A method for monitoring the movement of orthopedic patients

Abstract

This invention relates to the fields of medical big data analysis and orthopedic data monitoring, and discloses a method for monitoring the movement of orthopedic patients. The method includes: S1, collecting patient movement data; S2, performing inter-frame image processing on the collected data to extract the patient's gait features; S3, processing adjacent frames to obtain an original grayscale image, a histogram equalized image, and a denoised image; S4, fusing the original grayscale image, histogram equalized image, and denoised image to obtain a preprocessed image; S5, performing registration and difference calculation on the preprocessed image; S6, training and labeling the data using the inter-frame difference method; S7, constructing an orthopedic patient movement monitoring model; and S8, obtaining monitoring results based on the movement monitoring model. The beneficial effects of this invention are: through multi-dimensional dynamic analysis of patient movement details, it achieves full-process, refined tracking of patient movement behavior.

Description

Technical Field

[0001] This invention relates to the fields of medical big data analysis and orthopedic data monitoring, specifically to a method for monitoring and early warning of orthopedic patient movement. Background Technology

[0002] In the recovery process after orthopedic fracture fixation surgery, especially for special groups such as patients with osteoporosis and some elderly people with fragile bones, measures such as restricting the movement of these vulnerable groups or mandatory bed rest are generally taken to reduce the risk of re-injury.

[0003] Traditional patient activity assessment mainly relies on manual observation, regular examinations, or fixed-point measurements based on wearable sensors. These methods have problems such as strong subjectivity, poor continuity, and low patient compliance, making it difficult to achieve full-process and refined tracking of patient movement behavior. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for monitoring the movement of orthopedic patients, which achieves full-process and refined tracking of patient movement behavior through multi-dimensional dynamic analysis of patient movement details.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] A method for monitoring the movement of orthopedic patients, comprising the following steps:

[0007] S1. Collect patient movement data, perform inter-frame image processing on the collected data, and extract the patient's gait features;

[0008] S2. Process the acquired adjacent frame images to obtain the original grayscale image, histogram equalized image, and denoised result image;

[0009] S3. Perform feature fusion on the original grayscale image, histogram equalized image, and denoised result image to obtain the preprocessed image;

[0010] S4. Perform registration and difference calculation on the preprocessed image;

[0011] S5. Use the inter-frame difference method to train and label the data;

[0012] S6. Construct a movement monitoring model for orthopedic patients;

[0013] S7. Monitoring results are obtained based on the mobile monitoring model.

[0014] Furthermore, in S2, a weighted average method is used to calculate the grayscale value to obtain the original grayscale image;

[0015] The cumulative distribution function is used to map the original gray values ​​to a new distribution, resulting in a histogram-equalized image;

[0016] A nonlocal mean denoising method is adopted, which uses the redundant information of similar structures in the image to perform weighted average denoising, preserves edge details, and matches the range of similar blocks to obtain the denoised image.

[0017] Furthermore, the feature fusion method described in S3 is calculated by adding the pixel values ​​of corresponding pixel positions and then taking the average.

[0018] Furthermore, S4 specifically includes:

[0019] Align the preprocessed images corresponding to two consecutive frames to the same spatial coordinate system to eliminate geometric offset caused by differences in shooting angle, position or time;

[0020] By traversing all feature descriptors of the two graphs, the Hamming distance is calculated, and the top 50% of matching pairs with the smallest distance are retained to reduce mismatch interference.

[0021] The difference between two consecutive frames of images after quantization registration is used to identify obvious displacement features.

[0022] The frame difference method was used to process the data of two consecutive frames of original images to obtain the movement change data of the first orthopedic patient;

[0023] The frame difference method was used to process the original grayscale data corresponding to two consecutive frames of images to obtain the second orthopedic patient movement change data;

[0024] The histogram equalization image data corresponding to two consecutive frames were processed using the frame difference method to obtain the movement change data of the third orthopedic patient.

[0025] The frame difference method was used to process the denoised image data corresponding to two consecutive frames to obtain the movement change data of the fourth orthopedic patient.

[0026] The preprocessed image data corresponding to two consecutive frames were processed using the inter-frame difference method to obtain the movement change data of the fifth orthopedic patient.

[0027] The Canny edge detection image data corresponding to two consecutive frames were processed using the frame difference method to obtain the movement change data of the sixth orthopedic patient.

[0028] The ORB feature point image data corresponding to two consecutive frames were processed using the inter-frame difference method to obtain the movement change data of the seventh orthopedic patient.

[0029] Furthermore, the movement change data of the first, second, third, fourth, fifth, sixth, and seventh orthopedic patients, along with the pixel-level difference heatmap, are fused to obtain training sample data for orthopedic patient movement monitoring.

[0030] Furthermore, S6 constructs a motion monitoring model for orthopedic patients, including:

[0031] Estimation of topological manifold kernel density in orthopedic patient movement monitoring data;

[0032] The bandwidth parameter is adaptively selected by minimizing the error function;

[0033] Forward propagation processing of data for training convolutional neural networks;

[0034] Convolution kernel weights are updated based on topological manifold kernel density estimation;

[0035] In electromagnetic field-guided optimization, the loss function is calculated;

[0036] Learning rate adaptively adjusted based on changes in the loss function of convolutional neural networks;

[0037] The stopping condition for the iteration of convolutional neural network training is determined when the relative rate of change of the validation set loss function is less than a threshold in three consecutive iterations. If the model is considered to have converged, the iteration is stopped, indicating that the training of the convolutional neural network model is complete.

[0038] Furthermore, the movement data is derived from video recordings of the patient's gait and movement patterns, and during the data collection process, the patient is instructed to walk within a standardized gait analysis area.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] 1. This invention performs inter-frame difference analysis on multiple source image data such as original image, grayscale image, histogram equalized image, denoised image, edge detection image, and ORB feature map, and merges them into a unified training sample to achieve multi-dimensional dynamic analysis of patient action details.

[0041] 2. This invention utilizes kernel density estimation to analyze the manifold structure of data in a high-dimensional feature space, thereby dynamically focusing on the detailed learning of dense regions. Compared with traditional convolutional neural network methods, this method can avoid training bias under high-dimensional sparse data and significantly improve the model's sensitivity to subtle gait changes.

[0042] 3. This invention optimizes the weight update path by simulating the multi-force controlled behavior of particles in an electromagnetic field, preventing them from getting trapped in local optima, guiding the network to perform a more effective global search in a high-dimensional space, and improving training robustness.

[0043] 4. This invention automatically adjusts the kernel density estimation bandwidth based on minimizing the error function, taking into account both local details and global smoothness. Compared with the fixed bandwidth method, it is more adaptable to changes in dense / sparse regions of data and improves the model's generalization ability.

[0044] 5. This invention dynamically adjusts the learning rate based on the rate of change of the loss function, effectively suppressing oscillations and overfitting in the later stages of training, and making up for the problem that conventional fixed decay strategies are insufficient in responding to complex data scenarios. Attached Figure Description

[0045] Appendix Figure 1 This is a flowchart of the present invention;

[0046] Appendix Figure 2 This is a data example of two consecutive raw images, illustrating the process of movement in an orthopedic patient;

[0047] Appendix Figure 3 It consists of the original grayscale image, histogram equalized image, and denoised result image corresponding to the first original image frame;

[0048] Appendix Figure 4 These are the original grayscale image, histogram equalized image, and denoised result image corresponding to the second frame of the original image;

[0049] Appendix Figure 5 These are the preprocessed image corresponding to the first original image, the Canny edge detection image, and the ORB feature point (green) image;

[0050] Appendix Figure 6 These are the preprocessed image corresponding to the original image of the second frame, the Canny edge detection image, and the ORB feature point (green) image;

[0051] Appendix Figure 7 It is a pixel-level difference heatmap obtained from data of two consecutive frames of images;

[0052] Appendix Figure 8 It is a three-dimensional heat map;

[0053] Appendix Figure 9 It is a contour map

[0054] Appendix Figure 10 This is a comparison chart of the relationship between dynamic adjustment of the learning rate and training loss. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] To facilitate understanding of this embodiment, a method for monitoring the movement of orthopedic patients disclosed in this embodiment will first be described in detail. Figure 1 A flowchart of a method for monitoring the movement of orthopedic patients disclosed in an embodiment of the present invention is shown, as follows: Figure 1 As shown, this method for monitoring the movement of orthopedic patients includes the following steps:

[0057] S1. Collect patient movement data, perform inter-frame image processing on the collected data, and extract the patient's gait features;

[0058] In the orthopedic patient mobility monitoring method proposed in this invention, the data acquisition method relies on video recording of the patient's gait and movement patterns, using fixed cameras or handheld video equipment, and is carried out in a standardized laboratory environment or rehabilitation center.

[0059] Video capture uses an RGB three-channel camera, with the frame rate typically set at 60fps to ensure the continuity of gait movements and the clarity of details;

[0060] During data collection, patients are instructed to walk within a standardized gait analysis area. Cameras can be positioned at different angles, such as front, side, and back, to capture complete gait data. The laboratory environment is equipped with fixed marker lines or reference objects. Figure 2 The example shown is a data sample of two consecutive frames of original images illustrating the process of movement in an orthopedic patient.

[0061] After acquisition, the video data will undergo inter-frame image processing, which involves analyzing consecutive frames to extract the patient's gait characteristics. Specifically, a keyframe extraction method is used, which selects keyframes at specific moments in the gait cycle (such as when the foot touches the ground, leaves the ground, or swings) to reduce redundant information.

[0062] To eliminate imaging interference, unify image representation, and enhance the recognizability of key structures (skeleton / marker points), the data is preprocessed as follows:

[0063] S2. Process the acquired adjacent frame images to obtain the original grayscale image, histogram equalized image, and denoised result image;

[0064] Specifically, the acquired adjacent frame images are first processed to grayscale, converting the RGB three-channel image into a single-channel grayscale image, removing redundant color information, simplifying subsequent calculation complexity, and reducing memory usage;

[0065] This invention uses a weighted average method to calculate grayscale values, obtaining the original grayscale image, as follows:

[0066] I gray =0.299R img +0.587G img +0.114B img

[0067] In the formula, I gray R represents the grayscale pixel value. img G represents the pixel value of the first channel of the image. img B represents the second channel pixel value of the image. img The third channel pixel value of the image

[0068] Furthermore, histogram equalization is performed to redistribute pixel grayscale values, thereby expanding the dynamic range and enhancing contrast.

[0069] This invention uses a cumulative distribution function to map the original gray values ​​to a new distribution, resulting in a histogram-equalized image.

[0070] Furthermore, this invention employs a nonlocal means denoising method, utilizing redundant information from similar structures in the image for weighted average denoising while preserving edge details. The search window of the nonlocal means denoising method is 21×21 pixels, matching the range of similar blocks to obtain the denoised image, as shown below. Figure 3 and Figure 4 As shown.

[0071] S3. Perform feature fusion on the original grayscale image, histogram equalized image, and denoised result image to obtain the preprocessed image;

[0072] The feature fusion method uses weighted feature processing, specifically by adding the pixel values ​​of corresponding pixel positions and then taking the average.

[0073] Furthermore, to locate key feature points, continuous edges are extracted from the preprocessed image using the Canny edge detection method to obtain the Canny edge detection image.

[0074] Furthermore, the ORB feature point detection (Oriented FAST and Rotated BRIEF) method is used to process the preprocessed image to quickly locate corner points (such as bone corners and marker center points) and extract texture features around the feature points using binary encoding.

[0075] Examples of preprocessed images, Canny edge detection images, and ORB feature point (green) images are as follows: Figure 5 and Figure 6 As shown.

[0076] S4. Perform registration and difference calculation on the preprocessed image;

[0077] Align the preprocessed images corresponding to two consecutive frames to the same spatial coordinate system to eliminate geometric offset caused by differences in shooting angle, position or time;

[0078] By traversing all feature descriptors of the two graphs, the Hamming distance (ORB is a binary descriptor) is calculated, and the top 50% of matching pairs with the smallest distance are retained to reduce mismatch interference.

[0079] Furthermore, by quantifying the differences between two consecutive registered images and identifying obvious displacement features, this invention employs a pixel-level difference processing method to process the two consecutive registered images, obtaining a pixel-level difference heatmap, such as... Figure 7 As shown, the Jet color coding difference value is used in practice. Warm colors (red / yellow) represent areas with high difference. The distribution statistics of the difference value are obtained by statistical histogram to determine whether there are concentrated abnormal areas (such as the right tail of the histogram).

[0080] S5. Use the inter-frame difference method to train and label the data;

[0081] The frame difference method was used to process the data of two consecutive frames of original images to obtain the movement change data of the first orthopedic patient;

[0082] The frame difference method was used to process the original grayscale data corresponding to two consecutive frames of images to obtain the second orthopedic patient movement change data;

[0083] The histogram equalization image data corresponding to two consecutive frames were processed using the frame difference method to obtain the movement change data of the third orthopedic patient.

[0084] The frame difference method was used to process the denoised image data corresponding to two consecutive frames to obtain the movement change data of the fourth orthopedic patient.

[0085] The preprocessed image data corresponding to two consecutive frames were processed using the inter-frame difference method to obtain the movement change data of the fifth orthopedic patient.

[0086] The Canny edge detection image data corresponding to two consecutive frames were processed using the frame difference method to obtain the movement change data of the sixth orthopedic patient.

[0087] The ORB feature point (green) image data corresponding to two consecutive frames were processed using the frame difference method to obtain the movement change data of the seventh orthopedic patient.

[0088] The movement change data of the first, second, third, fourth, fifth, sixth, and seventh orthopedic patients, along with pixel-level difference heatmaps, are fused to obtain training sample data for orthopedic patient movement monitoring.

[0089] The feature fusion method uses weighted feature processing, specifically by adding the pixel values ​​of corresponding pixel positions and then taking the average.

[0090] The frame difference method is a method to obtain the contour of a moving target by performing a difference operation on two adjacent frames or several frames apart in a video image sequence. It subtracts the pixel values ​​of two adjacent frames or several frames apart in the video stream and performs thresholding on the subtracted images to extract the moving region.

[0091] Furthermore, the training sample data for monitoring the movement of orthopedic patients was labeled manually based on two consecutive frames of original images. The labeling categories included: small amplitude (safe), medium amplitude (warning), and large amplitude (high risk), for a total of three categories.

[0092] S6, Orthopedic Patient Motion Monitoring Model Training

[0093] This invention uses a convolutional neural network for monitoring the movement of orthopedic patients. The training process of the convolutional neural network is as follows:

[0094] S601. Perform topological manifold kernel density estimation on orthopedic patient mobility monitoring data.

[0095] The training dataset, composed of orthopedic patient movement monitoring data, is vectorized to form a high-dimensional feature space. These data often exhibit uneven distribution and local high-density clustering in orthopedic patient movement monitoring.

[0096] Conventional techniques typically use overall statistical distribution or simple distance metrics for data processing. However, they are difficult to accurately grasp key areas when faced with high-dimensional sparsity and local clustering, thus ignoring subtle distributional differences in patient movement characteristics.

[0097] This invention characterizes the local density of each data point in high-dimensional space through topological manifold kernel density estimation, thereby identifying potential manifold structures. Kernel density estimation of data points makes subsequent feature extraction and classification more targeted. The calculation method is expressed as follows:

[0098]

[0099] In the formula, is the kernel density estimate of the data points; x is the input of the function, representing the data points input to the convolutional neural network; n is the total number of training data; x i Let be the feature vector of the i-th data point; i is a positive integer; K(·) is the kernel function (often a Gaussian kernel); h is the bandwidth parameter.

[0100] The aforementioned topological manifold kernel density estimation refers to the use of kernel density estimation methods to characterize the local density of each data point in a high-dimensional data space, taking into account the complex and variable local features of orthopedic patient mobility monitoring data. By identifying densely clustered areas of data points, the intrinsic topological structure of the data can be revealed.

[0101] The manifold structure refers to the essential geometric shape or layout of orthopedic patient movement monitoring data in a high-dimensional space. These structures are manifested through the local similarity and continuity of the data.

[0102] The data point refers to a point in a high-dimensional feature space that represents a single observation instance, specifically a single data point from a motion monitoring of an orthopedic patient.

[0103] s602, Optimization of bandwidth parameters.

[0104] In the high-dimensional scenario of orthopedic patient movement monitoring, the bandwidth parameter directly determines the balance between the local and global aspects of the convolutional neural network kernel function. Different bandwidths will affect the smoothness and detail preservation of density estimation.

[0105] Conventional techniques typically fix an empirical bandwidth or select it through simple cross-validation, which can easily lead to inaccurate estimation in sparse regions and over-smoothing in dense regions, failing to balance local precision with global overview.

[0106] This invention adaptively selects the bandwidth parameter by minimizing the error function, ensuring that important features are preserved while avoiding excessive smoothing in the high-dimensional distribution of orthopedic patient movement data. The calculation method is expressed as follows:

[0107]

[0108] In the formula, The optimized bandwidth parameters; x j Let j be the feature vector of the j-th data point; j is a positive integer. This is the true or prior density estimate for the i-th data point.

[0109] In one embodiment, such as Figure 8 and Figure 9As shown, through a composite visualization of 3D heatmaps and contour lines, the kernel density estimation error distribution of the traditional fixed bandwidth selection method and the adaptive bandwidth optimization of this invention in different data density regions were compared. The experiment focused on analyzing the error control capability of orthopedic patient movement monitoring data in high-density clustered areas and sparse areas. Conventional methods are prone to oversmoothing errors in dense data areas and estimation distortion in sparse areas. The adaptive mechanism of this invention establishes a dynamic correlation between bandwidth parameters and local density, effectively suppressing noise interference in sparse areas while maintaining the detail accuracy in high-density areas. The distribution of advantageous regions in the 3D heatmap intuitively demonstrates the adaptive matching capability of this method for complex data structures, and the location of the optimal working point in the contour map verifies the rationality of the parameter optimization strategy.

[0110] Furthermore, the objective of bandwidth search is quantified using a bandwidth optimization loss function, and the calculation method is expressed as follows:

[0111]

[0112] In the formula, L h The bandwidth optimization loss function can be obtained by minimizing this value to obtain the bandwidth best suited to the distribution of movement characteristics of orthopedic patients. Let be the kernel density estimate for the i-th data point.

[0113] S603, Forward propagation processing of data for training convolutional neural networks.

[0114] This invention achieves deep feature learning of inter-frame difference feature maps through convolutional neural networks. It relies on a combination of convolutional and pooling layers to capture key regions at multiple levels and perform dimensionality reduction. Furthermore, the fully connected layers use ReLU activation to reduce irrelevant information while preserving motion details. The implementation methods of convolutional operations, pooling operations, and ReLU activation are as follows:

[0115]

[0116] p(x) = max(P ch *x)

[0117] h(x) = max(0,x)

[0118] In the formula, f(x) is the output of the convolutional layer; x is the input of the function, representing the data points input to the convolutional neural network; W k The weights of the k-th convolutional kernel are: K is the number of convolutional kernels; k is a positive integer; * denotes a convolution operation; b is the bias term; p(x) is the output of the pooling layer; P ch is the pooling window size; h(x) is the result after processing with the ReLU activation function.

[0119] s604. Convolution kernel weight update based on topological manifold kernel density estimation.

[0120] When training convolutional neural networks, the distribution density differences of orthopedic patient movement data in different regions can cause traditional gradient updates to ignore local details. This is because ordinary gradient descent is often guided by the overall error and cannot take into account the information value brought by the different local densities.

[0121] Conventional methods rely purely on the gradient of backpropagation, which can gradually learn key information. However, when dealing with particularly dense or sparse local distributions, the adjustment is often not timely enough, and there may be insufficient learning of features in the more important regions.

[0122] This invention incorporates gradient information obtained from topological manifold kernel density estimation into the update of convolutional kernels, enabling the network to adaptively focus on key movement regions with high density within the manifold. The convolutional kernel weights are updated by superimposing the manifold gradient after the standard gradient descent term, as expressed below:

[0123]

[0124] In the formula, The weights of the k-th convolutional kernel in the (t+1)-th iteration are: η is the kernel weight of the k-th convolutional kernel in the t-th iteration; t Let be the learning rate for the t-th iteration of the convolutional neural network; Let be the gradient of the loss function of the convolutional neural network with respect to the weights of the k-th convolutional kernel; η is the gradient term based on the topological manifold kernel density estimation; L is the loss function of the convolutional neural network; λ is the manifold density influence factor. Preferably, λ is set to 0.1 and η is set to 0.01.

[0125] The gradient term based on topological manifold kernel density estimation refers to the additional gradient added to the weight update of the convolutional neural network. This additional gradient is based on the local density estimate of each data point in the manifold. This gradient term allows the network to pay more attention to key regions with high density values ​​in the distribution of orthopedic patient movement monitoring data during training, thereby improving the ability to identify and learn these regions.

[0126] s605. Calculate the loss function for electromagnetic field force-guided optimization.

[0127] The high-dimensional parameter space in orthopedic patient mobility monitoring is extremely complex. Relying on conventional gradient descent methods may lead to local optima or premature convergence to certain weight configurations during convergence, failing to fully explore the global solution space.

[0128] While conventional methods can increase the search range to some extent by using momentum terms, they may still be unable to escape the constraints of local minima in complex high-dimensional distributions or noisy environments.

[0129] This invention simulates the dynamic behavior of particles during training by guiding optimization with electromagnetic field forces. This allows the weights to obtain more flexible search momentum in high-dimensional space, enabling the convolutional neural network to simultaneously consider global and local information during training, thereby better avoiding getting trapped in local optima. This can be expressed as:

[0130]

[0131] In the formula, F represents the combined force acting on the particle, characterizing the multi-directional guiding forces acting on the weight parameters of the convolutional neural network during the optimization process in high-dimensional space. These forces include electric force (driving the convolutional kernel weight parameters of the convolutional neural network to move in the direction of loss reduction), magnetic force (constraining the divergence of the update direction of the convolutional kernel weight parameters of the convolutional neural network), manifold density gradient force (adjusting the search path according to the density of orthopedic patient movement monitoring data), and weight inertia force (preserving the historical update trend of the convolutional kernel weights of the convolutional neural network); q represents the charge of the particle, characterizing the weight parameter's weight in the current optimization stage. The importance weights are dynamically calculated based on their feature contributions in the convolutional layer; E is the electric field strength, representing the pulling strength of the gradient direction of the loss function of the convolutional neural network on the update of the convolutional kernel weight parameters; v is the particle velocity, representing the update momentum of the convolutional kernel weight parameters in historical iterations; B is the magnetic field strength, representing the orthogonal constraint component of the update direction of the convolutional kernel weight parameters of the convolutional neural network, used to avoid getting trapped in local optima; γ is the manifold density gradient influence coefficient; α is the convolutional kernel weight adjustment parameter; ΔW is the change in the convolutional kernel weights between the current iteration and the previous iteration. Preferably, γ is set to 0.3 and α is set to 0.2.

[0132] The electromagnetic field force-guided optimization refers to the process of training a convolutional neural network using orthopedic patient movement monitoring data. By simulating the motion of charged particles in an electromagnetic field, the optimization process of the convolutional kernel weights is transformed into a dynamic system driven by multiple physical forces, thereby achieving a more robust global search in a high-dimensional parameter space.

[0133] The dynamic behavior of the particles refers to the fact that, under the complex and varied action pattern distribution in orthopedic data, the kernel weight parameters of the convolutional neural network adjust their update direction and step size during training based on the comprehensive forces (electric field, magnetic field, and manifold force) calculated in real time, while retaining historical momentum to maintain stability.

[0134] The electric field strength characterizes the pulling force of the gradient direction of the loss function of the convolutional neural network on the updating of the convolutional kernel weight parameters. It simultaneously considers the impact of global error reduction and local data density on weight updates, ensuring that the pulling effect of the gradient is appropriately amplified in areas with high local density of orthopedic patient movement monitoring data, thereby making weight updates more sensitive to key areas. The calculation method is expressed as follows:

[0135]

[0136] In the formula, This represents the gradient of the loss function of the convolutional neural network with respect to the kernel weights. δ represents the kernel density estimate for the data points. ce δ is the density weighting coefficient. Preferably, δ ce Set it to 0.2.

[0137] The particle velocity characterizes the update momentum of the convolutional neural network kernel weight parameters in historical iterations. To smooth the update process and retain historical information, an exponentially weighted moving average method can be used for updating, as expressed in the following manner:

[0138] v t =μ ce ·v t-1 +(1-μ)·ΔW

[0139] In the formula, v t-1 v is the particle velocity in the (t-1)th iteration; t μ is the particle velocity in the t-th iteration; ce ΔW is the momentum preservation coefficient; ΔW is the change in the convolution kernel weights between the current iteration and the previous iteration. Preferably, μ ce Set it to 0.9.

[0140] The magnetic field strength represents the orthogonal constraint component of the kernel weight parameter update direction of the convolutional neural network, used to avoid getting trapped in local optima. It is achieved by extracting the orthogonal component of the loss gradient in the current velocity direction, stripping away the component along the existing momentum direction from the gradient, and retaining only the part orthogonal to the velocity direction. This provides a directional correction for the weights during updates, avoiding over-reliance on a single gradient direction. The calculation method is expressed as follows:

[0141]

[0142] In the formula, <·,·> represent the vector inner product operation.

[0143] The charge of the particle reflects the contribution of the current convolutional kernel to feature extraction. Its calculation considers both the gradient response strength of the convolutional kernel and the local density information of the corresponding data point. Specifically, it is implemented through a normalization calculation method based on the Sigmoid function, expressed as:

[0144]

[0145] In the formula, Sig(·) is the Sigmoid activation function; This represents the gradient of the convolution kernel with respect to the weights at the current input, reflecting the strength of the feature contribution; a va This is the first adjustment coefficient used to balance gradient response and data density; a vb This is a second adjustment coefficient used to balance the gradient response and data density. Preferably, a va Set to 0.2, a vb Set it to 0.3.

[0146] Furthermore, the loss function of the convolutional neural network is calculated by combining the comprehensive force experienced by the particle and the bandwidth optimization loss function. By combining the manifold density weighting term and the electromagnetic force regularization term, joint optimization of the high-dimensional parameter space is achieved. The calculation method is expressed as follows:

[0147]

[0148] In the formula, L h The bandwidth-optimized loss function is defined by L; the loss function of the convolutional neural network is defined by β; the electromagnetic force regularization coefficient controls the contribution of the combined force to the total loss; and F... k represents the combined electromagnetic force corresponding to the kth convolutional kernel; ||·|| is the L2 norm; ∈ represents the classification loss weight; y is the true label of the orthopedic patient movement monitoring data; For predictive labels of orthopedic patient movement monitoring data using convolutional neural networks; Let be the cross-entropy loss function. Preferably, β is set to 0.2, and ∈ is set to 0.3.

[0149] s606. Adaptive adjustment of the learning rate based on changes in the loss function of a convolutional neural network.

[0150] When training a model using a training dataset composed of orthopedic patient movement monitoring data, noise interference or abnormal data distribution may be encountered in the later stages of model training (when the model is close to fitting). This may cause training oscillations or overfitting if a large learning rate is maintained.

[0151] Conventional methods typically employ a fixed decay strategy or manually reduce the learning rate at certain predetermined stages of training. They lack a fine-grained awareness of real-time loss changes and cannot effectively handle scenarios with large fluctuations.

[0152] This invention adaptively adjusts the learning rate based on changes in the loss function of a convolutional neural network, enabling the network to gradually reduce its stride and perform more refined parameter searches in later stages. Specifically, it adjusts the learning rate for the next iteration by using the relative rate of change of the loss function and gradient information, thus achieving flexible responses to unstable situations. This can be expressed as follows:

[0153]

[0154] In the formula, η t η is the learning rate for the t-th iteration of the convolutional neural network; t+1 λ is the learning rate for the (t+1)th iteration of the convolutional neural network; μ is the first learning rate adjustment factor; μ is the second learning rate adjustment factor; ΔL t L represents the change in the loss function during the t-th iteration; t Let be the loss function for the t-th iteration; Let λ be the gradient of the loss function in the t-th iteration. Preferably, λ is set to 0.3 and μ is set to 0.5.

[0155] In one embodiment, such as Figure 10 As shown, the dynamic coupling relationship between the learning rate adjustment strategy and loss reduction is analyzed through biaxial curve analysis. Conventional methods use mechanical decay, which maintains a high learning rate in the later stages of training (40-60 rounds), causing loss oscillations. This invention dynamically adjusts the loss by sensing the rate of change in loss in real time, automatically reducing the step size when the model is close to convergence, while retaining the gradient direction correction capability. In the critical stage (labeled area), it effectively suppresses overshoot, making the loss curve present a more stable exponential decay shape.

[0156] S607, Determining the stopping condition for iterative training of convolutional neural networks.

[0157] If the relative rate of change of the validation set loss function is less than a threshold (e.g., 0.1%) in three consecutive iterations, the model is considered to have converged, and the iteration is stopped, indicating that the training of the convolutional neural network model is complete.

[0158] S7. Obtain monitoring results based on the mobile monitoring model.

[0159] The trained motion monitoring model, i.e., the convolutional neural network model, was used to classify the motion monitoring of orthopedic patients, and the classification results corresponded to three categories: small amplitude (safe), medium amplitude (warning), and large amplitude (high risk).

[0160] In one embodiment, the structure of a convolutional neural network is as follows:

[0161] Input layer: The input layer takes in the movement monitoring data of orthopedic patients, and each data point represents a specific movement behavior.

[0162] Convolutional and pooling layers: A convolutional neural network contains multiple (e.g., 3) convolutional layers, each of which uses multiple (e.g., 4) convolutional kernels to extract features from the input data;

[0163] Each convolutional layer is followed by a pooling layer to reduce the size of the feature map while retaining key features.

[0164] Fully connected layer: The high-order features output by the pooling layer are flattened into a one-dimensional vector and fed into one or two fully connected layers for further feature integration.

[0165] Output layer: The output layer contains 3 neurons, corresponding to the three categories of small amplitude (safe), medium amplitude (warning), and large amplitude (high risk);

[0166] By normalizing the network output (e.g., using the Softmax function), the output of the network is transformed into three probability values, and the class corresponding to the neuron with the highest probability value is taken as the final output class.

[0167] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring the movement of orthopedic patients, characterized in that, Includes the following steps: S1. Collect patient movement data, perform inter-frame image processing on the collected data, and extract the patient's gait features; S2. Process the acquired adjacent frame images to obtain the original grayscale image, histogram equalized image, and denoised result image; S3. Perform feature fusion on the original grayscale image, histogram equalized image, and denoised result image to obtain the preprocessed image; S4. Perform registration and difference calculation on the preprocessed image; S5. The data is trained and labeled using the inter-frame difference method; S6. Construct a movement monitoring model for orthopedic patients; S7. Obtain monitoring results based on the mobile monitoring model; Among them, S6 constructs a movement monitoring model for orthopedic patients, including: The topological manifold kernel density of orthopedic patient mobility monitoring data is estimated. The training dataset composed of orthopedic patient mobility monitoring data is vectorized to form a high-dimensional feature space. The local density of each data point in the high-dimensional space is characterized by the topological manifold kernel density estimation. The bandwidth parameter is adaptively selected by minimizing the error function, and the target of the bandwidth search is quantified by the bandwidth optimization loss function. Forward propagation processing of data for training convolutional neural networks; Convolutional kernel weight updates are performed based on topological manifold kernel density estimation, with additional gradients added to the weight updates of the convolutional neural network, the additional gradients being based on the local density estimate of each data point in the manifold; In electromagnetic field force-guided optimization, the loss function is calculated. During the training of convolutional neural networks using orthopedic patient movement monitoring data, the motion law of charged particles in the electromagnetic field is simulated, and the optimization process of convolutional kernel weights is transformed into a dynamic system driven by multiple physical forces. The convolutional kernel weight parameters of the convolutional neural network are updated in direction and step size according to the comprehensive force calculated in real time during the training process. The loss function of the convolutional neural network is calculated by combining the comprehensive force and bandwidth of the particles. Learning rate adaptively adjusted based on changes in the loss function of convolutional neural networks; The stopping condition for the iteration of convolutional neural network training is determined when the relative rate of change of the validation set loss function is less than a threshold in three consecutive iterations. If the model is considered to have converged, the iteration is stopped, indicating that the training of the convolutional neural network model is complete.

2. The method for monitoring the movement of orthopedic patients according to claim 1, characterized in that, S2 The grayscale values ​​are calculated using a weighted average method to obtain the original grayscale image; The cumulative distribution function is used to map the original gray values ​​to a new distribution, resulting in a histogram-equalized image; The nonlocal mean denoising method is adopted, which uses the redundant information of similar structures in the image to perform weighted average denoising, preserves edge details, and matches the range of similar blocks to obtain the denoised image.

3. The method for monitoring the movement of orthopedic patients according to claim 1, characterized in that, The feature fusion method described in S3 calculates the value by adding the pixel values ​​of corresponding pixel positions and then taking the average.

4. The method for monitoring the movement of orthopedic patients according to claim 1, characterized in that, S4 specifically includes: Align the preprocessed images corresponding to two consecutive frames to the same spatial coordinate system to eliminate geometric offset caused by differences in shooting angle, position or time; By traversing all feature descriptors of the two graphs, the Hamming distance is calculated, and the top 50% of matching pairs with the smallest distance are retained to reduce mismatch interference. The difference between two consecutive frames of images after quantization registration is used to identify obvious displacement features.

5. The method for monitoring the movement of orthopedic patients according to claim 1, characterized in that, S5 specifically includes: The frame difference method was used to process the data of two consecutive frames of original images to obtain the movement change data of the first orthopedic patient; The frame difference method was used to process the original grayscale data corresponding to two consecutive frames of images to obtain the second orthopedic patient movement change data; The frame difference method was used to process the histogram equalized image data corresponding to two consecutive frames to obtain the movement change data of the third orthopedic patient. The frame difference method was used to process the denoised image data corresponding to two consecutive frames to obtain the movement change data of the fourth orthopedic patient. The preprocessed image data corresponding to two consecutive frames were processed using the inter-frame difference method to obtain the movement change data of the fifth orthopedic patient. The Canny edge detection image data corresponding to two consecutive frames were processed using the inter-frame difference method to obtain the movement change data of the sixth orthopedic patient. The ORB feature point image data corresponding to two consecutive frames were processed using the inter-frame difference method to obtain the movement change data of the seventh orthopedic patient.

6. The method for monitoring the movement of orthopedic patients according to claim 5, characterized in that, The movement change data of the first, second, third, fourth, fifth, sixth, and seventh orthopedic patients, along with pixel-level difference heatmaps, are fused to obtain training sample data for orthopedic patient movement monitoring.

7. The method for monitoring the movement of orthopedic patients according to claim 1, characterized in that, The movement data comes from video recordings of the patient's gait and movement patterns, and during the data collection process, the patient is asked to walk in a standardized gait analysis area as instructed.

Images