Intelligent identification method and device for rotator cuff injury

Abstract

The invention discloses an intelligent identification method and device for rotator cuff injury, and relates to the technical field of rehabilitation medicine, and the method comprises the steps: constructing a data collection module, collecting rotator cuff motion data in a motion capture mode, an electromyographic signal mode and a medical image mode, and carrying out the time-space alignment and fusion processing of the collected data through a Kalman filter; carrying out biomechanical modeling on the basis of the fused data, carrying out intelligent identification analysis on a biomechanical model by adopting improved convolution operation, carrying out spatial information fusion in combination with a graph convolution network, and outputting global motion state representation by self-adaptive expansion convolution; and based on the global motion state representation, dividing risk levels, predicting damage risks in real time, and providing an early warning signal. According to the method, the association mechanism of the multi-dimensional sensor data and the risk level is established, so that the shoulder joint movement risk can be accurately evaluated in real time. The shoulder joint injury prevention effect is remarkably improved, and the sports injury risk is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of rehabilitation medicine, and in particular to a method and a device for intelligently identifying rotator cuff injuries. Background Art

[0002] In modern society, as people pay more and more attention to health and exercise, more and more people participate in various sports activities, fitness training and rehabilitation exercises. However, this has also led to an increase in the incidence of sports injuries, among which shoulder joint injury is the most common problem. The shoulder joint is one of the joints with the largest range of motion and the most complex structure in the human body, and is widely involved in daily life and sports. Due to its high flexibility, the shoulder joint is easily injured due to overuse, incorrect posture or other external factors during exercise. Common shoulder joint injuries include rotator cuff tears, acromion impingement, shoulder dislocation, etc. This type of injury not only affects sports performance, but may also lead to long-term functional disorders, and in severe cases even require surgical treatment and long-term rehabilitation. Therefore, how to effectively prevent shoulder joint injuries has become an important topic in sports science and rehabilitation medicine.

[0003] However, the existing shoulder joint motion monitoring system still faces some technical bottlenecks. First, how to accurately extract the motion characteristics of the shoulder joint and perform risk assessment in complex motion scenes remains challenging. The movement of the shoulder joint involves multiple planes and dimensions, and simple sensor data can hardly fully reflect its dynamic characteristics. Secondly, the real-time and personalization capabilities of the existing system are weak, and the ordinary threshold model cannot adapt to the motion patterns and physiological characteristics of different users, resulting in frequent false alarms. Finally, the traditional system lacks support for real-time feedback, and it is difficult for users to obtain timely guidance and protection during exercise. The existence of these problems limits the application effect and popularity of existing technologies in motion monitoring and injury prevention. Summary of the invention

[0004] In view of the problems existing in the existing intelligent identification methods for rotator cuff injuries, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is the deficiencies of the existing monitoring systems in terms of accuracy, real-time performance, personalized adaptability and injury prevention effect.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides an intelligent identification method for rotator cuff injuries, which includes: constructing a data acquisition module to collect rotator cuff motion data from motion capture, electromyographic signals and medical images, and using a Kalman filter to perform spatiotemporal alignment and fusion processing on the collected data; performing biomechanical modeling based on the fused data, and using an improved convolution operation to perform intelligent identification and analysis on the biomechanical model, introducing multi-scale convolution operations, combining graph convolution networks to perform spatial information fusion, and adaptively expanding convolution to output a global motion state representation; based on the global motion state representation, dividing the risk level, predicting the injury risk in real time, and providing an early warning signal.

[0007] As a preferred solution of the intelligent identification method for rotator cuff injury described in the present invention, the data acquisition module includes a motion capture submodule, an electromyographic signal acquisition submodule and a medical image acquisition submodule; the motion capture submodule adopts a micro-inertial sensor array, and arranges sensors at key anatomical positions of the acromion, humerus and scapula to collect three-dimensional acceleration, angular velocity, and Euler angle parameters; the electromyographic signal acquisition submodule adopts a surface electromyographic electrode array to monitor the deltoid, supraspinatus, infraspinatus, and teres minor muscle groups; the medical image acquisition submodule adopts MRI or ultrasound real-time imaging to scan T1WI, T2WI, and proton density weighted sequences.

[0008] As a preferred solution of the intelligent identification method of rotator cuff injury of the present invention, the spatiotemporal alignment and fusion processing of data are realized based on the Kalman filter combined with the hardware synchronization trigger mechanism; the state transfer equation of the Kalman filter is expressed as: ; in, It's time The state variable vector, is the state transition matrix, It's time The state variable vector, is the control input matrix, is the control vector, is the process noise; the observation equation is expressed as: ; in, is the measurement vector, is the observation matrix, is the measurement noise; the Kalman filter updates two variables at each step: the predicted state and the updated state, expressed as: ; ; in, is the predicted state estimate, Indicates at time The updated state estimate at is the forecast covariance matrix, is the updated covariance matrix, is the process noise covariance matrix, represents the transpose operation; the update step is expressed as: ; ; ; in: is the updated state estimate, which means that after introducing the current observation After that, adjust the obtained state estimate; is the updated covariance matrix, is the Kalman gain, is the measurement noise covariance matrix, is the identity matrix.

[0009] As a preferred solution of the intelligent identification method for rotator cuff injury of the present invention, the kinematic characteristics, dynamic characteristics and imaging characteristics of shoulder joint movement are analyzed by feature extraction based on the multimodal data aligned in time and space; the kinematic characteristics are calculated by IMU sensor data to obtain the angle of the joint. , angular velocity and acceleration ; The dynamic features are based on electromyographic signals to estimate muscle torque and joint force; the imaging features are extracted from MRI or ultrasound images to obtain morphological features, signal intensity and texture features of tissues around joints. The texture features are extracted through gray-level co-occurrence matrices, including contrast, entropy and correlation indicators, which are used to describe the complexity of tissue structure.

[0010] As a preferred solution of the intelligent identification method for rotator cuff injury described in the present invention, the improved convolution operation is a deep learning framework that adopts a temporal convolutional network and a long short-term memory network combined with an attention mechanism, and introduces multi-scale convolution operations, combines graph convolution networks for spatial information fusion, and adaptive dilated convolutions and dynamic convolution kernel designs to enhance the model's expressive ability in processing multimodal temporal data; the improved network architecture is expressed as follows: input layer: multimodal data input, including IMU sensor data, electromyographic signals, and image data; multi-scale adaptive convolution layer: convolution operations are performed at different time scales, combined with adaptive dilated convolutions and dynamic convolution kernels; graph convolution layer: processes the spatial relationship between joints and learns the spatial characteristics of joint movements; spatiotemporal fusion layer: fuses the output of multi-scale temporal convolution with the output of graph convolution to extract spatiotemporal joint features; classification layer: performs final action mode and injury mode classification through a fully connected layer or an RNN layer.

[0011] As a preferred solution of the rotator cuff injury intelligent identification method of the present invention, wherein: the spatial information fusion combined with the graph convolution network includes: setting the shoulder joint motion data to be represented as a graph , where V is the node set and E is the edge set, indicating the mutual connection between different joints, then the operation of graph convolution is expressed as: ; in, Is a node In the The feature vector of the layer, Neighbor node In the The feature vector of the layer; Is a node The set of neighbor nodes; It is The learnable weight matrix of the layer; is the normalization coefficient; is a nonlinear activation function, Representation Node neighbor nodes.

[0012] As a preferred solution of the rotator cuff injury intelligent identification method of the present invention, the multi-scale convolution operation is to introduce multiple convolution kernels in each layer of the network to process time series data of different scales respectively; assuming that the time series data is ,in is the sequence length, and the standard formula for the convolution operation is: ; in, For the time steps of input data; is the weight of the convolution kernel; is the size of the convolution kernel, Represents the time step The output of the following; In order to capture both short-term and long-term information at different time scales, multiple convolution kernels are defined. , the size of each convolution kernel is ; Therefore, the improved multi-scale convolution operation is: ; in, It is The convolution kernel is in time step Output on; It is The weight of the convolution kernel; It is The size of the convolution kernel, Indicates the number of convolution kernels; Concatenate the convolution outputs of all scales to get the final output of multi-scale convolution: ; The adaptive dilated convolution introduces an adaptive function , based on the data of the current time step To dynamically adjust the expansion factor, expressed as: ; The adaptive dilated convolution formula is expressed as: ; in, Through the adaptive function Dynamically adjusted expansion factor; is the weight of the convolution kernel; The dynamic convolution kernel is defined by a generating function , according to the input data The features of generate the weights of the convolution kernel, expressed as: ; in, According to the time step Input data Dynamically generated weights; The convolution operation is expressed as: ; in, According to the time step The first convolution kernel weights; It is a linear transformation or nonlinear network used to generate the weights of the convolution kernel based on the input data.

[0013] In the second aspect, the present invention provides an intelligent identification system for rotator cuff injuries, which includes: an acquisition module for collecting rotator cuff motion data, and using a Kalman filter to perform spatiotemporal alignment and fusion processing on the data; an identification module for performing biomechanical modeling based on the fused data, using an improved convolution operation, performing intelligent identification analysis, and combining a graph convolutional network to output a global motion state representation; an early warning module for dividing risk levels based on the global motion state representation, predicting injury risks in real time, and providing early warning signals.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: the processor implements any step of the above-mentioned method for intelligent identification of rotator cuff injuries when executing the computer program.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: the computer program, when executed by a processor, implements any step of the above-mentioned method for intelligent identification of rotator cuff injuries.

[0016] The beneficial effects of the present invention are as follows: the present invention can achieve real-time and accurate assessment of shoulder joint movement risks by establishing a correlation mechanism between multi-dimensional sensor data and risk levels. Not only can potential injury risks be discovered in a timely manner, but the warning threshold can also be dynamically adjusted according to the actual situation of the user. The accuracy and personalization of the warning can be continuously improved. The effect of shoulder joint injury prevention is significantly improved, a more accurate and personalized protection plan is provided for users, and the risk of sports injuries is effectively reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 Flow chart of the intelligent identification method for rotator cuff injury. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0019] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0021] Example 1: Reference Figure 1 , which is the first embodiment of the present invention, provides an intelligent identification method for rotator cuff injury, comprising: S1: Build a data acquisition module to collect rotator cuff motion data from motion capture, electromyographic signals and medical imaging, and use the Kalman filter to perform spatiotemporal alignment and fusion processing on the collected data.

[0022] Deploy medical sensors to collect rotator cuff motion data and identify anatomical landmarks. Functional testing is required before collection, and care should be taken to avoid signal interference between sensors. Construct a data acquisition module, including a motion capture submodule, an electromyography signal acquisition submodule, and a medical image acquisition submodule.

[0023] The motion capture submodule uses a miniature inertial sensor array, with sensors placed at key anatomical locations such as the acromion, humerus, and scapula to collect parameters such as three-dimensional acceleration, angular velocity, and Euler angles.

[0024] The EMG signal acquisition submodule uses a surface EMG electrode array to monitor the deltoid, supraspinatus, infraspinatus, teres minor and other muscle groups.

[0025] The medical image acquisition submodule uses MRI or ultrasound real-time imaging to scan T1WI, T2WI, proton density weighted and other sequences.

[0026] Perform preliminary preprocessing on the collected data, including noise suppression and image correction; Noise suppression uses Gaussian noise processing, determines mean filtering and Gaussian filtering, and uses Fourier transform filtering to eliminate stripe artifacts. After preliminary processing, random granular noise, abrupt white or black dots, and stripes generated during the scanning process in the image are eliminated to obtain clear, high-quality image data.

[0027] The data acquired by different acquisition submodules are fused to generate accurate and unified shoulder joint motion information. The data fusion process is divided into three steps: spatiotemporal alignment, feature extraction, and data dimension reduction and feature selection.

[0028] Since the sampling rates, data formats, and timing benchmarks of IMU, electromyography sensors, and medical imaging devices are different, spatiotemporal alignment is used to ensure data consistency. A hardware synchronization trigger mechanism is used in conjunction with a Kalman filter to achieve spatiotemporal alignment and fusion processing of data.

[0029] Hardware synchronization trigger mechanism: During the data acquisition phase, all sensors start sampling at the same time through the hardware synchronization trigger module to ensure that the same starting time is maintained within the system.

[0030] Kalman filter: used to fuse signals from different sensors on the same time basis and reduce noise interference.

[0031] The state transition equation of the Kalman filter is expressed as: ; in, It's time The state variable vector includes motion state information such as joint angle, angular velocity, acceleration, etc. is the state transition matrix, describing the transition from time arrive system dynamics. It's time The state variable vector, is the control input matrix, is the control vector, is the process noise.

[0032] The observation equation is expressed as: ; in, is a vector of measurements, including IMU, EMG signals, and imaging data; is the observation matrix, which maps the state to the measurement space. is the measurement noise.

[0033] The Kalman filter updates two variables at each step: the predicted state and the updated state, expressed as: ; ; in, is the predicted state estimate, Indicates at time The updated state estimate at is the forecast covariance matrix, is the updated covariance matrix, is the process noise covariance matrix, Represents a transpose operation.

[0034] The update step is expressed as: ; ; ; in: is the updated state estimate, which means that after introducing the current observation After that, adjust the obtained state estimate; is the updated covariance matrix, is the Kalman gain, is the measurement noise covariance matrix, is the identity matrix.

[0035] The Kalman filter is used to reduce the impact of different sampling frequencies and measurement errors to ensure high-precision fusion of data in a unified space-time framework. Based on the multimodal data after space-time alignment, feature extraction is used to analyze the kinematic, dynamic and imaging characteristics of the shoulder joint movement for subsequent biomechanical modeling and risk assessment.

[0036] Calculate the angle of the joint through IMU sensor data , angular velocity and acceleration .

[0037] angle The calculation is based on the Euler angle formula, which converts the sensor data into rotation angle information, expressed as: ; in, and The accelerometer is and Direction of acceleration.

[0038] Estimation of muscle torque based on electromyographic signals The muscle torque calculation formula is: ; in, represents the muscle torque, is the length of the lever arm, It is the muscle contraction force, which is obtained through the mapping relationship between electromyographic signal and force.

[0039] The morphological features, signal intensity and texture features of periarticular tissues are extracted from MRI or ultrasound images. The texture features can be extracted through the gray-level co-occurrence matrix, including indicators such as contrast, entropy and correlation, which are used to describe the complexity of tissue structure.

[0040] Furthermore, by reducing the dimension and selecting features to simplify the data structure and retaining the key features, principal component analysis (PCA) is used to map high-dimensional features to low-dimensional space to reduce redundant information, which can be expressed as: ; in, is the feature covariance matrix, is the eigenvector, is the mean vector, Indicates the number of samples.

[0041] RFE gradually reduces the feature dimension by recursively training the model and eliminating the features with the smallest weights.

[0042] In each iteration, the weight of each feature is calculated and the least important features are removed until the optimal feature set is reached.

[0043] The L1 regularization method achieves sparse feature selection by adding the L1 norm term to the loss function. The regularized loss function is: ; in, represents the objective function, which is used to measure the difference between the model prediction results and the actual results; and are the true value and the predicted value respectively, is the weight coefficient of the model, is the regularization coefficient, which is used to control the degree of sparsity. Represents the number of features in the model.

[0044] The multimodal data are fused and processed by the above method to obtain accurate shoulder joint motion characteristics.

[0045] S2: Based on the fused data, biomechanical modeling is performed, and improved convolution operations are used to intelligently identify and analyze the biomechanical model. Multi-scale convolution operations are introduced, and spatial information fusion is performed in combination with graph convolution networks. Adaptive dilated convolution is used to output a global motion state representation.

[0046] Finite element analysis is used to construct a mechanical model of the shoulder joint, and personalized anatomical structure reconstruction is performed based on individual imaging data. The complex joint structure and muscle tissue are divided into multiple finite units, and each unit is given corresponding material properties to simulate the real mechanical properties of the tissue.

[0047] The basic equation of the finite element model is expressed as: ; in, represents the stiffness matrix, reflecting the mechanical properties of the material, Represents the displacement vector, which represents the displacement of each unit under the action of external force. Represents the external force vector, which represents the external force acting on a joint or tissue.

[0048] In order to achieve personalized anatomical structure reconstruction, it is necessary to extract the geometric features of joints and soft tissues from the individual's imaging data, and set different material properties according to different tissues to effectively reflect the actual biomechanical properties in the model.

[0049] The muscle force and joint stress of the shoulder joint during movement are estimated through dynamic force analysis, which is achieved by using the inverse dynamics method. Through inverse dynamics analysis, the muscle strength and stress distribution of each joint under specific movement are calculated to provide reference data for subsequent injury risk assessment.

[0050] In order to effectively learn the temporal features of multimodal data and identify the action patterns and potential injury patterns in shoulder joint movement, a deep learning framework of temporal convolutional network and long short-term memory network combined with attention mechanism is adopted. Multi-scale convolution operation is introduced, and spatial information fusion is performed with graph convolutional network. Adaptive dilated convolution and dynamic convolution kernel design are used to enhance the model's expressive ability in processing multimodal temporal data, especially for the recognition of complex shoulder joint movement patterns and injury patterns.

[0051] The improved network architecture is expressed as: Input layer: multimodal data input, including IMU sensor data, electromyography signals and image data.

[0052] Multi-scale adaptive convolution layer: performs convolution operations at different time scales, combining adaptive dilated convolution and dynamic convolution kernels.

[0053] Graph convolution layer: processes the spatial relationship between joints and learns the spatial features of joint motion.

[0054] Spatiotemporal fusion layer: The output of multi-scale temporal convolution is fused with the output of graph convolution to extract spatiotemporal joint features.

[0055] Classification layer: The final action mode and damage mode classification is performed through the fully connected layer or RNN layer.

[0056] In a standard temporal convolutional network, the receptive field of the convolution kernel is fixed, and the convolution operation is performed only at one scale. To improve this, multiple convolution kernels are introduced in each layer of the network to process temporal data of different scales.

[0057] Assume that the time series data is ,in is the sequence length, and the standard formula for the convolution operation is: ; in, For the time steps of input data; is the weight of the convolution kernel; is the size of the convolution kernel, Represents the time step The following output.

[0058] In order to capture both short-term and long-term information at different time scales, multiple convolution kernels are defined. , the size of each convolution kernel is ; Therefore, the improved multi-scale convolution operation is: ; in, It is The convolution kernel is in time step Output on; It is The weight of the convolution kernel; It is The size of the convolution kernel, Indicates the number of convolution kernels.

[0059] Concatenate the convolution outputs of all scales to get the final output of multi-scale convolution: ; By defining multiple convolution kernels, the model can capture both short-term and long-term features in the same layer, improving the ability to process complex time series data.

[0060] In order to enhance the model's adaptability to different time steps, an adaptive dilation convolution is set so that the dilation factor It is no longer fixed, but dynamically adjusted according to the characteristics of each time step. Specifically: Introducing an adaptive function , based on the data of the current time step To dynamically adjust the expansion factor, expressed as: ; The adaptive dilated convolution formula is expressed as: ; in, Through the adaptive function Dynamically adjusted expansion factor; is the weight of the convolution kernel.

[0061] By introducing an adaptive function, the model is allowed to dynamically adjust the expansion factor according to the characteristics of the input data at each time step, thereby more flexibly capturing characteristics in different time ranges.

[0062] Furthermore, in shoulder joint motion, different action patterns and electromyographic signals may correspond to different feature patterns. Therefore, the weights of the convolution kernel can be dynamically adjusted according to the input data. The introduction of dynamic convolution kernels enables the model to adaptively learn the parameters of the convolution kernel according to the input features.

[0063] Define a generating function , according to the input data The features of generate the weights of the convolution kernel, expressed as: ; in, According to the time step Input data Dynamically generated weights; The convolution operation is then expressed as: ; in, According to the time step The first convolution kernels; It is a linear transformation or nonlinear network used to generate the weights of the convolution kernel based on the input data.

[0064] The motion data of the shoulder joint is not only a time series, but also involves spatial relationships. The graph convolutional network is used to process the spatial information and is combined with the temporal convolution for joint learning.

[0065] Set the shoulder joint motion data to be represented as a graph , where V is the node set and E is the edge set, indicating the mutual connection between different joints, then the operation of graph convolution is expressed as: ; in, Is a node In the The feature vector of the layer; Is a node The set of neighbor nodes; It is The learnable weight matrix of the layer; is the normalization coefficient; is a nonlinear activation function, Representation Node neighbor nodes.

[0066] Through graph convolution, the spatial relationship between different joints is captured and learned. The output of the graph convolution network is combined with the output of the temporal convolution network, and the spatiotemporal features are learned at the same time to improve the recognition ability of shoulder joint movement patterns and injury patterns. The output features of the combined temporal convolution and graph convolution are combined to integrate the spatiotemporal information to form a global shoulder joint movement state representation, which is used to predict the risk of injury in motion in real time and provide timely warning signals.

[0067] S3: Based on the global motion state representation, risk levels are divided, injury risks are predicted in real time, and early warning signals are provided.

[0068] In order to prevent sports injuries of the shoulder joint, the risk of the action is evaluated and classified in real time. By analyzing the multimodal data, the risk level is divided into the following three categories: Low risk: This means that the current movement pattern is within the normal range, meets biomechanical specifications, and will not have a negative impact on the shoulder joint.

[0069] Medium risk: This means that the current action does not exceed the safety range, but there may be potential injury risks due to incorrect posture, frequent repetition of movements or muscle fatigue.

[0070] High risk: This means that the current movement is beyond the normal range of motion of the shoulder joint or may cause acute injury and requires immediate intervention.

[0071] If the status indicates that the joint movement angle is within the normal physiological range of the shoulder joint and does not exceed the natural activity limit of the human body, the muscle activity intensity shows that the load is at a normal level without signs of excessive tension or fatigue, the movement rhythm, repetition frequency and stress conform to normal biomechanical laws, and no excessive stress load is applied to the joints; the acceleration, angular velocity and other sensor data are stable, the movements are coordinated, and no abnormal movements or fluctuations are detected, then it is judged to be a low-risk state, that is, a normal state.

[0072] If the status indicates that the joint movement angle is close to but not beyond the normal physiological activity range, it may be too close to the limit; the electromyographic signal shows that the muscle load begins to increase, and mild fatigue or muscle tension occurs, prompting the user to pay attention to avoid further stimulation; the repetition frequency gradually increases, and the movement pattern is unstable, which may lead to local stress concentration or increased joint load; the acceleration or angular velocity data fluctuates greatly, and the movement is no longer completely smooth, then it is judged to be a medium-risk state.

[0073] Furthermore, a mild warning will be issued, suggesting to adjust the posture, reduce the frequency of movement or slow down the speed of exercise. If the fatigue signal is obvious, it is recommended to take a proper rest or do light rehabilitation training to avoid further increasing the risk.

[0074] If the status indicates that the joint movement angle has exceeded the normal physiological limit, the out-of-range movement may cause joint injury; the electromyographic signal shows an obvious stress response, the muscle load is significantly too large, and may be accompanied by muscle spasms or abnormal tension, indicating an impending acute muscle injury; the movement pattern shows rapid acceleration or incoordination, the frequency is too high and not properly adjusted, significantly increasing the probability of joint and muscle injury; the acceleration or angular velocity data fluctuates violently, which may be accompanied by sudden movement impact, indicating the risk of acute injury, then it is a high-risk state.

[0075] Furthermore, a strong warning will be issued, requiring you to stop the current action immediately to avoid further damage. At this time, an emergency notification will be given through sound and light signals or vibration feedback, and it will be recommended to stop training and take relaxation or first aid measures to prevent further damage.

[0076] In summary, the present invention can achieve real-time and accurate assessment of shoulder joint movement risks by establishing an association mechanism between multi-dimensional sensor data and risk levels. It integrates multiple sensor data such as angle, acceleration, electromyographic signal and angular velocity, sets a scientific threshold system based on physiological standards and individual characteristics, and adopts a low, medium and high three-level risk assessment mechanism. It can not only detect potential injury risks in a timely manner, but also dynamically adjust the warning threshold according to the actual situation of the user. It significantly improves the effect of shoulder joint injury prevention, provides users with a more accurate and personalized protection plan, and effectively reduces the risk of sports injuries.

[0077] This embodiment further provides an intelligent identification system for rotator cuff injuries, including: An acquisition module is used to collect rotator cuff motion data and use a Kalman filter to perform spatiotemporal alignment and fusion processing on the data; The recognition module is used to perform biomechanical modeling based on the fused data, use improved convolution operations to perform intelligent recognition analysis, and combine the graph convolution network to output the global motion state representation; The early warning module is used to classify risk levels according to the global motion state representation, predict damage risks in real time, and provide early warning signals.

[0078] This embodiment also provides a computer device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the intelligent identification method for rotator cuff injuries proposed in the above embodiment.

[0079] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0080] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for intelligently identifying rotator cuff injuries as proposed in the above embodiment is implemented.

[0081] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0082] Embodiment 2: This embodiment provides an intelligent identification method for rotator cuff injury. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0083] A group of healthy volunteers and a group of volunteers with potential shoulder injury risk were selected, and the two groups of volunteers performed conventional shoulder joint exercise and high-intensity exercise respectively. Each group of volunteers performed shoulder joint movement monitoring and risk assessment using the present invention and the prior art respectively.

[0084] Test system 1: The multi-sensor fusion system of the present invention combines angle sensors, acceleration sensors, electromyography sensors, and angular velocity sensors to perform spatiotemporal feature learning and risk assessment through multi-scale adaptive convolutional networks and graph convolutional networks.

[0085] Test system 2: Existing shoulder joint monitoring system based on a single sensor and traditional simple threshold judgment method.

[0086] Basic data such as shoulder joint range of motion, electromyography, acceleration, etc. were collected from volunteers to build personalized physiological thresholds.

[0087] The volunteers performed regular exercise and high-intensity exercise respectively, and the two systems were used to record the real-time sensor data of each volunteer and the system's judgment on the risk level. The risk assessment, reaction time, false alarm rate, and missed alarm rate of the two systems were analyzed. The experimental results are shown in Table 1: Table 1 Risk assessment comparison results index The present invention Single sensor system Response time 450 ms 1200 ms Risk assessment false positive rate 5.7% 19.6% Risk assessment underreporting rate 2.5% 15.2% Mild discomfort rate 5.2% 22.1% Incidence of sports injuries 0% 8.6% The multi-sensor fusion system of the present invention has significant advantages in shoulder joint motion monitoring and risk assessment. Compared with the existing single-sensor system, the present invention shows better performance in terms of accuracy, real-time performance, personalized adjustment effect and injury prevention effect.

[0088] Specifically, the false alarm rate and missed alarm rate of the present invention are much lower than those of the prior art, indicating that it can more accurately identify potential risks in sports and reduce misjudgments. At the same time, the response time of the present invention is relatively fast, and warnings can be issued in time to provide faster protection. Most importantly, in terms of injury prevention effects, the present invention significantly reduces the incidence of mild discomfort and sports injuries among volunteers, further verifying its effectiveness in practical applications. Therefore, the present invention can provide users with a more accurate, real-time, and personalized shoulder joint protection solution, significantly improve sports safety, and has broad application prospects.

[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent identification method for rotator cuff injury, characterized in that: include: Construct a data acquisition module to collect rotator cuff motion data from motion capture, electromyographic signals, and medical imaging, and use the Kalman filter to perform spatiotemporal alignment and fusion processing on the collected data; Based on the fused data, biomechanical modeling is performed, and an improved convolution operation is used to perform intelligent recognition and analysis on the biomechanical model, a multi-scale convolution operation is introduced, and a graph convolution network is combined to perform spatial information fusion, and an adaptive dilated convolution is used to output a global motion state representation; Based on the global motion state representation, risk levels are divided, injury risks are predicted in real time, and early warning signals are provided.

2. The intelligent identification method for rotator cuff injury according to claim 1, characterized in that: The data acquisition module includes a motion capture submodule, an electromyographic signal acquisition submodule and a medical image acquisition submodule; The motion capture submodule uses a micro inertial sensor array, and arranges sensors at key anatomical positions of the acromion, humerus, and scapula to collect three-dimensional acceleration, angular velocity, and Euler angle parameters; The electromyographic signal acquisition submodule uses a surface electromyographic electrode array to monitor the deltoid, supraspinatus, infraspinatus, and teres minor muscle groups; The medical image acquisition submodule adopts MRI or ultrasound real-time imaging to scan T1WI, T2WI, and proton density weighted sequences.

3. The intelligent identification method for rotator cuff injury according to claim 2, characterized in that: Based on the Kalman filter combined with the hardware synchronization trigger mechanism, the spatiotemporal alignment and fusion processing of data are realized; The state transfer equation of the Kalman filter is expressed as: ； in, It's time The state variable vector, is the state transition matrix, It's time The state variable vector, is the control input matrix, is the control vector, is the process noise; The observation equation is expressed as: ； in, is the measurement vector, is the observation matrix, is the measurement noise; The Kalman filter updates two variables at each step: the predicted state and the updated state, expressed as: ； ； in, is the predicted state estimate, Indicates at time The updated state estimate at is the forecast covariance matrix, is the updated covariance matrix, is the process noise covariance matrix, Represents a transpose operation; The update step is expressed as: ； ； ； in: is the updated state estimate, which means that after introducing the current observation After that, adjust the obtained state estimate; is the updated covariance matrix, is the Kalman gain, is the measurement noise covariance matrix, is the identity matrix.

4. The intelligent identification method for rotator cuff injury according to claim 3, characterized in that: Based on the spatiotemporally aligned multimodal data, analyzing the kinematic characteristics, dynamic characteristics, and imaging characteristics of the shoulder joint motion by feature extraction; The kinematic feature is to calculate the angle of the joint through the IMU sensor data. , angular velocity and acceleration ; The dynamic characteristics are based on the estimation of muscle torque and joint force by electromyographic signals; The imaging features are morphological features, signal intensity and texture features of periarticular tissue extracted from MRI or ultrasound images. The texture features are extracted through gray-level co-occurrence matrix, including contrast, entropy and correlation indicators, which are used to describe the complexity of tissue structure.

5. The intelligent identification method for rotator cuff injury according to claim 4, characterized in that: The improved convolution operation adopts a deep learning framework combining a temporal convolutional network and a long short-term memory network with an attention mechanism, introduces a multi-scale convolution operation, combines a graph convolutional network for spatial information fusion, and uses adaptive dilated convolution and dynamic convolution kernel design to enhance the model's expressiveness in processing multimodal temporal data. The improved network architecture is expressed as: Input layer: multimodal data input, including IMU sensor data, electromyographic signals and image data; Multi-scale adaptive convolution layer: performs convolution operations at different time scales, combining adaptive dilated convolution and dynamic convolution kernels; Graph convolution layer: processes the spatial relationship between joints and learns the spatial features of joint motion; Spatiotemporal fusion layer: fuses the output of multi-scale temporal convolution with the output of graph convolution to extract spatiotemporal joint features; Classification layer: The final action mode and damage mode classification is performed through the fully connected layer or RNN layer.

6. The intelligent identification method for rotator cuff injury according to claim 5, characterized in that: The method of combining a graph convolutional network to perform spatial information fusion includes: Set the shoulder joint motion data to be represented as a graph , where V is the node set and E is the edge set, indicating the mutual connection between different joints, then the operation of graph convolution is expressed as: ； in, Is a node In the The feature vector of the layer, Neighbor node In the The feature vector of the layer; Is a node The set of neighbor nodes; It is The learnable weight matrix of the layer; is the normalization coefficient; is a nonlinear activation function, Representation Node neighbor nodes.

7. The intelligent identification method for rotator cuff injury according to claim 6, characterized in that: The multi-scale convolution operation is to introduce multiple convolution kernels in each layer of the network to process time series data of different scales respectively; Assume that the time series data is ,in is the sequence length, and the standard formula for the convolution operation is: ； in, For the time steps of input data; is the weight of the convolution kernel; is the size of the convolution kernel, Represents the time step The output of the following; In order to capture both short-term and long-term information at different time scales, multiple convolution kernels are defined. , the size of each convolution kernel is ; Therefore, the improved multi-scale convolution operation is: ； in, It is The convolution kernel is in time step Output on; It is The weight of the convolution kernel; It is The size of the convolution kernel, Indicates the number of convolution kernels; Concatenate the convolution outputs of all scales to get the final output of multi-scale convolution: ； The adaptive dilated convolution introduces an adaptive function , based on the data of the current time step To dynamically adjust the expansion factor, expressed as: ； The adaptive dilated convolution formula is expressed as: ； in, Through the adaptive function Dynamically adjusted expansion factor; is the weight of the convolution kernel; The dynamic convolution kernel is defined by a generating function , according to the input data The features of generate the weights of the convolution kernel, expressed as: ； in, According to the time step Input data Dynamically generated weights; The convolution operation is expressed as: ； in, According to the time step The first convolution kernel weights; It is a linear transformation or nonlinear network used to generate the weights of the convolution kernel based on the input data.

8. An intelligent identification system for rotator cuff injuries, based on the intelligent identification method for rotator cuff injuries according to any one of claims 1 to 7, characterized in that: include: An acquisition module is used to collect rotator cuff motion data and use a Kalman filter to perform spatiotemporal alignment and fusion processing on the data; The recognition module is used to perform biomechanical modeling based on the fused data, use improved convolution operations to perform intelligent recognition analysis, and combine the graph convolution network to output the global motion state representation; The early warning module is used to classify risk levels according to the global motion state representation, predict damage risks in real time, and provide early warning signals.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent identification method for rotator cuff injury described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent identification method for rotator cuff injury described in any one of claims 1 to 7 are implemented.

Images