Shock test method for rotator cuff injury based on sensor data analysis

Abstract

The invention discloses a rotator cuff injury impact test method based on sensor data analysis, and relates to the technical field of intelligent medical equipment, and the method comprises the steps: inputting a joint contracture data set into a generative adversarial network model, carrying out pressure propagation mode analysis through a three-dimensional convolution layer, carrying out energy hot spot aggregation through an attention mechanism layer, and carrying out pressure propagation mode analysis through a three-dimensional convolution layer; the method comprises the following steps: generating a pressure field comprehensive vector of a contracture joint, extracting gradient amplitude and energy density characteristics in the pressure field comprehensive vector, performing neighborhood smoothing through Gaussian average filtering, obtaining an acoustic impedance value matrix, synchronously performing spatial continuous mapping, obtaining an acoustic impedance distribution diagram, and obtaining an acoustic impedance value of the contracture joint according to the acoustic impedance distribution diagram. And performing dynamic compensation on the pressure field comprehensive vector through an acoustic transmission equation to obtain a corrected pressure field comprehensive vector, and converting the corrected pressure field comprehensive vector into an impact control signal through a PID (Proportion Integration Differentiation) controller. According to the invention, through the generative adversarial network model and the PID controller, the multi-physical field data of the joint contracture area can be integrated, and the frequency and pulse width of the shock wave therapeutic apparatus can be adjusted.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical equipment, and in particular to a rotator cuff injury impact testing method based on sensor data analysis. Background Art

[0002] With the continuous advancement of medical technology, shockwave therapy, as a non-invasive treatment method, has been widely used in the fields of orthopedics and rehabilitation medicine. Shockwave therapy transmits high-energy sound waves into the human body, stimulating tissue repair and regeneration, making it particularly suitable for treating joint contractures. In recent years, with the development of sensor technology and artificial intelligence, intelligent shockwave therapy devices based on embedded sensors have gradually become a research hotspot. Intelligent shockwave therapy devices based on embedded sensors can monitor patients' physiological feedback in real time and dynamically adjust treatment parameters based on specific conditions, thereby improving treatment effectiveness and patient comfort.

[0003] Traditional shockwave therapy methods typically focus on a single physical quantity (such as pressure or energy density), failing to fully reflect the complex biomechanical properties of the joint area, limiting treatment accuracy. Furthermore, existing shockwave therapy devices employ fixed parameter settings and lack the flexibility to adapt to changing treatment needs, making them particularly inadequate for diverse patients and pathological conditions. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a rotator cuff injury impact testing method based on sensor data analysis to solve the problems of insufficient physical quantity data fusion and lack of dynamic adjustment mechanism of existing shock wave therapy devices.

[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 a rotator cuff injury impact testing method based on sensor data analysis, comprising: inputting a joint contracture dataset into a generative adversarial network model, performing pressure propagation pattern analysis in a three-dimensional convolutional layer, and performing energy hotspot aggregation in an attention mechanism layer to generate a comprehensive pressure field vector of the contracture joint; Extract the gradient amplitude and energy density features from the integrated pressure field vector, perform neighborhood smoothing through Gaussian average filtering, obtain the acoustic impedance value matrix, and simultaneously perform spatial continuous mapping to obtain the acoustic impedance distribution map; According to the acoustic impedance distribution diagram, the pressure field comprehensive vector is dynamically compensated through the acoustic transmission equation to obtain the corrected pressure field comprehensive vector, which is then converted into an impact control signal through the PID controller. The shock control signal drives the shock wave therapy device to perform targeted impact on the contracted joint area, obtains the joint range of motion parameters and joint resistance, and dynamically adjusts the frequency and pulse width of the shock wave therapy device according to the joint range of motion parameters and joint resistance.

[0007] As a preferred embodiment of the rotator cuff injury impact testing method based on sensor data analysis of the present invention, the joint contracture data set includes muscle pressure data, joint resistance data and energy density data of the contracted joint.

[0008] As a preferred embodiment of the rotator cuff injury impact testing method based on sensor data analysis of the present invention, generating pressure field data of the contracture joint specifically includes the following steps: In the GAN framework, the three-dimensional convolutional layer and the attention mechanism layer are called and initialized, and the residual connection is applied simultaneously for stacking to build a generative adversarial network model; The joint contracture dataset is input into the generative adversarial network model, and the 3D convolutional layer performs multi-scale spatial differentiation through separable convolution kernels to obtain the pressure propagation feature tensor; The attention mechanism layer performs weight distribution and aggregation of energy hotspot areas through global pooling to form an energy aggregation matrix; The pressure propagation characteristic tensor and the energy aggregation matrix are integrated through the gating unit to generate the comprehensive vector of the pressure field of the contracture joint.

[0009] As a preferred solution of the rotator cuff injury impact testing method based on sensor data analysis of the present invention, the step of obtaining the acoustic impedance value matrix specifically includes the following steps: Directional angle normalization is used to enhance the biomechanical characteristics of the pressure field integrated vector, extract the gradient amplitude characteristics, and simultaneously apply Morlet wavelet transform to perform energy spectrum decomposition and extract energy density characteristics; Gaussian average filtering performs directional smoothing and boundary preservation on the gradient amplitude features and energy density features, and performs dynamic acoustic impedance mapping to form an acoustic impedance value matrix.

[0010] As a preferred solution of the rotator cuff injury impact testing method based on sensor data analysis described in the present invention, the obtaining of the acoustic impedance distribution map refers to applying thin plate spline interpolation to perform three-dimensional surface reconstruction on the acoustic impedance value matrix, and performing spatial continuous mapping through morphological closing operations to generate an acoustic impedance distribution map.

[0011] As a preferred solution of the rotator cuff injury impact testing method based on sensor data analysis of the present invention, the conversion into an impact control signal by a PID controller specifically includes the following steps: Extract the spatially varying parameters of the acoustic impedance distribution map and solve the acoustic transmission equation using the finite difference method to obtain the pressure wave attenuation matrix; The adaptive Kalman filter algorithm is used to estimate the state of the pressure wave attenuation matrix and obtain the attenuation compensation coefficient. Based on the attenuation compensation coefficient, the pressure field comprehensive vector is subjected to real-time error compensation and phase correction, and the corrected pressure field comprehensive vector is synchronously output. The PID controller performs frequency domain feature analysis on the corrected pressure field integrated vector and performs pulse width-frequency conversion to form an impact control signal.

[0012] As a preferred embodiment of the rotator cuff injury impact testing method based on sensor data analysis of the present invention, the acquisition of joint range of motion parameters and joint resistance specifically includes the following steps: The shock control signal is converted into an analog driving voltage through a DAC converter to stimulate the shock wave therapy device to generate focused shock waves; The focus of the focused shock wave is phase-encoded and spatially matched with the joint contracture target area, and the joint contracture target area is impacted through a piezoelectric ceramic array; The built-in six-dimensional force sensor is used to monitor the joint reaction force in real time, and the isokinetic dynamometer is used to track the joint movement trajectory to obtain the joint motion parameters and joint resistance.

[0013] As a preferred embodiment of the rotator cuff injury impact testing method based on sensor data analysis of the present invention, the dynamic adjustment of the frequency and pulse width of the shock wave therapy device specifically includes the following steps: The least squares method was used to perform time domain alignment and polynomial fitting on the joint range of motion parameters and joint resistance to form the joint resistance-angle curve; Extract dynamic mechanical parameters from the resistance-angle curve to obtain stiffness coefficient and viscosity coefficient; According to the stiffness coefficient, the shock wave therapy device is phase reconfigured to obtain the frequency adjustment parameters, and the frequency closed-loop adjustment is performed through the PID control algorithm; According to the viscosity coefficient, the wavelet threshold algorithm is used to perform damping adaptation to form an optimized duty cycle, and the hysteresis compensator is used to correct the pulse width in real time.

[0014] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the rotator cuff injury impact testing method based on sensor data analysis as described in the first aspect of the present invention is implemented.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the rotator cuff injury impact testing method based on sensor data analysis as described in the first aspect of the present invention.

[0016] The present invention achieves the following beneficial effects: By using a generative adversarial network model, it achieves comprehensive integration and precise analysis of multi-physics field data (such as muscle pressure, joint resistance, and energy density) in the joint contracture region. Furthermore, a PID controller is used for real-time dynamic adjustment, automatically adjusting the frequency and pulse width of the shockwave therapy device based on changes in joint range of motion parameters and joint resistance, ensuring optimal treatment results for different patients and pathological conditions. 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 use in the description of 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 paying any creative work.

[0018] Figure 1 Flowchart of the rotator cuff injury impact testing method based on sensor data analysis.

[0019] Figure 2 Flowchart of the process of generating the comprehensive vector of the pressure field.

[0020] Figure 3 Flowchart of the process of obtaining acoustic impedance distribution map.

[0021] Figure 4 This is the flow chart for the generation and adjustment of the shock control signal. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. 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.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figure 1 , is an embodiment of the present invention, which provides a rotator cuff injury impact testing method based on sensor data analysis, comprising the following steps: S1, the joint contracture dataset is input into the generative adversarial network model, the three-dimensional convolution layer performs pressure propagation pattern analysis, and the attention mechanism layer performs energy hotspot aggregation to generate a comprehensive pressure field vector of the contracture joint area.

[0026] See also Figure 1 , specifically including the following steps: S1.1, collecting a joint contracture dataset, which includes muscle pressure data, joint resistance data, and energy density data of the contracted joint; Muscle pressure data was collected using an electromyograph. First, the electromyograph electrodes were attached to the contracted joints to record the electrophysiological activity of the contracted joints in different states, specifically changes in electrophysiological indicators such as the F wave, M wave, H reflex, and T reflex. Particular attention was paid to the H reflex latency and the ratio of the maximum H reflex amplitude to the maximum M wave amplitude. The changes in electrophysiological indicators recorded by the electromyograph and the H / M wave amplitude ratio were integrated to obtain muscle electrical activity data. Joint resistance data is collected using an isokinetic dynamometer. The patient's contracted joint is fixed to the test arm of the isokinetic dynamometer, and the initial test speed and range of motion are set for the isokinetic dynamometer to begin measuring joint resistance. During the test, the isokinetic dynamometer applies resistance at a constant speed while recording changes in resistance within the contracted joint's range of motion. These resistance changes are then integrated using the isokinetic dynamometer's built-in Biodex software to generate joint resistance data. Energy density data is collected using an ultrasound imager. First, an ultrasonic coupling agent is applied to the contracted joint area, and a high-frequency ultrasound probe is placed at the contracted joint for scanning to obtain an ultrasound image of the contracted joint. Based on the ultrasound image, the built-in sound field energy analysis unit of the ultrasound imager is used to calculate and display the energy density distribution at different depths in real time to obtain energy density data.

[0027] S1.2. Preprocess the joint contracture dataset. For muscle pressure data, Butterworth bandpass filtering is used to extract the effective frequency band to eliminate high-frequency noise, and median filtering is used to suppress pulse interference to ensure the smoothness of the muscle pressure data. Multi-scale noise reduction is performed on the joint resistance data through wavelet packet decomposition, and a dynamic baseline correction algorithm is simultaneously applied to perform drift elimination to prevent measurement errors. Gaussian normalization is used to unify the dimensions of the energy density data to eliminate sensor differences, and outliers are removed based on the 3σ criterion to ensure the reasonable distribution of the energy density data. The preprocessed muscle pressure data, joint resistance data, and energy density data were aligned and normalized using Z-score standardization to obtain a preprocessed joint contracture dataset.

[0028] S1.3, build a generative adversarial network model and train it. Specifically, in the GAN framework, the nn.Convd parameter is used to call and initialize the 3D convolution layer: separable convolution kernel is used for downsampling, LeakyReLU activation function is set to prevent gradient disappearance, and convolution kernel weights are initialized and distributed through He; the nn.Sequential function is used to call and initialize the attention mechanism layer: global average pooling is used to perform dimensionality compression, and the Sigmoid function is used for spatial attention weight activation, and the quantity normalization unit is used simultaneously for feature channel fusion; the initialized dimensional convolution layer and attention mechanism layer are applied with residual connection for skip connection and layer normalization, and parameterized stacking is performed through depthwise separable convolution, and the constructed generative adversarial network model is output; Next, the generative adversarial network model was trained. Specifically, the joint contracture dataset was divided into a sample set, a training set, and a validation set. On the sample set, the Wasserstein loss function was used to alternately update the parameters of the generative adversarial network model to obtain the adversarial loss. On the validation set, the adversarial loss was smoothed using a sliding window, and the AdamW optimizer was applied to dynamically adjust the learning rate to generate a stably converged training curve. On the validation set, the peak detection algorithm was used to locate the extreme points of the stably converged training curve to obtain the FID score. When the FID score reached the convergence threshold for five consecutive rounds, training was terminated and the trained generative adversarial network model was output. It should be noted that the convergence threshold is defined based on the moving standard deviation of the historical FID scores and its value range is [12.0, 13.0].

[0029] S1.4. Obtain the comprehensive vector of the contracture joint pressure field through the generative adversarial network model. Specifically, the joint contracture dataset is input into the generative adversarial network model through the DataLoader interface of PyTorch. The three-dimensional convolution layer performs multi-scale spatial feature extraction on the input data through the separable convolution kernel to obtain the primary feature tensor. The three-dimensional coordinates of the primary feature tensor are downsampled on the X-axis, Y-axis, and Z-axis according to a fixed step size (such as step size 2) to form a biomechanical feature map. At the same time, the LeakyReLU function is used to perform nonlinear feature activation on the biomechanical feature map to generate an enhanced feature representation. Then, the pressure propagation pattern in the enhanced feature representation is captured through three layers of cascaded 3×3×3 void convolution. The pressure propagation pattern output by each layer is normalized and then multi-scale spatial differentiation is performed using the Sobel operator to obtain a 128-dimensional pressure propagation feature tensor. It should be noted that the Sobel operator is called and used through nn.Conv3d in PyTorch; The attention mechanism layer performs global spatial pooling on the biomechanical feature map, compressing the three-dimensional features of each channel in the biomechanical feature map into a 1×1×1 scalar to generate a channel energy vector. The channel energy vector is then nonlinearly transformed through two fully connected layers and activated using a Sigmoid function to generate normalized channel attention weights. The Multiply function is then called to multiply the channel attention weights with the biomechanical feature map channel by channel to highlight the characteristic responses of the energy hotspot area. A 3×3×3 convolution kernel is then used to perform cross-space weight aggregation to form an energy aggregation matrix. The pressure propagation feature tensor and the energy aggregation matrix are concatenated in the channel dimension, and then the channel number is compressed through a 1×1×1 convolution kernel to obtain a compressed fused feature tensor. The compressed fused feature tensor is then integrated using a dual-pathway structure: the main path uses a 3×3×3 depthwise separable convolution to extract spatial features and output a three-dimensional feature map, while the auxiliary path performs gated channel attention allocation through global average pooling and two fully connected layers to generate a gated weight vector. The gating unit is applied to multiply the three-dimensional feature map by the gated weight vector channel by channel to obtain a channel-weighted feature response map, and the GeLU activation function is used to enhance nonlinearity. LayerNorm is used simultaneously for normalization to generate a comprehensive pressure field vector of the contracted joint.

[0030] S2, extract the gradient amplitude and energy density characteristics in the pressure field comprehensive vector, and perform neighborhood smoothing through Gaussian average filtering to obtain the acoustic impedance value matrix, and simultaneously perform spatial continuous mapping to obtain the acoustic impedance distribution map.

[0031] See also Figure 3 , specifically including the following operations: S2.1, using directional angle normalization to enhance the biomechanical characteristics of the pressure field integrated vector, extracting the gradient amplitude characteristics, and simultaneously applying Morlet wavelet transform to perform energy spectrum decomposition and extract energy density characteristics; First, trilinear interpolation is used to perform three-dimensional spatial interpolation on the pressure field integrated vector to obtain standardized resolution voxel data. Principal component analysis is used to decompose the standardized resolution voxel data according to the x-axis, y-axis, and z-axis of the three-dimensional space to establish a local polar coordinate system. The pressure gradient direction in the pressure field integrated vector is extracted and mapped to the local polar coordinate system to generate a direction-encoded gradient field. The direction-encoded gradient field is normalized by the inverse tangent function to obtain a normalized direction angle feature map. Amplitude separation is performed on the normalized direction angle feature map to obtain gradient amplitude features. Next, based on the ISO biomechanical frequency band range, the pressure field integrated vector was divided into multiple characteristic frequency bands, and the frequency band parameters of the characteristic frequency bands were extracted. The frequency band parameters were subjected to three-dimensional complex convolution using the Morlet wavelet basis, and orthogonal decomposition was performed using the Hilbert transform kernel to generate multi-scale complex analysis results. The multi-scale complex analysis results were subjected to frequency band decomposition using three-dimensional convolution, and frequency domain conversion was performed using the fast Fourier transform. The power spectrum formula was simultaneously used to perform energy density quantification to obtain the energy density distribution of each frequency band. Based on the energy density distribution, feature dimensionality reduction was performed using principal component analysis, and dynamic range compression was performed using logarithmic transformation to obtain energy density characteristics. It should be noted that the ISO biomechanical frequency band refers to the vibration-sensitive frequency range of human tissue, which is defined by the International Organization for Standardization (ISO) technical committee through clinical vibration testing experiments and ranges from 0.5 Hz to 80 Hz.

[0032] S2.2: Gaussian average filtering performs directional smoothing and boundary preservation on the gradient amplitude and energy density features, and performs dynamic acoustic impedance mapping to form an acoustic impedance value matrix. Specifically, the gradient amplitude features are anisotropically weighted by Gaussian average filtering to obtain gradient amplitude sensitivity weights. Based on the gradient amplitude sensitivity weights, the gradient amplitude features are directional smoothed using a separable Gaussian convolution kernel, and boundary identification is performed using an edge detection algorithm. A sharpening filter is simultaneously applied to perform boundary enhancement to preserve the boundary of the ligament attachment surface. The energy density features are subjected to frequency-domain adaptive directional smoothing using Gaussian average filtering: a frequency-domain threshold is set to divide the energy density features into high-frequency energy density features and low-frequency energy features; the high-frequency energy density features are subjected to narrow-kernel smoothing using Gaussian average filtering to obtain high-frequency energy density smoothing results; the high-frequency energy density smoothing results are subjected to edge sharpening to preserve the boundaries of the contracted articular cartilage microstructure; the low-frequency energy density features are subjected to wide-kernel smoothing using Gaussian average filtering to generate low-frequency energy density smoothing results; the low-frequency energy density smoothing results are subjected to motion artifact suppression to preserve the boundaries of the main ligament structure; It should be noted that the frequency domain threshold is defined based on the ISO 2631-1 vibration standard and has a value range of 0.5 Hz -5 Hz; In the dynamic acoustic impedance mapping stage, the gradient amplitude features after smoothing and boundary preservation are concatenated with the energy density features to obtain a fused biomechanical feature tensor; the finite difference method is used to extract the acoustic impedance parameters in the fused biomechanical feature tensor, and the acoustic impedance parameters are sampled through three-dimensional spline interpolation to obtain a spatially continuous impedance distribution; the spatially continuous impedance distribution is dynamically mapped using the spatial weighted method, and finally an acoustic impedance value matrix is ​​generated.

[0033] S2.3, thin plate spline interpolation is applied to perform three-dimensional surface reconstruction on the acoustic impedance value matrix, and spatial continuous mapping is performed through morphological closing operations to generate an acoustic impedance distribution map. Specifically, thin plate spline interpolation is first applied to divide the acoustic impedance value matrix into 512×512×300 grid nodes, and key feature points are extracted through distance-weighted sampling to obtain control points; radial basis functions are used to fit the three-dimensional spatial deformation field of the control points to obtain initial deformation field parameters; gradient descent optimization is then used to perform parameter fine-tuning on the initial deformation field parameters to establish a spatial mapping relationship. Based on the spatial mapping relationship, continuous surface generation is performed through three-dimensional B-spline surface fitting to obtain the initial acoustic impedance surface; Then, the initial acoustic impedance surface is subjected to morphological closing operation to obtain the acoustic impedance distribution probability value. The specific mathematical formula is as follows: ; in, represents the probability value of acoustic impedance distribution, represents the surface steepness adjustment coefficient, represents the normalized acoustic impedance value, Represents positional parameters, represents the standard deviation adjustment parameter; It should be noted that the surface steepness adjustment coefficient is defined based on the gradient amplitude distribution of the initial acoustic impedance surface, and its value range is 5-12; the position parameter is defined based on the actual measurement value range of the initial acoustic impedance surface, and its value range is 0.5-0.8; the standard deviation adjustment parameter is defined based on the local heterogeneity of the initial acoustic impedance surface, and its value range is 0.1-0.3; The probability field of acoustic impedance distribution probability value is smoothed by spatial filtering, and anisotropic diffusion algorithm is used for spatial continuous mapping to generate acoustic impedance distribution map.

[0034] S3, according to the acoustic impedance distribution diagram, the pressure field comprehensive vector is dynamically compensated through the acoustic transmission equation, the corrected pressure field comprehensive vector is obtained, and it is converted into an impact control signal through the PID controller.

[0035] See also Figure 4 , specifically including the following operations: S3.1. Extract the spatially varying parameters of the acoustic impedance distribution map and solve the acoustic transmission equation using the finite difference method to obtain the pressure wave attenuation matrix. Specifically, a sliding window is used to divide the acoustic impedance distribution map into multiple time windows. The ratio of the standard deviation to the mean of the acoustic impedance values ​​within each window is calculated to obtain the spatial coefficient of variation. The spatial coefficient of variation is then subjected to local anisotropic fitting using Gaussian weighted smoothing to obtain a spatially varying parameter field. The spatially varying parameter field is then tensor-concatenated with the acoustic impedance distribution map to form a four-dimensional feature tensor containing the impedance values ​​and the spatial coefficient of variation. According to the four-dimensional characteristic tensor, the finite difference method is used to solve the acoustic transmission equation: the four-dimensional characteristic tensor is discretized by the second-order central difference through the finite difference method, and the difference format is established. The impedance gradient term of the difference format is processed by upwind difference to ensure numerical stability, and the boundary condition is set as an absorbing non-reflecting boundary; according to the difference format, the conjugate gradient solution of the acoustic transmission equation is iteratively solved to generate the complex pressure field distribution. According to the complex pressure field distribution, the attenuation analysis of different frequencies is performed to obtain the attenuation gradient value. The specific mathematical formula is as follows, ; in, represents the attenuation gradient value, represents the analysis frequency index, represents the maximum analysis frequency, represents the minimum analysis frequency, The weight coefficient representing the analysis frequency; Represents the single-frequency attenuation coefficient of the analysis frequency; It should be noted that the weight coefficient of the analysis frequency is defined based on the normalized power spectrum density of the complex pressure field distribution, and its value range is [0,1]. The single-frequency attenuation coefficient of the analysis frequency is defined based on the logarithmic energy attenuation rate, and its value range is [0,50]. Matrix normalization is used to compress the dynamic range of the attenuation gradient value and output the pressure wave attenuation matrix.

[0036] S3.2, using the adaptive Kalman filter algorithm to perform state estimation on the pressure wave attenuation matrix and obtain the attenuation compensation coefficient. Specifically, in the initialization phase, the adaptive Kalman filter algorithm calculates the plane wave approximate solution of the pressure wave attenuation matrix to obtain the initial state vector; the pressure wave attenuation matrix is ​​downsampled by quarter-spaced sampling to form an observation vector; the noise covariance matrix of the initial state vector and the observation vector is dynamically adjusted and calibrated using an adaptive mechanism to generate a state transfer matrix; In the prediction phase, the adaptive Kalman filter algorithm parameterizes the state transfer matrix, synchronously updates the state estimate, and generates a priori state estimate matrix. In the update phase, the variational Bayesian method is used to perform noise recalibration and bilateral filtering optimization on the priori state estimate matrix, and finally outputs the attenuation compensation coefficient. Based on the traditional Kalman filter framework, the adaptive Kalman filter algorithm uses an adaptive mechanism to dynamically adjust the process noise covariance matrix and the observation noise covariance matrix, thereby significantly improving the filtering accuracy and robustness in non-stationary environments. It is particularly suitable for physical field estimation problems with spatial variation characteristics.

[0037] S3.3, based on the attenuation compensation coefficient, perform real-time error compensation and phase correction on the pressure field integrated vector, and synchronously output the corrected pressure field integrated vector. Specifically, the attenuation compensation coefficient is resampled to the same resolution as the pressure field integrated vector through cubic spline interpolation, and the attenuation compensation coefficient and the pressure field integrated vector of the same resolution are subjected to amplitude-phase decomposition to obtain amplitude attenuation characteristics; based on the amplitude attenuation characteristics, a weighted compensation algorithm is used to perform multi-scale error compensation; The instantaneous phase of the amplitude attenuation feature is extracted through Hilbert transform, and the instantaneous phase is quantified by the phase offset using the sound wave propagation delay formula to obtain the delay compensation amount. According to the delay compensation amount, phase correction is performed through complex phase rotation to output the corrected pressure field comprehensive vector.

[0038] S3.4, the PID controller performs frequency domain feature analysis on the corrected pressure field integrated vector and performs pulse width-frequency conversion to form an impact control signal. Specifically, the corrected pressure field integrated vector is first converted to the frequency domain through fast Fourier transform (FFT), and the Hanning window function is used to suppress spectrum leakage to generate frequency domain features. The frequency domain features are input into the PID controller, and the proportional coefficient adjusts the gain coefficient in real time according to the frequency domain features to obtain the proportional correction amount. The integral coefficient compensates for the accumulated error of the frequency domain features through integral operation to generate an integral compensation amount. The differential coefficient performs third-order difference calculation on the frequency domain features to predict the oscillation trend and form a differential suppression amount. The proportional correction amount, integral compensation amount and differential suppression amount are integrated through weighted superposition to output a frequency domain correction signal; the frequency domain correction signal is converted from pulse width to frequency through pulse width modulation to obtain a square wave signal with adjustable duty cycle, which is then band-pass filtered and power amplified to form an impulse control signal.

[0039] S4, the shock control signal drives the shock wave therapy device to perform targeted impact on the contracted joint area, obtains the joint range of motion parameters and joint resistance, and dynamically adjusts the frequency and pulse width of the shock wave therapy device according to the joint range of motion parameters and joint resistance.

[0040] The specific operations include: S4.1, converting the shock control signal into an analog drive voltage via a DAC converter to stimulate the shock wave therapy device to produce focused shock waves. Specifically, the shock control signal is converted into an analog voltage signal via a high-precision DAC converter. The analog voltage signal is then processed through a three-stage amplification circuit: a low-noise op amp is used for voltage pre-amplification in the front stage, a high-voltage op amp is used for voltage boosting in the middle stage, and a linear power amplifier is used in the final stage to amplify the voltage to the analog drive voltage required for treatment. The analog driving voltage is transmitted to the intelligent sensor network built into the shock wave therapy device through a coaxial cable. The intelligent sensor network uses the inverse piezoelectric effect to convert the shock control signal into mechanical vibration. The generated shock wave is focused by the ellipsoid reflector to form a focused shock wave.

[0041] S4.2: Phase encoding and spatial matching of the focus of the focused shock wave with the joint contracture target area is performed, and the target area is impacted by a piezoelectric ceramic array. Specifically, ultrasound positioning is used to perform three-dimensional acoustic field scanning of the focused shock wave, and the focus is reconstructed using a time-domain deconvolution algorithm to obtain the focus of the focused shock wave. Next, an MRI machine is used to perform high-resolution imaging of the contracted joint, and the target area edge is extracted to generate the joint contracture target area. The focus of the focused shock wave and the target area of ​​the joint contracture are tracked in real time through ultrasonic echolocation. Dynamic wavefront correction is performed based on the real-time position using a digital phase retarder to complete phase encoding. Based on the phase encoding, spatial matching is performed using a multimodal image fusion algorithm to obtain the registered contracture joint treatment coordinates. According to the treatment coordinates of the contracture joint, the piezoelectric ceramic array in the shock wave therapy device is activated for multi-focus time-sequential excitation to obtain a customized shock wave sequence; the customized shock wave sequence is used to deliver precise energy to the target area of ​​the joint contracture.

[0042] S4.3, synchronously uses the embedded six-dimensional force sensor to monitor the joint reaction force in real time, and combines it with the isokinetic dynamometer to track the joint movement trajectory to obtain the joint motion parameters and joint resistance. In the specific operation, first use the embedded six-dimensional force sensor of the shock wave therapy device to monitor the joint reaction force in real time and perform three-dimensional trajectory tracking to generate joint motion parameters; use the isokinetic dynamometer to measure the joint resistance and obtain real-time joint resistance; synchronize the joint motion parameters and joint resistance through timestamp matching, and eliminate high-frequency noise through Butterworth low-pass filtering to ensure the data synchronization accuracy of the intelligent sensor network.

[0043] S4.4, use the least squares method to perform time domain alignment and polynomial fitting of the joint range of motion parameters and joint resistance to form a joint resistance-angle curve. In the specific operation, the joint range of motion parameters and joint resistance are first aligned in the time domain through cubic spline interpolation to obtain a unified time reference; the least squares method is used to perform linear regression on the time-aligned joint range of motion parameters and joint resistance to obtain preliminary fitting coefficients; a fifth-order polynomial fitting is performed on the preliminary fitting coefficients using a curve fitting tool (such as OriginLab). During the fitting process, the Tukey weighting function is used to identify and eliminate deviation outliers in the preliminary fitting coefficients to suppress outlier interference and generate a polynomial function; after fitting, interval sampling is applied to the polynomial function to perform curve discretization output to obtain the joint resistance-angle curve.

[0044] S4.5, extract dynamic mechanical parameters from the resistance-angle curve to obtain the stiffness coefficient and viscosity coefficient. Dynamically adjust the frequency and pulse width of the shock wave therapy device using the stiffness coefficient and viscosity coefficient. Specifically, first, differentiate the joint resistance-angle curve and extract the slope of the curve as the dynamic stiffness coefficient. Simultaneously, perform phase demodulation on the joint resistance-angle curve using Hilbert transform to obtain a phase lag spectrum. Use a peak detection algorithm to extract the dominant frequency from the phase lag spectrum and obtain the viscosity coefficient. The extracted dynamic stiffness coefficient is input into an embedded intelligent sensor network, which drives the shock wave therapy device to perform resonant frequency mapping and obtain the fundamental frequency value. A digital phase accumulator is then used to perform phase accumulation on the fundamental frequency value to form a phase reconfiguration signal. Based on the phase reconfiguration signal, a direct digital frequency synthesizer is used to perform sine wave synthesis to form an adjustable drive signal. The adjustable drive signal is sampled in real time using a phase difference detection circuit to obtain the frequency adjustment parameter. The frequency adjustment parameter is input into the PID controller, and the frequency closed-loop adjustment is performed through the PID control algorithm: the proportional coefficient dynamically adjusts the gain coefficient according to the frequency adjustment parameter, the integral coefficient accumulates the resonant frequency deviation of the frequency adjustment parameter, and the differential coefficient suppresses the high-frequency noise of the frequency adjustment parameter through low-pass filtering, and finally outputs the adjusted frequency; A wavelet threshold algorithm is used to perform multi-scale decomposition of the viscosity coefficient to obtain the frequency band energy spectrum. Soft thresholding is performed on the frequency band energy spectrum to extract the characteristic frequency band energy distribution. The characteristic frequency band energy distribution is mapped to the PWM duty cycle adjustment value using a standard energy normalization formula. Based on the PWM duty cycle adjustment value, damping adaptation is performed using a digital comparator, and amplitude normalization is performed using a sigmoid limiter function to obtain the optimized duty cycle. A lag compensator is used to perform phase advance compensation on the optimized duty cycle to obtain a time domain correction signal, and edge triggering is performed on the time domain correction signal to generate a driving pulse; the leading and trailing edges of the driving pulse are dynamically corrected to output the optimized pulse width.

[0045] 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 rotator cuff injury impact testing method based on sensor data analysis proposed in the above embodiment.

[0046] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0047] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the rotator cuff injury impact testing method based on sensor data analysis proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0048] In summary, this invention achieves comprehensive integration and precise analysis of multi-physics field data (such as muscle pressure, joint resistance, and energy density) in joint contracture areas through a generative adversarial network model. Furthermore, a PID controller is used for real-time dynamic adjustment, automatically adjusting the frequency and pulse width of the shockwave therapy device based on changes in joint range of motion parameters and joint resistance, ensuring optimal treatment results for different patients and pathologies.

[0049] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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. A rotator cuff injury impact testing method based on sensor data analysis, characterized by: include: The joint contracture dataset is input into the generative adversarial network model. The three-dimensional convolution layer analyzes the pressure propagation pattern, and the attention mechanism layer aggregates energy hotspots to generate a comprehensive pressure field vector of the contracture joint. Extract the gradient amplitude and energy density features from the integrated pressure field vector, perform neighborhood smoothing through Gaussian average filtering, obtain the acoustic impedance value matrix, and simultaneously perform spatial continuous mapping to obtain the acoustic impedance distribution map; According to the acoustic impedance distribution diagram, the pressure field comprehensive vector is dynamically compensated through the acoustic transmission equation to obtain the corrected pressure field comprehensive vector, which is then converted into an impact control signal through the PID controller. The shock control signal drives the shock wave therapy device to perform targeted impact on the contracted joint area, obtains the joint range of motion parameters and joint resistance, and dynamically adjusts the frequency and pulse width of the shock wave therapy device according to the joint range of motion parameters and joint resistance.

2. The rotator cuff injury impact testing method based on sensor data analysis according to claim 1, characterized in that: The joint contracture data set includes muscle pressure data, joint resistance data, and energy density data of the contracted joint.

3. The rotator cuff injury impact testing method based on sensor data analysis according to claim 1, characterized in that: Generating the comprehensive pressure field vector of the contracted joint specifically includes the following steps: In the GAN framework, the three-dimensional convolutional layer and the attention mechanism layer are called and initialized, and the residual connection is applied simultaneously for stacking to build a generative adversarial network model; The joint contracture dataset is input into the generative adversarial network model, and the 3D convolutional layer performs multi-scale spatial differentiation through separable convolution kernels to obtain the pressure propagation feature tensor; The attention mechanism layer performs weight distribution and aggregation of energy hotspot areas through global pooling to form an energy aggregation matrix; The pressure propagation characteristic tensor and the energy aggregation matrix are integrated through the gating unit to generate the comprehensive vector of the pressure field of the contracture joint.

4. The rotator cuff injury impact testing method based on sensor data analysis according to claim 3, characterized in that: The step of obtaining the acoustic impedance value matrix specifically includes the following steps: Directional angle normalization is used to enhance the biomechanical characteristics of the pressure field integrated vector, extract the gradient amplitude characteristics, and simultaneously apply Morlet wavelet transform to perform energy spectrum decomposition and extract energy density characteristics; Gaussian average filtering performs directional smoothing and boundary preservation on the gradient amplitude features and energy density features, and performs dynamic acoustic impedance mapping to form an acoustic impedance value matrix.

5. The rotator cuff injury impact testing method based on sensor data analysis according to claim 1, characterized in that: The obtaining of the acoustic impedance distribution map refers to performing three-dimensional surface reconstruction on the acoustic impedance value matrix by applying thin plate spline interpolation, and performing spatial continuous mapping by morphological closing operation to generate the acoustic impedance distribution map.

6. The rotator cuff injury impact testing method based on sensor data analysis according to claim 1, characterized in that: The conversion into an impulse control signal by a PID controller specifically includes the following steps: Extract the spatially varying parameters of the acoustic impedance distribution map and solve the acoustic transmission equation using the finite difference method to obtain the pressure wave attenuation matrix; The adaptive Kalman filter algorithm is used to estimate the state of the pressure wave attenuation matrix and obtain the attenuation compensation coefficient. Based on the attenuation compensation coefficient, the pressure field comprehensive vector is subjected to real-time error compensation and phase correction, and the corrected pressure field comprehensive vector is synchronously output. The PID controller performs frequency domain feature analysis on the corrected pressure field integrated vector and performs pulse width-frequency conversion to form an impact control signal.

7. The rotator cuff injury impact testing method based on sensor data analysis according to claim 6, characterized in that: The obtaining of the joint range of motion parameters and joint resistance specifically includes the following steps: The shock control signal is converted into an analog driving voltage through a DAC converter to stimulate the shock wave therapy device to generate focused shock waves; The focus of the focused shock wave is phase-encoded and spatially matched with the joint contracture target area, and the joint contracture target area is impacted through a piezoelectric ceramic array; The built-in six-dimensional force sensor is used to monitor the joint reaction force in real time, and the isokinetic dynamometer is used to track the joint movement trajectory to obtain the joint motion parameters and joint resistance.

8. The rotator cuff injury impact testing method based on sensor data analysis according to claim 1, characterized in that: The dynamic adjustment of the frequency and pulse width of the shock wave therapy device specifically includes the following steps: The least squares method was used to perform time domain alignment and polynomial fitting on the joint range of motion parameters and joint resistance to form the joint resistance-angle curve; Extract dynamic mechanical parameters from the resistance-angle curve to obtain stiffness coefficient and viscosity coefficient; According to the stiffness coefficient, the shock wave therapy device is phase reconfigured to obtain the frequency adjustment parameters, and the frequency closed-loop adjustment is performed through the PID control algorithm; According to the viscosity coefficient, the wavelet threshold algorithm is used to perform damping adaptation to form an optimized duty cycle, and the hysteresis compensator is used to correct the pulse width in real time.

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 rotator cuff injury impact testing method based on sensor data analysis according to any one of claims 1 to 8 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 rotator cuff injury impact testing method based on sensor data analysis according to any one of claims 1 to 8 are implemented.

Images