Postoperative patient limb function recovery intelligent assessment and rehabilitation training path planning system

Abstract

The application discloses a postoperative patient limb function recovery intelligent evaluation and rehabilitation training path planning system, which comprises a multi-modal data acquisition module for acquiring postoperative limb movement data of a patient, and an intelligent evaluation module for calculating a limb function recovery index FR multidimensional precise evaluation based on the movement data; multi-modal sensor data (IMU, EMG, depth camera) are fused to cover whole-dimensional information such as kinematics and muscle function, the data integrity is improved by more than 60%, a dynamic model is used for tracking a recovery trend, and the evaluation precision is significantly improved compared with a traditional scheme. Personalized intelligent training: a personalized scheme containing 10-15 actions is generated through reinforcement learning, and a rehabilitation cycle is shortened by 23%; training parameters are dynamically adjusted in real time, and fatigue is warned, the fatigue occurrence rate is reduced from 65% to 39%, and safety and compliance are improved.

Description

Technical Field

[0001] This invention relates to the recovery of limb function in postoperative patients, specifically to an intelligent assessment and rehabilitation training pathway planning system for the recovery of limb function in postoperative patients. Background Technology

[0002] Postoperative limb function recovery is a crucial aspect of clinical rehabilitation. Traditional rehabilitation programs rely heavily on physician experience and face several technical limitations.

[0003] Insufficient accuracy in assessment: Existing methods mostly use a single sensor (such as an orthogonal sensor or muscle strength tester) or subjective scales, which can only capture single-point static data such as joint range of motion and muscle strength. They cannot integrate multi-dimensional information such as movement trajectory and muscle coordination control, resulting in a high rate of missed diagnoses (such as muscle compensation behavior after knee surgery). Furthermore, they lack dynamic correction over time, and the correlation between assessment results and actual functional improvement is only 0.52, making it difficult to guide precise intervention.

[0004] Homogeneous training programs: They rely on standardized combinations of movements (such as flexion and extension training with fixed frequency and intensity) without dynamic adjustment based on individual fatigue thresholds and differences in neural control. This results in 60% of patients experiencing fatigue injury or compensatory movements due to inappropriate intensity, extending the rehabilitation period by an average of more than 35 days. Furthermore, the lack of a real-time monitoring mechanism leads to a fatigue-related adverse event rate as high as 65%, and a risk identification delay of more than 15 minutes.

[0005] Limited integration of technology and clinical application: Analysis of muscle function remains at the level of superficial features of electromyography signals (such as root mean square value), and cannot analyze deep neuromuscular synergy patterns (such as flexor hyperactivity in hemiplegic patients), with an intervention window lag of 5-7 days; due to insufficient case data, initial assessment of new patients in small and medium-sized hospitals takes more than 2 hours, and the error rate in small sample scenarios reaches 30%, which limits the popularization of the technology.

[0006] Inefficient hardware and system collaboration: Traditional equipment is bulky (such as isokinetic muscle strength testers weighing >50kg), only supports use in fixed scenarios, cannot cover the needs of early postoperative bedside assessment, and the assessment data is disconnected from the training plan, requiring physicians to manually integrate the data. A single case plan takes 40 minutes to develop, and it is difficult to connect with the hospital information system in real time, resulting in low clinical efficiency.

[0007] To address the aforementioned issues, this technical solution utilizes multimodal data fusion (IMU, EMG, depth camera), intelligent algorithm modeling (reinforcement learning, transfer learning), and lightweight engineering design to construct an "assessment-planning-adjustment" closed-loop system. This system solves the problems of traditional rehabilitation, such as reliance on experience, low efficiency, and poor safety, thereby achieving precise, personalized, and intelligent postoperative rehabilitation. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned technologies and provide an intelligent assessment and rehabilitation training path planning system for postoperative patients' limb function recovery.

[0009] To address the aforementioned technical problems, the present invention provides a system for intelligent assessment and rehabilitation training pathway planning for postoperative limb function recovery in patients.

[0010] include:

[0011] A multimodal data acquisition module is used to acquire postoperative limb movement data of patients;

[0012] The intelligent assessment module calculates the Limb Function Recovery Index (FRI) based on the motion data. The calculation formula is as follows:

[0013]

[0014] in:

[0015] n represents the number of evaluation dimensions;

[0016] w i Let be the weight coefficient of the i-th dimension.

[0017] V i Let be the measured value of the i-th dimension;

[0018] V i,min and V i,max These are the minimum and maximum values ​​of the i-th dimension, respectively;

[0019] α i Adjust parameters for the dimension specificity index;

[0020] β i This is the time decay coefficient;

[0021] t represents the current time;

[0022] t i This is the reference time point for the i-th dimension.

[0023] As an improvement, the multimodal data acquisition module includes:

[0024] Inertial Measurement Unit (IMU) is used to collect the three-dimensional acceleration of the limbs a(t) =

[0025] [a x (t),a y (t),a z [(t)] and angular velocity ω(t)=[ω x (t),ω y (t),ω z (t)];

[0026] Surface electromyography (EMG) sensors are used to acquire muscle activity signals EMG(t) = [EMG1(t), EMG2(t), ..., EMG... m (t)];

[0027] A depth camera is used to acquire the limb movement trajectory P(t)=[x(t),y(t),z(t)];

[0028] Where t is the time variable and m is the number of electromyography (EMG) sensors.

[0029] As an improvement, the intelligent evaluation module further includes:

[0030] Kinematic feature extraction unit calculates joint range of motion (ROM). j :

[0031]

[0032] in:

[0033] j is the joint index;

[0034] θ j (t) represents the angle of joint j at time t;

[0035] θ j,rest Let be the resting angle of joint j;

[0036] [t1,t2] represents the motion period.

[0037] As an improvement, the intelligent evaluation module also includes:

[0038] The muscle synergy feature analysis unit calculates the muscle synergy coefficient C through nonnegative matrix factorization (NMF). ij :

[0039]

[0040] in:

[0041] This is an electromyography signal matrix;

[0042] For muscle synergy matrix;

[0043] This is the activation coefficient matrix;

[0044] ||·|| F It is the Frobenius norm;

[0045] ||·|| 1,2 It is a mixed norm;

[0046] λ1 and λ2 are regularization parameters;

[0047] r represents the number of collaborative modes.

[0048] As an improvement, the rehabilitation training path planning module generates personalized training plans based on reinforcement learning algorithms, and solves the following optimization problems:

[0049]

[0050] in:

[0051] π is the strategy function;

[0052] τ represents the state-action sequence;

[0053] γ is the discount factor;

[0054] R(s t ,a t ,s t+1 ) is from state s t Perform action a t Transition to state s t+1 The reward function;

[0055] T represents the upper limit of the time step.

[0056] As an improvement, the formula for calculating the reward function R is as follows:

[0057] R(s t ,a t ,s t+1 )=w1·ΔFRI+w2·IntensityPenalty+w3

[0058] SymmetryScore

[0059] in:

[0060] ΔFRI=FRI(s t+1 )-FRI(s t () represents the change in the functional recovery index;

[0061] This is a training intensity penalty item;

[0062] Scoring is based on limb symmetry;

[0063] I t Training intensity;

[0064] I opt To achieve the optimal training intensity;

[0065] σ is the penalty distribution parameter;

[0066] L t and R t These are the feature vectors of left and right limb movement, respectively;

[0067] w1, w2, and w3 are weighting coefficients.

[0068] As an improvement, the system further includes:

[0069] The dynamic adjustment module updates the training scheme based on real-time monitoring data and estimates the patient's state x using Kalman filtering. t :

[0070]

[0071] in:

[0072] x t This is the system state vector;

[0073] z t For observation vectors;

[0074] A is the state transition matrix;

[0075] B is the control input matrix;

[0076] H is the observation matrix;

[0077] w t and v t These are process noise and observation noise, respectively.

[0078] As an improvement, the dynamic adjustment module uses a particle filter algorithm to optimize the training parameters θ, which is achieved through the following iterative process:

[0079]

[0080] in:

[0081] i is the particle index;

[0082] N is the number of particles;

[0083] q represents the proposal distribution;

[0084] p is the probability density function;

[0085] For particle weights;

[0086] z 1:t This is the observation sequence up to time t.

[0087] As an improvement, the system also includes:

[0088] The fatigue detection module calculates the muscle fatigue index FI using wavelet transform.

[0089]

[0090] in:

[0091] W(f,t) represents the wavelet transform coefficients of the electromyographic signal;

[0092] [f1,f2] represents the frequency range;

[0093] t1 and t2 are the start and end times of the motion, respectively.

[0094] As an improvement, the system employs a transfer learning framework to achieve cross-patient knowledge transfer, optimized through the following loss function:

[0095]

[0096] in:

[0097]

[0098] Ω(θ) is the model complexity regularization term;

[0099] N and M are the number of samples in the source domain and the target domain, respectively;

[0100] The true label for sample i belonging to category c;

[0101] To predict probabilities;

[0102] D stands for the domain discriminator;

[0103] φ(x) is the feature extractor;

[0104] λ and μ are equilibrium parameters;

[0105] θ represents the model parameters.

[0106] The advantages of this invention compared with the prior art are: multi-dimensional accurate assessment: it integrates multi-modal sensor data (IMU, EMG, depth camera) to cover all dimensions of information such as kinematics and muscle function, improves data integrity by more than 60%, and uses dynamic models to track recovery trends, significantly improving assessment accuracy compared with traditional solutions.

[0107] Personalized intelligent training: Through reinforcement learning, personalized plans containing 10-15 movements are generated, shortening the rehabilitation cycle by 23%; real-time dynamic adjustment of training parameters and fatigue warnings reduce the fatigue incidence rate from 65% to 39%, improving safety and compliance.

[0108] Cross-technology integration and innovation: Muscle synergy analysis can detect functional abnormalities 3-5 days in advance, and transfer learning can shorten the initialization time for new patients from 2 hours to 15 minutes, solving the problem of scarce clinical data.

[0109] Engineering and practical advantages: Lightweight hardware supports deployment in multiple scenarios, real-time computing latency is <50ms, and after connecting to the hospital system, the time for doctors to formulate treatment plans is reduced from 40 minutes / case to 5 minutes / case, greatly improving efficiency.

[0110] Significant clinical results: joint range of motion measurement error <2°, fatigue-related adverse events reduced by 40%, patients' voluntary movement ability improved by 35%, and a safe and efficient closed-loop rehabilitation system was constructed. Attached Figure Description

[0111] Figure 1 This is a schematic diagram of the intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery of patients according to the present invention. Detailed Implementation

[0114] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0115] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0116] It is understood that spatial relation terms such as "below," "under," "below," "below," "above," "over," etc., can be used here to describe the relationship between one element or feature shown in the figure and other elements or features. It should be understood that, in addition to the orientation shown in the figure, spatial relation terms also include different orientations of the device in use and operation. For example, if the device in the figure is flipped, the element or feature described as "below" or "under" or "below" of the other element or feature will be oriented "over" the other element or feature. Therefore, the exemplary terms "below" and "under" can include both upper and lower orientations. Furthermore, the device may also include other orientations, such as being rotated 90 degrees or other orientations, and the spatial descriptive terms used herein will be interpreted accordingly.

[0117] It should be noted that when one element is considered to be "connected" to another element, it can be directly connected to the other element or connected to the other element through an intermediary element. In the following embodiments, "connection" should be understood as "electrical connection," "communication connection," etc., if the connected circuits, modules, units, etc., have the transmission of electrical signals or data between them.

[0118] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising,” “including,” or “having,” etc., specify the presence of the stated feature, whole, step, operation, component, part, or combination thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof.

[0119] Referring to the attached diagram, the postoperative patient limb function recovery intelligent assessment and rehabilitation training pathway planning system includes:

[0120] A multimodal data acquisition module is used to acquire postoperative limb movement data of patients;

[0121] The intelligent assessment module calculates the Limb Function Recovery Index (FRI) based on the motion data. The calculation formula is as follows:

[0122]

[0123] in:

[0124] n represents the number of evaluation dimensions;

[0125] w i Let be the weight coefficient of the i-th dimension.

[0126] V i Let be the measured value of the i-th dimension;

[0127] V i,min and V i,max These are the minimum and maximum values ​​of the i-th dimension, respectively;

[0128] α i Adjust parameters for the dimension specificity index;

[0129] β i This is the time decay coefficient;

[0130] t represents the current time;

[0131] t i This is the reference time point for the i-th dimension.

[0132] As an improvement, the multimodal data acquisition module includes:

[0133] Inertial Measurement Unit (IMU) is used to collect the three-dimensional acceleration of the limbs a(t) =

[0134] [a x (t),a y (t),a z [(t)] and angular velocity ω(t)=[ω x (t),ω y (t),ω z (t)];

[0135] Surface electromyography (EMG) sensors are used to acquire muscle activity signals EMG(t) = [EMG1(t), EMG2(t), ..., EMG... m (t)];

[0136] A depth camera is used to acquire the limb movement trajectory P(t)=[x(t),y(t),z(t)];

[0137] Where t is the time variable and m is the number of electromyography (EMG) sensors.

[0138] As an improvement, the intelligent evaluation module further includes:

[0139] Kinematic feature extraction unit calculates joint range of motion (ROM). j :

[0140]

[0141] in:

[0142] j is the joint index;

[0143] θ j (t) represents the angle of joint j at time t;

[0144] θ j,rest Let be the resting angle of joint j;

[0145] [t1,t2] represents the motion period.

[0146] As an improvement, the intelligent evaluation module also includes:

[0147] The muscle synergy feature analysis unit calculates the muscle synergy coefficient C through nonnegative matrix factorization (NMF). ij :

[0148]

[0149] in:

[0150] This is an electromyography signal matrix;

[0151] For muscle synergy matrix;

[0152] This is the activation coefficient matrix;

[0153] ||·|| F It is the Frobenius norm;

[0154] ||·|| 1,2 It is a mixed norm;

[0155] λ1 and λ2 are regularization parameters;

[0156] r represents the number of collaborative modes.

[0157] As an improvement, the rehabilitation training path planning module generates personalized training plans based on reinforcement learning algorithms, and solves the following optimization problems:

[0158]

[0159] in:

[0160] π is the strategy function;

[0161] τ represents the state-action sequence;

[0162] γ is the discount factor;

[0163] R(s t ,a t ,s t+1 ) is from state s t Perform action a t Transition to state s t+1 The reward function; T is the upper limit of the time step.

[0164] As an improvement, the formula for calculating the reward function R is as follows:

[0165] R(s t ,a t ,s t+1 )=w1·ΔFRI+w2·IntensityPenalty+w3

[0166] SymmetryScore

[0167] in:

[0168] ΔFRI=FRI(s t+1 )-FRI(s t () represents the change in the functional recovery index;

[0169] This is a training intensity penalty item;

[0170] Scoring is based on limb symmetry;

[0171] I t Training intensity;

[0172] I opt To achieve the optimal training intensity;

[0173] σ is the penalty distribution parameter;

[0174] L t and R t These are the feature vectors of left and right limb movement, respectively;

[0175] w1, w2, and w3 are weighting coefficients.

[0176] As an improvement, the system further includes:

[0177] The dynamic adjustment module updates the training scheme based on real-time monitoring data and estimates the patient's state x using Kalman filtering. t :

[0178]

[0179] in:

[0180] x t This is the system state vector;

[0181] z t For observation vectors;

[0182] A is the state transition matrix;

[0183] B is the control input matrix;

[0184] H is the observation matrix;

[0185] w t and v t These are process noise and observation noise, respectively.

[0186] As an improvement, the dynamic adjustment module uses a particle filter algorithm to optimize the training parameters θ, which is achieved through the following iterative process:

[0187]

[0188] in:

[0189] i is the particle index;

[0190] N is the number of particles;

[0191] q represents the proposal distribution;

[0192] p is the probability density function;

[0193] For particle weights;

[0194] z 1:t This is the observation sequence up to time t.

[0195] As an improvement, the system also includes:

[0196] The fatigue detection module calculates the muscle fatigue index FI using wavelet transform.

[0197]

[0198] in:

[0199] W(f,t) represents the wavelet transform coefficients of the electromyographic signal;

[0200] [f1,f2] represents the frequency range;

[0201] t1 and t2 are the start and end times of the motion, respectively.

[0202] As an improvement, the system employs a transfer learning framework to achieve cross-patient knowledge transfer, optimized through the following loss function:

[0203]

[0204] in:

[0205] The task loss function;

[0206] Domain adaptation loss;

[0207] Ω(θ) is the model complexity regularization term;

[0208] N and M are the number of samples in the source domain and the target domain, respectively;

[0209] The true label for sample i belonging to category c;

[0210] To predict probabilities;

[0211] D stands for the domain discriminator;

[0212] φ(x) is the feature extractor;

[0213] λ and μ are equilibrium parameters;

[0214] θ represents the model parameters.

[0215] I. System Overall Architecture:

[0216] The postoperative limb function recovery intelligent assessment and rehabilitation training path planning system provided in this embodiment includes a multimodal data acquisition module, an intelligent assessment module, a rehabilitation training path planning module, a dynamic adjustment module, a fatigue detection module, and a transfer learning module. These modules are connected via a data bus to achieve real-time acquisition of patient limb movement data, functional assessment, training program generation, and dynamic optimization.

[0217] II. Implementation of the Multimodal Data Acquisition Module:

[0218] (i) Inertial Measurement Unit (IMU): Hardware deployment: A miniature IMU sensor (model: InvenSense MPU-9250) is worn at key nodes of the patient's limbs (such as wrist, elbow, and knee joints), with the sampling frequency set to 100Hz.

[0219] Data output: Real-time output of three-dimensional acceleration signal a(t) = [a x (t),a y (t),a z [(t)], in m / s 2 ; Three-dimensional angular velocity signal ω(t)=[ω x (t),ω y (t),ω z [(t)], in rad / s. Where t is a time variable, synchronized by the system clock.

[0220] (ii) Surface electromyography (EMG) sensor: Electrode arrangement: Ag / AgCl electrodes are attached to the surface of the target muscle group (such as the biceps brachii and quadriceps femoris). The number of sensors m is determined according to the number of muscle groups being evaluated (typical value m = 8).

[0221] Signal processing: A bandpass filter (5-500Hz) is used to remove noise, and the output muscle activity signal EMG(t) = [EMG1(t), EMG2(t), ..., EMG...]. m [(t)], in μV.

[0222] (III) Depth camera: Equipment selection: Use an Intel RealSense D435i depth camera, installed above the rehabilitation training area, with a sampling frequency of 30Hz.

[0223] Trajectory Reconstruction: Obtaining the three-dimensional coordinates P(t) of limb joints using triangulation.

[0224] [x(t),y(t),z(t)], the coordinate system is based on the patient's body midline as the origin, with an accuracy of 1mm.

[0225] III. Implementation of the Intelligent Assessment Module:

[0226] (I) Calculation of the Functional Recovery Index (FRI):

[0227] Based on the formula described in claim 1:

[0228]

[0229] Assessment dimensions: n is usually 5-8, including joint range of motion, muscle strength, movement speed, symmetry, endurance, etc.

[0230] Weighting coefficient: w i Determined using the Analytic Hierarchy Process (AHP), satisfying... For example, the joint range of motion weight w1 = 0.3, and the muscle strength w2 = 0.25.

[0231] Time decay: β i Based on clinical experience, such as short-term assessment of β-carotene. i =0.1 / day, long-term evaluation day β i =0.05 / day; t i This is the time point for the first postoperative assessment.

[0232] (II) Calculation of Joint Range of Motion (ROM):

[0233] By the formula described in claim 3:

[0234]

[0235] Angle measurement: θ j (t) is calculated using the IMU data fusion quaternion algorithm, θ j,rest The average joint angle is measured over 5 consecutive seconds while the patient is at rest.

[0236] Movement cycle: [t1,t2] is automatically identified by motion detection algorithms (such as dynamic time warping), for example, the elbow flexion and extension cycle is 10-15 seconds.

[0237] (III) Analysis of Muscle Synergy Characteristics:

[0238] The nonnegative matrix factorization (NMF) algorithm of claim 4 is adopted:

[0239]

[0240] Matrix definition: This is the electromyography signal matrix (T is the number of sampling points). The muscle coordination matrix (r takes 2-4 coordination modes) is used. This is the activation coefficient matrix.

[0241] Optimized solution: Iterative solution using alternating nonnegative least squares (ANLS) with regularization parameters λ1 = 0.1 and λ2 = 0.05.

[0242] IV. Implementation Method of Rehabilitation Training Path Planning Module:

[0243] (I) Generation of reinforcement learning training programs:

[0244] The optimization problem based on claim 5:

[0245]

[0246] State space: s t Includes FRI, ROM, EMG features, and training intensity I. t etc., with dimensions of 15-20.

[0247] Action space: a t The training movement type (such as flexion and extension, abduction), resistance intensity, and duration are discretized into 5×3×3 combinations.

[0248] Reward function: Calculated according to claim 6:

[0249] R(s t ,a t ,s t+1 )

[0250] =w1·ΔFRI+w2·IntensityPenalty+w3

[0251] SymmetryScore

[0252] Where ΔFRI is the difference in FRI between adjacent time points, and IntensityPenalty uses a Gaussian penalty function (I0). opt =60% of maximum muscle strength, σ=10), SymmetryScore is calculated by the Euclidean distance of the left and right limb movement trajectories.

[0253] (II) Dynamic Adjustment Module:

[0254] Kalman filter state estimation:

[0255] The state-space model of claim 7 is adopted:

[0256] (x_t=Ax_{t-1}+Bu_{t-1}+w_{t-1}\

[0257] z_t=Hx_t+v_t

[0258] \end{cases}

[0259] - $x_t$ contains state variables such as joint angles, muscle activation, and fatigue index; - $z_t$ contains observation data from IMU, EMG, and depth cameras; - the noise covariance matrices $Q$ and $R$ are estimated statistically from historical data.

[0260] The number of particles N = 1000, the proposed distribution q is a Gaussian distribution, and the initial parameter θ0 is determined by the critical parameter.

[0261] V. Implementation method of fatigue detection module:

[0262] Based on the wavelet transform algorithm of claim 9:

[0263]

[0264] Frequency range: [f1,f2]=[20,50]Hz (mid-frequency band related to muscle fatigue);

[0265] Wavelet basis selection: The db4 wavelet is used for 5-level decomposition to calculate the energy ratio at the start (t1) and end (t2) of motion. When FI>1.5, fatigue warning is triggered.

[0266] VI. Implementation of the Transfer Learning Module:

[0267] Based on the loss function of claim 10:

[0268]

[0269] Mission losses: Cross-entropy loss is used to classify patient functional levels (C = 5).

[0270] Domain Adaptation: Adversarial training aligns the feature distributions of the source domain (healthy individuals / historical patients) with those of the target domain (current patients). The domain discriminator D employs a 3-layer fully connected network.

[0271] Parameter optimization: Using the Adam optimizer with λ = 0.5 and μ = 0.001, cross-patient rehabilitation knowledge transfer was achieved.

[0272] VII. System Workflow:

[0273] Data acquisition: Multimodal sensors acquire limb motion data in real time at a frequency of 100Hz;

[0274] Feature extraction: Calculates 50+ features including ROM, muscle coordination coefficient, and motion trajectory;

[0275] Functional assessment: Real-time recovery index is generated using the FRI formula, and threshold settings are set (e.g., FRI < 0.4 triggers high-intensity training);

[0276] Path planning: The reinforcement learning model outputs a personalized training plan (containing 10-15 action sequences);

[0277] Dynamic adjustment: Training parameters are updated every 5 minutes using Kalman filtering and particle filtering;

[0278] Fatigue monitoring: FI is calculated after each training session. If the threshold is exceeded, the training intensity is reduced by 20%.

[0279] Knowledge transfer: When new patients are admitted, the initialization time is shortened by using a pre-trained model based on 1,000+ historical cases.

[0280] VIII. Verification of Implementation Examples:

[0281] A test was conducted on 50 postoperative patients in the rehabilitation department of a tertiary hospital. The system demonstrated a Pearson correlation coefficient of 0.89 between the FRI (Frequency Rating Indicator) and the clinician's score. The training program generation time was less than 30 seconds, and the incidence of patient fatigue decreased by 40% after dynamic adjustment. The experiment proves that this system, through multi-dimensional mathematical modeling and intelligent algorithms, achieves accurate assessment and personalized intervention for postoperative rehabilitation.

[0282] Multi-dimensional and precise assessment, breaking through the limitations of traditional single-modality methods:

[0283] Multimodal data fusion technology: Through the coordinated acquisition of data by inertial measurement units (IMU), surface electromyography (EMG) sensors, and depth cameras, it achieves full-dimensional data coverage of limb kinematics (3D trajectory, joint angles), muscle function (electromyographic activity, synergistic patterns), and biomechanics (acceleration, angular velocity). Compared with traditional single-sensor assessment, data integrity is improved by more than 60%, avoiding assessment bias caused by noise from a single signal.

[0284] For example, combining millimeter-precise joint trajectory data from depth cameras with muscle activation data from electromyography signals can accurately identify compensatory behaviors of small muscle groups, such as the synergistic imbalance between the quadriceps and hamstrings after knee surgery, which is easily missed by traditional methods, while this system has a detection sensitivity of 92%.

[0285] Dynamic Time Decay and Weight Optimization Model: Based on Formula Introducing the time decay coefficient β i And the weights w of the Analytic Hierarchy Process (AHP) i This approach allows assessment results to reflect current functional status while dynamically tracking postoperative recovery trends. Clinical validation shows that the Pearson correlation coefficient between FRI and physician subjective scores reaches 0.89, a 37% improvement compared to traditional static scoring models, providing a quantitative basis for precise rehabilitation intervention.

[0286] Intelligent algorithms drive personalized training, improving rehabilitation efficiency:

[0287] Reinforcement Learning Dynamic Path Planning: Solving using a reinforcement learning framework By mapping patient status (FRI, ROM, muscle synergy coefficient, etc., 20+ dimensions) to training movements (type, intensity, duration), a high-dimensional model is generated, producing personalized programs containing 10-15 movement sequences. Compared to experience-based fixed programs, the training program adaptation efficiency is improved by 50%, and the average patient recovery period is shortened by 23%.

[0288] Core advantage: Introducing symmetric scoring into the reward function The addition of intensity penalties automatically avoids the risks of unilateral compensation and overtraining injuries, improving safety by 40% compared to traditional methods.

[0289] Real-time dynamic adjustment and fatigue early warning: Based on a dual feedback mechanism of Kalman filtering (state estimation) and particle filtering (parameter optimization), training parameters are updated every 5 minutes to achieve adaptive adjustment of the patient's real-time state (joint angle, fatigue index, muscle activation). When the fatigue index... When the intensity decays automatically (e.g., reducing resistance by 20%), the fatigue rate during training is reduced from 65% in traditional programs to 39%, significantly improving patient compliance.

[0290] Integrating interdisciplinary technologies to build an intelligent rehabilitation ecosystem:

[0291] In-depth analysis of muscle synergistic features: using the nonnegative matrix factorization (NMF) algorithm Two to four synergistic patterns are extracted from 8-channel electromyography (EMG) signals to accurately identify postoperative abnormalities in muscle nerve control (such as hypersynergism of flexor muscles in hemiplegic patients). Compared with traditional root mean square (RMS) EMG analysis, it can detect muscle function degeneration trends 3-5 days earlier, providing an advantage for early intervention.

[0292] Transfer learning accelerates clinical applications: utilizing transfer learning loss functions Based on a pre-trained model with 1,000+ historical cases, the initialization time for new patients is reduced from 2 hours to 15 minutes using traditional methods. Furthermore, the evaluation accuracy decreases by only 4% in small sample (<10 cases) scenarios, effectively addressing the problem of scarce clinical data and promoting the rapid deployment of the system in primary hospitals.

[0293] Engineering implementation leverages advantages and ensures clinical applicability:

[0294] Multimodal hardware lightweight deployment: It adopts commercial-grade sensors (such as InvenSense MPU-9250, IntelRealSense D435i), with a single device weighing less than 200g. The sampling frequency reaches 100Hz (IMU) and 30Hz (depth camera). It supports flexible deployment in multiple scenarios such as bedside and rehabilitation training area, and is compatible with assistive devices such as wheelchairs and walkers, adapting to the full cycle needs of postoperative patients from early bed rest to later gait training.

[0295] Real-time computing and clinical decision support: The core algorithms (FRI computing, reinforcement learning inference) are processed in real time on the NVIDIA Jetson AGX Orin edge computing platform (latency <50ms), and the training scheme generation time is <30 seconds. It supports seamless integration with hospital information systems (HIS) and electronic medical records (EMR), providing physicians with a visual interface for the entire process of "assessment-planning-adjustment", significantly reducing the time cost of manually formulating plans (from 40 minutes / case to 5 minutes / case).

[0296] Social benefits and clinical value:

[0297] Verified using 50 postoperative patients at a top-tier hospital, this system achieves the following:

[0298] Functional assessment accuracy: Joint range of motion (ROM) measurement error <2°, muscle synergy pattern recognition accuracy 91%;

[0299] Training safety: The incidence of fatigue-related adverse events decreased by 40%, and the detection rate of compensatory movements increased by 55%;

[0300] Rehabilitation efficiency: The average training cycle was shortened by 28 days, and the improvement rate of patients' voluntary motor ability (FIM score) was 35% faster than that of the traditional method.

[0301] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A smart assessment and rehabilitation training pathway planning system for postoperative limb function recovery, characterized in that, include: A multimodal data acquisition module is used to acquire postoperative limb movement data of patients; The intelligent assessment module calculates the Limb Function Recovery Index (FRI) based on the motion data. The calculation formula is as follows: in: n represents the number of evaluation dimensions; w i Let be the weight coefficient of the i-th dimension. V i Let be the measured value of the i-th dimension; V i,min and V i,max These are the minimum and maximum values ​​of the i-th dimension, respectively; α i Adjust parameters for the dimension specificity index; β i This is the time decay coefficient; t represents the current time; t i This is the reference time point for the i-th dimension.

2. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 1, characterized in that, The multimodal data acquisition module includes: Inertial Measurement Unit (IMU) is used to collect the three-dimensional acceleration of the limbs a(t) = [a x (t),a y (t),a z [(t)] and angular velocity ω(t)=[ω x (t),ω y (t),ω z (t)]; Surface electromyography (EMG) sensors are used to acquire muscle activity signals EMG(t) = [EMG1(t), EMG2(t), ..., EMG... m (t)]; A depth camera is used to acquire the limb movement trajectory P(t)=[x(t),y(t),z(t)]; Where t is the time variable and m is the number of electromyography (EMG) sensors.

3. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 2, characterized in that, The intelligent evaluation module further includes: Kinematic feature extraction unit calculates joint range of motion (ROM). j : in: j is the joint index; θ j (t) represents the angle of joint j at time t; θ j,rest Let be the resting angle of joint j; [t1,t2] represents the motion period.

4. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 3, characterized in that, The intelligent evaluation module also includes: The muscle synergy feature analysis unit calculates the muscle synergy coefficient C through nonnegative matrix factorization (NMF). ij : in: This is a matrix of electromyographic signals; For muscle synergy matrix; This is the activation coefficient matrix; ||·|| F It is the Frobenius norm; ||·|| 1,2 It is a mixed norm; λ1 and λ2 are regularization parameters; r represents the number of collaborative modes.

5. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 1, characterized in that, The rehabilitation training pathway planning module generates personalized training plans based on reinforcement learning algorithms, and solves the following optimization problems: in: π is the strategy function; τ represents the state-action sequence; γ is the discount factor; R(s t ,a t ,s t+1 ) is from state s t Perform action a t Transition to state s t+1 The reward function; T represents the upper limit of the time step.

6. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 5, characterized in that, The formula for calculating the reward function R is as follows: R(s t ,a t ,s t+1 ) =w1·ΔFRI+w2·IntensityPenalty+w3 SymmetryScore in: ΔFRI=FRI(s t+1 )-FRI(s t () represents the change in the functional recovery index; This is a training intensity penalty item; Scoring is based on limb symmetry; I t Training intensity; I opt To achieve the optimal training intensity; σ is the penalty distribution parameter; L t and R t These are the feature vectors of left and right limb movement, respectively; w1, w2, and w3 are weighting coefficients.

7. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 1, characterized in that, The system further includes: The dynamic adjustment module updates the training scheme based on real-time monitoring data and estimates the patient's state x using Kalman filtering. t : in: x t This is the system state vector; z t For observation vectors; A is the state transition matrix; B is the control input matrix; H is the observation matrix; w t and v t These are process noise and observation noise, respectively.

8. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 7, characterized in that, The dynamic adjustment module uses a particle filter algorithm to optimize the training parameters θ, which is achieved through the following iterative process: in: i is the particle index; N is the number of particles; q represents the proposal distribution; p is the probability density function; For particle weights; z 1:t This is the observation sequence up to time t.

9. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 1, characterized in that, The system also includes: The fatigue detection module calculates the muscle fatigue index FI using wavelet transform. in: W(f,t) represents the wavelet transform coefficients of the electromyographic signal; [f1,f2] represents the frequency range; t1 and t2 are the start and end times of the motion, respectively.

10. The intelligent assessment and rehabilitation training pathway planning system for postoperative limb function recovery as described in claim 1, characterized in that, The system employs a transfer learning framework to achieve cross-patient knowledge transfer, optimized using the following loss function: in: The task loss function; Ω(θ) represents the domain adaptation loss; Ω(θ) represents the model complexity regularization term. N and M are the number of samples in the source domain and the target domain, respectively; The true label for sample i belonging to category c; To predict probabilities; D stands for the domain discriminator; φ(x) is the feature extractor; λ and μ are equilibrium parameters; θ represents the model parameters.

Images