Method and system for evaluating symmetry of walking rehabilitation based on plantar pressure distribution

Abstract

The application relates to the technical field of gait analysis, and discloses a walking rehabilitation symmetry evaluation method and system based on plantar pressure distribution. The method trains a plantar pressure reference model according to historical gait data, collects real-time walking data of a patient to obtain a plantar pressure distribution thermogram and pressure distribution characteristic parameters, carries out fuzzy inference processing on the thermogram and the characteristic parameters, divides plantar pressure areas and screens out effective evaluation areas, obtains posture data and kinematic chain joint data of a diabetic foot patient, calculates the soft probability of high and low shoulders based on a mixed Gaussian model, quantizes the equivalent moment interference quantity of high and low shoulder postures on the plantar pressure distribution, combines the original asymmetric moment of the plantar pressure and the equivalent moment interference quantity, calculates a walking rehabilitation symmetry index, and comprehensively evaluates the walking symmetry of the patient. The application distinguishes the influence of posture compensation and foot pathological changes on the plantar pressure, and improves the accuracy and reliability of the walking rehabilitation symmetry evaluation of the diabetic foot patient.

Description

Technical Field

[0001] This invention relates to the field of gait analysis technology, and in particular to a method and system for assessing the symmetry of walking rehabilitation based on plantar pressure distribution. Background Technology

[0002] Assessing gait symmetry in rehabilitation is a crucial step in monitoring functional recovery in diabetic foot patients, as plantar pressure distribution analysis objectively reflects lower limb weight-bearing and motor control characteristics. However, existing gait analysis methods often focus on local foot parameters, such as peak pressure, pressure center trajectory, and correlation coefficients between left and right foot pressure distributions, neglecting the holistic characteristics of the human body as a complete kinetic chain. Current clinical assessment methods primarily utilize data acquisition systems based on pressure plates, gait mats, or wearable insoles, combined with various symmetry indices for quantitative judgment. However, these techniques have the following limitations: First, abnormal plantar pressure can originate from pathological changes in the foot itself, or indirectly caused by postural compensations such as pelvic tilt and scoliosis in the upper kinetic chain; existing methods fail to effectively distinguish between these two causes. Second, the impact of uneven shoulders on center of gravity distribution and lower limb load transmission lacks quantitative description, leading to misjudgments of pressure shifts originating from postural compensation as worsening of foot injury, resulting in false-positive assessments. Therefore, there is an urgent need for a gait rehabilitation symmetry assessment method and system based on plantar pressure distribution to improve the accuracy and clinical applicability of gait symmetry assessment in diabetic foot patients. Summary of the Invention

[0003] To overcome the shortcomings of traditional diabetic foot risk assessment, which fails to distinguish between postural compensation due to uneven shoulders and pathological changes in the foot, resulting in misjudgment of pressure abnormalities, this invention provides a method and system for assessing the symmetry of walking rehabilitation based on plantar pressure distribution.

[0004] The technical implementation scheme of the present invention is as follows: a symmetry assessment method for walking rehabilitation based on plantar pressure distribution, comprising the following steps:

[0005] S1: Train a plantar pressure benchmark model based on historical gait data, collect real-time walking data of patients, and use the plantar pressure benchmark model to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters;

[0006] S2: Process the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and screen effective plantar assessment regions;

[0007] S3: Obtain postural data and kinetic chain joint data of diabetic foot patients, and calculate the probability of uneven shoulder type based on the postural data;

[0008] S4: Quantify the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data;

[0009] S5: Assess the symmetry of walking rehabilitation in diabetic foot patients based on the equivalent moment interference and the effective assessment area of ​​the foot.

[0010] Preferably, the step of training a plantar pressure benchmark model based on historical gait data and collecting real-time walking data from patients to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters using the plantar pressure benchmark model includes: the historical gait data includes spatiotemporal sequence data of plantar pressure, gait cycle phase data, and weight-normalized data of healthy individuals; after preprocessing the historical gait data to construct a dataset, a plantar pressure benchmark model is trained through a variational autoencoder network; real-time walking data from patients is collected; and a plantar pressure distribution heatmap and pressure distribution characteristic parameters are obtained based on the real-time walking data using the plantar pressure benchmark model. The pressure distribution characteristic parameters include peak pressure in the plantar region, pressure time integral, pressure center trajectory offset, and correlation coefficient between left and right foot pressure distribution.

[0011] Preferably, the step of processing the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and screening effective plantar assessment regions includes: calculating the pressure concentration index and pressure offset direction angle of the plantar region based on the pressure distribution characteristic parameters; processing the plantar pressure distribution heatmap through fuzzy reasoning based on the pressure concentration index and pressure offset direction angle to obtain plantar pressure regions, corresponding pressure anomaly factors and durations; determining regions where the pressure anomaly factor is less than the pressure anomaly threshold and the duration is greater than the duration threshold as effective plantar assessment regions, thereby obtaining a set of effective plantar assessment regions.

[0012] Preferably, the acquisition of postural data and kinetic chain joint data of diabetic foot patients, and the calculation of the soft probability of uneven shoulder type based on the postural data, includes: the postural data includes bilateral acromion height difference, trunk tilt angle, pelvic tilt angle, cervical spine deviation angle, root mean square value of surface electromyography of the upper trapezius muscles bilaterally, passive range of motion of both shoulder joints, and shoulder and neck pain score; after standardizing the postural data, a postural feature vector is constructed; the uneven shoulder type includes functional uneven shoulder, structural uneven shoulder, and pain-compensating uneven shoulder; and the total number of uneven shoulder types is statistically obtained. Using a historical patient posture dataset labeled with physician-defined shoulder types, a Gaussian mixture model was used to train and fit the standard posture feature mean vectors and standard posture feature covariance matrices for different shoulder types. A support vector machine recursive feature elimination method was then employed for training to obtain the importance coefficient of each posture indicator in distinguishing different types of shoulder unevenness. The importance coefficients were normalized and used as weight vectors for different shoulder types. The baseline probability of shoulder unevenness was statistically obtained, and the soft probability of shoulder unevenness in diabetic foot patients was calculated by combining the posture feature vectors.

[0013] Preferably, the step of calculating the soft probability of uneven shoulder type in diabetic foot patients by combining body posture feature vectors includes:

[0014] The soft probability formula for uneven shoulders is:

[0015] ;

[0016] In the formula, Given a body feature vector High and low shoulder type soft probability, Uneven shoulders The transpose of the weight vector, For body shape feature vectors, Uneven shoulders The basic probability of occurrence. Uneven shoulders The standard body posture feature mean vector, Uneven shoulders The inverse matrix of the standard body shape feature covariance matrix. for transpose, This represents the total number of uneven shoulders. Uneven shoulders The transpose of the weight vector, Uneven shoulders The basic probability of occurrence. Uneven shoulders The standard body posture feature mean vector, Uneven shoulders The inverse matrix of the standard body shape feature covariance matrix.

[0017] Preferably, the step of quantifying the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data includes: the kinetic chain joint data includes the number of compensating joints in the kinetic chain, the actual range of motion of each joint and the normal reference range of motion of each joint, obtaining the patient's weight, and calculating the equivalent asymmetric torque by combining the kinetic chain joint data, the difference in acromion height between the two sides and the trunk tilt angle.

[0018] Preferably, the step of calculating the equivalent asymmetric torque by combining the kinetic chain joint data, the bilateral acromion height difference, and the trunk tilt angle includes:

[0019] The formula for the equivalent asymmetric torque is:

[0020] ;

[0021] In the formula, For equivalent asymmetric torque, For the patient's weight, It is the acceleration due to gravity. The difference in height between the two acromions. The angle of trunk tilt. This represents the total number of uneven shoulders. Given a body feature vector High and low shoulder type soft probability, Uneven shoulders The conduction gain coefficient, For kinetic chain conduction;

[0022] The formula for the kinetic chain conduction term is:

[0023] ;

[0024] In the formula, For kinetic chain conduction, The number of joints used to compensate for kinetic chain defects. For the first The compensatory weight of each joint For the first Normal reference range of motion for each joint. For the first The actual range of motion of each joint.

[0025] Preferably, the step of assessing the symmetry of gait rehabilitation for diabetic foot patients based on the equivalent torque interference and the effective plantar assessment area includes: obtaining the gait cycle duration based on the patient's real-time gait data; obtaining the actual pressure of the effective plantar assessment area of ​​the left foot and the actual pressure of the effective plantar assessment area of ​​the right foot based on the plantar pressure distribution heatmap and the set of effective plantar assessment areas, and determining the sign function of the actual offset direction of the plantar pressure center; obtaining the patient's foot length to calculate the original asymmetric torque of plantar pressure; continuously statistically analyzing the original asymmetric torque of plantar pressure for multiple gait cycles; calculating the periodic standard deviation of the original asymmetric torque of plantar pressure; determining the sign function of the horizontal offset direction of the center of gravity caused by the difference in acromion height between the two sides; calculating the gait rehabilitation symmetry index by combining the original asymmetric torque of plantar pressure and the equivalent asymmetric torque; and evaluating the gait rehabilitation symmetry of the patient based on the gait rehabilitation symmetry index.

[0026] The formula for the original asymmetric moment of plantar pressure is:

[0027] ;

[0028] In the formula, The original asymmetric torque of plantar pressure, For the patient's foot length, The duration of the gait cycle. For the first Risk weights for each effective assessment area of ​​the sole of the foot. A set of effective assessment areas for the soles of the feet. For the left foot Each effective assessment area of ​​the sole of the foot is located in The real pressure at all times It is a fractional exponent. For the right foot Each effective assessment area of ​​the sole of the foot is located in The actual pressure at all times It is the integral variable.

[0029] Preferably, the calculation of the gait rehabilitation symmetry index by combining the original asymmetric torque and the equivalent asymmetric torque of plantar pressure includes:

[0030] The formula for the symmetry index of walking rehabilitation is:

[0031] ;

[0032] In the formula, The symmetry index for walking rehabilitation. The original asymmetric torque of plantar pressure, For equivalent asymmetric torque, The sign function for the horizontal shift of the center of gravity caused by uneven shoulders. The sign function represents the actual offset direction of the plantar pressure center. To adjust the index, This is the scaling factor. The periodic standard deviation of the original asymmetric moment of plantar pressure. It is a very small positive number. This is the sensitivity coefficient. The duration of the left foot's stance phase. The duration of the right foot's support phase.

[0033] Preferably, the symmetry assessment system for gait rehabilitation based on plantar pressure distribution further includes:

[0034] The data acquisition module trains a plantar pressure benchmark model based on historical gait data, collects real-time walking data of patients, and uses the plantar pressure benchmark model to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters.

[0035] The region filtering module processes the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and filters effective plantar assessment regions.

[0036] The probability calculation module acquires the body posture data and kinetic chain joint data of diabetic foot patients, and calculates the soft probability of the patient's uneven shoulder type based on the body posture data.

[0037] The interference calculation module quantifies the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data.

[0038] The symmetry assessment module assesses the symmetry of walking rehabilitation in diabetic foot patients based on the equivalent moment interference and the effective assessment area of ​​the sole.

[0039] The beneficial effects of this invention are:

[0040] This invention breaks through the limitations of traditional gait analysis, which only focuses on local parameters of the foot. It organically integrates the identification of uneven shoulder posture type with the assessment of plantar pressure distribution. By collecting multi-dimensional posture data such as bilateral acromion height difference, trunk lateral tilt angle, and pelvic tilt angle, and combining them with a Gaussian mixture model to calculate the soft probability of functional, structural, and pain-compensating uneven shoulders, it can accurately distinguish whether the abnormal plantar pressure is caused by foot disease itself or upper kinetic chain postural compensation, thus improving the assessment bias caused by etiological confusion.

[0041] This invention constructs a quantitative calculation model for the equivalent interference torque of shoulder height difference on plantar pressure. Based on the soft probability of posture, the degree of joint compensation in the kinetic chain, and the conduction gain coefficient, the formula for the equivalent asymmetric torque is derived. The center of gravity shift caused by trunk tilt and shoulder height difference is transformed into a measurable mechanical interference quantity, which enables the subsequent symmetry assessment to effectively eliminate the pseudo-asymmetric components of posture compensation, thereby improving the objectivity and clinical credibility of the assessment results.

[0042] This invention designs a gait rehabilitation symmetry index that includes a fractional-order index and a dynamic adjustment factor, and introduces a screening process for effective plantar assessment areas. Through fuzzy reasoning, transient or low-risk pressure areas are eliminated, while areas with persistent abnormalities and high risk weights are retained for calculation. At the same time, the standard deviation of the cycle of multiple gait cycles and the difference in the duration of the support phase are combined to comprehensively reflect the patient's gait stability and symmetry. Attached Figure Description

[0043] Figure 1 This is a flowchart of the symmetry assessment method for walking rehabilitation based on plantar pressure distribution according to the present invention.

[0044] Figure 2 This is a structural diagram of the symmetry assessment system for walking rehabilitation based on plantar pressure distribution, as described in this invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1: A method for assessing the symmetry of gait rehabilitation based on plantar pressure distribution, such as... Figure 1 As shown, it includes the following steps:

[0047] S1: Train a plantar pressure benchmark model based on historical gait data, collect real-time walking data of patients, and use the plantar pressure benchmark model to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters;

[0048] Historical gait data includes spatiotemporal sequence data of plantar pressure, gait cycle phase data, and weight-normalized data of healthy individuals. After preprocessing the historical gait data to construct a dataset, a plantar pressure benchmark model is trained through a variational autoencoder network. Real-time walking data of patients is collected, and plantar pressure distribution heatmaps and pressure distribution characteristic parameters are obtained from the plantar pressure benchmark model based on the real-time walking data. The pressure distribution characteristic parameters include peak pressure in the plantar region, pressure time integral, pressure center trajectory offset, and correlation coefficient between left and right foot pressure distribution.

[0049] It should be further explained that the historical gait data used for training comes from plantar pressure information collected from healthy individuals during normal walking. Specifically, it includes three categories: spatiotemporal sequence data of plantar pressure, gait cycle phase data, and weight-normalized data. The spatiotemporal sequence data of plantar pressure records the pressure changes in each plantar region at different times during walking in healthy individuals. The gait cycle phase data is used to label the standing and swing phase intervals within each gait cycle. The weight-normalized data is used to eliminate the influence of individual weight differences on the absolute pressure value, thus enabling the model to learn a pressure distribution pattern independent of body weight. After collecting the above historical gait data, preprocessing operations are performed. Preprocessing includes data cleaning to remove noise points and missing values ​​from the collection process, time series alignment to ensure the comparability of time points with the same phase in different gait cycles, and data augmentation to expand the diversity of training samples. The preprocessed historical gait data is used to construct the training dataset, and a variational autoencoder network is used for model training. The variational autoencoder network consists of an encoder and a decoder. The encoder maps the input spatiotemporal sequence of plantar pressure to probability distribution parameters in the latent space, while the decoder reconstructs the original pressure distribution from the latent variables. During training, by minimizing the reconstruction error and the KL divergence between the latent space distribution and the standard normal distribution, the network can effectively learn the latent patterns and typical characteristics of plantar pressure distribution in healthy individuals. After training, the network parameters are saved to obtain a baseline model of plantar pressure.

[0050] Real-time plantar pressure data was collected from patients walking on a force-measuring platform with a natural gait. This collected real-time walking data was input into a pre-trained plantar pressure baseline model. The model extracted feature representations of the real-time data in the latent space based on its encoder, then reconstructed the pressure distribution using a decoder and output a corresponding plantar pressure distribution heatmap. The model extracted four types of pressure distribution feature parameters from the real-time data. The first type is the peak pressure of the plantar region, i.e., the maximum pressure value reached by each segment of the plantar region within a complete gait cycle. The second type is the pressure-time integral, representing the total accumulated pressure of each plantar region during the support phase, reflecting the total load borne by that region throughout the support phase. The third type is the pressure center trajectory offset, i.e., the degree of deviation of the trajectory of the point of application of the resultant plantar pressure force during gait from the healthy baseline position. The fourth type is the correlation coefficient of the left and right foot pressure distribution, obtained by calculating the correlation between the pressure value sequences of corresponding regions of the left and right feet; the closer this coefficient is to 1, the more symmetrical the pressure distribution of both feet. The heatmap and the four types of feature parameters together form the input basis for subsequent evaluation processes.

[0051] S2: Process the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and screen effective plantar assessment regions;

[0052] The pressure concentration index and pressure offset direction angle of the plantar region are calculated based on the pressure distribution characteristic parameters. The plantar pressure distribution heat map is processed by fuzzy reasoning based on the pressure concentration index and pressure offset direction angle to obtain the plantar pressure region, the corresponding pressure anomaly factor and duration. The region with pressure anomaly factor less than the pressure anomaly threshold and duration greater than the duration threshold is determined as the effective plantar assessment region, and the set of effective plantar assessment regions is obtained.

[0053] It needs further explanation that two intermediate indices were calculated based on the obtained pressure distribution characteristic parameters. The first intermediate indices is the pressure concentration index of the plantar region. This index characterizes the degree of concentration of pressure distribution in each divided area of ​​the plantar region. It is calculated by statistically processing the pressure values ​​of each sampling point in the region to obtain the variance and kurtosis coefficient of the pressure values. The larger the variance, the higher the dispersion of pressure in the region. The kurtosis coefficient reflects the sharpness of the pressure distribution curve. The pressure concentration index of the region is obtained by weighting and combining the variance and kurtosis coefficient. Principal components are extracted from the variance and kurtosis coefficients calculated for all regions in the historical gait samples. The loading coefficients of each index in the first principal component are normalized and used as weights. The higher the index value, the more concentrated the pressure is in a local location within the region. The second intermediate indices are the pressure offset direction angle. This angle represents the offset direction of the plantar pressure center on the horizontal plane relative to the geometric center line of the foot. Specifically, it is calculated by calculating the horizontal and vertical coordinate offset of the pressure center point, and then using inverse trigonometric functions to obtain the angle between the offset direction and the long axis of the foot.

[0054] After obtaining the two intermediate indicators mentioned above, the plantar pressure distribution heatmap is partitioned using fuzzy inference. The first step of fuzzy inference is to map the pressure concentration index and pressure offset direction angle as input variables to pre-defined fuzzy sets. The pressure concentration index corresponds to three fuzzy subsets: small, medium, and large; the pressure offset direction angle corresponds to three fuzzy subsets: negative, zero, and positive. The second step is to perform inference based on a fuzzy rule base developed by experts. The fuzzy rule base contains nine inference rules. Rule 1: If the pressure concentration index is small and the pressure offset direction angle is negative, the output plantar pressure region is the lateral hindfoot region, and the pressure anomaly factor is low. Rule 2: If the pressure concentration index is small and the pressure offset direction angle is zero, the output plantar pressure region is the uniform region of the entire foot, and the pressure anomaly factor is low. Rule 3: If the pressure concentration index is small and the pressure offset direction angle is positive, the output plantar pressure region is the medial hindfoot region, and the pressure anomaly factor is low. Rule 4: If the pressure concentration index is medium and the pressure offset angle is negative, the output plantar pressure area is the lateral midfoot region, and the pressure anomaly factor is medium. Rule 5: If the pressure concentration index is medium and the pressure offset angle is zero, the output plantar pressure area is the midfoot arch region, and the pressure anomaly factor is medium. Rule 6: If the pressure concentration index is medium and the pressure offset angle is positive, the output plantar pressure area is the medial midfoot region, and the pressure anomaly factor is medium. Rule 7: If the pressure concentration index is high and the pressure offset angle is negative, the output plantar pressure area is the lateral forefoot region, and the pressure anomaly factor is high. Rule 8: If the pressure concentration index is high and the pressure offset angle is zero, the output plantar pressure area is the central forefoot region, and the pressure anomaly factor is high. Rule 9: If the pressure concentration index is high and the pressure offset angle is positive, the output plantar pressure area is the medial forefoot region, and the pressure anomaly factor is high. The output confidence scores of all activated rules out of the nine rules are synthesized and defuzzified. The synthesis operation uses a weighted average method, summing the output pressure anomaly factors using the output confidence scores of each rule as weights to obtain the comprehensive pressure anomaly factor value for each plantar pressure region. The defuzzification operation converts the comprehensive pressure anomaly factor into a clear output value, and the duration is determined based on the number of cycles in which the region continuously satisfies the above rules during a continuous gait cycle.

[0055] After completing the division of plantar pressure zones and assigning values ​​to the pressure abnormality factors and durations for each zone, the process proceeds to the screening of effective assessment zones. The screening process employs a dual-condition joint judgment strategy. The first condition is that the pressure abnormality factor is less than a pre-set pressure abnormality threshold; the second condition is that the duration is greater than a pre-set duration threshold. For each plantar pressure zone, both conditions are compared simultaneously, and only zones that meet both conditions are considered effective plantar pressure assessment zones. After completing the above screening for all plantar pressure zones, the selected zones are merged to form a set of effective plantar pressure assessment zones. This set serves as the basis for calculating the gait rehabilitation symmetry index in subsequent step S5. The method for setting the pressure abnormality threshold is as follows: historical gait data from a large number of healthy individuals and diabetic foot patients are collected. The pressure abnormality factor for each plantar region of each sample is calculated, and the corresponding clinical gold standard judgment result for that region is obtained simultaneously. For example, a label indicating normal or abnormal pressure distribution is given by a professional physician based on imaging examinations, foot skin condition, and ulcer risk assessment. Using abnormal pressure factors as diagnostic indicators and the clinical gold standard as a reference variable, receiver operating characteristic (ROC) curves were plotted. The abnormal pressure factor value corresponding to the maximum Youden index was selected as the abnormal pressure threshold. The method for setting the persistence threshold is as follows: plantar pressure data of healthy individuals were collected for multiple consecutive gait cycles under normal walking conditions. The number of cycles in which the abnormal pressure state persisted in each region was calculated, and the distribution pattern of these persistent cycles was statistically analyzed. The 95th percentile of this distribution was used as the persistence threshold.

[0056] S3: Obtain postural data and kinetic chain joint data of diabetic foot patients, and calculate the probability of uneven shoulder type based on the postural data;

[0057] Postural data includes bilateral acromion height difference, trunk tilt angle, pelvic tilt angle, cervical spine deviation angle, root mean square value of surface electromyography of the upper trapezius muscles bilaterally, passive range of motion of both shoulder joints, and shoulder and neck pain scores. After standardization of the postural data, postural feature vectors are constructed. The uneven shoulder types include functional uneven shoulder, structural uneven shoulder, and pain-compensated uneven shoulder. The total number of uneven shoulder types is statistically obtained. Using a historical patient postural dataset labeled with uneven shoulder types by physicians, a Gaussian mixture model is used to train and fit the standard postural feature mean vectors and standard postural feature covariance matrices for different uneven shoulder types. Support vector machine recursive feature elimination is used for training to obtain the importance coefficient of each postural indicator for distinguishing various types of uneven shoulder. The importance coefficients are normalized and used as weight vectors for different uneven shoulder types. The baseline probability of uneven shoulder types is statistically obtained, and the soft probability of uneven shoulder type in diabetic foot patients is calculated by combining the postural feature vectors.

[0058] The soft probability formula for uneven shoulders is:

[0059] ;

[0060] In the formula, Given a body feature vector High and low shoulder type soft probability, Uneven shoulders The transpose of the weight vector, For body shape feature vectors, Uneven shoulders The basic probability of occurrence. Uneven shoulders The standard body posture feature mean vector, Uneven shoulders The inverse matrix of the standard body shape feature covariance matrix. for transpose, This represents the total number of uneven shoulders. Uneven shoulders The transpose of the weight vector, Uneven shoulders The basic probability of occurrence. Uneven shoulders The standard body posture feature mean vector, Uneven shoulders The inverse matrix of the standard body shape feature covariance matrix.

[0061] It should be further explained that postural data and kinetic chain joint data of diabetic foot patients were acquired. Postural data was used for subsequent soft probability calculations, while kinetic chain joint data was reserved for quantification of equivalent torque interference in step S4. Postural data collection was completed simultaneously using a surface electromyography (EMG) instrument and a joint motion measurement goniometer, specifically including seven indicators: The first is the bilateral acromion height difference, i.e., the vertical height difference between the highest points of the left and right acromions relative to the horizontal plane when standing, reflecting the overall asymmetry of the shoulders; the second is the trunk tilt angle, i.e., the angle between the trunk midline and the vertical line of gravity in the coronal plane, characterizing the degree of trunk tilt to the left or right; the third is the pelvic tilt angle, i.e. The angle between the line connecting the highest points of the left and right iliac crests of the pelvis and the horizontal plane reflects the pelvic tilt state; the fourth item is the cervical spine lateral deviation angle, that is, the angle between the cervical spine midline and the trunk midline in the coronal plane, reflecting the degree of lateral flexion of the neck; the fifth item is the root mean square value of electromyography on the surface of the upper trapezius muscles on both sides, which is obtained by collecting the electrical signals of muscle activity through electrode pads attached to the surface of the upper trapezius muscles and calculating the root mean square value, reflecting the tension of the muscles in a static state; the sixth item is the passive range of motion of the shoulder joints, that is, the maximum range of motion of the joints measured by the examiner when passively moving the patient's shoulder joints; the seventh item is the shoulder and neck pain score, which is subjectively rated by the patient according to the visual analog scale.

[0062] After collecting the above seven raw body posture data, the data were standardized because the dimensions and numerical ranges of each indicator are different. The standardization process adopted the Z-score method, that is, for each body posture indicator, its mean and standard deviation were calculated for all collected samples, and then the mean of the indicator was subtracted from the raw value of each sample and then divided by the standard deviation, so that each standardized indicator follows a standard normal distribution with a mean of zero and a standard deviation of 1. The seven standardized body posture indicators are arranged in order to form a seven-dimensional body posture feature vector, which fully reflects the patient's body posture asymmetry characteristics in the coronal plane.

[0063] Uneven shoulders are classified into three categories: functional uneven shoulders, structural uneven shoulders, and pain-compensating uneven shoulders. Functional uneven shoulders are usually caused by long-term poor posture or muscle imbalances, with no abnormalities in the skeletal structure itself. Structural uneven shoulders originate from skeletal morphological changes such as scoliosis or pelvic tilt, and have a fixed anatomical basis. Pain-compensating uneven shoulders are caused by protective postural compensation due to pain in the shoulder, neck, or upper limbs. To calculate the soft probability of a patient belonging to each type of uneven shoulders, a classification model was pre-established. The dataset used for model training came from historical patient posture data explicitly labeled with uneven shoulder types by physicians. During training, a Gaussian mixture model was used for fitting. The Gaussian mixture model assumes that the posture feature vector of each type of uneven shoulders follows a multivariate Gaussian distribution, and the distribution parameters of each category are estimated iteratively using the expectation-maximization algorithm. After fitting, the following are obtained: the mean vector of standard postural features for each type of uneven shoulders. This vector has the same dimension as the postural feature vector and represents the center of postural features of typical patients in this type. The covariance matrix of standard postural features for each type of uneven shoulders is also obtained. This matrix is ​​a 7x7 symmetric matrix, with diagonal elements reflecting the dispersion of each postural indicator within the class and off-diagonal elements reflecting the correlation between indicators. Using a historical patient postural dataset labeled with uneven shoulder types by physicians, a support vector machine (SVM) recursive feature elimination method is employed for training to obtain the importance coefficient of each postural indicator for distinguishing different types of uneven shoulders. The importance coefficients are normalized and used as weight vectors. A linear kernel SVM is then used to train all postural indicators to obtain the following results. The absolute value of the weight coefficient for each indicator is used as the initial importance score, and the current cross-validation accuracy is recorded. Then, a recursive elimination process begins. After each training round, the posture indicator with the smallest absolute value of its weight coefficient is removed. Before removal, the absolute value of the weight coefficients of each remaining indicator is recorded. This absolute value is multiplied by the reciprocal of the total number of remaining indicators and added to the cumulative importance score of the corresponding indicator. Then, the support vector machine is retrained using the remaining indicators, and the weight coefficients are updated. This process is repeated, recording the cross-validation accuracy in each round. The recursion terminates when the cross-validation accuracy shows its first decrease compared to the previous round or fails to improve for three consecutive rounds. The subset of indicators retained at this point is the optimal feature subset. For all posture indicators that have been retained at least once during the recursive process, their final cumulative importance score is used as the original importance coefficient. The original importance coefficients of all indicators are divided by their sum to obtain a normalized weight vector, ensuring that the sum of all weights equals one. Based on the sample proportions of various types of uneven shoulders in historical datasets, the basic probability of occurrence of each type of uneven shoulders is statistically obtained. This probability reflects the prior likelihood of various types of uneven shoulders appearing in the population without observing specific body characteristics.

[0064] Based on the parameters obtained from the above training and the current patient's body shape feature vector, the soft probability of the current patient belonging to each type of uneven shoulders is calculated using the Bayesian posterior probability formula. The base probability in the numerator represents the prior proportion of that type in the population. For each type of uneven shoulders, the dot product of the weight vector of that type and the body shape feature vector is calculated. This is used to linearly weight different body shape indicators, reflecting the bias direction of each type of uneven shoulders in the feature space. The negative half of the squared Mahalanobis distance is used to measure the similarity between the body shape feature vector and the standard mean vector of that type. The Mahalanobis distance normalizes the variance and correlation of each indicator through the inverse of the covariance matrix, so that indicators with different dimensions and different correlations can be compared on a uniform scale. Finally, the denominator is normalized so that the sum of the soft probabilities of all categories equals one, thus obtaining the soft probability of the patient belonging to each type of uneven shoulders given the body shape features.

[0065] S4: Quantify the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data;

[0066] The kinetic chain joint data includes the number of compensatory joints in the kinetic chain, the actual range of motion of each joint, and the normal reference range of motion of each joint. The patient's weight is obtained, and the equivalent asymmetric torque is calculated by combining the kinetic chain joint data, the difference in acromion height between the two sides, and the trunk tilt angle.

[0067] The formula for the equivalent asymmetric torque is:

[0068] ;

[0069] In the formula, For equivalent asymmetric torque, For the patient's weight, It is the acceleration due to gravity. The difference in height between the two acromions. The angle of trunk tilt. This represents the total number of uneven shoulders. Given a body feature vector High and low shoulder type soft probability, Uneven shoulders The conduction gain coefficient, For kinetic chain conduction;

[0070] The formula for the kinetic chain conduction term is:

[0071] ;

[0072] In the formula, For kinetic chain conduction, The number of joints used to compensate for kinetic chain defects. For the first The compensatory weight of each joint For the first Normal reference range of motion for each joint. For the first The actual range of motion of each joint.

[0073] It should be further explained that after obtaining the soft probabilities of patients belonging to three types of uneven shoulder—functional uneven shoulder, structural uneven shoulder, and pain-compensated uneven shoulder—the equivalent torque interference quantity generated by uneven shoulder posture on plantar pressure distribution was quantified based on these soft probabilities and previously acquired kinetic chain joint data. The physical meaning of this equivalent torque interference quantity is as follows: the difference in acromion height caused by uneven shoulders induces lateral tilting of the trunk in the coronal plane. This tilted trunk causes the body's center of gravity to shift to one side. This shift in the center of gravity, through compensatory transmission in the kinetic chain joints, ultimately alters the left-right distribution of plantar pressure. The final effect of this series of mechanical transmissions can be equivalent to an asymmetric torque acting on the plantar surface. The kinetic chain joint data specifically includes three sub-items: the number of compensatory joints in the kinetic chain, the actual range of motion of each joint, and the normal reference range of motion of each joint. The number of compensatory joints in the kinetic chain refers to the total number of joints along the kinetic chain path from the shoulder to the foot that cause compensatory postural adjustments due to uneven shoulders. These joints typically include the cervical, thoracic, lumbar, hip, knee, and ankle joints, with the specific number determined by clinical physical examination. The actual range of motion of each joint refers to the range of angles that each compensatory joint can actually achieve in the current posture, measured using a goniometer or inertial measurement unit during the patient's standing and walking positions. The normal reference range of motion for each joint refers to the standard reference values ​​for the corresponding joint range of motion in healthy individuals of the same age and sex, obtained from anatomical or rehabilitation medicine reference data.

[0074] After obtaining the above data, the equivalent asymmetric moment is calculated by combining the patient's weight, the difference in acromion height between the two sides, and the trunk tilt angle. The formula for the equivalent asymmetric moment is constructed based on the fundamental principles of mechanics. The product of the patient's weight and the acceleration due to gravity constitutes the gravity term, and the product of the difference in acromion height between the two sides and the tangent of the trunk tilt angle constitutes the horizontal lever arm term. Multiplying the two together yields the original gravitational overturning moment caused by trunk tilt. The weighted summation term within square brackets in the formula is a weighted combination of three types of soft probabilities of shoulder height and unevenness and their corresponding conduction gain coefficients. This is used to correct for the differences in plantar pressure transmission efficiency among different body posture types. The higher the soft probability and the larger the conduction gain coefficient, the higher the weight of the actual influence of that body posture on plantar pressure. The kinetic chain transmission term summarizes the attenuation effect of each compensating joint on torque transmission through a multiplication. The attenuation factor of each joint is determined by the degree of loss of the actual range of motion relative to the normal reference range. The more severe the loss of the range of motion, the smaller the attenuation factor. At the same time, it is multiplied by the compensation weight of the joint to reflect the difference in importance of different joints in the transmission path. The total transmission efficiency is obtained by multiplying the attenuation factors of all joints. The physical meaning of the entire equivalent asymmetric torque formula is that it starts from the asymmetry of the shoulder, is transmitted step by step along the kinetic chain, and after being corrected by postural type and attenuated by joint compensation, it acts on the sole of the foot as an equivalent unbalanced torque. In the formula, the method for setting the conduction gain coefficient is as follows: Postural simulation experiments are designed for three types of uneven shoulders. A group of subjects with typical functional uneven shoulders, structural uneven shoulders, and pain-compensated uneven shoulders are recruited. Data on the bilateral acromion height difference, trunk tilt angle, and plantar pressure distribution of each group of subjects in natural standing and walking states are collected. The measured plantar pressure asymmetric torque is used as the dependent variable, and other known variables in the formula are used as independent variables. The conduction gain coefficient corresponding to each type of uneven shoulder is obtained by fitting multiple nonlinear regression. During the fitting process, the least squares method is used to minimize the sum of squared residuals between the model prediction value and the measured value.

[0075] The specific calculation of the kinetic chain transmission term is accomplished using the kinetic chain transmission term formula. The physical meaning of the kinetic chain transmission term is: the comprehensive transmission efficiency coefficient after adjustment by the range of motion of each compensating joint, as the original eccentric torque generated by uneven shoulders is transmitted downwards along the spine and lower limb joints. Limited by the number of transmission joints and their range of motion, the kinetic chain transmission term is a bounded value. The kinetic chain transmission term formula uses a multiplication operation to evaluate the collective attenuation effect of each compensating joint on torque transmission. For each joint, the difference between the normal reference range of motion and the actual range of motion is first calculated. Then, the absolute value of the difference is added to the normal reference range of motion as the denominator. This ratio is normalized to between -1 and +1, multiplied by the compensating weight of that joint, and subtracted from the value to obtain the transmission efficiency coefficient of that joint. When the actual range of motion equals the normal reference range, the difference is zero, and the transmission efficiency coefficient is one, indicating that the joint has no attenuation in torque transmission. When the actual range of motion deviates significantly from the normal reference range, the absolute value of the difference is close to the normal reference range, and the transmission efficiency coefficient is close to one minus the compensation weight, indicating that the joint causes attenuation. Multiplying the transmission efficiency coefficients of all joints yields the total transmission efficiency of the entire kinetic chain. The higher the efficiency value, the smoother the transmission of the high and low shoulder torque in the kinetic chain. The lower the efficiency value, the more limited the compensation of multiple joints, hindering the effective transmission of torque. In the formula, the method for setting the joint compensation weight is as follows: Based on the biomechanical transmission theory of the human kinetic chain, an expert group composed of multiple rehabilitation medicine experts and biomechanics experts uses the Delphi method to score the importance of each joint in the high and low shoulder torque transmission process. The scoring level is divided into four levels: very important, important, average, and unimportant. After collecting multiple rounds of expert scores, the average score of each joint is calculated and normalized to obtain the compensation weight of each joint.

[0076] S5: Assess the symmetry of walking rehabilitation in diabetic foot patients based on the equivalent moment interference and the effective assessment area of ​​the foot.

[0077] Based on the patient's real-time walking data, the gait cycle duration is obtained. The actual pressure in the effective assessment areas of the left and right feet is obtained according to the plantar pressure distribution heatmap and the set of effective plantar assessment areas. The sign function of the actual offset direction of the plantar pressure center is determined. The patient's foot length is obtained to calculate the original asymmetric torque of plantar pressure. The original asymmetric torque of plantar pressure is continuously statistically analyzed for multiple gait cycles. The periodic standard deviation of the original asymmetric torque of plantar pressure is calculated. The sign function of the horizontal offset direction of the center of gravity caused by uneven shoulders is determined based on the bilateral acromion height difference. The gait rehabilitation symmetry index is calculated by combining the original asymmetric torque of plantar pressure and the equivalent asymmetric torque. The gait rehabilitation symmetry index is used to evaluate the patient's gait rehabilitation symmetry.

[0078] The formula for the original asymmetric moment of plantar pressure is:

[0079] ;

[0080] In the formula, The original asymmetric torque of plantar pressure, For the patient's foot length, The duration of the gait cycle. For the first Risk weights for each effective assessment area of ​​the sole of the foot. A set of effective assessment areas for the soles of the feet. For the left foot Each effective assessment area of ​​the sole of the foot is located in The actual pressure at all times It is a fractional exponent. For the right foot Each effective assessment area of ​​the sole of the foot is located in The actual pressure at all times It is the integral variable.

[0081] The formula for the symmetry index of walking rehabilitation is:

[0082] ;

[0083] In the formula, The symmetry index for walking rehabilitation. The original asymmetric torque of plantar pressure, For equivalent asymmetric torque, The sign function for the horizontal shift of the center of gravity caused by uneven shoulders. The sign function represents the actual offset direction of the plantar pressure center. To adjust the index, This is the scaling factor. The periodic standard deviation of the original asymmetric moment of plantar pressure. It is a very small positive number. This is the sensitivity coefficient. The duration of the left foot's stance phase. The duration of the right foot's support phase.

[0084] It should be further explained that after completing the quantitative calculation of the equivalent asymmetric torque, a comprehensive assessment of the symmetry of gait rehabilitation in diabetic foot patients is conducted based on the interference of this equivalent torque and the set of effective plantar assessment areas obtained in the previous screening. This assessment process extracts the gait cycle duration from the patient's real-time gait data. The gait cycle duration refers to the time interval between two consecutive heel strikes on the same side, reflecting the temporal rhythm characteristics of the patient's gait. Based on the plantar pressure distribution heatmap and the set of effective plantar assessment areas obtained in step S2, the actual pressure values ​​of each effective assessment area of ​​the left foot at each moment in the gait cycle and the actual pressure values ​​of each effective assessment area of ​​the right foot at the same moment are extracted from the heatmap. While extracting the pressure values, the sign function of the actual offset direction of the plantar pressure center is determined according to the overall offset direction of the plantar pressure center. The sign function takes a value of positive or negative one, where positive one indicates that the pressure center is generally biased to the left and negative one indicates that the pressure center is generally biased to the right. The patient's foot length data is obtained. Foot length is the horizontal distance between the posterior edge of the heel and the tip of the longest toe when standing. This data is used to convert the pressure difference dimension into the torque dimension.

[0085] After the above preparations are completed, the original pressure asymmetry between the patient's two feet is calculated using the original asymmetric moment formula for plantar pressure. This formula integrates the instantaneous pressure difference between the left and right feet over time. The instantaneous pressure difference is calculated by first calculating the weighted contribution values ​​for the left and right feet for each region in the effective assessment area set. The contribution value for each region is the fractional power of the actual pressure in that region multiplied by its risk weight. The introduction of the fractional exponent ensures that the pressure values, after exponentiation, retain the numerical order of the pressure values ​​while compressing the dominance of maximum values ​​on the results, resulting in a more balanced contribution from each region and a more sensitive reflection of minute pressure changes. The summation of the left and right foot contributions from all regions and the absolute value of the difference yield the overall pressure difference between the left and right feet at that moment. This difference is integrated over the entire gait cycle and multiplied by the foot length and divided by the gait cycle duration. This transforms the average pressure difference over time into a value with torque dimensions, reflecting the overall asymmetry of pressure distribution in both feet throughout the gait cycle. In the formula, the fractional-order index is set as follows: a portion of samples is randomly selected from historical data as the training set. The index is then used for grid search with a step size of 0.01 between 0.1 and 0.9. For each candidate index value, the original asymmetric torque of plantar pressure for all training samples is calculated. The Spearman correlation coefficient between this torque and the standard level of plantar pressure asymmetry determined by clinicians is used as the evaluation index. The index that maximizes the correlation coefficient is selected as the optimal fractional-order index. The risk weights were set as follows: based on the clinical epidemiological data of diabetic foot ulcers, the frequency of each plantar region involved in ulcer cases was statistically analyzed. This frequency was used as the basic risk coefficient for that region. Then, the risk weights were corrected by the proportion of peak pressure in the plantar weight distribution of each region. Finally, the risk weights of all effective plantar assessment regions were normalized to make the sum of the risk weights equal to one.

[0086] To assess the stability and consistency between gait cycles, the original plantar pressure asymmetry torque is statistically analyzed over multiple gait cycles, typically more than ten consecutive cycles. The standard deviation of these torque values ​​is then calculated to obtain the periodic standard deviation of the original plantar pressure asymmetry torque. This standard deviation reflects the fluctuation range of plantar pressure asymmetry between different gait cycles during the patient's walking process; a larger standard deviation indicates a more unstable walking rhythm and load distribution.

[0087] Before calculating the symmetry index of gait rehabilitation, a sign function is determined based on the difference in acromion height between the two sides to indicate the direction of the horizontal shift in the center of gravity caused by uneven shoulders. This sign function is determined as follows: when the left acromion is higher than the right, it indicates that the center of gravity is relatively shifted to the left, and the sign function is positive one; when the right acromion is higher than the left, the center of gravity is relatively shifted to the right, and the sign function is negative one. The core calculation of the gait rehabilitation symmetry index is performed using the formula, which consists of a correction ratio term and a time symmetry penalty factor. The numerator of the correction ratio term is the absolute value of the difference between the original asymmetric torque of plantar pressure and the product of the equivalent asymmetric torque and the two directional sign functions. The product of the two directional sign functions is used to determine the direction of the center of gravity shift caused by uneven shoulders and the direction of the plantar pressure center shift. If the directions are consistent, the equivalent asymmetric torque is directly subtracted from the original torque; if the directions are inconsistent, the equivalent asymmetric torque acts as a superimposed correction. The denominator is the sum of the original asymmetric torque, the scaled periodic standard deviation, and the adjustment exponent raised to the power of a minimum positive number. The introduction of the periodic standard deviation means that the greater the fluctuation between gait cycles, the larger the denominator and the smaller the symmetry index, reflecting the impact on gait stability. The minimum positive number ensures that the denominator is always greater than zero. Subtracting this ratio from the numerical value yields the basic symmetry score. The time symmetry penalty factor is calculated based on the difference in the duration of the stance phase between the left and right feet. The greater the difference in stance phase duration, the further the penalty factor deviates from one. Multiplying this factor by the baseline score yields the final symmetry index. The moderating index in the formula is set as follows: using a combination of grid search and cross-validation in historical sample data, the moderating index is iterated between 0.5 and 3.0 in steps of 0.1, selecting the value that maximizes the Pearson correlation coefficient between the calculated gait rehabilitation symmetry index and the clinical rehabilitation assessment scale score. The scaling factor is set as follows: based on the distribution range of the standard deviation of the original asymmetric moment of plantar pressure during the cycle in the collected historical gait data, the ratio of the 95th quantile to the 5th quantile of this distribution is taken, and then the reciprocal is taken. This ensures that the contribution of the scaled standard deviation of the cycle in the denominator is on the same order of magnitude as the original asymmetric moment term. The sensitivity coefficient is set as follows: based on the normal reference range of the difference in support phase duration widely used in clinical gait assessment. This reference range is usually the ratio of the difference in support phase duration between the left and right feet to the total support phase duration in healthy individuals, which should not exceed five percent.When setting this coefficient, a clinically acceptable tolerance point for symmetry penalty is determined. For example, when the ratio of differences in the duration of the stance phase is 5%, the time symmetry penalty factor should be maintained above 0.95 to ensure that slight differences produce almost no penalty; when the ratio of differences in the duration of the stance phase reaches 15%, the penalty factor should drop below 0.5 to reflect the serious impact of significant time asymmetry on rehabilitation evaluation. Based on the above two clinical constraints, this coefficient is substituted as an unknown into the expression for the exponential penalty factor, and two equations are established and their geometric mean is taken to obtain the set value of the sensitivity coefficient. The level of gait rehabilitation symmetry is evaluated based on the value of the gait rehabilitation symmetry index. This index is between zero and one; the closer the index is to one, the better the gait symmetry, and the closer it is to zero, the worse the gait symmetry. By comparing the relative magnitudes of the equivalent asymmetric torque and the original asymmetric torque of the plantar pressure, it is possible to further determine whether the dominant factor affecting the patient's gait symmetry is upper kinetic chain postural compensation or pathological changes in the foot itself, providing a more accurate decision-making basis for targeted adjustments to the clinical rehabilitation program.

[0088] Example 2: Based on Example 1, a symmetry assessment system for walking rehabilitation based on plantar pressure distribution, such as... Figure 2 As shown, it also includes:

[0089] The data acquisition module trains a plantar pressure benchmark model based on historical gait data, collects real-time walking data of patients, and uses the plantar pressure benchmark model to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters.

[0090] The region filtering module processes the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and filters effective plantar assessment regions.

[0091] The probability calculation module acquires the body posture data and kinetic chain joint data of diabetic foot patients, and calculates the soft probability of the patient's uneven shoulder type based on the body posture data.

[0092] The interference calculation module quantifies the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data.

[0093] The symmetry assessment module assesses the symmetry of walking rehabilitation in diabetic foot patients based on the equivalent moment interference and the effective assessment area of ​​the sole.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for assessing the symmetry of gait rehabilitation based on plantar pressure distribution, characterized in that, Includes the following steps: S1: Train a plantar pressure benchmark model based on historical gait data, collect real-time walking data of patients, and use the plantar pressure benchmark model to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters; S2: Process the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and screen effective plantar assessment regions; S3: Obtain postural data and kinetic chain joint data of diabetic foot patients, and calculate the probability of uneven shoulder type based on the postural data; S4: Quantify the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data; S5: Assess the symmetry of walking rehabilitation in diabetic foot patients based on the equivalent moment interference and the effective assessment area of ​​the foot.

2. The method for assessing the symmetry of walking rehabilitation based on plantar pressure distribution according to claim 1, characterized in that, The process of training a plantar pressure benchmark model based on historical gait data and collecting real-time walking data from patients to obtain plantar pressure distribution heatmaps and pressure distribution characteristic parameters using the plantar pressure benchmark model includes: the historical gait data including spatiotemporal sequence data of plantar pressure, gait cycle phase data, and weight-normalized data of healthy individuals; after preprocessing the historical gait data to construct a dataset, the plantar pressure benchmark model is trained through a variational autoencoder network; real-time walking data from patients is collected; and plantar pressure distribution heatmaps and pressure distribution characteristic parameters are obtained based on the real-time walking data using the plantar pressure benchmark model. The pressure distribution characteristic parameters include peak pressure in the plantar region, pressure time integral, pressure center trajectory offset, and correlation coefficient between left and right foot pressure distribution.

3. The method for assessing the symmetry of walking rehabilitation based on plantar pressure distribution according to claim 1, characterized in that, The process of processing the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and screening effective plantar assessment regions includes: calculating the pressure concentration index and pressure offset direction angle of the plantar region based on the pressure distribution characteristic parameters; processing the plantar pressure distribution heatmap through fuzzy inference based on the pressure concentration index and pressure offset direction angle to obtain plantar pressure regions, corresponding pressure anomaly factors and durations; determining regions where the pressure anomaly factor is less than the pressure anomaly threshold and the duration is greater than the duration threshold as effective plantar assessment regions, thus obtaining a set of effective plantar assessment regions.

4. The method for assessing the symmetry of walking rehabilitation based on plantar pressure distribution according to claim 1, characterized in that, The process involves acquiring postural data and kinetic chain joint data from diabetic foot patients, and calculating the soft probability of uneven shoulder type based on the postural data. This includes: the postural data comprising bilateral acromion height difference, trunk tilt angle, pelvic tilt angle, cervical spine deviation angle, root mean square values ​​of surface electromyography (EMG) of the upper trapezius muscles bilaterally, passive range of motion of both shoulder joints, and shoulder and neck pain scores; constructing a postural feature vector after standardizing the postural data; and identifying uneven shoulder types including functional uneven shoulder, structural uneven shoulder, and pain-compensating uneven shoulder, and statistically determining the total number of uneven shoulder types. Using a historical patient posture dataset labeled with physician-defined shoulder types, a Gaussian mixture model was used to train and fit the standard posture feature mean vectors and standard posture feature covariance matrices for different shoulder types. A support vector machine recursive feature elimination method was then employed for training to obtain the importance coefficient of each posture indicator in distinguishing between various shoulder types. The importance coefficients were normalized and used as weight vectors for different shoulder types. The baseline probability of shoulder type occurrence was statistically obtained, and the soft probability of shoulder type in diabetic foot patients was calculated by combining the posture feature vectors.

5. The method for assessing the symmetry of gait rehabilitation based on plantar pressure distribution according to claim 4, characterized in that, The calculation of the soft probability of uneven shoulder type in diabetic foot patients by combining body posture feature vectors includes: the soft probability formula for uneven shoulder type is: ； In the formula, Given a body feature vector High and low shoulder type soft probability, Uneven shoulders The transpose of the weight vector, For body shape feature vectors, Uneven shoulders The basic probability of occurrence. Uneven shoulders The standard body posture feature mean vector, Uneven shoulders The inverse matrix of the standard body shape feature covariance matrix. for transpose, This represents the total number of uneven shoulders. Uneven shoulders The transpose of the weight vector, Uneven shoulders The basic probability of occurrence. Uneven shoulders The standard body posture feature mean vector, Uneven shoulders The inverse matrix of the standard body shape feature covariance matrix.

6. The method for assessing the symmetry of walking rehabilitation based on plantar pressure distribution according to claim 1, characterized in that, The method of quantifying the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of uneven shoulder type and the kinetic chain joint data includes: the kinetic chain joint data includes the number of compensating joints in the kinetic chain, the actual range of motion of each joint and the normal reference range of motion of each joint, obtaining the patient's weight, and calculating the equivalent asymmetric torque by combining the kinetic chain joint data, the difference in acromion height between the two sides and the trunk tilt angle.

7. The method for assessing the symmetry of walking rehabilitation based on plantar pressure distribution according to claim 6, characterized in that, The calculation of the equivalent asymmetric torque by combining the kinetic chain joint data, the bilateral acromion height difference, and the trunk tilt angle includes: The formula for the equivalent asymmetric torque is: ； In the formula, For equivalent asymmetric torque, For the patient's weight, It is the acceleration due to gravity. The difference in height between the two acromions. The angle of trunk tilt. This represents the total number of uneven shoulders. Given a body feature vector High and low shoulder type soft probability, Uneven shoulders The conduction gain coefficient, For kinetic chain conduction; The formula for the kinetic chain conduction term is: ； In the formula, For kinetic chain conduction, The number of joints used to compensate for kinetic chain defects. For the first The compensatory weight of each joint For the first Normal reference range of motion for each joint. For the first The actual range of motion of each joint.

8. The method for assessing the symmetry of gait rehabilitation based on plantar pressure distribution according to claim 1, characterized in that, Assessing the symmetry of gait rehabilitation in diabetic foot patients based on the equivalent torque interference and the effective plantar assessment area includes: obtaining the gait cycle duration based on the patient's real-time gait data; obtaining the actual pressure of the effective plantar assessment area of ​​the left foot and the actual pressure of the effective plantar assessment area of ​​the right foot based on the plantar pressure distribution heatmap and the set of effective plantar assessment areas, and determining the sign function of the actual offset direction of the plantar pressure center; obtaining the patient's foot length to calculate the original asymmetric torque of plantar pressure; continuously statistically analyzing the original asymmetric torque of plantar pressure over multiple gait cycles; calculating the periodic standard deviation of the original asymmetric torque of plantar pressure; determining the sign function of the horizontal offset direction of the center of gravity caused by uneven shoulders based on the bilateral acromion height difference; calculating the gait rehabilitation symmetry index by combining the original asymmetric torque of plantar pressure and the equivalent asymmetric torque; and evaluating the gait rehabilitation symmetry of the patient based on the gait rehabilitation symmetry index. The formula for the original asymmetric moment of plantar pressure is: ； In the formula, The original asymmetric torque of plantar pressure, For the patient's foot length, The duration of the gait cycle. For the first Risk weights for each effective assessment area of ​​the sole of the foot. A set of effective assessment areas for the soles of the feet. For the left foot Each effective assessment area of ​​the sole of the foot is located in The actual pressure at all times It is a fractional exponent. For the right foot Each effective assessment area of ​​the sole of the foot is located in The actual pressure at all times It is the integral variable.

9. The method for assessing the symmetry of walking rehabilitation based on plantar pressure distribution according to claim 8, characterized in that, The calculation of the gait rehabilitation symmetry index by combining the original asymmetric torque and the equivalent asymmetric torque of plantar pressure includes: The formula for the symmetry index of walking rehabilitation is: ； In the formula, The symmetry index for walking rehabilitation. The original asymmetric torque of plantar pressure, For equivalent asymmetric torque, The sign function for the horizontal shift of the center of gravity caused by uneven shoulders. The sign function represents the actual offset direction of the plantar pressure center. To adjust the index, This is the scaling factor. The periodic standard deviation of the original asymmetric moment of plantar pressure. It is a very small positive number. This is the sensitivity coefficient. The duration of the left foot's stance phase. The duration of the right foot's support phase.

10. A symmetry assessment system for gait rehabilitation based on plantar pressure distribution, used to implement the symmetry assessment method for gait rehabilitation based on plantar pressure distribution as described in any one of claims 1-9, characterized in that, Also includes: The data acquisition module trains a plantar pressure benchmark model based on historical gait data, collects real-time walking data of patients, and uses the plantar pressure benchmark model to obtain a plantar pressure distribution heatmap and pressure distribution characteristic parameters. The region filtering module processes the plantar pressure distribution heatmap and pressure distribution characteristic parameters to obtain plantar pressure regions, corresponding pressure anomaly factors and durations, and filters effective plantar assessment regions. The probability calculation module acquires the body posture data and kinetic chain joint data of diabetic foot patients, and calculates the soft probability of the patient's uneven shoulder type based on the body posture data. The interference calculation module quantifies the equivalent torque interference of uneven shoulders on plantar pressure based on the soft probability of the uneven shoulder type and the kinetic chain joint data. The symmetry assessment module assesses the symmetry of walking rehabilitation in diabetic foot patients based on the equivalent moment interference and the effective assessment area of ​​the sole.

Images