Intelligent orthopedic traction method, system and device

Abstract

The invention belongs to the technical field of orthopedic traction, and discloses an intelligent orthopedic traction method, system and device. The method comprises the steps that clinical data and DICOM images are collected, anatomical parameters are extracted through three-dimensional reconstruction, and an initial scheme set is obtained through a parameter recommendation model in combination with the clinical data; original signals are drawn, collected and preprocessed through the initial parameters to obtain biomechanical characteristics, and the biomechanical characteristics are input into a fear analysis model to obtain the fear level; counting and analyzing the original signal after traction to obtain a muscle fatigue result; evaluating a curative effect based on an original signal and a daily fatigue result, optimizing the digital twins by combining biomechanical characteristics and an evaluation result, and obtaining an optimized traction scheme by combining a reinforcement learning optimization scheme; according to the invention, collaborative optimization of orthopedic traction treatment on physiological recovery and psychological care is realized, and the personalized treatment level and rehabilitation effect are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of orthopedic traction, and more particularly, to an intelligent orthopedic traction method, system and device. Background Art

[0002] Orthopedic traction is a commonly used method for treating orthopedic diseases such as fractures and dislocations. Traditional orthopedic traction methods have many shortcomings. For example, most orthopedic traction beds are manually controlled for traction strength, which makes it inconvenient to precisely control the strength during the stretching process, and can easily cause secondary damage to the stretched part. Various existing traction methods use fixed brackets with traction weights, which have potential hidden dangers of misaligned traction line angles. At the same time, continuous static traction can easily increase the occurrence of complications such as thrombosis and soft tissue damage. In addition, when performing orthopedic traction, medical staff lack better auxiliary methods to improve the success rate and recovery before and after traction, and cannot guarantee the recovery speed of the patient's affected limb. Different symptoms require different traction strengths, and medical staff need to spend a lot of energy to control the traction equipment.

[0003] The Chinese patent application with publication number CN117860374A discloses an artificial intelligence-based orthopedic traction auxiliary method and system: S1: Data classification: access the current hospital's database, process previous orthopedic traction data, and classify the limbs that have been pulled and the causes of the disease for subsequent treatment; S2: Data reference: upload the patient's affected limb data, access the database, search and process it according to its cause of disease, and generate a three-dimensional map of the uploaded affected limb condition, and extract and compare the generated three-dimensional map features with the features of similar cases. After the comparison is completed, the treatment data and recovery data of similar cases are referenced; S3: Processing of auxiliary plans: produce images of traction methods based on the treatment and recovery data of similar cases for medical staff to use. Perform auxiliary inspection, including the displacement distance of the traction device, the spacing adjustment of the traction device, the displacement direction of the traction device and the pressure control when the traction device is in use. When the traction device is in use, the pressure value needs to be extracted, and through experiments and adjustments, the pressure is judged to be qualified or not, so as to carry out traction work on the patient's affected limb; S4: Auxiliary care: medical staff log in to the system to handle and record the patient's care, including the patient's blood condition, temperature, skin condition, traction equipment condition, placement of the affected limb and protective treatment of complications; S5: Measurement and processing of the patient's affected limb: measure the length of the patient's affected limb and healthy limb and compare them, so as to process the traction equipment according to the measurement, including adjusting the weight of the traction equipment and controlling the reset timing.

[0004] Although the above method can meet most scenarios, research and practical application of the above method and existing technology have found that the above method and existing technology have at least the following defects:

[0005] Psychological fear can cause abnormal muscle resistance and affect cooperation. The above method does not take into account the patient's psychological fear caused by treatment, making it difficult for the traction strategy to adapt to the patient's psychology, reducing patient compliance and rehabilitation effects.

[0006] In view of this, the present invention proposes an intelligent orthopedic traction method, system and device to solve the above problems. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an intelligent orthopedic traction method, comprising:

[0008] Collect clinical data and DICOM images, perform 3D reconstruction on the DICOM images, and extract anatomical parameters; input the clinical data and anatomical parameters into the parameter recommendation model to obtain the initial solution set;

[0009] The parameters in the initial solution set are used as initial execution parameters for traction, an original signal set during the traction process is collected, the original signal set is preprocessed to obtain a signal set, the signal set is analyzed to obtain biomechanical characteristics, and the biomechanical characteristics are input into a fear analysis model to obtain a fear level;

[0010] After each set of traction, statistical analysis was performed on each set of original signals to obtain muscle fatigue results;

[0011] The efficacy of the original signal set and daily muscle fatigue results is evaluated, and the digital twin is dynamically optimized according to the biomechanical characteristics and evaluation results to obtain a dynamic twin. The scheme is optimized in combination with the reinforcement learning algorithm to obtain an optimized traction scheme.

[0012] Furthermore, the signal set includes grip force signals, traction force signals and surface electromyography signals of muscle groups of left and right hands;

[0013] Methods for obtaining biomechanical characteristics include:

[0014] The average grip force signal of the left and right hands is taken to obtain the average grip force signal of both hands;

[0015] Calculate the average value and standard deviation of the average grip force signals of both hands within a preset period, calculate the ratio of the standard deviation to the average value, and obtain the grip force fluctuation rate;

[0016] Calculate the time offset corresponding to the maximum value of the cross-correlation function between the average grip force signal of both hands and the traction force signal to obtain the response phase difference;

[0017] Calculate the covariance of the average grip force signal and the traction force signal of both hands, as well as the standard deviation of the traction force signal within a preset time period, and calculate the Pearson correlation coefficient between the average grip force signal and the traction force signal of both hands to obtain force coordination;

[0018] When the traction force signal remains constant, the peak value of the average grip force signal of both hands is identified, and the time required for the peak value to decay to half of the peak value is calculated to obtain the relaxation delay time;

[0019] The marginal probability density functions of the average grip force signals of both hands, the marginal probability density functions of the traction force signals, and the marginal probability density functions of the average grip force signals of both hands and the traction force signals are calculated to obtain the mutual information entropy; the marginal probability density functions are obtained by kernel density estimation.

[0020] Perform wavelet packet decomposition on the surface electromyographic signal of the muscle group to obtain B frequency bands. Each frequency band corresponds to a different muscle group state. Calculate the energy of each frequency band and take the normalized energy of the fatigue frequency band as the frequency band energy.

[0021] The biomechanical characteristics were obtained by splicing the grip force fluctuation rate, response phase difference, force coordination, relaxation delay time, mutual information entropy and frequency band energy.

[0022] Further, the method of obtaining muscle fatigue results includes:

[0023] Extract the peak values ​​of the left and right hand grip force signals from the Kth group of original signals. When the peak values ​​of the left and right hand grip force signals decay to M% of the initial maximum value, it is determined to be muscle fatigue.

[0024] Calculate the average of the peak values ​​of the surface electromyographic signals of the flexor muscle group during the first F times, and combine it with the flexor muscle group adjustment factor to calculate the fatigue threshold of the flexor muscle group;

[0025] Calculate the average of the peak values ​​of the surface electromyographic signals of the adductor muscle group during the first F times, and combine it with the adductor muscle group regulation factor to calculate the fatigue threshold of the adductor muscle group;

[0026] If the surface electromyographic signal of the flexor muscle group is not higher than the fatigue threshold of the flexor muscle group, it is determined to be flexor muscle group fatigue;

[0027] If the surface electromyographic signal of the adductor muscle group is not higher than the fatigue threshold of the adductor muscle group, it is determined to be adductor muscle group fatigue.

[0028] Furthermore, the method for evaluating the efficacy of the original signal set and daily muscle fatigue results includes:

[0029] The average fear level of the previous day and the average fear level of the current day were counted to calculate the fear improvement rate;

[0030] The average of the peak grip strength of the left and right hands in each group on the previous day and the average of the peak grip strength of the left and right hands on the current day were calculated to obtain the muscle growth rate, including the muscle growth rate of the flexor group and the muscle growth rate of the adductor group.

[0031] The difference between the growth rates of the flexor group and the adductor group was calculated to obtain the muscle recovery difference index;

[0032] Obtain the curve of the change in traction force versus traction displacement and fit it using the Kelvin-Voigt viscoelastic model. Calculate the muscle and ligament stiffness based on the Kelvin-Voigt viscoelastic model. Calculate the muscle and ligament stiffness of the previous day and the current day, and calculate the change in muscle and ligament stiffness.

[0033] The fear improvement rate, muscle growth rate, muscle group recovery difference index and muscle and ligament stiffness changes are calculated to generate an efficacy report.

[0034] Furthermore, the method of obtaining the dynamic twin includes:

[0035] The Hooke elastic model was used to correct the ligament stiffness and obtain the corrected ligament stiffness;

[0036] The muscle stiffness is corrected according to the muscle viscoelastic coefficient, muscle growth rate and muscle group recovery difference index to obtain the corrected muscle stiffness;

[0037] The traction force signal was time discretized to calculate the rate of change of the traction displacement over time. A fitting equation for the traction force signal with respect to the muscle viscoelastic coefficient was established based on the Kelvin-Voigt viscoelastic model, and the least squares method was used to obtain the muscle and ligament viscoelastic correction coefficients.

[0038] The digital twin is updated according to the corrected ligament stiffness, corrected muscle stiffness and muscle and ligament viscoelasticity correction coefficients to obtain a dynamic twin.

[0039] Furthermore, the method for obtaining an optimized traction solution includes:

[0040] Define the state space: composed of a state vector, which includes biomechanical characteristics, fear level, muscle fatigue results, fear improvement rate, muscle growth rate, muscle group recovery difference index and stiffness change;

[0041] Define the action space: It consists of action vectors, which include traction force increment, traction angle increment and traction mode;

[0042] Design reward function: The reward function is obtained by weighted calculation of safety reward, efficacy reward, and comfort reward;

[0043] Initialize the state vector, randomly select an action vector, perform traction simulation in the dynamic twin, and obtain the next round of state vector and reward function value; generate experience data containing the current state vector, the selected action vector, the reward function value, and the next round of state vector, and store it in the experience replay pool;

[0044] The DQN model randomly samples W pieces of experience data from the experience replay pool, calculates the target action value function value and the current action value function value, minimizes the loss function, and updates the current network parameters through back propagation; the target action value function value is based on the reward function value and the dynamic twin body in state When traversing all action vectors The action value function value corresponding to the action vector that maximizes the action value function value is calculated;

[0045] The trained DQN model is used to collect the patient's state vector after actual traction, calculate the action value function values ​​of all action vectors, select the action vector with the largest action value function value, and calculate the parameters corresponding to the optimized traction plan.

[0046] Furthermore, in the reward function, a safety reward is calculated based on the simulated maximum stress of the ligaments in the dynamic twin and the safety threshold of the human spinal ligaments;

[0047] The therapeutic effect reward is calculated based on the difference index between the next round of muscle group growth rate predicted by the dynamic twin and the next round of muscle group recovery predicted by the dynamic twin;

[0048] The comfort bonus is calculated based on the current average fear level and the viscoelastic coefficient of the muscle group before and after correction.

[0049] Furthermore, the clinical data include diagnosis results, age, weight and past medical history.

[0050] Furthermore, the method of obtaining anatomical parameters includes:

[0051] Each DICOM image is traversed sequentially in the order of CT scans, and a three-dimensional voxel grid is constructed based on the size and resolution of the DICOM image;

[0052] Traverse the entire 3D voxel grid voxel by voxel; for each voxel, read the grayscale value of its eight vertices and compare it with the pre-determined bone grayscale threshold to determine the vertex state;

[0053] According to the voxel vertex state, the corresponding isosurface topology structure is searched in the topology table to generate the corresponding triangular patch model;

[0054] After generating the preliminary triangular face model, it is optimized to obtain a three-dimensional reconstructed image;

[0055] The 3D reconstructed image is used as the input of the convolutional neural network to obtain a feature map, which is then used as the input of the segmentation model to obtain anatomical parameters; the anatomical parameters include intervertebral disc height, spinal Cobb angle, facet joint spacing, and the coordinates of the anatomical attachment points of the adductor and flexor muscles.

[0056] Furthermore, the initial program set includes an initial traction program and an initial upper limb exercise program. The initial traction program includes initial traction force, traction angle, traction duration and traction mode; the initial upper limb exercise program includes exercise mode, initial resistance and exercise parameters.

[0057] An intelligent orthopedic traction system, which implements the intelligent orthopedic traction method, comprises:

[0058] Acquisition and analysis module: collects clinical data and DICOM images, performs 3D reconstruction on DICOM images, and extracts anatomical parameters; inputs clinical data and anatomical parameters into the parameter recommendation model to obtain an initial solution set;

[0059] Fear analysis module: traction is performed using the parameters in the initial solution set as the initial execution parameters. The original signal set during the traction process is collected, pre-processed, and then analyzed to obtain biomechanical characteristics. The biomechanical characteristics are input into the fear analysis model to obtain the fear level;

[0060] Fatigue analysis module: After each set of traction, statistical analysis is performed on each set of original signals to obtain muscle fatigue results;

[0061] Traction optimization module: Evaluate the efficacy of the original signal set and daily muscle fatigue results, dynamically optimize the digital twin based on the biomechanical characteristics and evaluation results to obtain a dynamic twin, and combine the reinforcement learning algorithm to optimize the solution to obtain an optimized traction solution.

[0062] An intelligent orthopedic traction device, wherein a boom rope serves as a load-bearing base component for suspending the traction rope, one end of the traction rope is connected to the boom rope via a rotating mechanism, and the other end is fixedly connected to a lifting ring gripper; a pressure detection film is fitted on the outer surface of the lifting ring gripper; the device implements the intelligent orthopedic traction method described above.

[0063] The technical effects and advantages of the intelligent orthopedic traction method, system and device provided by the present invention are:

[0064] The present invention collects clinical data and DICOM images, extracts anatomical parameters through 3D reconstruction, and inputs them into the parameter recommendation model to generate an initial solution set, laying a personalized foundation for adapting to the patient's physiological and psychological state; during the traction process, the original signal set is collected and preprocessed, and the biomechanical characteristics are analyzed to obtain the biomechanical characteristics, which are input into the fear analysis model to quantify the fear level, breaking the limitation of existing methods that do not consider psychological fear, and accurately identifying the fear-muscle abnormality confrontation relationship; after each group of traction, the muscle fatigue result is determined by the peak attenuation of grip force and the threshold of electromyographic signal to avoid misjudging the muscle tension caused by fear as normal fatigue; based on the original signal and daily fatigue results, the fear improvement rate, muscle growth rate and other indicators are calculated to carry out efficacy evaluation, clearly quantifying the psychological and The association between fear and physiological recovery is analyzed; the ligament / muscle stiffness and viscoelastic coefficient are corrected by combining the Hooke elastic model and the Kelvin-Voigt model, and the dynamic twin that incorporates the influence of fear is updated to provide an accurate virtual environment for strategy simulation; finally, the traction plan is optimized through the state space containing fear-related indicators, the reward function that takes into account comfort rewards, and the DQN model to generate a strategy that can adapt to the patient's psychological fear state; it effectively alleviates the abnormal muscle confrontation caused by fear, improves the patient's cooperation and compliance, and solves the problem of mismatch of traction strategies and poor rehabilitation effects caused by ignoring psychological fear in existing methods, realizing the coordinated optimization of orthopedic traction treatment in physiological recovery and psychological care, and improving the level of personalized treatment and rehabilitation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of an intelligent orthopedic traction method according to the present invention;

[0066] Figure 2 This is a data flow diagram of an intelligent orthopedic traction method of the present invention;

[0067] Figure 3 This is a schematic diagram of the intelligent orthopedic split traction device of the present invention;

[0068] Figure 4 This is a schematic diagram of the intelligent orthopedic gantry traction device of the present invention;

[0069] Figure 5 A schematic diagram of an intelligent orthopedic traction system of the present invention;

[0070] The meaning of the marks in the figure: 1-rotating mechanism; 2-traction rope; 3-lifting ring grab; 4-pressure detection film; 5-jib rope. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0072] Example 1:

[0073] See also Figure 1 、 Figure 2 As shown, this embodiment provides an intelligent orthopedic traction method, comprising the following steps:

[0074] Collect clinical data and DICOM images, perform 3D reconstruction on the DICOM images, and extract anatomical parameters; input the clinical data and anatomical parameters into the parameter recommendation model to obtain the initial solution set;

[0075] Clinical data included diagnosis, age, weight, and past medical history;

[0076] Methods for obtaining anatomical parameters include:

[0077] Each DICOM image is traversed sequentially in the order of CT scans, and a three-dimensional voxel grid is constructed based on the size and resolution of the DICOM image;

[0078] Traverse the entire 3D voxel grid voxel by voxel; for each voxel, read the grayscale value of its eight vertices and compare it with the pre-determined bone grayscale threshold to determine the vertex state;

[0079] According to the voxel vertex state, the corresponding isosurface topology structure is searched in the topology table to generate the corresponding triangular patch model;

[0080] After generating the preliminary triangular face model, it is optimized to obtain a three-dimensional reconstructed image;

[0081] The 3D reconstructed image is used as the input of the convolutional neural network to obtain a feature map, which is then used as the input of the segmentation model to obtain anatomical parameters; the anatomical parameters include intervertebral disc height, spinal Cobb angle, facet joint spacing, and the coordinates of the anatomical attachment points of the adductor and flexor muscles.

[0082] The initial program set includes an initial traction program and an initial upper limb exercise program. The initial traction program includes initial traction force, traction angle, traction duration and traction mode; the initial upper limb exercise program includes exercise mode, initial resistance and exercise parameters.

[0083] By collecting clinical data including diagnostic results, age, weight and medical history, combined with anatomical parameters such as intervertebral height and spinal Cobb angle extracted through DICOM image three-dimensional reconstruction, the parameter recommendation model is input to generate an initial traction plan covering initial traction force, traction angle, etc. and an initial upper limb exercise plan including exercise mode, initial resistance, etc., from the source of precise and personalized plans, laying the foundation for the subsequent dynamic optimization strategy based on psychological fear factors. Through the early comprehensive construction of plans that fit the patient's physiological characteristics, the psychological fear that may be aggravated due to plan incompatibility can be alleviated to a certain extent, reducing the risk of abnormal muscle confrontation and affecting cooperation, helping to improve patient compliance and rehabilitation effects.

[0084] The parameters in the initial solution set are used as initial execution parameters for traction, an original signal set during the traction process is collected, the original signal set is preprocessed to obtain a signal set, the signal set is analyzed to obtain biomechanical characteristics, and the biomechanical characteristics are input into a fear analysis model to obtain a fear level;

[0085] Methods for obtaining signal sets include:

[0086] Filter, amplify, and perform analog-to-digital conversion on all original signals in the original signal set to obtain digital signals, and then add a unified time stamp to the data signals to obtain a signal set;

[0087] The signal set includes the grip force signal of the left and right hands, the traction force signal and the surface electromyography signal of the muscle group; the traction device can refer to Figure 3 、 Figure 4 ,in, Figure 3 and Figure 4 A pressure detection film 4 is attached to the ring grip 3 to detect the grip force signals of the left and right hands in real time; each traction rope 2 hanging on the boom rope 5 is equipped with a tension sensor to detect the traction force signal in real time; each end of the traction rope 2 and the boom rope 5 is equipped with a rotating mechanism 1, and the rope length can be manually adjusted to adapt to the traction of different patients; the surface electromyography signals of the muscle groups are collected by sensors directly attached to the surface of the patient's skin, which is not shown in the figure.

[0088] Methods for obtaining biomechanical characteristics include:

[0089] The average grip force signal of the left and right hands is taken to obtain the average grip force signal of both hands;

[0090] Calculate the average and standard deviation of the average grip force signals of both hands within a preset time period, such as 30 seconds, and calculate the ratio of the standard deviation to the average to obtain the grip force fluctuation rate;

[0091] Calculate the time offset corresponding to the maximum value of the cross-correlation function between the average grip force signal of both hands and the traction force signal to obtain the response phase difference; ,in, is the mathematical expectation; is the time offset; is the traction force signal at time t; is the average grip force signal of both hands at time t;

[0092] Calculate the covariance of the average grip force signal of both hands and the traction force signal, as well as the standard deviation of the traction force signal within the preset time period, and calculate the Pearson correlation coefficient of the average grip force signal of both hands and the traction force signal to obtain force coordination; such as force coordination ,in, Traction force signal Average grip force signal of both hands covariance of The traction force signal within the preset period The standard deviation of The average grip force signal of both hands during the preset period The standard deviation of

[0093] When the traction force signal remains constant, the peak value of the average grip force signal of both hands is identified, and the time required for the peak value to decay to half of the peak value is calculated to obtain the relaxation delay time;

[0094] Calculate the marginal probability density function of the average grip force signal of both hands, the marginal probability density function of the traction force signal, and the marginal probability density function of the average grip force signal of both hands and the traction force signal to obtain the mutual information entropy; the marginal probability density function can be obtained by kernel density estimation method; ,in, is the marginal probability density function of the average grip force signal and traction force signal of both hands, is the average grip force signal of both hands; It is the traction signal; for and The joint probability density function of and They are and The marginal probability density function of is the value of the average grip force signal of both hands, is the value of the traction force signal;

[0095] Perform wavelet packet decomposition on the surface electromyographic signal of the muscle group to obtain B frequency bands. Each frequency band corresponds to a different muscle group state, such as 1-4Hz for fatigue signal and 20-50Hz for active contraction signal. Calculate the energy of each frequency band and take the normalized energy of the fatigue frequency band as the frequency band energy. ,in, is the wavelet packet coefficient of the i-th frequency band; N is the number of sampling points in the window, and n is the index value of the number of sampling points in the window; the energy of each frequency band is normalized to obtain the normalized energy, such as the normalized energy ,in, is the normalized energy of the ith frequency band, is the frequency band energy of the i-th frequency band; the normalized energy corresponding to the fatigue band is selected as the frequency band energy;

[0096] The biomechanical characteristics were obtained by splicing the grip force fluctuation rate, response phase difference, force coordination, relaxation delay time, mutual information entropy and frequency band energy.

[0097] When traction is carried out with the parameters in the initial solution set as the initial execution parameters, the original signal set including the grip force, traction force and surface electromyography of the left and right hands is collected during the traction process. All the original signals are first filtered, amplified, analog-to-digital converted and time-stamped to obtain a synchronized signal set, and then key information is extracted from the signal set to analyze the biomechanical characteristics; the grip force of the left and right hands is averaged to obtain the average grip force signal of both hands, and the grip force fluctuation rate of the signal within the preset time period, the response phase difference with the traction force signal and the force coordination are calculated. The relaxation delay time is counted in the constant traction force stage, and the kernel density is used to calculate the biomechanical characteristics of the signal. The estimation method calculates the mutual information entropy of the average grip force and traction force signals of both hands. At the same time, the surface electromyographic signals of the muscle group are decomposed by wavelet packets and the normalized energy of the fatigue frequency band is taken as the frequency band energy. These features are spliced ​​into biomechanical characteristics and then input into the fear analysis model to obtain the fear level. This process can accurately quantify the abnormal muscle confrontation caused by the patient's psychological fear, breaking the limitation of existing methods that do not consider the patient's psychological fear during treatment, so that subsequent traction strategy adjustments can adapt to the patient's psychological state based on a clear fear level, reduce the problem of decreased cooperation caused by fear, and thus improve patient compliance and rehabilitation effects.

[0098] After each set of traction, statistical analysis was performed on each set of original signals to obtain muscle fatigue results;

[0099] Methods for achieving muscle fatigue results include:

[0100] Extract the peak values ​​of the left and right hand grip force signals from the Kth group of original signals. When the peak values ​​of the left and right hand grip force signals decay to M% of the initial maximum value, it is determined to be muscle fatigue.

[0101] Calculate the average of the peak values ​​of the surface electromyographic signals of the flexor muscle group during the first F times, and combine it with the flexor muscle group adjustment factor to calculate the fatigue threshold of the flexor muscle group;

[0102] Calculate the average of the peak values ​​of the surface electromyographic signals of the adductor muscle group during the first F times, and combine it with the adductor muscle group regulation factor to calculate the fatigue threshold of the adductor muscle group;

[0103] If the surface electromyographic signal of the flexor muscle group is not higher than the fatigue threshold of the flexor muscle group, it is determined to be flexor muscle group fatigue;

[0104] If the surface electromyographic signal of the adductor muscle group is not higher than the fatigue threshold of the adductor muscle group, it is determined to be adductor muscle group fatigue.

[0105] After each set of traction, the muscle fatigue results are obtained by performing statistical analysis on each set of original signals; the peak values ​​of the left and right hand gripping forces in the Kth set of original signals are extracted, and muscle fatigue is determined when the peak value decays to M% of the initial maximum value. M can be determined through large-sample clinical experiments and statistical analysis combined with clinical expert consensus, such as recruiting orthopedic traction patients with different conditions and physical conditions to carry out multiple sets of traction experiments, continuously monitoring the decay of the left and right hand gripping force peak values ​​over time in each set of traction, and recording the decay ratio of the gripping force peak value relative to the initial maximum value when it is clinically determined to be muscle fatigue; then, these experimental data are statistically analyzed; finally, combined with the consensus of experts in the field of orthopedic rehabilitation on the muscle fatigue judgment criteria, the specific percentage value of M is comprehensively determined to ensure that the threshold value can not only accurately reflect the physiological characteristics of muscle fatigue, but also adapt to the actual needs of clinical traction treatment. ; At the same time, the average of the peak values ​​of the surface electromyographic signals of the flexor and adductor muscle groups for the previous F times is calculated, and the fatigue thresholds of the flexor and adductor muscle groups are obtained by combining the corresponding muscle group regulation factors. Then, the fatigue status of the specific muscle groups is determined by judging whether the electromyographic signals of the two types of muscle groups are higher than the corresponding thresholds. This process can accurately distinguish between abnormal muscle confrontation caused by psychological fear and normal muscle fatigue: avoid misjudging muscle tension caused by fear as fatigue and blindly adjusting the traction load, and prevent the patient's fear from being aggravated by the continuous application of uncomfortable loads due to failure to identify real muscle fatigue, helping to more accurately adapt the traction strategy based on the fear level in the future, reducing the strategy mismatch problem caused by signal misjudgment, further alleviating the patient's fear caused by discomfort or misunderstanding, improving the treatment cooperation and compliance, and ultimately helping to improve the rehabilitation effect, making up for the defects of existing methods that do not consider psychological fear and are difficult to accurately adapt to the patient's psychology.

[0106] The efficacy of the original signal set and daily muscle fatigue results is evaluated, and the digital twin is dynamically optimized according to the biomechanical characteristics and evaluation results to obtain a dynamic twin. The scheme is optimized in combination with the reinforcement learning algorithm to obtain an optimized traction scheme.

[0107] Methods for evaluating efficacy of the raw signal set and daily muscle fatigue outcomes include:

[0108] The average fear level of the previous day and the average fear level of the day are counted to calculate the fear improvement rate; ,in, is the average fear level of the previous day; is the average fear level for the day;

[0109] The average of the peak grip strength of each group of left and right hands on the previous day and the average of the peak grip strength of the left and right hands on the current day were counted to calculate the muscle growth rate, including the muscle growth rate of the flexor group and the muscle growth rate of the adductor group; for example, the muscle growth rate of the flexor group , adductor muscle growth rate ,in, is the average peak value of the flexor group in the current training cycle; is the mean of the peak values ​​of the flexor group in the previous training cycle; is the average peak value of the adductor muscle group in the current training cycle; is the mean of the peak values ​​of the adductor muscle group in the previous training cycle;

[0110] The difference between the growth rates of the flexor group and the adductor group was calculated to obtain the muscle recovery difference index;

[0111] The traction force signal and the corresponding traction displacement are obtained and fitted by the Kelvin-Voigt viscoelastic model; ,in, It is the traction signal; for muscle and ligament stiffness; is the traction displacement; is the viscoelastic coefficient of muscles and ligaments; is the rate of change of traction displacement over time; muscle and ligament stiffness is calculated based on the Kelvin-Voigt viscoelastic model, and the muscle and ligament stiffness of the previous day and the muscle and ligament stiffness of the current day are counted to calculate the change of muscle and ligament stiffness; if the muscle and ligament stiffness changes ,in, The stiffness of muscles and ligaments on that day; is the muscle and ligament stiffness of the previous day;

[0112] The fear improvement rate, muscle growth rate, muscle group recovery difference index and muscle and ligament stiffness changes are calculated to generate an efficacy report.

[0113] The efficacy evaluation was conducted by analyzing the original signal set and daily muscle fatigue results. The fear improvement rate was calculated by calculating the average fear level of the previous day and the current day. The peak grip force values ​​of the flexor and adductor groups from the previous day and the current day were compared to obtain the muscle growth rate. The difference in muscle growth rate between the two muscle groups was calculated to obtain the muscle recovery difference index. The muscle and ligament stiffness was fitted using the Kelvin-Voigt viscoelastic model by combining the traction force and the corresponding traction displacement. The stiffness change was then compared between the previous day and the current day. Finally, these indicators were integrated to generate an efficacy report, which can clearly quantify the relationship between psychological fear improvement and physiological recovery: The fear improvement rate can be used to intuitively judge the adaptation effect of the current traction strategy to the patient's psychological state. It can also be combined with physiological indicators such as muscle growth rate and stiffness change to eliminate the misjudgment of physiological recovery caused by abnormal muscle confrontation caused by fear. At the same time, the muscle group recovery difference index can be used to identify whether fear has a tendency to affect the recovery of a certain muscle group, providing data support for subsequent targeted adjustments to the traction strategy, making up for the defects of existing methods that do not take psychological fear into consideration and are difficult to dynamically adapt to the patient's psychology, and then continuously optimizing the strategy to alleviate the patient's fear, reduce abnormal muscle confrontation, improve cooperation and compliance, and ultimately ensure the rehabilitation effect.

[0114] Methods for obtaining dynamic twins include:

[0115] The Hooke elastic model is used to correct the ligament stiffness to obtain the corrected ligament stiffness; ,in, To correct ligament stiffness; is the ligament stiffness of the previous round in the dynamic twin; is the change in ligament stiffness;

[0116] The muscle stiffness is corrected according to the muscle viscoelastic coefficient, muscle growth rate and muscle group recovery difference index to obtain the corrected muscle stiffness, such as the corrected muscle stiffness ,in, is muscle stiffness; is the muscle viscoelastic coefficient; is the muscle growth rate; It is the muscle recovery difference index;

[0117] The traction force signal is processed in time discretization to calculate the rate of change of traction displacement over time; the fitting equation of traction force signal with respect to muscle viscoelastic coefficient is established based on Kelvin-Voigt viscoelastic model, and the viscoelastic correction coefficient of muscle and ligament is obtained by least square method; if the traction force signal ,in, Correct stiffness for muscles and ligaments; is the traction displacement; is the correction factor for the viscoelasticity of muscles and ligaments; is the rate of change of traction displacement with time;

[0118] The digital twin is updated according to the corrected ligament stiffness, corrected muscle stiffness and muscle and ligament viscoelasticity correction coefficients to obtain a dynamic twin.

[0119] Ligament stiffness is corrected using the Hooke elastic model combined with ligament stiffness changes. Muscle stiffness is then corrected based on the muscle viscoelastic coefficient, muscle growth rate, and muscle group recovery difference index. The traction force signal is then discretized in time to calculate the rate of change of traction displacement. A fitting equation is established based on the Kelvin-Voigt model, and the least squares method is used to solve for the muscle and ligament viscoelastic correction coefficients. Finally, these corrected biomechanical parameters are combined to update the digital twin to generate a dynamic twin. This method can transform the abnormal muscle antagonism caused by psychological fear into quantifiable changes in muscle and ligament stiffness and viscoelasticity, which are then incorporated into the twin simulation. The dynamic twin accurately replicates the impact of fear-induced muscle tension on the traction process, avoiding the limitations of existing methods that fail to simulate real physiological states due to their lack of consideration of fear. Subsequent traction strategy simulation optimization based on this twin can predict in advance the adaptation effect of different strategies to the fear-muscle antagonism correlation state, providing a precise virtual environment support for developing traction strategies adapted to the patient's psychological state. This can mitigate the exacerbation of abnormal muscle antagonism and decreased compliance caused by mismatch between strategy and fear state, thereby improving patient compliance and rehabilitation outcomes.

[0120] Methods for obtaining an optimized traction solution include:

[0121] Define the state space: It consists of a state vector, which includes biomechanical characteristics, fear level, muscle fatigue results, fear improvement rate, muscle growth rate, muscle group recovery difference index and muscle and ligament stiffness changes;

[0122] Define the action space: It consists of action vectors, which include traction force increment, traction angle increment and traction mode;

[0123] Design reward function: The reward function is obtained by weighted calculation of safety reward, efficacy reward, and comfort reward;

[0124] Safety rewards are obtained based on the simulated maximum stress of the ligaments in the dynamic twin and the safety threshold of the human spinal ligaments; such as safety rewards ,in, is the simulated maximum stress of the ligament in the dynamic twin; It is the safety threshold of human spinal ligaments;

[0125] The efficacy reward is calculated based on the difference index between the next round of muscle group growth rate predicted by the dynamic twin and the next round of muscle group recovery predicted by the dynamic twin; such as safety reward ,in, and Both are the next round of muscle growth rate predicted by the dynamic twin; The next round of muscle group recovery difference index predicted for the dynamic twin;

[0126] The comfort bonus is calculated based on the current average fear level and the viscoelastic coefficient of the muscle group before and after correction; such as the comfort bonus ,in, is the current average fear level; and is the viscoelastic coefficient of the muscle group before and after correction;

[0127] Initialize the state vector, randomly select an action, perform traction simulation in the dynamic twin, and obtain the next round of state vector and reward function value; generate experience data containing the current state vector, selected action, reward function value, and next round of state vector, and store it in the experience replay pool;

[0128] The DQN model randomly samples W pieces of experience data from the experience replay pool, calculates the target action value function value and the current action value function value, minimizes the loss function, and updates the current network parameters through back propagation; the target action value function value is based on the reward function value and the dynamic twin body in state When traversing all possible actions , the action value function value corresponding to the action with the largest action value function value is calculated;

[0129] The trained DQN model is used to collect the patient's actual state after traction, calculate the action value function values ​​of all actions, select the action with the largest action value function value, and calculate the parameters corresponding to the optimized traction plan.

[0130] In the process of obtaining the optimized traction plan, the biomechanical characteristics, fear level, muscle fatigue results, fear improvement rate, muscle growth rate, muscle recovery difference index and muscle and ligament stiffness changes are incorporated into the state space to ensure that the optimization process fully considers the relationship between psychological fear and physiological state; the traction force increment, traction angle increment and traction mode are used as the action space to provide a specific direction for strategy adjustment; a weighted reward function including safety reward, efficacy reward and comfort reward is designed, among which the comfort reward is calculated based on the current average fear level and the viscoelastic coefficient of the muscle group before and after correction, so that the patient's psychological feelings are taken into account simultaneously when optimizing the plan, avoiding the defect of existing methods ignoring psychological factors; then, by initializing the state vector and Traction simulation is performed in the affected dynamic twin to generate experience data and store it in the experience replay pool. The DQN model is used to sample the experience data to calculate the target and current action value function values, and the loss function is minimized to update the network parameters. The model is allowed to learn the adaptation rules of psychological fear-physiological response-traction action during training. Finally, the trained DQN model is used in combination with the patient's actual state after traction to select the action with the largest action value function value to obtain the optimized parameters. The optimized traction plan generated in this way can accurately adapt to the patient's psychological fear state; it avoids the abnormal muscle confrontation caused by fear due to improper strategy, and improves the patient's cooperation by taking comfort into consideration, effectively making up for the problem that the existing method does not consider the strategy mismatch caused by psychological fear, thereby ensuring the rehabilitation effect.

[0131] Example 2:

[0132] This embodiment provides a traction correction method, including:

[0133] The initial traction force in the initial traction plan is used as the target traction force. The traction force signal of the pre-processed synchronized signal set is monitored in real time. When the following conditions are met: if the difference between the traction force signal of the pre-processed synchronized signal set and the target traction force is greater than V% of the target value, and the duration exceeds H seconds, and the fear level is not higher than level 1;

[0134] It is determined to be a traction force offset, and a telescopic adjustment instruction is output according to the offset direction.

[0135] Example 3:

[0136] See also Figure 5 As shown, this embodiment provides an intelligent orthopedic traction system, including:

[0137] Acquisition and analysis module: collects clinical data and DICOM images, performs 3D reconstruction on DICOM images, and extracts anatomical parameters; inputs clinical data and anatomical parameters into the parameter recommendation model to obtain an initial solution set;

[0138] Fear analysis module: traction is performed using the parameters in the initial solution set as the initial execution parameters. The original signal set during the traction process is collected, pre-processed, and then analyzed to obtain biomechanical characteristics. The biomechanical characteristics are input into the fear analysis model to obtain the fear level;

[0139] Fatigue analysis module: After each set of traction, statistical analysis is performed on each set of original signals to obtain muscle fatigue results;

[0140] Traction optimization module: Evaluate the efficacy of the original signal set and daily muscle fatigue results, dynamically optimize the digital twin based on the biomechanical characteristics and evaluation results to obtain a dynamic twin, and combine the reinforcement learning algorithm to optimize the solution to obtain an optimized traction solution.

[0141] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent orthopedic traction method, characterized in that: include: Collect clinical data and DICOM images, perform 3D reconstruction on DICOM images, and extract anatomical parameters; Input clinical data and anatomical parameters into the parameter recommendation model to obtain an initial solution set; The parameters in the initial solution set are used as initial execution parameters for traction, an original signal set during the traction process is collected, the original signal set is preprocessed to obtain a signal set, the signal set is analyzed to obtain biomechanical characteristics, and the biomechanical characteristics are input into a fear analysis model to obtain a fear level; After each set of traction, statistical analysis was performed on each set of original signals to obtain muscle fatigue results; The efficacy of the original signal set and daily muscle fatigue results is evaluated, and the digital twin is dynamically optimized according to the biomechanical characteristics and evaluation results to obtain a dynamic twin. The scheme is optimized in combination with the reinforcement learning algorithm to obtain an optimized traction scheme.

2. The intelligent orthopedic traction method according to claim 1, characterized in that: The signal set includes grip force signals, traction force signals and surface electromyography signals of muscle groups of left and right hands; Methods for obtaining biomechanical characteristics include: The average grip force signal of the left and right hands is taken to obtain the average grip force signal of both hands; Calculate the average value and standard deviation of the average grip force signals of both hands within a preset period, calculate the ratio of the standard deviation to the average value, and obtain the grip force fluctuation rate; Calculate the time offset corresponding to the maximum value of the cross-correlation function between the average grip force signal of both hands and the traction force signal to obtain the response phase difference; Calculate the covariance of the average grip force signal and the traction force signal of both hands, as well as the standard deviation of the traction force signal within a preset time period, and calculate the Pearson correlation coefficient between the average grip force signal and the traction force signal of both hands to obtain force coordination; When the traction force signal remains constant, the peak value of the average grip force signal of both hands is identified, and the time required for the peak value to decay to half of the peak value is calculated to obtain the relaxation delay time; The marginal probability density functions of the average grip force signals of both hands, the marginal probability density functions of the traction force signals, and the marginal probability density functions of the average grip force signals of both hands and the traction force signals are calculated to obtain the mutual information entropy; the marginal probability density functions are obtained by kernel density estimation. Perform wavelet packet decomposition on the surface electromyographic signal of the muscle group to obtain B frequency bands. Each frequency band corresponds to a different muscle group state. Calculate the energy of each frequency band and take the normalized energy of the fatigue frequency band as the frequency band energy. The biomechanical characteristics were obtained by splicing the grip force fluctuation rate, response phase difference, force coordination, relaxation delay time, mutual information entropy and frequency band energy.

3. The intelligent orthopedic traction method according to claim 2, characterized in that: Methods for achieving muscle fatigue results include: Extract the peak values ​​of the left and right hand grip force signals from the Kth group of original signals. When the peak values ​​of the left and right hand grip force signals decay to M% of the initial maximum value, it is determined to be muscle fatigue. Calculate the average of the peak values ​​of the surface electromyographic signals of the flexor muscle group during the first F times, and combine it with the flexor muscle group adjustment factor to calculate the fatigue threshold of the flexor muscle group; Calculate the average of the peak values ​​of the surface electromyographic signals of the adductor muscle group during the first F times, and combine it with the adductor muscle group regulation factor to calculate the fatigue threshold of the adductor muscle group; If the surface electromyographic signal of the flexor muscle group is not higher than the fatigue threshold of the flexor muscle group, it is determined to be flexor muscle group fatigue; If the surface electromyographic signal of the adductor muscle group is not higher than the fatigue threshold of the adductor muscle group, it is determined to be adductor muscle group fatigue.

4. The intelligent orthopedic traction method according to claim 3, characterized in that: Methods for evaluating efficacy of the raw signal set and daily muscle fatigue outcomes include: The average fear level of the previous day and the average fear level of the current day were counted to calculate the fear improvement rate; The average of the peak grip strength of the left and right hands in each group on the previous day and the average of the peak grip strength of the left and right hands on the current day were calculated to obtain the muscle growth rate, including the muscle growth rate of the flexor group and the muscle growth rate of the adductor group. The difference between the growth rates of the flexor group and the adductor group was calculated to obtain the muscle recovery difference index; Obtain the curve of the change in traction force versus traction displacement and fit it using the Kelvin-Voigt viscoelastic model. Calculate the muscle and ligament stiffness based on the Kelvin-Voigt viscoelastic model. Calculate the muscle and ligament stiffness of the previous day and the current day, and calculate the change in muscle and ligament stiffness. The fear improvement rate, muscle growth rate, muscle group recovery difference index and muscle and ligament stiffness changes are calculated to generate an efficacy report.

5. The intelligent orthopedic traction method according to claim 4, characterized in that: Methods for obtaining dynamic twins include: The Hooke elastic model was used to correct the ligament stiffness and obtain the corrected ligament stiffness; The muscle stiffness is corrected according to the muscle viscoelastic coefficient, muscle growth rate and muscle group recovery difference index to obtain the corrected muscle stiffness; The traction force signal was time discretized to calculate the rate of change of the traction displacement over time. A fitting equation for the traction force signal with respect to the muscle viscoelastic coefficient was established based on the Kelvin-Voigt viscoelastic model, and the least squares method was used to obtain the muscle and ligament viscoelastic correction coefficients. The digital twin is updated according to the corrected ligament stiffness, corrected muscle stiffness and muscle and ligament viscoelasticity correction coefficients to obtain a dynamic twin.

6. The intelligent orthopedic traction method according to claim 5, characterized in that: Methods for obtaining an optimized traction solution include: Define the state space: composed of a state vector, which includes biomechanical characteristics, fear level, muscle fatigue results, fear improvement rate, muscle growth rate, muscle group recovery difference index and stiffness change; Define the action space: It consists of action vectors, which include traction force increment, traction angle increment and traction mode; Design reward function: The reward function is obtained by weighted calculation of safety reward, efficacy reward, and comfort reward; Initialize the state vector, randomly select an action vector, perform traction simulation in the dynamic twin, and obtain the next round of state vector and reward function value; generate experience data containing the current state vector, the selected action vector, the reward function value, and the next round of state vector, and store it in the experience replay pool; The DQN model randomly samples W pieces of experience data from the experience replay pool, calculates the target action value function value and the current action value function value, minimizes the loss function, and updates the current network parameters through back propagation; the target action value function value is based on the reward function value and the dynamic twin body in state When traversing all action vectors The action value function value corresponding to the action vector that maximizes the action value function value is calculated; The trained DQN model is used to collect the patient's state vector after actual traction, calculate the action value function values ​​of all action vectors, select the action vector with the largest action value function value, and calculate the parameters corresponding to the optimized traction plan.

7. The intelligent orthopedic traction method according to claim 6, characterized in that: In the reward function, a safety reward is calculated based on the simulated maximum stress of the ligament in the dynamic twin and the safety threshold of the human spinal ligament; The therapeutic effect reward is calculated based on the difference index between the next round of muscle group growth rate predicted by the dynamic twin and the next round of muscle group recovery predicted by the dynamic twin; The comfort bonus is calculated based on the current average fear level and the viscoelastic coefficient of the muscle group before and after correction.

8. The intelligent orthopedic traction method according to claim 7, characterized in that: The clinical data included diagnosis, age, weight and medical history.

9. The intelligent orthopedic traction method according to claim 8, characterized in that: Methods for obtaining anatomical parameters include: Each DICOM image is traversed sequentially in the order of CT scans, and a three-dimensional voxel grid is constructed based on the size and resolution of the DICOM image; Traverse the entire 3D voxel grid voxel by voxel; for each voxel, read the grayscale value of its eight vertices and compare it with the pre-determined bone grayscale threshold to determine the vertex state; According to the voxel vertex state, the corresponding isosurface topology structure is searched in the topology table to generate the corresponding triangular patch model; After generating the preliminary triangular face model, it is optimized to obtain a three-dimensional reconstructed image; The 3D reconstructed image is used as the input of the convolutional neural network to obtain a feature map, which is then used as the input of the segmentation model to obtain anatomical parameters; the anatomical parameters include intervertebral disc height, spinal Cobb angle, facet joint spacing, and the coordinates of the anatomical attachment points of the adductor and flexor muscles.

10. The intelligent orthopedic traction method according to claim 9, characterized in that: The initial program set includes an initial traction program and an initial upper limb exercise program. The initial traction program includes initial traction force, traction angle, traction duration and traction mode; the initial upper limb exercise program includes exercise mode, initial resistance and exercise parameters.

11. An intelligent orthopedic traction system, implementing an intelligent orthopedic traction method according to any one of claims 1 to 10, characterized in that: include: Acquisition and analysis module: collect clinical data and DICOM images, perform 3D reconstruction of DICOM images, and extract anatomical parameters; Input clinical data and anatomical parameters into the parameter recommendation model to obtain an initial solution set; Fear analysis module: traction is performed using the parameters in the initial solution set as the initial execution parameters. The original signal set during the traction process is collected, pre-processed, and then analyzed to obtain biomechanical characteristics. The biomechanical characteristics are input into the fear analysis model to obtain the fear level; Fatigue analysis module: After each set of traction, statistical analysis is performed on each set of original signals to obtain muscle fatigue results; Traction optimization module: Evaluate the efficacy of the original signal set and daily muscle fatigue results, dynamically optimize the digital twin based on the biomechanical characteristics and evaluation results to obtain a dynamic twin, and combine the reinforcement learning algorithm to optimize the solution to obtain an optimized traction solution.

12. An intelligent orthopedic traction device, implementing the intelligent orthopedic traction method according to any one of claims 1 to 10.

Images