechanical conduction chain topology network, update the prediction parameters of the stress distribution thermogram, and the update trigger condition is:

[0131]

[0132] Among them, ΔBoneDensity is the bone density change, and θ threshold is the offset threshold of the joint range of motion;

[0133] Pathological model verification: The generated three-dimensional dynamic bone model is registered and compared with the postoperative CT images in the orthopedic PACS system. The Hausdorff distance algorithm is used to calculate the matching degree between the predicted contour of the model and the real bone morphology. When the error exceeds 1.2 mm, the parameter calibration process is automatically started, and a verification report containing a confidence score is output to the multi-modal training generation module.

[0134] Hausdorff distance algorithm: The Hausdorff distance algorithm is used to register and compare the generated three-dimensional dynamic bone model with the postoperative CT images in the orthopedic PACS system. The Hausdorff distance calculation formula is:

[0135]

[0136] where A and B are two sets of point sets, and ∥x - y∥ is the distance between points.

[0137] If the error exceeds 1.2 mm, the parameter calibration process is started, and a verification report containing the confidence score C confidence is output.

[0138] The multi-modal training generation module includes:

[0139] Multi-source data fusion: Receive the three-dimensional dynamic bone model and real-time physiological monitoring data from the three-dimensional pathological modeling module, extract the peak stress area coordinates of the dynamic stress distribution heat map, the trabecular fracture risk value of the microstructural damage model, the environmental sensor temperature and humidity data, and the patient's real-time blood oxygen saturation and electromyogram signals. Through the spatio-temporal alignment algorithm, the mechanical loading target area coordinates are spatially matched with the micro-damage area to generate a fusion data set containing biomechanical constraint conditions and physiological state parameters;

[0140] Spatio-temporal alignment algorithm: Receive the three-dimensional dynamic bone model and real-time physiological monitoring data from the three-dimensional pathological modeling module, perform spatio-temporal alignment, and generate a fusion data set containing biomechanical constraint conditions and physiological state parameters. Through the spatio-temporal alignment algorithm, the mechanical loading target area coordinates P load are spatially matched with the micro-damage area coordinates P damage , and the formula is:

[0141] P aligned = T alignment (P load , P damage );

[0142] where, T alignment is the spatio-temporal alignment transformation function, and Paligned is the fused spatial coordinates, including biomechanical constraints (such as stress gradient, damage risk) and physiological state parameters (such as blood oxygen saturation, electromyogram signal). The output fused dataset D fusion includes the coordinates of the mechanical loading target area, the fracture risk value of the microstructural damage area, real-time physiological monitoring data (blood oxygen, electromyogram signal, etc.) and environmental sensor data;

[0143] Mechanical loading scheme generation: Based on the coordinates of the peak stress area in the fused dataset, use the inverse dynamics model to calculate the activation threshold of the target muscle group, and combine the stress gradient change rate data of the three-dimensional dynamic bone model to generate variable impedance electromagnetic damping parameters through the fuzzy PID control algorithm;

[0144] Specifically include: Determine the initial load value according to the intensity attenuation gradient of the abnormal bone density area, set the safety threshold based on the pain trigger point in the joint range of motion feature vector, and output a mechanical loading scheme including an electromagnetic pulse sequence with a frequency of 0.5 - 5 Hz, a damping coefficient in the range of 0.2 - 1.8 N·s / m, and an action duration of 15 - 45 minutes;

[0145] Inverse dynamics model calculation: Based on the coordinates P of the peak stress area in the fused dataset peak , calculate the activation threshold T of the target muscle group through the inverse dynamics model activation , expressed as:

[0146]

[0147] where f is the inverse dynamics model function, is the stress gradient change rate, representing the distribution of stress in the bone;

[0148] Fuzzy PID control algorithm to generate damping parameters: Use the fuzzy PID control algorithm to generate the variable impedance electromagnetic damping parameter Z of the electromagnetic pulse damping , and the formula is:

[0149]

[0150] where e(t) is the current error (such as the difference between the target and the actual stress), K p , K i , K d are the proportional, integral, and differential gains of the fuzzy PID controller, and Z damping is the generated damping coefficient, representing the force of the electromagnetic pulse on the joint;

[0151] The output mechanical loading scheme S loading includes a frequency f EM in the range (0.5 - 5 Hz), a damping coefficient Z dampingRange (0.2 - 1.8 N·s / m) and action duration t duration (15 - 45 minutes);

[0152] S loading =(f EM , Z damping , t duration );

[0153] Nutritional supplement optimization: Analyze the trabecular bone porosity data of the three - dimensional bone density distribution map, combine the serum calcium, phosphorus concentration and vitamin D level in the real - time metabolism monitoring data, construct a differential equation of bone metabolism kinetics, and calculate the daily nutritional supplement dose using a model predictive control algorithm;

[0154] Specifically, determine the calcium and phosphorus supplement gradient according to the energy dissipation coefficient of the stress distribution heat map, adjust the collagen intake based on the pore connectivity index of the microstructure damage model, and generate a dynamic nutrition plan including supplement type, administration time window and dose curve;

[0155] Analyze the trabecular bone porosity data φ trabecular and the serum calcium and phosphorus concentration C calcium , C phosphate , and construct a differential equation of bone metabolism kinetics:

[0156]

[0157] where M bone represents bone mineral mass, α1, α2 are the efficacy coefficients of nutritional supplements, and β1, β2 are the related parameters of bone resorption;

[0158] Use the model predictive control algorithm to calculate the daily nutritional supplement dose D nutrition , and optimize the intake of calcium, phosphorus and collagen. The calculation formula is as follows:

[0159] D nutrition = MPC(M bone , φ trabecular , C calcium , C phosphate );

[0160] where D nutrition is the dynamic nutritional supplement dose, and M bone , φ trabecular , C calcium , C phosphate are input parameters.

[0161] The output dynamic nutrition plan S nutrition includes supplement type, administration time window and dose curve, specifically:

[0162] S nutrition=(Supplements,t window ,D dose );

[0163] Among them, D dose represents the daily dose distribution;

[0164] Psychological intervention matching: Input the emotional stability score of the psychological assessment matrix into the pre-trained psychological state classifier, construct a virtual reality scene decision tree based on the reinforcement learning algorithm, and select the corresponding intervention mode according to the rehabilitation stage identifier;

[0165] When the anxiety index ≥ 7 points, load the forest meditation scene. When the rehabilitation confidence index < 5 points, activate the social incentive module, and output a psychological intervention parameter package including the scene type, exposure duration (10 - 30 minutes), and interaction frequency (3 - 10 times / day);

[0166] Psychological state classifier: Input the emotional stability score S emotion of the psychological assessment matrix into the pre-trained psychological state classifier, and the classification results are the anxiety index A anxiety and the rehabilitation confidence index C confidence , use the reinforcement learning algorithm to generate a decision tree, and select the corresponding intervention mode according to the rehabilitation stage identifier C phase ;

[0167] If the anxiety index A anxiety ≥ 7 or the rehabilitation confidence index C confidence < 5, then load the corresponding virtual reality scene;

[0168] Reinforcement learning algorithm to generate intervention parameters: The intervention parameter package P intervention generated by the reinforcement learning algorithm includes the scene type S scene , exposure duration t exposure and interaction frequency f interaction , specifically:

[0169] P intervention =(S scene ,t exposure ,f interaction );

[0170] Among them, S scene may be "forest meditation" or "social incentive", t exposure is the exposure duration of the virtual reality scene (10 - 30 minutes), and f interaction is the interaction frequency (3 - 10 times / day).

[0171] The multi-modal training generation module also includes:

[0172] Transfer learning parameter adjustment: Establish a pre-trained network containing 100,000 orthopedic rehabilitation cases. Input the mechanical loading scheme, nutritional supplement scheme, and psychological intervention parameter package into the feature mapping layer. Align the distribution differences between the target patient and the source domain data through the domain adaptation algorithm. Specifically, use the Gradient Reversal Layer (GRL) to process the individual physiological feature offset. Finally, output a training parameter set containing personalized impedance curves, nutritional supplement gradient functions, and VR scene trigger conditions;

[0173] Multi-modal parameter collaborative optimization: Construct a three-objective optimization model containing mechanical loading intensity, nutritional supplement dosage, and psychological intervention effect. Use the NSGA-II algorithm to solve the Pareto optimal solution set. The constraint conditions include: the joint torque does not exceed 80% of the pain trigger point threshold, the blood calcium concentration is maintained in the range of 2.1 - 2.6 mmol / L, and the fluctuation range of the psychological stress index ≤ ±15%. Output the time collaboration matrix of each scheme parameter;

[0174] Dynamic parameter encapsulation: Synchronize the optimized mechanical loading electromagnetic pulse sequence, nutritional supplement gradient function, and virtual reality scene trigger conditions on the time axis. Generate a personalized training parameter set containing the training intensity (0 - 100%) at each time period within a 24-hour cycle, nutritional agent types (I - IV), and psychological intervention modes (A - D). The parameter set is encapsulated in JSON-LD format and transmitted to the rehabilitation effect prediction module after attaching a data verification code.

[0175] The rehabilitation effect prediction module includes:

[0176] Spatio-temporal data preprocessing: Receive the personalized training parameter set from the multi-modal training generation module. Parse the electromagnetic pulse sequence timestamps, nutritional supplement gradient functions, and VR scene trigger conditions contained therein. Convert the discrete parameter sequence into equally spaced time step data through the sliding window algorithm. Use the Z-score normalization method to perform feature encoding on the mechanical loading intensity, nutritional agent types, and psychological intervention modes, and generate a 128-dimensional time series input vector with a unified dimension;

[0177] Sliding window algorithm and Z-score normalization: Perform spatio-temporal alignment and normalization processing on the discrete data in the personalized training parameter set from the multi-modal training generation module. Convert the unequally spaced electromagnetic pulse sequence timestamps, nutritional supplement gradient functions, and VR scene trigger conditions into equally spaced time step data through the sliding window algorithm:

[0178]

[0179] Among them, S(t) is the original training parameter sequence, and Δt is the set time step. It is the processed equidistant time-step data. Then, each parameter is feature-encoded by the Z-score normalization method, and the normalization formula is:

[0180]

[0181] where μ is the mean of the training parameters, σ is the standard deviation, and Z(t) is the normalized feature-encoding vector. Finally, a 128-dimensional time series input vector is generated, denoted as:

[0182] X(t) = [Z1(t), Z2(t), …, Z 128 (t)];

[0183] LSTM-GAN Network Construction: Build a hybrid neural network architecture with dual-channel inputs. The generator network consists of 4 layers of bidirectional LSTM, with each layer containing 256 memory units. It receives the encoded training parameter sequence and attaches the electromyogram signal spectral features in the real-time physiological monitoring data. The discriminator network uses a 3-layer temporal convolutional neural network. The real rehabilitation effect data and the generated prediction data are input for adversarial training, and a gated attention mechanism is set at the output layer to focus on key time nodes;

[0184] Generator Network: The generator consists of 4 layers of bidirectional LSTM networks, with each layer containing 256 memory units. It receives the encoded training parameter sequence X(t) and the electromyogram signal spectral features M(t) in the real-time physiological monitoring data and inputs them into the bidirectional LSTM network:

[0185] H gen = BiLSTM(X(t), M(t));

[0186] where H gen is the hidden state of the generator network;

[0187] Discriminator Network: The discriminator network uses a 3-layer temporal convolutional neural network. The real rehabilitation effect data Y real (t) and the generated prediction data Y gen (t) are used for adversarial training:

[0188] Y gen (t) = TCN(H gen );

[0189] The output of the discriminator network is a probability value, indicating whether the predicted data is real data;

[0190] Gated Attention Mechanism: At the output layer, a gated attention mechanism is set to focus on the key time node t k , and the prediction accuracy is optimized through attention weighting:

[0191] A(tk ) = Attention(Y gen (t));

[0192] Bone callus growth kinetics modeling: Embed the bone metabolism kinetics equation in the hidden state layer of the generator network, calculate the calcium and phosphorus deposition rate according to the nutrient supplementation gradient function, combine the electromagnetic pulse frequency and damping coefficient in the mechanical loading parameters, and simulate the three-dimensional growth process of the bone callus through a differential equation solver, and output time-varying prediction parameters including the density of new trabecular bone, mineralization rate, and the proportion of bone callus volume;

[0193] Bone metabolism kinetics equation: Embed the bone metabolism kinetics equation in the hidden state layer of the generator network and calculate the calcium and phosphorus deposition rate R CaP (t) as a key parameter for bone callus growth, and a differential equation solver is used to simulate the three-dimensional growth process of the bone callus:

[0194]

[0195] where V bone is the bone callus volume, and f loading (t) is the influencing factor of mechanical loading.

[0196] Output the following time-varying prediction parameters:

[0197] Density of new trabecular bone (g / cm^3): ρ trab (t);

[0198] Mineralization rate (μm / day):

[0199] Proportion of bone callus volume (%): θ bone (t);

[0200] Prediction of mechanical property recovery: Input the stress distribution heat map of the three-dimensional dynamic bone model into the auxiliary feature channel of the discriminator network, calculate the bone stiffness recovery coefficient using the strain energy density integration method, combine the pore connectivity index decay curve of the microstructural damage model, and map the mechanical property recovery trajectory through a radial basis function neural network, and output a time series of mechanical recovery indicators including the maximum load capacity, elastic modulus, and fatigue life;

[0201] Stress recovery coefficient and strain energy density: Input the stress distribution heat map σ stress (t) of the three-dimensional dynamic bone model into the discriminator network, and calculate the bone stiffness recovery coefficient K stiffness (t) using the strain energy density integration method, expressed as:

[0202] K stiffness (t) = ∫ V σ stress (t)dV;

[0203] Microstructure damage attenuation: Combining the pore connectivity index decay curve C damage in the microstructure damage model, calculate the mechanical recovery trajectory through a radial basis function neural network:

[0204] L restoration (t) = RBF(K stiffness (t), C damage (t));

[0205] The output mechanical recovery indicators include:

[0206] Maximum load capacity (N): F max (t);

[0207] Elastic modulus (GPa): E(t);

[0208] Fatigue life (cycles): N fatigue (t);

[0209] Multi-modal result fusion: Align the time-varying prediction parameters with the time series of mechanical recovery indicators in space and time, use the evidence theory fusion algorithm to calculate the confidence weights of each indicator, construct a third-order rehabilitation effect tensor containing the time dimension, space dimension, and indicator dimension, and extract key features through tensor decomposition to generate a stage rehabilitation effect matrix;

[0210] Evidence theory fusion algorithm: Align the time-varying prediction parameters and mechanical recovery indicators in space and time, and use the evidence theory fusion algorithm to calculate the confidence weights of each indicator:

[0211]

[0212] Construct a third-order rehabilitation effect tensor T rehab (t), including the time dimension (0 - 12 weeks), space dimension (bone partition), and indicator dimension (biological / mechanical / metabolic);

[0213] T rehab (t) = Tensor(W(t), t);

[0214] Post-processing of prediction results: Conduct a medical rationality check on the generated stage rehabilitation effect matrix. When it is detected that the mutation of the bone callus volume ratio exceeds 20% / week or the recovery rate of the elastic modulus exceeds the anatomical limit, start the constraint satisfaction algorithm to correct abnormal data points, and finally output a visual prediction report containing the predicted values of the weekly bone callus growth rate, mechanical property recovery curve, and compliance probability within the next 8 weeks.

[0215] The dynamic parameter adjustment module includes:

[0216] Multi-source data input: Receive the stage rehabilitation effect matrix from the rehabilitation effect prediction module and the temperature and humidity data collected by the environmental sensor, extract the callus growth rate prediction value, mechanical property recovery curve, environmental temperature compensation coefficient, and humidity attenuation factor contained therein, and perform time series alignment on the rehabilitation effect prediction data and environmental parameters through the time axis synchronization algorithm to generate a fusion input data set containing the biomechanical response lag time and environmental interference sensitivity index;

[0217] Environment-physiology coupling modeling: Based on the fusion input data set, construct a non-linear mapping model between temperature and humidity parameters and bone metabolism rate, and use the bivariate polynomial regression algorithm to calculate the temperature compensation coefficient and humidity attenuation factor to generate a theoretical rehabilitation rate benchmark curve after environmental parameter correction;

[0218] Non-linear mapping model: Based on the relationship between temperature, humidity and bone metabolism rate, use the bivariate polynomial regression algorithm to calculate the temperature compensation coefficient α(T) and humidity attenuation factor β(H):

[0219] Calculation formula for temperature compensation coefficient: α(T) = 0.02(T - 25) 2 + 0.8 (T is Celsius temperature);

[0220] Calculation formula for humidity attenuation factor: (H is relative humidity percentage);

[0221] The generated corrected theoretical rehabilitation rate benchmark curve is expressed as:

[0222] R adjusted (t) = α(T)·R baseline (t) + β(H)·R baseline (t);

[0223] Where, R adjusted (t) is the corrected rehabilitation rate, and R baseline (t) is the theoretical benchmark curve;

[0224] Adaptive particle swarm initialization: Define the solution space dimension of the particle swarm optimization algorithm as a two-dimensional vector of [training intensity parameter, nutritional supplement dose]. The training intensity parameter includes electromagnetic pulse frequency and damping coefficient, and the nutritional supplement dose includes calcium and phosphorus supplement gradient and vitamin D intake. Initialize the particle swarm size to 200, and the position vector of each particle corresponds to a set of adjustable parameter combinations;

[0225] Multi-objective Fitness Function Construction: Establish a three-objective evaluation system including the rehabilitation progress deviation degree, physiological safety index, and environmental fitness. Among them, the rehabilitation progress deviation degree is calculated as the root mean square error of the predicted callus volume ratio and the theoretical reference curve. The physiological safety index is jointly determined by the joint torque safety threshold (not exceeding 80% of the pain trigger point) and the blood calcium concentration range (2.1 - 2.6 mmol / L). The environmental fitness calculates the dynamic safety margin based on the energy consumption rate after temperature and humidity compensation;

[0226] Particle Swarm Iterative Optimization: Adopt a dynamic inertia weight adjustment strategy to update the particle velocity and position. The inertia weight linearly decreases from 0.9 to 0.4. The individual learning factor and the swarm learning factor are set to 1.8 and 2.2 respectively. Calculate the non-dominated sorting rank and crowding distance of each particle in each iteration, and retain the feasible solutions that satisfy the joint torque constraint and blood calcium concentration constraint in the Pareto front solution set;

[0227] Multi-objective Decision-making: Select the optimization strategy based on the rehabilitation stage identifier. Minimize the physiological safety risk preferentially in the inflammation period, and focus on optimizing the rehabilitation progress deviation degree in the repair period. Use the entropy weight TOPSIS method to select the optimal parameter combination from the Pareto solution set, and output the electromagnetic pulse frequency adjustment amount (±0.3 Hz), damping coefficient correction value (±0.25 N·s / m), and calcium and phosphorus supplement gradient update amount (±50 mg / day);

[0228] Dynamic Parameter Encapsulation: Divide the optimized training intensity parameters and nutritional supplement doses into time windows, generate a dynamic training parameter package including the morning high-intensity training segment (06:00 - 08:00), the afternoon nutritional supplement window (12:00 - 14:00), and the evening rehabilitation maintenance period (18:00 - 20:00), and encapsulate and transmit it to the full-cycle management module using the Protobuf protocol.

[0229] The full-cycle management module includes:

[0230] Multi-source Data Integration: Receive the dynamic training parameter package from the dynamic parameter adjustment module, the stage rehabilitation effect matrix from the rehabilitation effect prediction module, and the microstructural damage model from the three-dimensional pathological modeling module. Extract the electromagnetic pulse frequency adjustment amount, calcium and phosphorus supplement gradient update amount, predicted callus volume ratio, and pore connectivity index. Align the discrete parameters to the unified time axis through the timestamp alignment algorithm to generate a full-dimensional rehabilitation time series dataset including training intensity - nutritional supplement - repair progress;

[0231] Visual rehabilitation calendar generation: Map the full-dimensional rehabilitation time-series data set to the augmented reality interface, construct a four-dimensional visualization view (three-dimensional space + time dimension) based on the three-dimensional dynamic bone model, where the electromagnetic pulse frequency is displayed as a dynamic waveform diagram with an accuracy of 0.1 Hz, the calcium and phosphorus supplementation gradient is converted into a heat map covering the bone surface, and the bone callus growth rate is presented in the form of a growth animation. Embed the morning high-intensity training segment, afternoon nutrition window, and evening maintenance period time partitions divided based on Claim 6 to generate an interactive rehabilitation calendar that supports gesture control;

[0232] Data fingerprint generation: Structurally process the multi-dimensional rehabilitation requirement parameters, personalized training parameter sets, and stage rehabilitation effect matrices for each rehabilitation cycle. Use the SHA-3-256 hash algorithm to calculate the data fingerprints layer by layer. First, generate the leaf node hash values of the original data of each module, and then gradually aggregate them through the Merkle tree structure to generate the global data fingerprint. Each data block contains a triple of timestamp (accurate to milliseconds), device signature (ECDSA algorithm), and the hash value of the previous block;

[0233] Blockchain evidence storage: Write the global data fingerprint into the medical consortium chain network, and use the Practical Byzantine Fault Tolerance (PBFT) consensus mechanism for distributed storage. The blockchain nodes include hospital servers (40% weight), medical insurance institution nodes (30% weight), and patient mobile terminals (30% weight). Each data write requires more than 2 / 3 of the node verification signatures for validity. The evidence storage content includes the hash value of the rehabilitation plan execution record, abnormal event markers, and model correction logs.

[0234] This invention covers any substitutions, modifications, equivalent methods, and solutions made within the essence and scope of this invention. To enable the public to have a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments of this invention. However, those skilled in the art can fully understand this invention without these detailed descriptions. Additionally, to avoid unnecessary confusion to the essence of this invention, well-known methods, processes, flows, components, and circuits are not described in detail.

[0235] The above are only the preferred embodiments of this invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of this invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this invention.

Claims

1. An orthopedic full-cycle rehabilitation management system based on artificial intelligence, characterized in that, It includes a rehabilitation needs assessment module, a three-dimensional pathological modeling module, a multi-modal training generation module, a rehabilitation effect prediction module, a dynamic parameter adjustment module, and a full-cycle management module, where; Rehabilitation needs assessment module: Collects patients' bone density data, joint range of motion data, gait mechanics data, and psychological assessment data through multi-modal sensors, and outputs a multi-dimensional rehabilitation needs parameter set; Three-dimensional pathological modeling module: Receives the multi-dimensional rehabilitation needs parameters, and generates a three-dimensional dynamic bone model including a stress distribution heat map and a microstructural damage model based on the reverse deduction algorithm of the bone mechanics conduction chain; Multi-modal training generation module: According to the three-dimensional dynamic bone model and real-time physiological monitoring data, generates a personalized training parameter set including a mechanical loading plan, a nutritional supplement plan, and a psychological intervention plan through a transfer learning network; Rehabilitation effect prediction module: Inputs the personalized training parameter set into an LSTM-GAN hybrid network, and outputs a stage rehabilitation effect matrix including predicted values of callus growth rate and mechanical property recovery curves; Dynamic parameter adjustment module: Based on the temperature and humidity data collected by environmental sensors and the stage rehabilitation effect matrix, dynamically corrects the training intensity parameters and nutritional supplement doses through an adaptive particle swarm algorithm; Full-cycle management module: Integrates all corrected parameters to generate a visual rehabilitation calendar, and stores the data fingerprints of each stage through blockchain to form an immutable rehabilitation evidence chain.

2. The orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 1, wherein The rehabilitation needs assessment module includes: Bone density data collection: Performs spiral tomographic scanning on the affected bones of patients through a dual-energy X-ray absorptiometer to obtain bone mineral content distribution data, uses the Hounsfield value calibration algorithm to eliminate soft tissue interference, generates a three-dimensional bone density distribution map including the interface between cortical bone and cancellous bone, and outputs a coordinate set of bone density abnormal regions and intensity attenuation gradient parameters; Quantification of joint range of motion: Deploys a millimeter-wave radar array around the joints of patients, captures joint movement trajectories through the reflection signals of 60GHz band electromagnetic waves, reconstructs a three-dimensional kinematic model using Doppler phase resolution technology, calculates flexion angles, rotational angular velocities, and movement plane offsets, and generates a joint range of motion feature vector including maximum range of motion thresholds and pain trigger points; Gait mechanics analysis: Embeds a piezoelectric six-dimensional force sensor array in the soles of patients, continuously collects ground reaction force data during the gait cycle at a sampling frequency of 500Hz, eliminates motion artifacts through a wavelet denoising algorithm, and then calculates the lower limb joint torque distribution using an inverse dynamics model, and outputs a gait mechanics parameter set including the center of pressure trajectory, the proportion of stance phase time, and the left-right symmetry index; Psychological state assessment: Continuously monitors skin conductance responses and heart rate variability through a smart bracelet, combines the electronic anxiety-depression scale filled in daily, calculates the psychological stress index using a fuzzy logic algorithm, and simultaneously detects the blood oxygen level of the prefrontal cortex using a miniature near-infrared spectrometer to construct a psychological assessment matrix including emotional stability scores and rehabilitation confidence indices; Multi-dimensional parameter fusion: Input the three-dimensional bone density distribution map, joint range of motion feature vector, gait mechanics parameter set, and psychological assessment matrix into the feature-level fusion network for time-axis alignment and dimension normalization processing, and assign weight coefficients to each parameter through the attention mechanism, finally generating a multi-dimensional rehabilitation demand parameter set including biomechanical demand level, psychological intervention priority, and rehabilitation stage identifier.

3. The orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 2, wherein, The three-dimensional pathological modeling module includes: Multi-dimensional data parsing: Receive the multi-dimensional rehabilitation demand parameter set from the rehabilitation demand assessment module, extract the coordinate set of abnormal bone density regions, joint range of motion feature vector, gait mechanics parameter set, and psychological assessment matrix, perform three-dimensional spatial alignment of the bone density distribution map and the mechanical loading target area coordinates through the spatial coordinate system registration algorithm, and use the cubic spline interpolation method to fill the data acquisition blind area to generate a spatio-temporally unified multi-modal bone dataset; Inverse construction of the mechanical conduction chain: Based on the lower limb biomechanical conduction principle, with the flexion angle in the joint range of motion feature vector as the boundary condition, combined with the center of pressure trajectory data in the gait mechanics parameter set, use the finite element inverse solution method to deduce the bone stress conduction path and establish a mechanical conduction chain topological network including the principal stress direction vector and energy dissipation coefficient, where the intensity attenuation gradient parameter of the bone density distribution map is used to correct the material anisotropy parameter; Dynamic stress field reconstruction: Input the mechanical conduction chain topological network into the improved U-Net++ network, extract multi-scale stress features through cascaded dilated convolutional layers, and combine the electromyogram signals in the real-time physiological monitoring data to generate a dynamic stress distribution heat map including peak stress regions, stress gradient change rates, and fatigue accumulation effects; Micro-damage coupling modeling: Based on the cancellous bone porosity data of the three-dimensional bone density distribution map, use the fractal dimension algorithm to calculate the microstructural damage distribution, combine the peak stress region coordinates of the dynamic stress distribution heat map, and simulate the microcrack growth trajectory through the crack propagation finite element method to construct a three-dimensional microstructural damage model including pore connectivity index and trabecular bone fracture risk value; Multi-physical field fusion: Perform spatial registration on the dynamic stress distribution heat map and the three-dimensional microstructural damage model, introduce the temperature and humidity data collected by the environmental sensor as boundary conditions, and calculate the influence factor of the temperature gradient on the bone remodeling rate through the thermal-mechanical coupling algorithm to generate a three-dimensional dynamic bone model integrating biomechanical characteristics, microstructural state, and environmental parameters; Model dynamic update: Perform incremental training on the three-dimensional dynamic bone model according to the real-time rehabilitation data feedback by the full-cycle management module. If it is detected that the change in the coordinate set of abnormal bone density regions exceeds 5% or the offset of the joint range of motion feature vector exceeds the preset threshold, trigger the re-deduction of the mechanical conduction chain topological network and update the prediction parameters of the stress distribution heat map; Pathological model verification: The generated three-dimensional dynamic bone model is registered and compared with the postoperative CT images in the orthopedic PACS system. The Hausdorff distance algorithm is used to calculate the matching degree between the predicted contour of the model and the real bone morphology. When the error exceeds 1.2 mm, the parameter calibration process is automatically started, and a verification report including a confidence score is output to the multi-modal training generation module.

4. An orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 3, characterized in that, The multi-modal training generation module includes: Multi-source data fusion: Receive the three-dimensional dynamic bone model and real-time physiological monitoring data from the three-dimensional pathological modeling module, extract the peak stress area coordinates of the dynamic stress distribution heat map, the trabecular fracture risk value of the microstructural damage model, the temperature and humidity data of the environmental sensor, and the patient's real-time blood oxygen saturation and electromyogram signal. The mechanical loading target area coordinates and the micro-damage area are spatially matched through the spatio-temporal alignment algorithm to generate a fusion data set containing biomechanical constraint conditions and physiological state parameters; Mechanical loading scheme generation: Based on the peak stress area coordinates of the fusion data set, the inverse dynamics model is used to calculate the activation threshold of the target muscle group, and combined with the stress gradient change rate data of the three-dimensional dynamic bone model, the variable impedance electromagnetic damping parameter is generated through the fuzzy PID control algorithm; Nutritional supplement optimization: Analyze the trabecular porosity data of the three-dimensional bone density distribution map, combine the serum calcium and phosphorus concentrations and vitamin D levels in the real-time metabolism monitoring data, construct a differential equation of bone metabolism kinetics, and calculate the daily nutritional supplement dose using the model predictive control algorithm; Psychological intervention matching: Input the emotional stability score of the psychological assessment matrix into the pre-trained psychological state classifier, construct a virtual reality scene decision tree based on the reinforcement learning algorithm, and select the corresponding intervention mode according to the rehabilitation stage identifier.

5. The orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 4, characterized in that, The multi-modal training generation module further includes: Transfer learning parameter adjustment: Establish a pre-trained network containing 100,000 orthopedic rehabilitation cases, input the mechanical loading scheme, nutritional supplement scheme, and psychological intervention parameter package into the feature mapping layer, and align the distribution differences between the target patient and the source domain data through the domain adaptation algorithm. Finally, output a training parameter set containing personalized impedance curves, nutritional supplement gradient functions, and VR scene trigger conditions; Multi-modal parameter collaborative optimization: Construct a three-objective optimization model including mechanical loading intensity, nutritional supplement dose, and psychological intervention effect, and use the NSGA-II algorithm to solve the Pareto optimal solution set. The constraint conditions include: the joint torque does not exceed 80% of the pain trigger point threshold, the blood calcium concentration is maintained in the range of 2.1-2.6 mmol / L, and the psychological stress index fluctuation range ≤ ±15%. Output the time collaborative matrix of each scheme parameter; Dynamic parameter encapsulation: Synchronize the optimized mechanical loading electromagnetic pulse sequence, nutritional supplement gradient function, and virtual reality scene trigger conditions on the time axis to generate a personalized training parameter set containing the training intensity, nutritional agent type, and psychological intervention mode for each period within a 24-hour cycle. The parameter set is encapsulated in JSON-LD format and transmitted to the rehabilitation effect prediction module with an additional data verification code.

6. An orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 5, characterized in that, The rehabilitation effect prediction module includes: Spatio-temporal data preprocessing: Receive the personalized training parameter set from the multi-modal training generation module, parse the electromagnetic pulse sequence timestamps, nutrient supplement gradient functions, and VR scene trigger conditions contained therein, convert the discrete parameter sequence into equally spaced time-step data through a sliding window algorithm, and use the Z-score normalization method to perform feature encoding on the mechanical loading intensity, nutrient agent type, and psychological intervention mode to generate a 128-dimensional time series input vector with a unified dimension; LSTM-GAN network construction: Build a hybrid neural network architecture with dual-channel inputs. The generator network consists of 4 layers of bidirectional LSTMs, with each layer containing 256 memory units. It receives the encoded training parameter sequence and appends the electromyogram signal spectral features in the real-time physiological monitoring data. The discriminator network uses a 3-layer temporal convolutional neural network, and the real rehabilitation effect data and the generated prediction data are input for adversarial training, and a gated attention mechanism is set in the output layer; Callus growth kinetics modeling: Embed the bone metabolism kinetics equation in the hidden state layer of the generator network, calculate the calcium and phosphorus deposition rate according to the nutrient supplement gradient function, combine the electromagnetic pulse frequency and damping coefficient in the mechanical loading parameters, and simulate the three-dimensional growth process of the callus through a differential equation solver, and output the time-varying prediction parameters including the density of new trabeculae, mineralization rate, and the proportion of callus volume; Prediction of mechanical property recovery: Input the stress distribution heat map of the three-dimensional dynamic bone model into the auxiliary feature channel of the discriminator network, calculate the bone stiffness recovery coefficient using the strain energy density integration method, combine the pore connectivity index decay curve of the microstructural damage model, and map the mechanical property recovery trajectory through a radial basis function neural network, and output the time series of mechanical recovery indicators including the maximum load capacity, elastic modulus, and fatigue life; Multi-modal result fusion: Align the time-varying prediction parameters and the time series of mechanical recovery indicators spatiotemporally, use the evidence theory fusion algorithm to calculate the confidence weights of each indicator, construct a third-order rehabilitation effect tensor including the time dimension, space dimension, and indicator dimension, and extract key features through tensor decomposition to generate the stage rehabilitation effect matrix; Post-processing of prediction results: Perform medical rationality verification on the generated stage rehabilitation effect matrix. When it is detected that the mutation of the callus volume ratio exceeds 20% / week or the elastic modulus recovery rate exceeds the anatomical limit, start the constraint satisfaction algorithm to correct the abnormal data points, and finally output a visual prediction report including the predicted value of the callus growth rate per week within the next 8 weeks, the mechanical property recovery curve, and the compliance probability.

7. An orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 6, characterized in that, The dynamic parameter adjustment module includes: Multi-source data input: Receive the stage rehabilitation effect matrix from the rehabilitation effect prediction module and the temperature and humidity data collected by the environmental sensor, extract the predicted value of the callus growth rate, the mechanical property recovery curve, the environmental temperature compensation coefficient, and the humidity attenuation factor contained therein, and align the rehabilitation effect prediction data and the environmental parameters in time series through the time axis synchronization algorithm to generate a fusion input data set including the biomechanical response lag time and the environmental interference sensitivity index; Environmental-physiological coupling modeling: Based on the fused input data set, a nonlinear mapping model of temperature and humidity parameters and bone metabolism rate is constructed, and a bivariate polynomial regression algorithm is used to calculate the temperature compensation coefficient and humidity attenuation factor to generate a theoretical rehabilitation rate benchmark curve after environmental parameter correction; Adaptive particle swarm initialization: The solution space dimension of the particle swarm optimization algorithm is defined as a two-dimensional vector of [training intensity parameter, nutritional supplement dosage], where the training intensity parameter includes the electromagnetic pulse frequency and the damping coefficient, and the nutritional supplement dosage includes the calcium-phosphorus supplement gradient and the vitamin D intake. The size of the initialized particle swarm is 200, and the position vector of each particle corresponds to a set of adjustable parameter combinations; Construction of multi-objective fitness function: A three-objective evaluation system including rehabilitation progress deviation, physiological safety index, and environmental adaptability was established. The rehabilitation progress deviation was calculated as the root mean square error between the predicted callus volume ratio and the theoretical benchmark curve. The physiological safety index was determined by the joint torque safety threshold and the blood calcium concentration range. The environmental adaptability calculated the dynamic safety margin based on the energy consumption rate after temperature and humidity compensation. Particle swarm iterative optimization: A dynamic inertia weight adjustment strategy is used to update particle velocity and position. The inertia weight decreases linearly from 0.9 to 0.

4. The individual learning factor and group learning factor are set to 1.8 and 2.2, respectively. The non-dominated sorting level and crowding distance of each particle are calculated in each iteration, and the feasible solutions that meet the joint torque constraints and blood calcium concentration constraints are retained in the Pareto front solution set. Multi-objective decision-making: Select optimization strategies based on rehabilitation stage identifiers, prioritize minimizing physiological safety risks during the inflammatory period, focus on optimizing the rehabilitation progress deviation during the repair period, use the entropy weight TOPSIS method to select the optimal parameter combination from the Pareto solution set, and output the electromagnetic pulse frequency adjustment, damping coefficient correction value, and calcium-phosphorus supplementation gradient update; Dynamic parameter encapsulation: The optimized training intensity parameters and nutritional supplement dosage are divided into time windows to generate a dynamic training parameter package that includes the morning high-intensity training segment, the afternoon nutritional supplement window, and the evening rehabilitation maintenance period. The package is encapsulated using the Protobuf protocol and transmitted to the full-cycle management module.

8. An orthopedic full-cycle rehabilitation management system based on artificial intelligence according to claim 7, characterized in that, The full cycle management module includes: Multi-source data integration: Receive the dynamic training parameter package from the dynamic parameter adjustment module, the stage rehabilitation effect matrix from the rehabilitation effect prediction module, and the microstructure damage model from the three-dimensional pathology modeling module, extract the electromagnetic pulse frequency adjustment amount, the calcium-phosphorus supplementation gradient update amount, the predicted value of the callus volume ratio, and the pore connectivity index, and unify the discrete parameters on the time axis through the timestamp alignment algorithm to generate a full-dimensional rehabilitation time series data set including training intensity-nutritional supplementation-repair progress; Visual rehabilitation calendar generation: mapping the full-dimensional rehabilitation time series data set to an augmented reality interface, and constructing a four-dimensional visualization view based on a three-dimensional dynamic skeletal model; Data fingerprint generation: Structurally process the multi-dimensional rehabilitation requirement parameters, personalized training parameter sets, and stage rehabilitation effect matrices for each rehabilitation cycle. Use the SHA-3-256 hash algorithm to calculate the data fingerprint layer by layer. First, generate the leaf node hash values of the original data of each module, and then gradually aggregate them through the Merkle tree structure to generate the global data fingerprint; Blockchain evidence storage: Write the global data fingerprint into the medical consortium chain network and perform distributed storage using the Practical Byzantine Fault Tolerance (PBFT) consensus mechanism.

Images