A method for predicting optimal length of the hamstring muscle

Abstract

This invention belongs to the field of hamstring length prediction technology. Specifically, it relates to a method for predicting the optimal hamstring length. This method generates curves of knee flexion angle versus hamstring length and curves of knee flexion angle versus knee flexion force through preprocessing, and verifies and outputs accurate optimal hamstring length. Then, it constructs a prediction model consisting of a global prediction model and a portrait prediction sub-model based on a person's profile. Finally, it calculates the matching degree between the person being evaluated and the portrait prediction sub-model, selects the corresponding model, and outputs the final prediction result. This invention improves the accuracy of determining and predicting the optimal hamstring length through hyperbolic collaborative analysis and verification to correct biases. Furthermore, it utilizes the portrait sub-model to adapt to specific groups and combines a matching degree adaptive calling strategy to achieve an optimal balance between prediction accuracy and evaluation efficiency. This method is suitable for scenarios involving sports injury prevention, rehabilitation program development, and sports training guidance.

Description

Technical Field

[0001] This invention belongs to the field of hamstring length prediction technology, and more specifically, relates to a method for predicting the optimal length of the hamstring. Background Technology

[0002] The hamstrings are located on the back of the thigh and consist of four muscles: the long head of the biceps femoris, the short head of the biceps femoris, the semimembranosus, and the semitendinosus. The optimal length of the muscle determines the strain when it is stretched, reflecting its strain limit. Therefore, it is necessary to predict the optimal length of the muscle in order to better implement sports protection.

[0003] The existing Chinese invention patent CN121093548A discloses a method for self-sensing the length of a polymer twisted artificial muscle based on a coil structure. Specifically, it equates the artificial muscle to N equidistant circular current loops, establishes an inductance calculation formula by combining the Biot-Savart law and the principle of magnetic field superposition, and breaks through the limitation of the helix angle of the traditional solenoid model. Subsequently, through MATLAB numerical simulation and iterative calculation, a power function describing the relationship between inductance and length is fitted. Finally, in view of complex factors such as actual leakage magnetic field and skin effect, a parameter dynamic correction method based on a single initial inductance measurement is proposed, and rapid adaptive calibration is achieved by adjusting the parameters.

[0004] Although modeling and calculation methods have been proposed, most existing predictions of the optimal hamstring length are determined solely by the collected data, without verifying or correcting the determined optimal hamstring length. This results in a large error between the obtained optimal hamstring length and the true optimal hamstring length.

[0005] Secondly, the current method of establishing a mapping relationship by integrating the feature data of all subjects with the optimal hamstring length data results in insufficient prediction accuracy for well-defined subgroups and prolongs the prediction response time, making it difficult to apply to large-scale evaluation scenarios. Summary of the Invention

[0006] In view of this, in order to solve the above problems, a method for predicting the optimal length of the hamstring muscle is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: This invention provides a method for predicting the optimal length of the hamstring muscle, the method comprising: performing a knee flexion test on each subject at a preset angular velocity, observing and collecting the knee flexion force value and hamstring muscle length at each knee flexion angle in real time, and constructing a test parameter set.

[0008] The test parameter set is preprocessed to generate curves showing the relationship between knee flexion angle and hamstring length, and between knee flexion angle and knee flexion force, which are denoted as the first curve and the second curve, respectively.

[0009] The first and second curves are analyzed to obtain the optimal hamstring length represented by the corresponding curves. The two curves are then verified, and the final optimal hamstring length is output.

[0010] Based on the basic information and movement information of each subject, as well as the optimal hamstring length, a prediction model for the optimal hamstring length is constructed. The model consists of a global prediction model and several portrait prediction sub-models, and the portrait prediction sub-models are associated with the person's portrait.

[0011] Register the basic information and movement data of the person being evaluated, calculate the matching degree between the person and the portrait prediction sub-model, and output the final predicted optimal hamstring length of the person being evaluated based on the matching degree.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention corrects the one-sided bias of directly determining the optimal length of the hamstring muscle based on the collected hamstring muscle length by using the hyperbola collaborative analysis and verification method, and selects whether to retest and output the final result based on whether the verification can be passed, thereby improving the stability of the determination of the optimal length of the hamstring muscle.

[0013] (2) The present invention constructs a profile prediction sub-model based on the subject groups obtained by clustering, which can accurately capture the unique mapping relationship between the features and the optimal length within the group, thereby reducing the prediction error of the subdivided group. At the same time, the profile prediction sub-model is constructed for homogeneous small groups, and its model structure is more lightweight, thereby improving its training speed and shortening the prediction response time.

[0014] (3) The present invention calls the profile prediction sub-model when the matching degree is high and the global prediction model when the matching degree is not high. This ensures the high accuracy of most homogeneous groups and takes into account the universality of scattered samples, thus achieving the optimal balance between accuracy and efficiency. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the overall implementation process of the present invention;

[0017] Figure 2 This is a schematic diagram illustrating the construction process of the global prediction model of this invention. Detailed Implementation

[0018] 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.

[0019] This invention focuses on the prediction of optimal hamstring length as its primary application scenario, and provides a detailed analysis of specific implementation examples. Please refer to the provided text for more information. Figure 1 As shown, the present invention provides a method for predicting the optimal length of the hamstring muscle. The method includes: S1, performing knee flexion tests on each subject at a preset angular velocity, observing and collecting the knee flexion force value and hamstring muscle length (including the long head of the biceps femoris, the short head of the biceps femoris, the semimembranosus and the semitendinosus) at each knee flexion angle in real time, and constructing a test parameter set.

[0020] It should be noted that, in order to avoid prediction errors caused by gender differences, this method only collects the knee flexion force and hamstring length of subjects of a single gender at various knee flexion angles. This ensures that the subsequent multidimensional feature index dataset, optimal length ground truth dataset, and personnel profile are all constructed based on the same gender group, thereby eliminating the interference of differences in the physiological structure of the lower limb muscles of men and women, such as the distribution of muscle fiber types, the location of muscle attachment points, or mechanical properties, on the prediction results, and thus improving the prediction accuracy of the model.

[0021] It is also necessary to collect basic information and exercise information of the subjects. Basic information includes height, weight, age and history of sports injury, while exercise information includes strength rating parameters and muscle status rating parameters.

[0022] It should be added that the LabelEncoder tool was used to convert the two categorical variables, exercise level and history of sports injury, into consecutive integers from 0 to n-1, where n is the number of categories of the categorical variable.

[0023] As a preferred example, the strength rating parameters are obtained by collecting the measured values ​​of the maximum and average eccentric force of one lower limb in the Nordic Drop test using a force measuring instrument, as well as the measured values ​​of the maximum and average eccentric force of the contralateral lower limb, and then calculating them by linear weighted summation based on the weight coefficients of each measured value.

[0024] The aforementioned linear weighted summation is a prior art technique and will not be described or illustrated in detail in this invention.

[0025] It should be added that the weighting coefficients of the above measured values ​​can be set according to the type of event. For example, when an athlete's sport is an explosive power-dominated event, the key movements of explosive power-dominated events mainly depend on the instantaneous eccentric braking and force exertion of one lower limb. Therefore, the weight of one lower limb is greater than that of the contralateral lower limb, and the weight of the maximum eccentric force of the same part is greater than that of the average eccentric force. The weighting coefficient of the measured value of the maximum eccentric force of one lower limb is 0.35, the weighting coefficient of the measured value of the maximum eccentric force of the contralateral lower limb is 0.30, the weighting coefficient of the measured value of the average eccentric force of one lower limb is 0.18, and the weighting coefficient of the measured value of the average eccentric force of the contralateral lower limb is 0.17.

[0026] As a preferred example, the above muscle condition scoring parameters are composed of the measured values ​​of muscle thickness, pennate angle and muscle bundle length in the resting state obtained by dynamic ultrasound equipment, as well as the measured values ​​of muscle thickness, pennate angle and muscle bundle length in the stretched state.

[0027] Specifically, the process of collecting the knee flexion force and knee flexion angle includes: after the subject is fixed in a preset position, force measuring devices are fixed at the bottom and top of the lower leg area, and at the same time, an angle sensor is installed at the knee joint.

[0028] Under the instruction of the testers, the subjects performed knee flexion movements at a preset angular velocity, slowly flexing their knees from the fixed knee flexion angle in their current position to their maximum tolerable angle, and the real-time knee flexion force and knee flexion angle data during this process were recorded by a signal acquisition device.

[0029] Preferably, the preset body position is that the subject lies prone on the test bed with both hip joints naturally extended and the knee joints in a fully extended neutral position. The pelvis and torso are restrained by a fixation strap to avoid body position deviation during the test.

[0030] It should be noted that the above preset angular velocity is set with reference to the optimal hamstring length test in the industry, and for example, it can be set to 10 degrees per second.

[0031] Specifically, the process of collecting hamstring length includes: connecting the dynamic ultrasound device, force meter and angle sensor to record the hamstring length of the subject at different knee flexion angles in real time.

[0032] After integrating the data on hamstring length and time, knee flexion angle, and knee flexion force, a set of test parameters is constructed.

[0033] S2. Preprocess the test parameter set to generate curves showing the relationship between knee flexion angle and hamstring length, and between knee flexion angle and knee flexion force, which are denoted as the first curve and the second curve, respectively.

[0034] Specifically, the preprocessing of the test parameter set includes: calculating the Pearson correlation coefficient between the knee flexion angle and the corresponding hamstring length, and between the knee flexion angle and the corresponding knee flexion force value.

[0035] It should be added that the above-mentioned method for calculating the Pearson correlation coefficient between the knee flexion angle and the corresponding hamstring length, as well as the knee flexion angle and the corresponding knee flexion force value, is prior art and will not be shown or described in detail in this invention.

[0036] For each subject, it was determined whether all of their knee flexion angles met the following condition: all knee flexion angles were negatively correlated with hamstring length using the Pearson correlation coefficient.

[0037] The absolute values ​​of the Pearson correlation coefficients between all knee flexion angles and the corresponding hamstring lengths were greater than or equal to a preset correlation threshold.

[0038] The absolute value of the Pearson correlation coefficient between all knee flexion angles and the corresponding knee flexion force values ​​is greater than or equal to the preset correlation threshold.

[0039] If all conditions are met, the test parameter set will not be processed.

[0040] If any conditions are not met, the following rules shall apply.

[0041] If two or more conditions are not met, the collected data associated with that subject will be recorded as abnormal data and filtered.

[0042] If only one condition is not met, and the hamstring length corresponding to a certain knee flexion angle does not meet the condition, then the hamstring length associated with that angle is discarded.

[0043] If the knee flexion force value corresponding to a certain knee flexion angle does not meet the conditions, then the knee flexion force value associated with that angle will be discarded.

[0044] The calculation of the Pearson correlation coefficient is existing technology, and the calculation principle and derivation process will not be elaborated here.

[0045] It should be noted that the above-mentioned preset correlation thresholds are set according to the following rules: subjects with no history of sports injury in the past three months and whose muscle status score parameters meet the preset requirements are selected from the test samples and used as target test subjects. The knee flexion force and hamstring length of the target test subjects at each knee flexion angle are extracted, and a benchmark dataset is constructed. The Pearson correlation coefficient between the knee flexion angle and hamstring length and the knee flexion force for each subject in the benchmark dataset are calculated, thereby forming two sets of correlation coefficient distribution samples. The 25th percentile of the correlation coefficient distribution samples is used as the preset correlation threshold value of the corresponding Pearson correlation coefficient, and the results are output sequentially.

[0046] It should be noted that the process of generating the above-mentioned curve relating knee flexion angle and hamstring length is as follows: First, extract the knee flexion angle, hamstring length, and knee flexion force data with timestamps from the test parameter set, and arrange them in ascending order according to the time stamps to form a time-series data sequence. Then, pair the knee flexion angle and hamstring length corresponding to the same time stamp to generate a set of data pairs of knee flexion angle and hamstring length sorted by time. Finally, with the knee flexion angle as the x-axis and the hamstring length as the y-axis, connect the data points through linear interpolation or smooth fitting to generate the first curve.

[0047] It should be noted that the above-mentioned linear interpolation and smooth fitting are existing technologies and will not be shown or described in detail in this invention.

[0048] Meanwhile, the process of generating the curve relating knee flexion angle and knee flexion force is the same as the process of generating the curve relating knee flexion angle and hamstring length. Specifically, the knee flexion angle and knee flexion force corresponding to the same time stamp are paired to generate a set of data pairs of knee flexion angle and knee flexion force ordered by time. The knee flexion angle is used as the horizontal axis and the knee flexion force is used as the vertical axis. The data points are connected using the same fitting method as the first curve to generate the second curve.

[0049] It is important to note that the timestamp interval is consistent with the data collection frequency, so as to ensure that the curve can continuously reflect the dynamic correspondence between the angle and muscle length and force value during the knee flexion movement.

[0050] S3. Analyze the first curve and the second curve to obtain the optimal hamstring length represented by the corresponding curve, verify the two curves, and output the final optimal hamstring length.

[0051] Specifically, the analysis process of the first curve and the second curve includes: identifying the peak point (i.e. the point of maximum force) on the second curve; if there are multiple local peaks, the peak with the highest force and located in the middle region of the knee flexion angle (e.g., 30-60 degrees) is selected as the main peak, and the corresponding knee flexion angle is recorded as the peak force angle.

[0052] The hamstring length corresponding to the peak force angle is read from the first curve, and this length is used as the initial value of the optimal hamstring length obtained by the subject in the current test.

[0053] It should be noted that in the knee flexion test, the hamstring length gradually decreases as the knee flexion angle increases (showing a negative correlation); the length at which the muscle generates maximum active tension (i.e., the optimal length) does not correspond to the anatomically maximum stretch length (the length at the minimum knee flexion angle), but rather to the length of sarcomere overlap at the moment of maximum force. Therefore, this invention uses the peak force angle of the second curve as a benchmark and maps the muscle length at that angle from the first curve as the initial value of the physiologically optimal hamstring length.

[0054] Specifically, the verification process includes obtaining the maximum length of the hamstring muscle during the entire flexion process from the first curve (i.e., the maximum value on the curve, corresponding to the minimum knee flexion angle).

[0055] The ratio of the maximum length of the hamstring muscle of each subject during the entire flexion process to the corresponding initial value of the optimal length of the hamstring muscle is calculated and denoted as the optimal length representation ratio. Based on the optimal length representation ratio, an optimal length representation ratio sequence is constructed.

[0056] Calculate the standard deviation of the optimal length characterization ratio sequence.

[0057] The percentage of subjects whose optimal length representation ratio conforms to physiological norms is denoted as the percentage of qualified subjects. Physiological norms refer to the fact that, considering the inherent structural constraints of the sarcomere, the optimal length representation ratio is limited to a fixed range, empirically between 1.05 and 1.1.

[0058] If all of the following conditions are met, the verification is deemed successful: (1) The proportion of qualified subjects is greater than or equal to the preset proportion threshold.

[0059] It should be noted that, considering that there may be slight differences in the degree of muscle relaxation among different subjects, which are within the range of individual differences within a healthy range, the preset percentage threshold is set to 0.9 as a preferred example of the present invention.

[0060] (2) The standard deviation is less than or equal to the preset discrete threshold.

[0061] It should be added that, based on the fact that the range of variation in myofibril length in healthy individuals as recorded in medical standards is less than or equal to 2%, therefore, as a preferred embodiment of the present invention, the preset discrete threshold value is 0.02.

[0062] Otherwise, the verification is deemed unsuccessful, and the knee flexion test is repeated for each subject.

[0063] When the verification is successful, the initial value of the optimal hamstring length obtained by the subject in the current test will be taken as the subject's final optimal hamstring length.

[0064] S4. Based on the basic information and movement information of each subject and the optimal length of the hamstring, construct a prediction model for the optimal length of the hamstring. The model consists of a global prediction model and several portrait prediction sub-models, and the portrait prediction sub-models are associated with the portrait of the person.

[0065] For details, please refer to [link / reference]. Figure 2 As shown, the construction process of the global prediction model includes: establishing a multidimensional feature index dataset based on the basic information and motion information of each subject, and establishing an optimal length true value dataset based on the final optimal hamstring length of each subject.

[0066] It should be noted that min-max linear normalization is used to scale the feature values ​​and target values ​​of the training set to the range of [0, 1], and min-max linear normalization is also applied to the test set data to ensure that the scaling standard is consistent. At the same time, the scaled feature values ​​of the training set, the target value of the training set, the feature values ​​of the test set, and the target value of the test set are saved as independent XLSX format files for easy model calling.

[0067] It should be added that the above-mentioned minimum-maximum linear normalization is a prior art and will not be shown or described in detail in this invention.

[0068] The multidimensional feature index dataset and the optimal length ground truth dataset are divided into training set and test set according to a preset ratio, and the training set is used as the model training and validation dataset.

[0069] It should be added that, in order to balance the sufficiency of training data and the independence of the test set, as a preferred method, the multidimensional feature index dataset and the optimal length ground truth dataset are randomly divided into training set and test set in a ratio of 9:1, whereby the training set is used as the model training and validation dataset, and the test set is used for the final model evaluation.

[0070] The training and validation datasets for the model are split into three folds for validation, generating training and validation subsets.

[0071] The target hyperparameter configuration of the neural network is obtained through grid search hyperparameter optimization based on the training and validation subsets.

[0072] Based on the target hyperparameter configuration, a global prediction model is constructed. The global prediction model is a multi-layer fully connected neural network model, which consists of an input layer, a hidden layer, and an output layer.

[0073] The input layer consists of normalized feature values, and the number of neurons is the same as the number of feature values.

[0074] The hidden layers are either 1 or 2, with 25, 35 or 45 neurons per layer, depending on the results of the hyperparameter search.

[0075] The output layer consists of a single neuron that directly predicts the optimal length of a muscle.

[0076] The target hyperparameter configuration of the neural network is obtained through grid search hyperparameter optimization based on the training and validation subsets: normalized multidimensional feature values ​​are used as input, and the number of neurons is consistent with the feature value dimension. To avoid overfitting due to model parameter redundancy, two hidden layers are preferably set. The input feature dimension of this model is 11 dimensions. Preferably, the number of neurons is set to 2 to 5 times the input feature dimension. Considering the efficiency requirements of hyperparameter search, the number of neurons in each layer can preferably be 25, 35, or 45. To ensure that the gradient does not decay due to multi-layer propagation during backpropagation, and thus better adapt to the parameter update requirements of the two hidden layers, ensuring fast model convergence, the ReLU activation function is preferably used in the hidden layers. ReLU maps the part of the input less than 0 to 0, and the part greater than or equal to 0 remains unchanged. The weight parameters of the feature values ​​are initialized using the default random initialization method of TensorFlow 2.6. A single neuron output layer is set to directly predict the optimal length of the hamstring muscle.

[0077] A grid search method is used to further optimize hyperparameters. The searched hyperparameters include: number of hidden layers, number of neurons per layer, activation function type, optimizer, learning rate, and batch size. Triple-fold validation is used to evaluate the prediction performance of the model using each combination of hyperparameters. The specific validation and evaluation process includes: dividing the training set into three equal parts; during model training, selecting any two parts of the data for training each time, and using the remaining part as the validation set to evaluate the model's performance; repeating the training process three times; recording the validation loss obtained from the three training sessions and calculating the average; and using the set of hyperparameters with the smallest average of the three validation losses as the target hyperparameter configuration, recording the target hyperparameter configuration, including the number of hidden layers, number of neurons per layer, activation function type, optimizer, learning rate, and batch size.

[0078] Furthermore, constructing a global prediction model based on the target hyperparameter configuration includes: training the model according to the target hyperparameter configuration; during training, the mean squared error (MSE) is used as the loss function to calculate the deviation between the model's predicted values ​​and the actual values; the formula for calculating MSE is: In the formula, The number of training set samples. It is the first The actual value of each sample It is the first The predicted value for each sample.

[0079] It should be added that the actual values ​​mentioned above are the final optimal hamstring lengths in the corresponding samples, while the predicted values ​​are the optimal hamstring lengths output by the global prediction model in the corresponding samples.

[0080] Then, backpropagation is performed using the error backpropagation algorithm combined with gradient descent and the chain rule, and the model parameters are updated layer by layer to minimize the loss function. The specific backpropagation process is handled automatically by the TensorFlow framework. The model performs backpropagation calculation and model parameter updates by specifying the optimizer and loss function during compilation.

[0081] The aforementioned backpropagation algorithm, gradient descent method, and chain rule are all existing technologies and will not be described or illustrated in detail in this invention.

[0082] The final trained model is used as the global prediction model and saved as an HDF5 file for later use in predicting the risk of hamstring strain in the assessed person.

[0083] More specifically, the construction process of the portrait prediction sub-model includes: clustering the multidimensional feature index dataset and the optimal length true value dataset using a clustering algorithm to obtain each subject group.

[0084] It should be noted that the above clustering algorithm can preferably be the K-Means clustering algorithm. The K-Means clustering algorithm is existing technology and will not be shown or described in detail in this invention.

[0085] Calculate the standard deviation of the optimal hamstring length for each subject within each group.

[0086] For each group, select any subject as the target person and calculate the Euclidean distance between the target person and the other subjects for the optimal hamstring length.

[0087] Subjects whose Euclidean distance is less than the preset Euclidean distance are recorded as similar subjects to the target person, and the number of similar subjects to the target person is counted.

[0088] It should be added that the above-mentioned preset Euclidean distance is obtained in the following way: First, the optimal hamstring length data of all subjects in the group are extracted, then the Euclidean distance between each pair of subjects is calculated and a matrix is ​​constructed. Then, the mean and standard deviation of all pairwise Euclidean distances in the group are calculated, and finally the preset Euclidean distance is determined by the 3σ principle.

[0089] By sequentially identifying other subjects as target subjects, the number of similar subjects for all subjects is obtained. The ratio of this number of similar subjects to the total number of subjects in that group is taken as the similar subject ratio.

[0090] Determine whether the following conditions are met simultaneously: the proportion of similar subjects exceeds the preset proportion.

[0091] It should be noted that the above preset ratio is obtained based on empirical values; for example, the preset ratio is 0.2.

[0092] The standard deviation did not exceed the preset optimal hamstring length fluctuation range.

[0093] Furthermore, the aforementioned preset optimal hamstring length fluctuation range is achieved through the following technical solution: first, the quartiles and interquartile ranges of the optimal hamstring length for each group are calculated, and then the preset optimal hamstring length fluctuation range is determined by graphical reference rules.

[0094] It should be noted that the above figures are based on existing technology and will not be shown or described in detail in this invention.

[0095] If both of the above conditions are met, the group is designated as an independent prediction group. The basic information, movement information, and final optimal hamstring length of each subject in the group are extracted, and the prediction model for the independent prediction group is constructed in the same way as the global prediction model.

[0096] The basic information and movement information of each subject in the group were deduplicated. Based on the remaining information after deduplication, the value range of each information item was statistically obtained, and a subject profile corresponding to the optimal hamstring length in the group was generated.

[0097] The prediction model of the independent prediction group is associated with the subject profile to generate a profile prediction sub-model.

[0098] Furthermore, when constructing the prediction model of this independent prediction group in the same way as the global prediction model, it should be noted that the hyperparameters of the portrait prediction sub-model are determined through grid search. In order to adapt to the stability requirements of small sample training, preferably, the number of hidden layers is 2. By referring to the parameter scaling experience of the small sample model, preferably, the number of neurons in each layer is 0.5 to 0.6 times the number of neurons in the global model. Preferably, the learning rate is 1.2 to 1.5 times that of the global model, thereby ensuring that the model is lightweight and adaptable to small sample training.

[0099] In existing technologies, the prediction learning of optimal hamstring length needs to consider all subjects. The learning process averages the correlation between different group characteristics and optimal hamstring length, resulting in insufficient prediction accuracy for distinct subgroups. In contrast, the profile prediction sub-model is constructed based on subject groups obtained through clustering. It can accurately capture the unique mapping relationship between features and optimal length within the group, thereby reducing the prediction error of subgroups. At the same time, the profile prediction sub-model is built for small groups, and its model structure is more lightweight, thereby improving the training speed of the profile prediction sub-model and shortening the response time for predicting optimal hamstring length. By calling the profile prediction sub-model when the matching degree is high and calling the global prediction model when the matching degree is low, it ensures high accuracy for most homogeneous groups and takes into account the universality of scattered samples, achieving the optimal balance between accuracy and efficiency.

[0100] S5. Register the basic information and motion data of the person being evaluated, calculate the matching degree between the person and the portrait prediction sub-model, and output the final predicted optimal hamstring length of the person being evaluated based on the matching degree.

[0101] Specifically, the process of calculating the profile matching degree includes: extracting the basic information and motion data of the person being evaluated to form an information set of the person being evaluated.

[0102] The information set of the person being evaluated is matched and compared with the personnel profiles associated with each profile prediction sub-model.

[0103] If a certain information item of the person being evaluated falls within the value range of the corresponding information item in the subject profile, then the baseline matching degree of that information item is assigned a value of 1, and that information item is recorded as a matching information item.

[0104] The number of matching information items is counted and divided by the total number of information items to obtain the proportion of matching information items, which is denoted as the comprehensive benchmark matching degree.

[0105] Calculate the relative deviation between the value of each matching information item and the midpoint of its value range, and subtract the relative deviation from 1 to obtain the matching weight of that information item.

[0106] The matching weights of the matching information items are sorted in ascending order. Based on the sorting, the matching weights are divided into several weight intervals according to a preset interval weight. The number of matching information items in each interval is counted. The average value between the upper and lower limits of the weight interval with the most matching information items is taken as the target weight value.

[0107] Multiplying the overall baseline matching degree by the target weight value yields the final profile matching degree between the evaluated person and the current profile prediction sub-model.

[0108] It should be noted that the aforementioned preset interval weights are specifically divided into three equal parts.

[0109] Specifically, the process of obtaining the final predicted optimal hamstring length includes: extracting the portrait matching degree between the person being evaluated and each portrait sub-model, filtering out the maximum matching degree, and recording the portrait sub-model corresponding to the maximum matching degree as the target model.

[0110] If the maximum matching degree is at the preset high matching level, the feature set of the person being evaluated will be input into the target model, and the final predicted optimal hamstring length will be output.

[0111] If the target is not at the preset high matching level, the feature set of the person being evaluated will be input into the global prediction model, and the final predicted optimal hamstring length will be output.

[0112] It should be added that the aforementioned preset high matching level is obtained through the following method: The matching degree between all subjects and each portrait sub-model is collected, and the matching degree is processed by min-max linear normalization to obtain each matching degree value. Then, the matching degree is divided into several continuous intervals. For the samples in each interval, the optimal hamstring length is predicted using the corresponding portrait sub-model and the global model, and the root mean square error (RMSE) is calculated for each interval. The formula for calculating the root mean square error (RMSE) is as follows: In the formula, The number of samples within the interval. It is the first The actual value of the optimal hamstring length for each sample. It is the first The predicted optimal hamstring length of each sample is selected as the lowest matching degree value where the prediction error of the image sub-model is reduced by more than or equal to 20% compared with the global model. The lowest matching degree value and the corresponding upper limit of matching degree are defined as the matching degree threshold range corresponding to the preset high matching level.

[0113] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method of predicting optimal length of a hamstring muscle, characterized by, The method includes: Knee flexion tests were performed on each subject at a preset angular velocity. The knee flexion force and hamstring length at each knee flexion angle were observed and collected in real time to construct a set of test parameters. The test parameter set is preprocessed to generate curves showing the relationship between knee flexion angle and hamstring length, and between knee flexion angle and knee flexion force, which are denoted as the first curve and the second curve, respectively. The first and second curves are analyzed to obtain the optimal hamstring length represented by the corresponding curves. The two curves are then cross-validated to output the final optimal hamstring length. Based on the basic information and movement information of each subject, as well as the optimal length of the hamstring, a prediction model for the optimal length of the hamstring is constructed. The model consists of a global prediction model and several portrait prediction sub-models, and the portrait prediction sub-models are associated with the portrait of the person. Register the basic information and movement data of the person being evaluated, calculate the matching degree between the person and the portrait prediction sub-model, and output the final predicted optimal hamstring length of the person being evaluated based on the matching degree.

2. The method of claim 1, wherein: The process of acquiring the knee flexion force and knee flexion angle includes: After the subject was fixed in a preset position, force measuring devices were fixed at the bottom and top of the lower leg area, and an angle sensor was installed at the knee joint. Under the instruction of the testers, the subjects performed knee flexion movements at a preset angular velocity, slowly flexing their knees from the fixed knee flexion angle in their current position to their maximum tolerable angle, and the real-time knee flexion force and knee flexion angle data during this process were recorded by a signal acquisition device.

3. The method of claim 2, wherein: The process of collecting the hamstring length includes: The dynamic ultrasound equipment, force meter and angle sensor are connected to record the hamstring length of the subject at different knee flexion angles in real time. After integrating the data on hamstring length and time, knee flexion angle, and knee flexion force, a set of test parameters is constructed.

4. The method of claim 1, wherein: The preprocessing of the test parameter set includes: Calculate the Pearson correlation coefficients between the knee flexion angle and the corresponding hamstring length, and between the knee flexion angle and the corresponding knee flexion force. For each subject, determine whether all of their knee flexion angles meet all of the following conditions: The Pearson correlation coefficients for all knee flexion angles and hamstring lengths were negative. The absolute values ​​of the Pearson correlation coefficients between all knee flexion angles and the corresponding hamstring lengths were greater than or equal to a preset correlation threshold. The absolute value of the Pearson correlation coefficient between all knee flexion angles and the corresponding knee flexion force values ​​is greater than or equal to the preset correlation threshold. If all conditions are met, the test parameter set will not be processed. If any condition is not met, the following rules apply: If two or more conditions are not met, the collected data associated with that subject will be recorded as abnormal data and filtered. If only one condition is not met, the hamstring length or knee flexion force value corresponding to the knee flexion angle that does not meet the condition is eliminated.

5. The method for predicting the optimal hamstring length as described in claim 1, characterized in that: The analytical process for the first and second curves includes: Identify the peak points on the second curve. If there are multiple local peaks, select the peak with the highest force value that is located in the middle region of the knee flexion angle as the main peak, and record the corresponding knee flexion angle as the peak force value angle. The hamstring length corresponding to the peak force angle is read from the first curve and used as the initial value of the optimal hamstring length.

6. The method for predicting the optimal hamstring length as described in claim 1, characterized in that: The cross-validation process includes: Obtain the maximum length of the hamstring muscle throughout the entire flexion process from the first curve; Calculate the ratio of the maximum length corresponding to each subject to the initial value of the optimal hamstring length, and record it as the optimal length representation ratio. Based on the optimal length representation ratio, construct the optimal length representation ratio sequence. Calculate the standard deviation of the optimal length characterization ratio sequence; The percentage of subjects whose optimal length representation ratio conforms to physiological laws is recorded as the percentage of qualified subjects. Cross-validation is considered successful if all of the following conditions are met: The percentage of qualified subjects is greater than or equal to a preset percentage threshold; The standard deviation is less than or equal to a preset discrete threshold; Otherwise, the cross-validation is deemed unsuccessful, and the knee flexion test is repeated for each subject. When cross-validation is successful, the initial value of the optimal hamstring length is taken as the final optimal hamstring length for that subject.

7. The method for predicting the optimal hamstring length as described in claim 1, characterized in that: The construction process of the global prediction model includes: A multidimensional feature index dataset was established based on the basic information and movement information of each subject, and an optimal length true value dataset was established based on the final optimal hamstring length of each subject. The multidimensional feature index dataset and the optimal length true value dataset are divided into training set and test set according to a preset ratio, and the training set is used as the model training and validation dataset. The training and validation datasets for the model are split using a three-fold cross-validation process to generate training and validation subsets. The target hyperparameter configuration of the neural network is obtained through grid search hyperparameter optimization based on the training and validation subsets; A global prediction model is constructed based on the target hyperparameter configuration.

8. The method for predicting the optimal hamstring length as described in claim 7, characterized in that: The construction process of the portrait prediction sub-model includes: The subject groups were obtained by clustering the multidimensional feature index dataset and the optimal length true value dataset using a clustering algorithm. Calculate the standard deviation of the optimal hamstring length for each subject within each group; For each group, select any subject as the target person and calculate the Euclidean distance between the target person and the other subjects for the optimal hamstring length. Subjects whose Euclidean distance is less than the preset Euclidean distance are recorded as similar subjects to the target person, and the number of similar subjects to the target person is counted. Other subjects are successively taken as target subjects, and the number of similar subjects for all subjects is obtained. The ratio of the number of similar subjects to the total number of subjects in the group is taken as the similar subject ratio. Determine whether the following conditions are met simultaneously: The proportion of similar subjects among the subjects exceeded the preset proportion; The standard deviation did not exceed the preset optimal hamstring length fluctuation range; If both of the above conditions are met, the group is recorded as an independent prediction group. The basic information, movement information and the final optimal hamstring length of each subject in the group are extracted, and the prediction model of the independent prediction group is constructed in the same way as the global prediction model. The basic information and movement information of each subject in the group were deduplicated. Based on the remaining information after deduplication, the value range of each information item was statistically obtained, and a subject profile corresponding to the optimal hamstring length in the group was generated. The prediction model of the independent prediction group is associated with the subject profile to generate a profile prediction sub-model.

9. The method for predicting the optimal hamstring length as described in claim 8, characterized in that: The process of calculating the portrait matching degree includes: Extract the basic information and movement data of the person being evaluated to form the information set of the person being evaluated; The information set of the person being evaluated is matched and compared with the personnel profiles associated with each profile prediction sub-model; If a certain information item of the person being evaluated is within the value range of the corresponding information item in the subject profile, then the baseline matching degree of that information item is assigned a value of 1, and that information item is recorded as a matching information item. The number of matching information items is counted and divided by the total number of information items to obtain the proportion of matching information items, which is recorded as the comprehensive benchmark matching degree. Calculate the relative deviation between the value of each matching information item and the midpoint of its value range, and obtain the matching weight of the information item based on the relative deviation. The matching weights of the matching information items are sorted in ascending order. Based on the sorting, the matching weights are divided into several weight intervals according to the preset interval weights. The number of matching information items in each interval is counted. The median value corresponding to the weight interval with the most matching information items is taken as the target weight value. Multiplying the overall baseline matching degree by the target weight value yields the final profile matching degree between the evaluated person and the current profile prediction sub-model.

10. The method for predicting the optimal hamstring length as described in claim 1, characterized in that: The final process for obtaining the optimal predicted hamstring length includes: Extract the portrait matching degree between the evaluated person and each portrait sub-model, and select the maximum matching degree. Record the portrait sub-model corresponding to the maximum matching degree as the target model. If the maximum matching degree is at the preset high matching level, the feature set of the person being evaluated will be input into the target model, and the final predicted optimal hamstring length will be output. If the target is not at the preset high matching level, the feature set of the person being evaluated will be input into the global prediction model, and the final predicted optimal hamstring length will be output.

Images