Multi-signal fusion upper limb muscular tension increase detection method and device

Abstract

The invention discloses a multi-signal fusion upper limb muscular tension increase detection method and device, and relates to the technical field of muscular tension increase detection. The method comprises a data acquisition process, a parameter fitting process, a model training process and a muscular tension increase detection process, passive elbow flexion and elbow extension are carried out at a set speed by driving a forearm with an elbow joint as an axis, mechanical data, kinematics data and myoelectricity data in the movement process are collected, and after the collected various data are preprocessed, the obtained data are subjected to model training and muscular tension increase detection. And training the muscular tension grade prediction model by taking a clinical evaluation result of a doctor on the muscular tension grade as a label, and predicting the muscular tension grade and the contribution degree of each feature to the prediction result through the trained muscular tension grade prediction model. According to the multi-signal fusion upper limb muscular tension increase detection method and device provided by the invention, accurate quantification of muscular tension levels is realized, and meanwhile, the reliability and interpretability of a detection result are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of hypertonia detection, and particularly relates to a method and device for detecting hypertonia of the upper limb by multi-signal fusion. Background Art

[0002] Hypertonia refers to the abnormal increase in muscle tension and resistance, manifested as spasm and stiffness. Persistent increase will limit joint movement, cause secondary injury, affect joint and limb function, and limit daily life and activities. Traditional measurement relies on doctors manually stretching the diseased limb and evaluating by observing the range of joint movement and feeling the muscle impedance force, such as the Ashworth scale method. However, this method is relatively subjective, and the conclusions drawn by doctors with insufficient experience may not be accurate enough, and the whole set of evaluations also takes a relatively long time. Therefore, some objective auxiliary evaluation means are needed.

[0003] At present, the commonly used detection methods are mechanical evaluation method and sEMG evaluation method. The mechanical evaluation method infers the level of hypertonia by collecting the resistance during passive movement of the elbow joint through a force sensor, and its core lies in evaluating the resistance characteristics of the elbow joint. However, the muscle and joint system is a complex biomechanical system with significant nonlinearity and individual differences. A method that solely relies on joint resistance measurement is difficult to comprehensively analyze the complexity of hypertonia, especially in cases involving the coordinated action of multiple muscle groups or abnormal activities of deep muscles. In addition, this method mainly focuses on the change of resistance value and is difficult to capture the dynamic information of hypertonia at different movement speeds and ranges of patients, resulting in the evaluation result may not be completely consistent with the actual hypertonia state of the patient.

[0004] Secondly, the sEMG evaluation method records the myoelectric signals under the skin surface through an sEMG sensor (electromyography sensor) to reflect the muscle activation, contraction intensity, and nerve regulation state. However, due to the fact that hypertonia is usually accompanied by complex physiological and neural mechanism changes, the non-stationarity of sEMG signals and the sensitivity to external interference make the sEMG evaluation method for detecting hypertonia easily affected by artifacts and movement interference, resulting in misjudgment.

[0005] In addition, the data collected by using a force sensor and an sEMG sensor currently only shows the relationship between them and the hypertonia level by using a machine learning model, without exploring the influence degree of different features on the hypertonia level, and machine learning only outputs results without interpretability.

[0006] Therefore, the current evaluation methods have significant deficiencies in terms of comprehensiveness and interpretability of detection results. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method and device for detecting increased upper limb muscle tone through multi-signal fusion. By fusing multi-dimensional features such as mechanics, kinematics, and electromyography signals, accurate quantification of muscle tone levels is achieved, and the contribution degrees of features are ranked in combination with the SHAP framework, thereby enhancing the reliability and interpretability of the detection results.

[0008] In a first aspect, the present invention provides a method for detecting increased upper limb muscle tone through multi-signal fusion, including:

[0009] Data acquisition process: Drive the forearm to perform passive elbow flexion and extension around the elbow joint at a set speed, collect mechanical data and kinematic data during the movement, collect electromyography data during the movement through an electromyography sensor, and preprocess the collected various types of data;

[0010] Parameter fitting process: Based on the system motion differential equation of mechanical impedance characteristics and the mechanical impedance formula of the system, fit biomechanical characteristic parameters according to the preprocessed mechanical data and kinematic data; obtain electromyography characteristic parameters by using the sliding window method according to the preprocessed electromyography data;

[0011] Model training process: Based on the electromyography characteristic parameters and biomechanical characteristic parameters, use the doctor's clinical evaluation results of muscle tension levels as labels to train the muscle tension level prediction model, and the muscle tension level prediction model includes a classification model and SHAP, and SHAP is used to quantify the contribution of each feature to the model prediction;

[0012] Increased muscle tone detection process: Make the elbow joint of the detection object perform passive elbow flexion and extension at a set speed, collect mechanical data, kinematic data, and electromyography data and perform preprocessing. After fitting to obtain biomechanical characteristic parameters and electromyography characteristic parameters, input them into the trained muscle tension level prediction model to obtain the predicted muscle tension level and the contribution degree of each feature to the prediction result.

[0013] Further, after the parameter fitting process, a correlation analysis process is further included to perform Pearson correlation analysis on the biomechanical characteristic parameters and electromyography data, and select features with a correlation coefficient r > 0.8 and a significance level p < 0.01.

[0014] Further, in the data acquisition process, preprocessing the collected various types of data specifically includes: processing the mechanical data, acceleration data, and angle data by using the two-end removal algorithm; filtering the electromyography data, performing full-wave rectification on the filtered signal to ensure that the signal output is positive, and deriving the envelope through a low-pass filter.

[0015] Further, in the parameter fitting process, the system motion differential equation based on mechanical impedance characteristics is:

[0016]

[0017] Among them, T is the torque of the system, F is the mechanical data collected by the force sensor, is the acceleration data collected by the gyroscope, θ is the angle data of the encoder, K represents the elastic parameter, B represents the viscous parameter, C is the offset, t is the time, and l is the distance from the force sensor to the rotation center;

[0018] The mechanical impedance formula of the system is:

[0019] Z = K + Bω

[0020] Among them, ω is the angular frequency of the forearm movement;

[0021] The least squares parameter estimation method is used to find the optimal fitting parameters K, B, and C, and then K, B, C, Z, MRFE, and MREE are used as biomechanical characteristic parameters. Among them, MRFE is the maximum resistance of elbow flexion, MREE is the maximum resistance of elbow extension, and both MRFE and MREE are measured by the force sensor.

[0022] Furthermore, in the parameter fitting process, it includes plotting a torque-angle curve with the torque of elbow flexion and extension as the ordinate and the angle as the abscissa, and then calculating the curve area W enclosed by the force-angle curve as one of the biomechanical characteristic parameters.

[0023] Furthermore, the sEMG characteristic parameters are selected as the root mean square value RMS, the mean absolute value MAV, the simple square integral SSI, the variance of the electromyogram signal VAR, the average amplitude change ACC, the waveform length WL, the mean power frequency MPF, and the median frequency MF.

[0024] Furthermore, the forearm is driven to passively flex and extend the elbow at two or more set speeds with the elbow joint as the axis, and the corresponding electromyogram characteristic parameters and biomechanical characteristic parameters are obtained. After respectively selecting the corresponding characteristic parameter combinations at the specified speeds, they are input into the trained muscle tension level prediction model to obtain the predicted muscle tension level and the contribution degree of each characteristic to the prediction result.

[0025] In a second aspect, the present invention provides an upper limb muscle tension increase detection device for multi-signal fusion, which is used to execute the method described in the first aspect. The device includes:

[0026] A support base;

[0027] A forearm drive unit, which is connected to the support base through a rotating shaft, is used to be fixed to the forearm, and drives the forearm to passively flex and extend the elbow at a set speed with the elbow joint as the axis;

[0028] A force sensor is arranged on the forearm driving unit and is used for collecting mechanical data during movement.

[0029] A motion sensor is arranged on the forearm driving unit and is used for collecting kinematic data during movement.

[0030] An electromyogram sensor has electrodes arranged on the skin surfaces of the biceps brachii and triceps brachii of the upper arm and is used for collecting electromyogram data during movement.

[0031] A data processing unit is used for obtaining the data collected by the force sensor, the motion sensor, and the electromyogram sensor, processing the data, and then outputting the predicted muscle tension level and the contribution degree of each feature to the prediction result.

[0032] Furthermore, a control unit is further included and is used for controlling the forearm driving unit to drive the forearm to flex and extend the elbow passively at a set speed.

[0033] Furthermore, the motion sensor includes a gyroscope and an encoder. The gyroscope is used for collecting speed and acceleration data during movement; the encoder is used for collecting angle data during movement.

[0034] One or more technical solutions provided in the embodiments of the present invention have at least the following technical effects:

[0035] By performing feature processing on sEMG data, mechanical data, and kinematic data, biomechanical features and electromyogram features are extracted simultaneously. Different from the system that solely relies on sEMG signals, by capturing key biomechanical features, the sensitivity of sEMG signals to external interference is effectively reduced, thereby enhancing the redundancy and reliability of the data. At the same time, by combining the SVM model and the SHAP framework, in addition to outputting the patient's muscle tension level, the contribution degree of each feature can also be output, solving the inherent "black box" problem of machine learning models. This makes the classification results easier for doctors and patients to understand and accept.

[0036] The above description is only an overview of the technical solutions of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present invention more obvious and understandable, the following specifically illustrates the specific embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further described below with reference to the accompanying drawings in conjunction with embodiments.

[0038] Figure 1 It is the overall flowchart of the method in Embodiment 1 of the present invention;

[0039] Figure 2 It is the schematic diagram of the implementation process of the method in Embodiment 1 of the present invention;

[0040] Figure 3 Schematic diagram of the preprocessing result of the myoelectric data in the first embodiment of the present invention;

[0041] Figure 4 Schematic diagram of the torque-angle curve in the first embodiment of the present invention;

[0042] Figure 5 Schematic diagram of the overall process of the muscle tone level prediction model in the first embodiment of the present invention;

[0043] Figure 6 Schematic diagram of the prediction result of the muscle tension level in the first embodiment of the present invention;

[0044] Figure 7 Numerical graph of the Pearson correlation analysis in the first embodiment of the present invention;

[0045] Figure 8 Schematic diagram of the structure of the device in the second embodiment of the present invention. Detailed implementation manners

[0046] In the embodiment of the present invention, a method and a device for detecting increased upper limb muscle tone by multi-signal fusion are provided. By fusing multi-dimensional features such as mechanics, kinematics, and myoelectric signals, accurate quantification of the muscle tone level is achieved, and the SHAP framework is combined to rank the feature contribution degrees, thereby enhancing the reliability and interpretability of the detection results.

[0047] The technical solution in the embodiment of the present invention has the following general idea:

[0048] Design a method and a device for detecting increased upper limb muscle tone by fusing signals of multi-sensors (such as force sensors, myoelectric sensors, encoders, gyroscopes, etc.). Based on the collected data, a method for quantifying the muscle tone level (combining biomechanical features and sEMG features) is designed. Further, the correlation (Pearson correlation) between different features and the muscle tone level is analyzed to screen out key features, and SHAP analysis is introduced to rank the contribution degrees of the features input into the classification model (such as support vector machine SVM), enhancing the interpretability of SVM. At the same time, the device has portability (can be directly used on the hospital bed) and dynamic testing capabilities (testing modes with multiple angles and multiple speeds).

[0049] Embodiment 1

[0050] This embodiment provides a method for detecting increased upper limb muscle tone by multi-signal fusion, as shown in Figure 1 and Figure 2 and includes:

[0051] S1. Data acquisition process: Drive the forearm to perform passive elbow flexion and extension at a set speed with the elbow joint as the axis, collect mechanical data and kinematic data during the movement, collect electromyography data during the movement through an electromyography sensor, and preprocess various types of collected data; in a specific embodiment, mechanical data during the movement can be collected through a force sensor, acceleration data during the movement can be collected through a gyroscope, angular data during the movement can be collected through an encoder, and electromyography data during the movement can be collected through an sEMG sensor.

[0052] The movement of the elbow joint is mainly completed by the coordinated contraction of the biceps brachii and triceps brachii. The triceps brachii is mainly responsible for the elbow extension movement of the elbow joint, and the biceps brachii is mainly responsible for the elbow flexion movement of the elbow joint. Therefore, to predict the upper limb muscle tone level of a patient, at least biomechanical signals during the upper limb movement and electromyography signals of the biceps brachii and triceps brachii (collected by an sEMG sensor) are required as a basis. In this embodiment, kinematic, mechanical, and sEMG signals required for training the muscle tone prediction model are collected, and multi-dimensional features are fused to achieve accurate quantification of the muscle tone level.

[0053] In the data acquisition process of this embodiment, various types of collected data are preprocessed, specifically including: processing the mechanical data, acceleration data, and angular data using the two-end removal algorithm; filtering the electromyography data (designing a 6th-order Butterworth band-pass filter of 20 Hz to 500 Hz to extract relevant movement information in the sEMG signal and remove most of the interference contained in the signal), performing full-wave rectification on the filtered signal to ensure that the signal output is positive, and deriving the envelope through a low-pass filter to observe the degree of amplitude change, as Figure 3 shown, the abscissa is time, and the ordinate is the amplitude change.

[0054] S2. Parameter fitting process: Based on the system motion differential equation of the mechanical impedance characteristic and the mechanical impedance formula of the system, biomechanical characteristic parameters are fitted according to the preprocessed mechanical data and kinematic data; electromyography characteristic parameters are obtained using the sliding window method according to the preprocessed electromyography data.

[0055] In this embodiment, for sEMG data, the eigenvalue is calculated by using the sliding window method. The length of the sliding window is set to 200 units, and it is translated 10 units each time. Since it is difficult to fully reflect the characteristics of the measured sEMG signal through a single characteristic parameter for the passive elbow movement of patients with different degrees of increased muscle tone, multiple characteristic parameters are needed to more accurately evaluate the muscle tone level of patients. The sEMG characteristic parameters selected in the present invention are root mean square (RMS), mean absolute value (MAV), simple square integral (SSI), variance of EMG signal (VAR), average amplitude change (ACC), waveform length (WL), mean power frequency (MPF), and median frequency (MF).

[0056] In biomechanics, the dynamic characteristics of the human limb can be described in the form of mechanical impedance. Mechanical impedance is a method to describe the characteristics of the limb from a mechanical perspective, and it generally includes three parts: elasticity, viscosity, and inertia. The bending and stretching movement of the patient's arm under the action of an external force is similar to a single-degree-of-freedom low-frequency vibration system in vibration theory. Therefore, the system motion differential equation based on the mechanical impedance characteristics is shown in Equation (1):

[0057]

[0058] where T is the torque of the system, K represents the elastic parameter (unit: N·m / °), B represents the viscous parameter (unit: N·m·s / °), and t is time; and θ are the angular acceleration, acceleration, and angle measured during the movement, respectively.

[0059] Since the product of the inertia coefficient and the angular acceleration is usually a constant, the offset is T is equal to the pressure F applied to the patient's wrist multiplied by the distance l from the applied force to the axis of rotation (elbow joint). Therefore, the motion differential equation of the system shown in Equation (1) can be further simplified to Equation (2):

[0060]

[0061] where F is the mechanical data collected by the force sensor (unit: N), l is the distance from the force sensor to the center of rotation, and C is the offset (unit: N·m); can be measured by a gyroscope (unit: ° / s), and θ can be measured by an encoder (unit: °).

[0062] Since the mechanical impedance characteristics of the limb are closely related to the movement of the limb. When the patient's upper limb makes passive back-and-forth movements under an external force, it has elastic and viscous characteristics. The mechanical impedance of the elastic part is the elastic coefficient K, and the mechanical impedance of the viscous part is the viscous coefficient multiplied by the angular frequency Bω of the forearm's back-and-forth movement. Therefore, the mechanical impedance Z of the system is the sum of the two impedances, as shown in Equation (3):

[0063] Z = K + Bω (3)

[0064] Where ω is the angular frequency of the forearm movement; the angular frequency is the average angular frequency. The peak finding method is used to find each peak and valley in the angle output waveform, and then the average angular frequency is calculated based on the time.

[0065] To obtain Z, the least squares parameter estimation method is used to find the optimal fitting parameters K, B, and C. Therefore, there are a total of 7 biomechanical characteristic parameters in this embodiment, namely K, B, C, Z, W, MRFE, and MREE. Among them, MRFE is the maximum resistance to elbow flexion, MREE is the maximum resistance to elbow extension, and both MRFE and MREE are measured by a force sensor. W is the curve area enclosed by the force-angle curve. The torque-angle curve is plotted with the torque of elbow flexion and extension as the ordinate and the angle as the abscissa, as Figure 4 shown.

[0066] S3. Model training process: Based on the electromyography characteristic parameters and biomechanical characteristic parameters, using the doctor's clinical assessment results of muscle tension level as labels, train the muscle tension level prediction model. The muscle tension level prediction model includes a classification model and SHAP, and SHAP is used to quantify the contribution of each feature to the model prediction.

[0067] In this embodiment, two clinical doctors use the double-blind evaluation method to rate the muscle tension of the patient according to the Ashworth (MAS) scale (Grade 0, Grade I, Grade I+, Grade II, Grade III, Grade IV). The doctor's rating will be used as the variable for correlation analysis and the true label for the classification algorithm.

[0068] First, the statistically derived features are subjected to principal component analysis (PCA) for dimensionality reduction, retaining more than 95% of the variance information to improve the computational efficiency. Subsequently, a support vector machine (SVM) is used as the classification model, and the doctor's clinical assessment of the muscle tension level serves as the label for training the machine learning model. The dataset is divided into a training set and a test set in a ratio of 8:2. To improve the interpretability of the SVM model and clarify the contribution of each feature to the muscle tension level assessment, the present invention combines SMV with the SHAP (SHapley Additive explanation) framework. SHAP quantifies the contribution of each feature to the model prediction by using Shapley values, solving the inherent "black box" problem of machine learning models. The Shapley value of the feature X j in the SVM model is given by Equation (4):

[0069]

[0070] where N is the original feature set (the feature set after Pearson correlation analysis, including sEMG features and biomechanical features), is any subset of all possible feature combinations excluding X j , |S| is the number of features in the subset, |N| is the total number of all features, f(S) is the output of the machine learning model for the feature subset S, and f(S∪{j}) is the output of the machine learning model for the feature subset S plus the feature X j .

[0071] Therefore, the overall process of the muscle tension level prediction model is as shown in Figure 5 . Among them, sEMGi and Fi are the sEMG features and biomechanical features after Pearson correlation analysis respectively, obtaining the input [sEMGi, Fi], and the output is the predicted value Grade of the muscle tension level and the contribution degree Contribution of the features. The SHAP framework is used to reveal the correlation between the features and the muscle tension level, and the SVM model is combined to rank the contribution degrees of the features, thereby enhancing the reliability and interpretability of the detection results.

[0072] S4. Muscle tension increase detection process: The elbow joint of the detection object is flexed and extended passively at a set speed, mechanical data, kinematic data, and electromyographic data are collected and preprocessed. After fitting to obtain biomechanical feature parameters and electromyographic feature parameters, they are input into the trained muscle tension level prediction model to obtain the predicted muscle tension level and the contribution degree of each feature to the prediction result. The contribution degrees of each feature are as shown in Figure 6 .

[0073] In a preferred implementation, the features that can best reflect patients with different levels of increased muscle tone are found as the input of the muscle tone level prediction model. After the parameter fitting process, a correlation analysis process can also be included. Using the doctor's rating as the variable for correlation analysis, Pearson correlation analysis is performed on the biomechanical characteristic parameters and electromyography data, and a numerical graph is plotted as shown in Figure 7 shown. Finally, the features with a correlation coefficient r > 0.8 and a significance level p < 0.01 are selected. The features after screening in this embodiment are as shown in the ordinate of Figure 6 . Through the correlation analysis process, the number of parameters input for model training can be reduced while not affecting the accuracy of the detection results.

[0074] In addition, since muscle tone may vary at different movement speeds and ranges, the current detection methods lack dynamic evaluation at different movement speeds and ranges. Therefore, in a preferred implementation, the forearm is driven to passively flex and extend the elbow joint around the elbow joint axis at two or more set speeds to obtain the corresponding electromyography characteristic parameters and biomechanical characteristic parameters, so as to realize the acquisition of sEMG data, mechanical data, and kinematic data of the elbow joint at different angles and speeds, improving the dynamic performance of muscle tone level detection. Then, through analysis, the corresponding characteristic parameter combinations at the specified speed are respectively selected and input into the trained muscle tone level prediction model to obtain the predicted muscle tension level and the contribution degree of each feature to the prediction result. Combining with the test modes at different angles and speeds, the dynamic changes of muscle tone with movement range and speed are evaluated.

[0075] In a specific embodiment, sEMG data, mechanical data, and kinematic data at different speeds (15° / s, 20° / s, 25° / s) are collected, and then feature extraction is performed on these data.

[0076] Regarding the sEMG data: Previous studies have shown that it is difficult to distinguish between grade 0 and grade I of MAS when using electromyography for evaluation. To address this problem, the sEMG features of the biceps and triceps at different speeds (15° / s, 20° / s, 25° / s) are combined and classified. The results show that when the RMS, MAV, SSI, VAR, AAC, WL, MPF, and MF features of the biceps at 20° / s and the triceps at 25° / s are selected, they can significantly distinguish between MAS grade 0 and grade I subjects.

[0077] For mechanical data and kinematic data: They are distinguished by biomechanical feature W, which increases with the increase of the detection speed (15° / s - 25° / s). When the detection speed is 20° / s, with the increase of the muscle tone level (from grade 0 to grade II), the increase of W is more significant (the average increase between each grade is 17.47 N·m·°), indicating that the resistance generated by the patient's arm during passive movement under external force increases significantly at this time.

[0078] Based on the above analysis results, the finally collected data are the sEMG features of the biceps brachii at 20° / s, the biomechanical features at 20° / s, and the sEMG features of the triceps brachii at 25° / s for subsequent prediction of the muscle tone grade, which improves the dynamic performance of the muscle tone grade detection. It is also possible to adjust the set speed and collect data according to further analysis to achieve the best muscle tone grade detection effect.

[0079] Based on the same inventive concept, the present application also provides a device for implementing the method in Embodiment 1. For details, see Embodiment 2.

[0080] Embodiment 2

[0081] In this embodiment, a multi-signal fusion upper limb muscle tone increase detection device is provided for performing the method described in Embodiment 1, as Figure 8 shown, including:

[0082] Support base 1;

[0083] Forearm drive unit 2, connected to the support base 1 through a rotating shaft 3, for fixing with the forearm and driving the forearm to perform passive elbow flexion and extension around the elbow joint at a set speed;

[0084] Force sensor, arranged on the wrist support plate 4 of the forearm drive unit 2, for collecting mechanical data during the movement process;

[0085] Motion sensor, arranged on the forearm drive unit, for collecting kinematic data during the movement process;

[0086] Electromyography sensor (i.e., sEMG sensor), with electrodes arranged on the skin surfaces of the biceps brachii and triceps brachii of the upper arm, for collecting electromyography data during the movement process;

[0087] Data processing unit, for obtaining the data collected by the force sensor, motion sensor, and electromyography sensor and processing them, and then outputting the predicted muscle tension grade and the contribution degree of each feature to the prediction result.

[0088] The specific positions of the motion sensor, electromyography sensor, and data processing unit are not shown in the figure, as long as they are arranged at positions where the corresponding functions can be realized.

[0089] Preferably, it further includes a control unit for controlling the forearm drive unit to drive the forearm to flex and extend the elbow passively at a set speed.

[0090] In one implementation, the motion sensor includes a gyroscope and an encoder. The gyroscope is used to collect speed and acceleration data during the movement process; the encoder is used to collect angle data during the movement process. By combining the rotation of the rotation axis 3 with the encoder, the angle data of the elbow joint flexion and extension can be accurately obtained, and the resistance of the elbow joint flexion and extension can be obtained through the force sensor.

[0091] This device has portability (can be directly used on the hospital bed), safety (a threshold is set for the force sensor), and dynamic testing capabilities (testing modes with multiple angles and multiple speeds). It supports direct use on the hospital bed, improving the clinical applicability of the device; at the same time, a safety threshold is set using the force sensor to avoid potential risks caused by forced movement to patients and improve the safety of the device.

[0092] The process of using an upper limb muscle hypertonia detection device with multi-signal fusion is as follows:

[0093] First, wipe the skin at the highest points of the biceps brachii and triceps brachii on the upper arm with alcohol, and then attach the dual-channel sEMG sensors to the bellies of the biceps brachii and triceps brachii on the upper arm respectively. Each sEMG channel is aligned with the direction of the muscle fibers. Subsequently, the subject places the upper arm on the support base 1, places the wrist horizontally on the wrist support plate 4 of the forearm drive unit 2, and at the same time adjusts the rotation center of the elbow joint to be level with the rotation axis 3 of the forearm drive unit 2 and fixes it with an elastic band. The range of motion of the elbow joint can be set between 0° and 120°.

[0094] Then the test starts. The forearm drive unit 2 drives the forearm to flex and extend the elbow passively around the elbow joint at a set speed, and collects mechanical data, kinematic data, and EMG data during the movement process. It is possible to collect sEMG, mechanical, and kinematic data of the passive movement process (from elbow flexion to elbow extension) of patients with different muscle tone levels at different speeds (for example, divided into three speed levels: 1, 2, and 3), and send these data to the data processing unit. Through the data processing unit, various types of collected data are preprocessed, and then biomechanical characteristic parameters and EMG characteristic parameters are obtained by fitting. These are input into the trained muscle tone level prediction model to obtain the predicted muscle tension level and the contribution degree of each characteristic to the prediction result and display them, as Figure 6 shown.

[0095] Since the device introduced in the second embodiment of the present invention is a device used to implement the method of the first embodiment of the present invention, those skilled in the art can understand the specific structure and deformation of the device based on the method introduced in the first embodiment of the present invention, so it is not described here in detail. All devices used in the method of the first embodiment of the present invention belong to the scope of protection of the present invention.

[0096] The embodiment of the present invention realizes multi-dimensional signal fusion by combining force sensors, sEMG sensors, encoders and gyroscopes. The patient's current muscle activation state and contraction strength can be obtained through sEMG features, and the elasticity and viscosity of the patient's muscles can be further analyzed on the basis of the biomechanical characteristics that can reflect different muscle tension resistances, thereby enhancing the redundancy and reliability of the data. Unlike systems that rely solely on sEMG signals, by capturing key biomechanical features, the sensitivity of sEMG signals to external interference is effectively reduced. At the same time, by combining the SVM model with the SHAP framework, in addition to outputting the patient's muscle tension level, the contribution of each feature can also be output, solving the "black box" problem inherent in the machine learning model. Make the classification results easier for doctors and patients to understand and accept. Cooperate with test modes of different angles and speeds to evaluate the dynamic changes of muscle tension with range of motion and speed.

[0097] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more of the processes and / or blocks Figure 1 one or more of the processes and / or blocks Figure 1 specified in the function.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes and / or blocks Figure 1 one or more of the processes and / or blocks Figure 1 specified in the function.

[0101] Although the specific embodiments of the present invention have been described above, those skilled in the art of this technology should understand that the specific embodiments we described are illustrative only and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered by the scope of the claims of the present invention.

Claims

1. A multi-signal fusion upper limb muscle tension detection method, characterized in that: include: Data acquisition process: Drive the forearm to passively flex and extend the elbow at a set speed with the elbow joint as the axis, collect mechanical data and kinematic data during the movement, collect electromyographic data during the movement through electromyographic sensors, and pre-process the collected data; Parameter fitting process: Based on the system motion differential equation of mechanical impedance characteristics and the system mechanical impedance formula, the biomechanical characteristic parameters are obtained by fitting the preprocessed mechanical data and kinematic data; According to the preprocessed electromyographic data, the electromyographic characteristic parameters are obtained by using the sliding window method; Model training process: Based on electromyographic feature parameters and biomechanical feature parameters, the muscle tension level prediction model is trained using the doctor's clinical assessment results of the muscle tension level as labels. The muscle tension level prediction model includes a classification model and SHAP, which is used to quantify the contribution of each feature to the model prediction; Increased muscle tension detection process: The elbow joint of the test subject is made to passively flex and extend at a set speed, and the mechanical data, kinematic data and electromyographic data are collected and preprocessed. After fitting the biomechanical characteristic parameters and electromyographic characteristic parameters, they are input into the trained muscle tension level prediction model to obtain the predicted muscle tension level and the contribution of each feature to the prediction result.

2. The method according to claim 1, characterized in that: After the parameter fitting process, a correlation analysis process is also included, in which the biomechanical characteristic parameters and electromyographic data are subjected to Pearson correlation analysis to screen out the characteristics with correlation coefficient r>0.8 and significance level p<0.

01.

3. The method according to claim 1, characterized in that: During the data acquisition process, all types of collected data are preprocessed, including: using the two-end removal algorithm to process the mechanical data, acceleration data and angle data; filtering the electromyographic data, full-wave rectifying the filtered signal to ensure that the signal output is positive, and deriving the envelope curve through a low-pass filter.

4. The method according to claim 1, characterized in that: During the parameter fitting process, the system motion differential equation based on the mechanical impedance characteristics is: Where T is the torque of the system, F is the mechanical data collected by the force sensor, is the acceleration data collected by the gyroscope, θ is the angle data of the encoder, K represents the elastic parameter, B represents the viscosity parameter, C is the offset, t is the time, and l is the distance from the force sensor to the rotation center; The mechanical impedance formula of the system is: Z=K+Bω Where, ω is the angular frequency of the forearm movement; The least squares parameter estimation method was used to find the optimal fitting parameters K, B and C, and then K, B, C, Z, MRFE and MREE were used as biomechanical characteristic parameters, among which MRFE was the maximum resistance of elbow flexion and MREE was the maximum resistance of elbow extension. Both MRFE and MREE were measured by force sensors.

5. The method according to claim 4, characterized in that: The parameter fitting process also includes drawing a torque-angle curve with the torque of elbow flexion and extension as the ordinate and the angle as the abscissa, and then calculating the curve area W enclosed by the force-angle curve as one of the biomechanical characteristic parameters.

6. The method according to claim 1, characterized in that: The EMG characteristic parameters include root mean square value RMS, mean absolute value MAV, simple square integral SSI, EMG signal variance VAR, average amplitude change ACC, waveform length WL, mean power frequency MPF and median frequency MF.

7. The method according to claim 1, characterized in that: Drive the forearm to passively flex and extend the elbow at two or more set speeds with the elbow joint as the axis to obtain the corresponding electromyographic characteristic parameters and biomechanical characteristic parameters. Select the corresponding characteristic parameter combinations at the specified speeds and input them into the trained muscle tension level prediction model to obtain the predicted muscle tension level and the contribution of each feature to the prediction result.

8. A multi-signal fusion upper limb muscle tension detection device, characterized in that: For executing the method according to any one of claims 1 to 7, the device comprises: Support base; A forearm driving unit, connected to the support base via a rotating shaft, used to be fixed to the forearm and drive the forearm to passively flex and extend the elbow at a set speed with the elbow joint as the axis; A force sensor is provided on the forearm drive unit and is used to collect mechanical data during the movement; A motion sensor, disposed on the forearm drive unit, for collecting kinematic data of the motion process; Electromyographic sensor: electrodes are placed on the skin surface of the biceps and triceps of the upper arm to collect electromyographic data during exercise; The data processing unit is used to obtain and process the data collected by the force sensor, motion sensor and electromyography sensor, and then output the predicted muscle tension level and the contribution of each feature to the prediction result.

9. The device according to claim 8, characterized in that: The invention also comprises a control unit, which is used for controlling the forearm driving unit to drive the forearm to passively flex and extend the elbow at a set speed.

10. The device according to claim 8, characterized in that: The motion sensor includes a gyroscope and an encoder. The gyroscope is used to collect speed and acceleration data of the motion process; the encoder is used to collect angle data of the motion process.

Images