A data management system and method for application to a spinal correction apparatus

Abstract

The application discloses a data management system and method applied to a spine correction device, relates to the technical field of artificial intelligence, and generates a feedback score prediction model, initializes the model, and dynamically adjusts the model, collects original spine shape images of a user in real time, calculates real-time initialization control parameters of the spine correction device based on the feedback score prediction model and the initialization model, collects a real-time physiological parameter change rate and a real-time posture feature change value in a time window, calculates dynamic adjustment control parameters through the dynamic adjustment model, and adjusts the spine correction device according to the dynamic adjustment control parameters, so that the real-time performance, reliability and individual adaptability of the system are enhanced.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically a data management system and method for spinal correction devices. Background Technology

[0002] Most current spinal correction devices rely on traditional manual monitoring or simple sensor data acquisition methods, which can usually only provide data on the device's operating status or the user's basic posture. They also fail to effectively integrate multi-source information for in-depth analysis. Due to the lack of accurate real-time monitoring of the user's individual physiological characteristics and usage status, traditional correction methods often cannot achieve dynamic adjustment during device operation, resulting in a lack of personalized feedback in the system.

[0003] With the development of intelligent technology, more and more spinal correction devices are beginning to introduce data management and intelligent analysis technology. Most existing systems focus on the collection of single types of data, such as device operating parameters or physiological signals, lacking the integration and processing of multi-dimensional data. This makes it difficult to comprehensively and in real time evaluate the device's operating status and changes in user posture, and also makes it impossible to detect potential problems in a timely manner, affecting the accuracy of device optimization and feedback.

[0004] Therefore, there is a need for a system capable of real-time acquisition and intelligent analysis of multi-source data to improve the intelligence level of spinal correction devices in data processing, status assessment, and adaptive control, thereby enhancing the system's real-time performance, reliability, and individual adaptability. Summary of the Invention

[0005] The purpose of this invention is to provide a data management system and method for spinal correction devices to solve the problems raised in the prior art.

[0006] To address the aforementioned technical problems, the present invention provides the following technical solution: a data management method for spinal correction devices, the method comprising:

[0007] Step S100: Through the spinal correction device platform, obtain the user's original spinal morphology image from the historical usage records, and perform similarity calculation with the standard spinal morphology image to generate a feature grid and a set of image feature parameters.

[0008] Step S200: Based on the physiological information and posture images and videos in the historical usage records, calculate the physiological score and posture score corresponding to each time window, and then calculate the user feedback score for each time window. By using the average user feedback score of all historical usage records, determine the range of the average user feedback score corresponding to each feedback level.

[0009] Step S300: Set the training period, calculate the grid cell score of each feature grid corresponding to the record during the training period based on the feature grid and the set of image feature parameters, and combine it with the average user feedback score to generate a feedback score prediction model.

[0010] Step S400: Based on the range of the average user feedback score corresponding to each feedback level, determine the feedback level corresponding to each usage record, calculate the image score of each usage record, and combine it with the average user feedback score and the initialization control parameters of the spinal correction device to generate the initialization model for each feedback level.

[0011] Step S500: In a certain feedback level, calculate the rate of change of physiological parameters, the change value of posture characteristics, and the adjustment value of control parameters for each time window in the record, and generate a dynamic adjustment model for each feedback level in different time windows;

[0012] Step S600: Real-time acquisition of the user's original spinal morphology image; calculation of real-time initialization control parameters of the spinal correction device based on the feedback scoring prediction model and initialization model; acquisition of real-time physiological parameter change rate and real-time posture feature change value within the time window; calculation of dynamic adjustment control parameters through the dynamic adjustment model; and adjustment of the spinal correction device according to the dynamic adjustment control parameters.

[0013] Furthermore, step S100 includes:

[0014] Step S101: In the spinal correction device, an external physiological monitoring device is connected to monitor the user's physiological information during use. An image acquisition device is deployed on the spinal correction device to collect posture images and videos of the user during use. At the same time, the operating parameters of the spinal correction device during use are obtained. Based on time synchronization technology, the physiological information, posture images and videos, and operating parameters corresponding to the same timestamp are grouped into a timestamped information group. All information groups during use are summarized to generate a usage record, and the usage record is uploaded to the spinal correction device platform.

[0015] Step S102: Extract the original spinal morphology image before the user's use. Divide the original spinal morphology image into several grid units through feature mapping of a convolutional neural network. Extract several image parameters from each grid unit and number each grid unit to obtain a standard spinal morphology image. Extract several standard image parameters from each grid unit in the standard spinal morphology image. Compare the image parameters of each grid unit with the standard image parameters through a structural similarity algorithm. Calculate the absolute difference of each image parameter. Preset the weight of each image parameter. Calculate the grid unit score by weighted summation of the absolute differences of each image parameter. Preset a grid unit score threshold. Mark grid units that exceed the grid unit score threshold as abnormal grids.

[0016] Step S103: Obtain historical usage records, count the number of times each grid cell is marked as an abnormal grid, calculate the abnormal frequency of each grid cell, preset an abnormal frequency threshold, and mark grid cells that exceed the abnormal frequency threshold as feature grids.

[0017] Step S104: Collect historical usage records where the feature grid is an abnormal grid at least once, obtain the image parameters of each feature grid, normalize all image parameters, perform feature decomposition on the normalized image parameter matrix using principal component analysis, calculate the contribution of each image parameter based on the variance contribution rate of each image parameter in the principal components, summarize the contributions of each image parameter, calculate the average contribution of each image parameter, preset the number of image feature parameters to be 'a', sort the image parameters from high to low according to the average contribution, select the first 'a' image parameters as image feature parameters, and generate the feature grid and image feature parameter set;

[0018] By integrating external physiological monitoring devices, image acquisition devices, and device operation parameter acquisition modules into the spinal correction device, and using time synchronization technology to match the timestamps of physiological information, posture images and videos with the device operation parameters, a complete and accurate usage record is formed. This achieves multi-dimensional information fusion and acquisition of users during the correction process, improving the spatiotemporal consistency and analysis accuracy of the data.

[0019] By statistically analyzing historical usage records, the abnormal frequency of each grid cell is calculated, and an abnormal frequency threshold is set. The characteristic grids with high-frequency abnormalities are automatically identified, which can reflect the stable problem areas of users in the long-term correction process and provide data support for subsequent personalized correction solutions.

[0020] By using principal component analysis to decompose and calculate the contribution of image parameters in the feature grid, the most representative set of image feature parameters is automatically selected, thereby reducing feature dimensionality, reducing redundant data, and improving the computational efficiency and generalization ability of the model.

[0021] Furthermore, step S200 includes:

[0022] Step S201: Obtain physiological information from historical usage records, extract physiological parameters from historical usage records, perform normalization calculations on each physiological parameter, preset time windows, construct time series analysis models for normalized physiological parameters using the time windows as units, calculate the rate of change and stability index of physiological parameters within each time window, preset the weights of the rate of change and stability index, perform weighted summation of the rate of change and stability index, calculate the weight of each physiological parameter in different time windows, perform weighted summation of the normalized physiological parameters and the weights of the corresponding time windows, and calculate the physiological score for each time window;

[0023] Step S202: Acquire posture image videos from historical usage records. Using visual human motion capture technology, identify and extract continuous image frames corresponding to each posture change of the user. Extract posture features from each frame in the continuous image frames, obtain the change value of each posture feature, normalize the change value of each posture feature, construct a posture feature change vector for the normalized posture features in units of time windows, calculate the posture change rate and smoothness index within the time window, preset the weights of the posture change rate and smoothness index, and calculate the weight of each posture feature in different time windows by weighted summation. Calculate the posture score for each time window by weighted summation of the normalized posture feature change value and the weight of the corresponding time window.

[0024] Step S203: Combine the physiological scores and posture scores of the same time window, preset the weights of the physiological scores and posture scores, calculate the user feedback score of a certain time window by weighted summation of the physiological scores and posture scores, summarize the user feedback scores corresponding to all time windows, and calculate the average user feedback score of historical usage records.

[0025] Step S204: Summarize the average user feedback ratings from all historical usage records, and calculate the variance of the average user feedback ratings as b. If there are k feedback levels, the range for each feedback level is:

[0026] ;

[0027] Where D1 and D2 represent the minimum and maximum values ​​of the average user feedback rating, respectively;

[0028] Normalize and divide the physiological parameters in historical usage records into time windows, combine them with time series analysis models, calculate the rate of change and stability index of physiological parameters in different time windows, and obtain physiological scores by weighted summation. This can dynamically reflect the fluctuations and recovery trends of the user's physiological state during the correction process, and improve the timeliness and accuracy of physiological data analysis.

[0029] Visual human motion capture technology extracts continuous image frames from user posture images and videos, constructs posture feature change vectors, calculates posture change rate and smoothness index, and generates posture score by weighted summation. It can accurately capture subtle changes in user posture and motion stability, and realize the quantification and intelligent evaluation of posture dynamic features.

[0030] By weighted and fused physiological and postural scores within the same time window, a comprehensive user feedback score is obtained, which can fully reflect the physiological response and postural changes of users during the use of spinal correction equipment. This allows for the establishment of a multi-dimensional user experience and correction effect evaluation model, thereby improving the completeness and scientific nature of data analysis.

[0031] By statistically analyzing and calculating the variance of user feedback ratings in historical usage records, feedback level intervals are automatically generated based on the variance intervals. This achieves standardized hierarchical evaluation of user feedback, clearly distinguishing the feedback levels of different users or the same user at different stages. This facilitates personalized parameter adjustments and long-term tracking of corrective effects within the system.

[0032] Furthermore, step S300 includes:

[0033] Step S301: Select several consecutive days as the training period, obtain the usage records within the training period, obtain the original spinal morphology image in a certain usage record based on the feature grid and image feature parameter set, collect the image feature parameters of each feature grid, and calculate the grid cell score of each feature grid.

[0034] Step S302: Collect physiological information and posture images and videos from the usage records, calculate the average user feedback score of the usage records, combine the grid cell score of each feature grid with the average user feedback score to generate an image training set, use the grid cell score of each feature grid as input and the average user feedback score as output, train the image training set through a linear regression model to generate a feedback score prediction model.

[0035] By selecting several consecutive days as the training period, collecting usage records within the training period, and extracting the image feature parameters and corresponding grid unit scores for each feature grid, these objective image quantification indicators are combined with the average user feedback scores to establish an input-output correspondence, thus realizing the correlation modeling between spinal morphological changes and user subjective feedback.

[0036] Using grid cell ratings as input variables and average user feedback ratings as output targets, a feedback rating prediction model is generated through training a linear regression model. This model can quickly predict user feedback trends during the correction process based on new image features, enabling quantitative and predictable evaluation of the correction effect.

[0037] Collecting multiple usage records as training samples within a continuous time period ensures that the input features and feedback results have temporal continuity and diversity, which helps to reduce random errors and improve the stability and prediction accuracy of the model.

[0038] The feedback rating prediction model can calculate the predicted rating in real time during the user's new usage process, providing a basis for the system to automatically adjust the correction intensity, duration or posture correction parameters, thereby realizing intelligent control and personalized guidance of the correction process.

[0039] Furthermore, step S400 includes:

[0040] Step S401: Obtain the average user feedback rating of the usage records, and based on the range of feedback levels, determine the feedback level corresponding to each usage record, and extract the initialization control parameters of the spinal correction device corresponding to each usage record;

[0041] Step S402: The grid cell scores of each feature grid in the usage record are weighted and summed to calculate the image score of the usage record. The usage records of the same feedback level are summarized. The image score and the average user feedback score are used as inputs, and the initialization control parameters are used as outputs. The multilayer perceptron regression model is used for training to generate the initialization model for each feedback level.

[0042] By determining the feedback level of each usage record based on the user's average feedback rating and the preset feedback level range, and combining it with the initialization control parameters of the spinal correction device, the initialization control parameters corresponding to each usage record can be accurately matched, ensuring that the system's response to different user feedback can be quantitatively adjusted, further improving the accuracy and practicality of personalized control effects.

[0043] By weighted summing of the grid cell scores of each feature grid, the image score of the usage record is calculated, which can comprehensively evaluate the image features and posture changes of the user during the correction process. This serves as the basis for adjusting the system's control parameters, making the state analysis of the spinal correction system more detailed and comprehensive, and improving the reliability and detail of the image score.

[0044] Using image scores and average user feedback scores as inputs and initial control parameters as outputs, the system is trained using a multilayer perceptron regression model. This model can dynamically generate an initial model corresponding to each feedback level based on historical data, effectively realizing intelligent control of the spinal correction device. It can optimize the control strategy in real time according to changes in user feedback, thereby improving the user correction experience.

[0045] By aggregating usage records at the same feedback level and training them using a regression model, customized initialization models are generated for users with different feedback levels. This ensures that the correction process for each user group can be tailored to their feedback performance, thereby improving the adaptability and personalization of spinal correction equipment.

[0046] Furthermore, step S500 includes:

[0047] Step S501: In a set of usage records for a certain feedback level, collect the rate of change of physiological parameters and the change value of posture features for each time window in a certain usage record, and normalize the rate of change of physiological parameters and the change value of posture features.

[0048] Step S502: Obtain the adjustment value of each control parameter corresponding to each time window in the usage record, and normalize the adjustment value of each control parameter. Combine the normalized physiological parameter change rate, posture feature change value, and control parameter adjustment value of all usage records in the feedback level under the same time window to form a sample and construct a dynamic adjustment training set.

[0049] Step S503: Using the normalized physiological parameter change rate and posture feature change value as input, and the normalized control parameter adjustment value as output, a long short-term memory neural network is used for training to generate a dynamic adjustment model for each feedback level in different time windows.

[0050] By collecting the rate of change of physiological parameters and the value of change of posture features in each time window of the usage record, and combining them with the adjustment value of control parameters for normalization, the dynamic characteristics of the user's physiological and posture changes can be characterized in the time dimension, realizing the temporal modeling of the spinal correction process, and providing a highly timely data foundation for subsequent intelligent adjustment.

[0051] By constructing a dynamic adjustment training set using the rate of change of physiological parameters, the value of change of posture features, and the adjustment value of control parameters as samples in the same time window, the system can reflect the real-time response characteristics of the user during the correction process. This enables the system to dynamically adjust according to changes in the user's state, significantly improving the response speed and accuracy of the device to real-time posture and physiological changes.

[0052] By training time-series data with a long short-term memory neural network, the long-term dependency between physiological and posture changes can be effectively captured, avoiding the problem that traditional static models cannot reflect continuous time characteristics. This enables continuous optimization and adaptive adjustment of control parameters, improving the precision of correction control.

[0053] Independent dynamic adjustment models are trained for different feedback levels, enabling the system to match corresponding control strategies according to the user's feedback level and achieve differentiated dynamic responses in different correction stages, thereby improving the model's adaptability and control stability in multi-user, multi-stage application scenarios.

[0054] Furthermore, step S600 includes:

[0055] Step S601: Real-time acquisition of the user's original spinal morphology image; calculation of the real-time grid cell score of each feature grid based on the feature grid and image feature parameter set; input into the feedback score prediction model to calculate the feedback score prediction value.

[0056] Step S602: Based on the range of feedback levels, determine the real-time feedback level corresponding to the current user, calculate the real-time image score, and input the real-time image score and the predicted feedback score into the initialization model corresponding to the real-time feedback level to calculate the real-time initialization control parameters of the spinal correction device;

[0057] Step S603: Monitor the rate of change of real-time physiological parameters and the value of change of real-time posture features within the current time window, and perform normalization calculation. Input the normalized rate of change of real-time physiological parameters and the value of change of real-time posture features into the dynamic adjustment model of the real-time feedback level in the current time window, calculate the dynamic adjustment control parameters, and upload them to the spinal correction device. Perform dynamic adjustment on the spinal correction device according to the dynamic adjustment control parameters.

[0058] By acquiring the user's original spinal morphology images in real time, and calculating real-time grid cell scores based on feature grids and image feature parameter sets, and combining the feedback score prediction model to output real-time feedback score prediction values, the user's current spinal status can be reflected in real time, realizing intelligent perception and dynamic monitoring of the correction process.

[0059] The real-time image score and the predicted feedback score are input into the initialization model of the corresponding feedback level to calculate the real-time initialization control parameters. This enables the system to adaptively adjust the initial correction strategy according to the user's current feedback level, thereby ensuring that the control output accurately matches the user's actual state.

[0060] Real-time monitoring of the rate of change of physiological parameters and the value of change of posture characteristics, and inputting the normalized data into a dynamic adjustment model constructed by a long short-term memory neural network to generate dynamic adjustment control parameters, forming a closed-loop feedback system from "real-time perception - intelligent prediction - dynamic control", which effectively improves the dynamic response capability and control stability of the system.

[0061] Input features and control outputs can be updated in real time within each time window, enabling the spinal correction device to respond quickly to changes in the user's posture and physiological feedback, thereby achieving millisecond-level adaptive adjustment and significantly improving the safety, comfort, and accuracy of the correction process.

[0062] To better implement the above method, a data management system for spinal correction equipment is also proposed. The system includes an image analysis module, a feedback scoring module, a feedback scoring prediction model module, an initialization model module, a dynamic adjustment model module, and a real-time adjustment module.

[0063] Image analysis module: Through the spinal correction device platform, it acquires the user's original spinal morphology image from historical usage records, calculates the similarity with the standard spinal morphology image, and generates a feature grid and image feature parameter set;

[0064] Feedback scoring module: Based on physiological information and posture images and videos in historical usage records, calculate the physiological score and posture score corresponding to each time window, and then calculate the user feedback score for each time window. By using the average user feedback score of all historical usage records, determine the range of the average user feedback score corresponding to each feedback level.

[0065] Feedback rating prediction model module: Set the training period, calculate the grid cell score of each feature grid corresponding to the usage record during the training period based on the feature grid and image feature parameter set, and combine it with the average user feedback score to generate a feedback rating prediction model.

[0066] Initialization Model Module: Based on the range of average user feedback scores corresponding to each feedback level, determine the feedback level corresponding to each usage record, calculate the image score of each usage record, and combine it with the average user feedback score and the initialization control parameters of the spinal correction device to generate the initialization model for each feedback level.

[0067] Dynamic adjustment model module: At a certain feedback level, calculate the rate of change of physiological parameters, change value of posture characteristics, and adjustment value of control parameters for each time window within the record, and generate a dynamic adjustment model for each feedback level in different time windows;

[0068] Real-time adjustment module: Real-time acquisition of the user's original spinal morphology image, calculation of real-time initialization control parameters of the spinal correction device based on feedback scoring prediction model and initialization model, acquisition of real-time physiological parameter change rate and real-time posture feature change value within the time window, calculation of dynamic adjustment control parameters through dynamic adjustment model, and adjustment of the spinal correction device according to the dynamic adjustment control parameters.

[0069] Furthermore, the image analysis module includes units for identifying anomalous mesh cells and determining image feature parameters:

[0070] Identifying Abnormal Mesh Cells: In the spinal correction device, external physiological monitoring equipment is connected to monitor the user's physiological information during use. Image acquisition devices are deployed on the spinal correction device to collect posture images and videos of the user during use, and simultaneously acquire the operating parameters of the spinal correction device during use. Based on time synchronization technology, physiological information, posture images / videos, and operating parameters corresponding to the same timestamp are grouped into timestamped information groups. All information groups during use are summarized to generate a usage record, which is then uploaded to the spinal correction device platform. The original spinal morphology image before the user's use is extracted, and through feature mapping of a convolutional neural network, the... The original spinal morphology image is divided into several grid cells. Several image parameters are extracted from each grid cell and each grid cell is numbered to obtain a standard spinal morphology image. Several standard image parameters are extracted from each grid cell in the standard spinal morphology image. Using a structural similarity algorithm, the image parameters of each grid cell are compared with the standard image parameters to calculate the absolute difference of each image parameter. The weight of each image parameter is preset. The grid cell score is calculated by weighted summation of the absolute differences of each image parameter. A grid cell score threshold is preset. Grid cells that exceed the grid cell score threshold are marked as abnormal grids.

[0071] Determine image feature parameter units: Obtain historical usage records, count the number of times each grid unit is marked as an abnormal grid, calculate the abnormal frequency of each grid unit, preset an abnormal frequency threshold, and mark grid units exceeding the abnormal frequency threshold as feature grids. Collect historical usage records in which a feature grid appears as an abnormal grid at least once, obtain the image parameters of each feature grid, normalize all image parameters, perform eigenvalue decomposition on the normalized image parameter matrix using principal component analysis, calculate the contribution of each image parameter based on the variance contribution rate of each image parameter in the principal components, summarize the contributions of each image parameter, calculate the average contribution of each image parameter, preset the number of image feature parameters to be 'a', sort the image parameters from high to low according to the average contribution, select the top 'a' image parameters as image feature parameters, and generate feature grids and image feature parameter sets.

[0072] Furthermore, the feedback scoring module includes a physiological scoring unit and a posture scoring unit:

[0073] Physiological scoring unit: Obtain physiological information from historical usage records, extract physiological parameters from historical usage records, normalize each physiological parameter, preset time windows, construct time series analysis models for normalized physiological parameters using the time windows as units, calculate the rate of change and stability index of physiological parameters within each time window, preset the weights of the rate of change and stability index, perform weighted summation of the rate of change and stability index, calculate the weight of each physiological parameter in different time windows, and perform weighted summation of the normalized physiological parameters and the corresponding time window weights to calculate the physiological score for each time window;

[0074] The posture scoring unit acquires posture images and videos from historical usage records. Using visual human motion capture technology, it identifies and extracts consecutive image frames corresponding to each user's posture change. Posture features are extracted from each frame in the consecutive image frames, and the change value of each posture feature is obtained. The change value of each posture feature is normalized. Using a time window as the unit, a posture feature change vector is constructed from the normalized posture features. The posture change rate and smoothness index within the time window are calculated. Weights for the posture change rate and smoothness index are preset, and the posture change rate and smoothness index are weighted and summed to obtain the weight of each posture feature in different time windows. The normalized posture feature change value is weighted and summed with the corresponding weight of the time window to calculate the posture score for each time window.

[0075] Compared with the prior art, the beneficial effects of the present invention are: by analyzing multi-dimensional data such as user history, posture images, and real-time physiological information, a personalized feedback score prediction model and a dynamic adjustment model are generated. Compared with the traditional fixed parameter settings, the spinal correction device can be dynamically adjusted according to the user's real-time status and historical feedback, thereby achieving more accurate and personalized spinal correction.

[0076] By combining multiple factors such as image features, posture scores, and physiological scores for comprehensive evaluation, a feedback level is generated and the user's spinal correction device is dynamically adjusted. This comprehensive evaluation mechanism improves the scientific nature and reliability of the correction and reduces the errors or deficiencies that may be caused by a single factor.

[0077] By using a dynamic adjustment model to control the spinal correction device in real time and dynamically, it can respond more accurately to changes in the user's body. Traditional methods usually rely on fixed control parameters, while the dynamic adjustment of this invention can adaptively adjust the device according to real-time data, reducing manual intervention.

[0078] By optimizing control parameters using advanced machine learning algorithms such as multilayer perceptron regression models and long short-term memory neural networks, the prediction accuracy and robustness of the model are improved, which can significantly enhance the response speed and accuracy of the equipment and reduce unnecessary error adjustments. Attached Figure Description

[0079] Figure 1 This is a flowchart illustrating a data management method for spinal correction devices according to the present invention.

[0080] Figure 2 This is a schematic diagram of the structure of a data management system for spinal correction devices according to the present invention. Detailed Implementation

[0081] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0082] Please see Figure 1 and Figure 2 The present invention provides a technical solution: a data management system for spinal correction equipment, the method comprising:

[0083] Step S100: Obtain the original image and generate the feature mesh: The original spinal morphology image of the user in the historical usage record is obtained through the spinal correction device platform, and the similarity is calculated with the standard spinal morphology image to generate the feature mesh and image feature parameter set.

[0084] Step S100 includes:

[0085] Step S101: Collect synchronized physiological posture data: In the spinal correction device, connect an external physiological monitoring device to monitor the user's physiological information during use. Deploy an image acquisition device on the spinal correction device to collect posture images and videos of the user during use. At the same time, obtain the operating parameters of the spinal correction device during use. Based on time synchronization technology, combine the physiological information, posture images and videos, and operating parameters corresponding to the same timestamp into a timestamped information group. Summarize all information groups during use to generate a usage record, and upload the usage record to the spinal correction device platform.

[0086] Step S102: Mesh division and calculation of abnormal meshes: Extract the original spinal morphology image before user use, divide the original spinal morphology image into several mesh units through feature mapping of a convolutional neural network, extract several image parameters in each mesh unit and number each mesh unit to obtain a standard spinal morphology image, extract several standard image parameters in each mesh unit of the standard spinal morphology image, compare the image parameters of each mesh unit with the standard image parameters through a structural similarity algorithm, calculate the absolute difference of each image parameter, preset the weight of each image parameter, calculate the mesh unit score by weighted summation of the absolute differences of each image parameter, preset the mesh unit score threshold, and mark mesh units that exceed the mesh unit score threshold as abnormal meshes;

[0087] Step S103: Statistical analysis of abnormal frequency to determine feature grids: Obtain historical usage records, count the number of times each grid cell is marked as an abnormal grid, calculate the abnormal frequency of each grid cell, preset an abnormal frequency threshold, and mark grid cells that exceed the abnormal frequency threshold as feature grids.

[0088] Step S104: Extract feature parameters and generate a parameter set: Collect historical usage records where the feature grid is an abnormal grid at least once, obtain the image parameters of each feature grid, normalize all image parameters, perform feature decomposition on the normalized image parameter matrix using principal component analysis, calculate the contribution of each image parameter based on the variance contribution rate of each image parameter in the principal components, summarize the contributions of each image parameter, calculate the average contribution of each image parameter, preset the number of image feature parameters to be 'a', sort the image parameters from high to low according to the average contribution, select the first 'a' image parameters as image feature parameters, and generate the feature grid and image feature parameter set;

[0089] For example, a frontal image of the user's spine taken before use is obtained and divided into 8×8 grid units using a convolutional neural network;

[0090] In the first grid cell, the average grayscale value was 125.3, the edge intensity was 43.2, the texture energy was 0.215, and the contrast was 0.354.

[0091] Simultaneously, standard spinal morphology images are acquired, and each parameter is compared using a structural similarity algorithm:

[0092] Grayscale mean: User value 125.3, Standard value 122.8, Absolute difference 2.5, Weight 0.2;

[0093] Edge strength: User value 43.2, standard value 40.1, absolute difference 3.1, weight 0.3;

[0094] Texture energy: User value 0.215, Standard value 0.240, Absolute difference 0.025, Weight 0.25;

[0095] Contrast ratio: User value 0.354, Standard value 0.320, Absolute difference 0.034, Weighting value 0.25;

[0096] The calculated grid cell score is 0.2×2.5+0.3×3.1+0.25×0.025+0.25×0.034≈1.13;

[0097] If the grid cell scoring threshold is set to 1.0, then the first grid cell is marked as an abnormal grid.

[0098] Suppose we retrieve the most recent 30 usage records from the cloud platform and count the number of times each grid is marked as abnormal:

[0099] The grid number is G(3,4), the number of anomalies is 19, and the anomaly frequency is 0.63.

[0100] The grid number is G(3,5), the number of anomalies is 22, and the anomaly frequency is 0.73.

[0101] The grid number is G(4,4), the number of anomalies is 17, and the anomaly frequency is 0.57.

[0102] The grid number is G(5,4), the number of anomalies is 8, and the anomaly frequency is 0.27.

[0103] If the abnormal frequency threshold is set to 0.6, then G(3,4) and G(3,5) are the feature grids.

[0104] Step S200: Calculate the feedback score range: Based on the physiological information and posture images and videos in the historical usage records, calculate the physiological score and posture score corresponding to each time window, and then calculate the user feedback score for each time window. By using the average user feedback score of all historical usage records, determine the range of the average user feedback score corresponding to each feedback level.

[0105] Step S200 includes:

[0106] Step S201 Calculate physiological score: Obtain physiological information from historical usage records, extract physiological parameters from historical usage records, normalize each physiological parameter, preset time windows, construct a time series analysis model for the normalized physiological parameters using the time windows as units, calculate the rate of change and stability index of physiological parameters within each time window, preset the weights of the rate of change and stability index, perform weighted summation of the rate of change and stability index, calculate the weight of each physiological parameter in different time windows, perform weighted summation of the normalized physiological parameters and the weights of the corresponding time windows, and calculate the physiological score for each time window;

[0107] Step S202: Calculate posture score: Acquire posture images and videos from historical usage records. Using visual human motion capture technology, identify and extract continuous image frames corresponding to each posture change of the user. Extract posture features from each frame in the continuous image frames, obtain the change value of each posture feature, normalize the change value of each posture feature, construct a posture feature change vector for the normalized posture features in units of time windows, calculate the posture change rate and smoothness index within the time window, preset the weights of the posture change rate and smoothness index, and calculate the weight of each posture feature in different time windows by weighted summation. Calculate the posture score for each time window by weighted summation of the normalized posture feature change value and the weight of the corresponding time window.

[0108] Step S203 generates user feedback scores: Combine physiological scores and posture scores from the same time window, preset the weights of physiological scores and posture scores, calculate the user feedback score for a certain time window by weighted summation of physiological scores and posture scores, summarize the user feedback scores corresponding to all time windows, and calculate the average user feedback score of historical usage records.

[0109] Step S204: Divide the feedback level range: Summarize the average user feedback ratings from all historical usage records, calculate the variance of the average user feedback rating as b, and if there are k feedback levels, generate the range for each feedback level as follows:

[0110] ;

[0111] Where D1 and D2 represent the minimum and maximum values ​​of the average user feedback rating, respectively;

[0112] For example, the system extracted the following physiological parameters from the user's historical records: time window 0-5, heart rate 80, surface electromyography 35.6, and device pressure 12.2;

[0113] Time window 5-10, heart rate 82, surface electromyography 36.3, equipment pressure 12.5;

[0114] Time window 10-15, heart rate 81, surface electromyography 35.9, equipment pressure 12.3;

[0115] Time window 15-20, heart rate 78, surface electromyography 34.8, equipment pressure 12.1;

[0116] Time window 20-25, heart rate 77, surface electromyography 33.5, equipment pressure 11.8;

[0117] Time window 25-30, heart rate 79, surface electromyography 34.2, equipment pressure 12.0;

[0118] Normalize each physiological parameter.

[0119] The normalized values ​​for time windows 0-5 are 0.5, 0.12, and 0.24, respectively.

[0120] The normalized values ​​for time windows 5-10 are 0.6, 0.18, and 0.3, respectively.

[0121] The normalized values ​​for time windows 10-15 are 0.55, 0.14, and 0.26, respectively.

[0122] The normalized values ​​for time windows 15-20 are 0.4, 0.08, and 0.2, respectively.

[0123] The normalized values ​​for time windows 20-25 are 0.35, 0.04, and 0.13, respectively.

[0124] The normalized values ​​for time windows 25-30 are 0.45, 0.1, and 0.23, respectively.

[0125] Assuming a preset time window of 5 minutes, calculate the rate of change of physiological parameters, such as the rate of change of heart rate, and the stability, such as the standard deviation of heart rate. Assume the weights for the rate of change and stability indices are 0.6 and 0.4, respectively.

[0126] Heart rate variability rate (N / A), heart rate stability (N / A), and heart rate score (N / A) were all within the time window of 0-5.

[0127] The heart rate variability rate was 0.025, the heart rate stability was 0.2, and the heart rate score was 0.42 within the time window of 5-10.

[0128] The heart rate variability rate was -0.025, the heart rate stability was 0.25, and the heart rate score was 0.35 within the time window of 10-15.

[0129] During the time window of 15-20, the heart rate variability was -0.05, the heart rate stability was 0.3, and the heart rate score was 0.4.

[0130] The heart rate variability rate was -0.025, the heart rate stability was 0.25, and the heart rate score was 0.3 within the time window of 20-25.

[0131] The heart rate variability rate was 0.1, the heart rate stability was 0.3, and the heart rate score was 0.55 within the time window of 25-30.

[0132] The system uses visual human motion capture technology to extract posture feature changes in each time window, such as changes in spine angle and shoulder position. Assume the following posture feature data is extracted:

[0133] During the time window of 0-5, the spinal angle changed by 1.2 and the shoulder position changed by 0.3.

[0134] During the time window of 5-10, the spinal angle changed by 1.5 degrees and the shoulder position changed by 0.35 degrees.

[0135] Within a time window of 10-15, the spinal angle changed by 1.4 and the shoulder position changed by 0.3.

[0136] During the time window of 15-20, the spinal angle changed by 1.3 and the shoulder position changed by 0.2.

[0137] During the time window of 20-25, the spinal angle changed by 1.1 and the shoulder position changed by 0.15.

[0138] During the time window of 25-30 days, the spinal angle changed by 1.2 degrees and the shoulder position changed by 0.25 degrees.

[0139] By calculating the rate of change and smoothness of each pose feature, assuming the weights for the rate of change and smoothness are 0.7 and 0.3 respectively:

[0140] The spinal angle change rate was N / A, the shoulder position change rate was N / A, the posture smoothness was N / A, and the posture score was N / A for the time window 0-5.

[0141] The change rate of spinal angle was 0.25, the change rate of shoulder position was 0.167, the postural smoothness was 0.3, and the postural score was 0.575 within the time window of 5-10.

[0142] Within the time window of 10-15, the spinal angle change rate was -0.1, the shoulder position change rate was -0.05, the posture smoothness was 0.2, and the posture score was 0.425.

[0143] During the time window of 15-20, the spinal angle change rate was -0.05, the shoulder position change rate was -0.1, the posture smoothness was 0.15, and the posture score was 0.4.

[0144] During the time window of 20-25, the spinal angle change rate was -0.1, the shoulder position change rate was -0.05, the posture smoothness was 0.1, and the posture score was 0.3.

[0145] Within the time window of 25-30, the rate of change in spinal angle was 0.1, the rate of change in shoulder position was 0.083, the postural smoothness was 0.25, and the postural score was 0.55.

[0146] The physiological and postural scores were combined, and a weighted sum was used to calculate the user feedback score for each time window. The weights for the physiological and postural scores were assumed to be 0.6 and 0.4, respectively.

[0147] Physiological scores, posture scores, and user feedback scores were all N / A within the time window of 0-5.

[0148] During the time window of 5-10, the physiological score was 0.42, the posture score was 0.575, and the user feedback score was 0.495.

[0149] Within the time window of 10-15, the physiological score was 0.35, the posture score was 0.425, and the user feedback score was 0.387.

[0150] During the time window of 15-20, the physiological score was 0.4, the posture score was 0.4, and the user feedback score was 0.4.

[0151] During the time window of 20-25, the physiological score was 0.3, the posture score was 0.3, and the user feedback score was 0.3.

[0152] The physiological score, posture score, and user feedback score were all 0.55 within the time window of 25-30.

[0153] Step S300: Train the feedback rating prediction model: Set the training period, calculate the grid cell score of each feature grid corresponding to the usage record during the training period based on the feature grid and image feature parameter set, and combine it with the average user feedback score to generate the feedback rating prediction model.

[0154] Step S300 includes:

[0155] Step S301 Calculate the feature grid score: Select several consecutive days as the training period, obtain the usage records within the training period, obtain the original spinal morphology image in a certain usage record based on the feature grid and image feature parameter set, and collect the image feature parameters of each feature grid to calculate the grid cell score of each feature grid.

[0156] Step S302: Training the feedback rating prediction model: Collect physiological information and posture images and videos from the usage records, calculate the average user feedback rating of the usage records, combine the grid cell rating of each feature grid with the average user feedback rating to generate an image training set, use the grid cell rating of each feature grid as input and the average user feedback rating as output, train the image training set through a linear regression model to generate the feedback rating prediction model.

[0157] Step S400: Generate feedback level initialization model: Based on the range of the average user feedback score corresponding to each feedback level, determine the feedback level corresponding to each usage record, calculate the image score of each usage record, and combine it with the average user feedback score and the initialization control parameters of the spinal correction device to generate an initialization model for each feedback level.

[0158] Step S400 includes:

[0159] Step S401: Determine feedback level and control parameters: Obtain the average user feedback score of the usage records, and based on the range of feedback levels, determine the feedback level corresponding to each usage record, and extract the initialization control parameters of the spinal correction device corresponding to each usage record;

[0160] Step S402 Training and generating initialization model: The grid cell scores of each feature grid in the usage record are weighted and summed to calculate the image score of the usage record. The usage records of the same feedback level are summarized. The image score and the average user feedback score are used as inputs, and the initialization control parameters are used as outputs. The multilayer perceptron regression model is used for training to generate the initialization model for each feedback level.

[0161] For example, a feedback rating range of 0.0-0.45 is designated as level L1;

[0162] Feedback ratings range from 0.45 to 0.55, with the grade number L2.

[0163] Feedback ratings range from 0.55 to 0.70, with the grade number L3.

[0164] Average user feedback rating: On date 1, the average user feedback rating was 0.45, and the feedback level was L1;

[0165] On date 2, the average user feedback rating was 0.50, and the feedback level was L2.

[0166] On date 3, the average user feedback rating was 0.42, and the feedback level was L1.

[0167] On date 4, the average user feedback rating was 0.47, and the feedback level was L2.

[0168] On date 5, the average user feedback rating was 0.55, and the feedback level was L2.

[0169] On date 6, the average user feedback rating was 0.60, and the feedback level was L3.

[0170] On date 7, the average user feedback rating was 0.58, and the feedback level was L3.

[0171] Extract the initialization control parameters of each usage record.

[0172] The spinal correction device includes three main control channel parameters:

[0173] P1 represents the tension of the support module;

[0174] P2 represents the angle correction force;

[0175] P3 is the attitude adjustment delay;

[0176] The initialization control parameters for each record before use are as follows:

[0177] For date 1, P1 is 35, P2 is 8, and P3 is 450;

[0178] For date 2, P1 is 40, P2 is 9, and P3 is 430;

[0179] For date 3, P1 is 33, P2 is 7, and P3 is 470;

[0180] For date 4, P1 is 38, P2 is 8, and P3 is 440;

[0181] For date 5, P1 is 41, P2 is 9, and P3 is 420;

[0182] For date 6, P1 is 45, P2 is 10, and P3 is 400;

[0183] For date 7, P1 is 44, P2 is 10, and P3 is 410.

[0184] Step S500: Establish a dynamic adjustment model: In a certain feedback level, calculate the rate of change of physiological parameters, the change value of posture characteristics, and the adjustment value of control parameters for each time window within the record, and generate a dynamic adjustment model for each feedback level in different time windows.

[0185] Step S500 includes:

[0186] Step S501 Normalize the rate of change of physiological posture: In the set of usage records for a certain feedback level, collect the rate of change of physiological parameters and the change value of posture features for each time window in a certain usage record, and normalize the rate of change of physiological parameters and the change value of posture features.

[0187] Step S502: Construct a dynamic adjustment training set: Obtain the adjustment value of each control parameter corresponding to each time window in the usage record, and normalize the adjustment value of each control parameter. Combine the normalized physiological parameter change rate, posture feature change value, and control parameter adjustment value of all usage records in the feedback level under the same time window to form a sample and construct a dynamic adjustment training set.

[0188] Step S503: Train and generate a dynamic adjustment model: Using the normalized physiological parameter change rate and posture feature change value as input, and the normalized control parameter adjustment value as output, a long short-term memory neural network is used for training to generate a dynamic adjustment model for each feedback level in different time windows.

[0189] Step S600: Real-time generation and adjustment of control parameters: Real-time acquisition of the user's original spinal morphology image; calculation of real-time initialization control parameters of the spinal correction device based on the feedback scoring prediction model and the initialization model; acquisition of real-time physiological parameter change rate and real-time posture feature change value within the time window; calculation of dynamic adjustment control parameters through the dynamic adjustment model; and adjustment of the spinal correction device according to the dynamic adjustment control parameters.

[0190] Step S600 includes:

[0191] Step S601 Calculate the predicted feedback score: Collect the user's original spinal morphology image in real time, calculate the real-time grid cell score of each feature grid based on the feature grid and image feature parameter set, and input it into the feedback score prediction model to calculate the predicted feedback score.

[0192] Step S602 generates real-time initialization parameters: Based on the range of feedback levels, the real-time feedback level corresponding to the current user is determined, the real-time image score is calculated, and the real-time image score and the predicted feedback score are input into the initialization model corresponding to the real-time feedback level to calculate the real-time initialization control parameters of the spinal correction device.

[0193] Step S603: Calculate and apply dynamic adjustment parameters: Monitor the rate of change of real-time physiological parameters and the value of change of real-time posture characteristics within the current time window, and perform normalization calculations. Input the normalized rate of change of real-time physiological parameters and the value of change of real-time posture characteristics into the dynamic adjustment model of the real-time feedback level in the current time window, calculate the dynamic adjustment control parameters, and upload them to the spinal correction device. Perform dynamic adjustment on the spinal correction device according to the dynamic adjustment control parameters.

[0194] To better implement the above method, a data management system for spinal correction equipment is also proposed. The system includes an image analysis module, a feedback scoring module, a feedback scoring prediction model module, an initialization model module, a dynamic adjustment model module, and a real-time adjustment module.

[0195] Image analysis module: Through the spinal correction device platform, it acquires the user's original spinal morphology image from historical usage records, calculates the similarity with the standard spinal morphology image, and generates a feature grid and image feature parameter set;

[0196] The image analysis module includes units for identifying abnormal mesh cells and units for determining image feature parameters.

[0197] Identifying Abnormal Mesh Cells: In the spinal correction device, external physiological monitoring equipment is connected to monitor the user's physiological information during use. Image acquisition devices are deployed on the spinal correction device to collect posture images and videos of the user during use, and simultaneously acquire the operating parameters of the spinal correction device during use. Based on time synchronization technology, physiological information, posture images / videos, and operating parameters corresponding to the same timestamp are grouped into timestamped information groups. All information groups during use are summarized to generate a usage record, which is then uploaded to the spinal correction device platform. The original spinal morphology image before the user's use is extracted, and through feature mapping of a convolutional neural network, the... The original spinal morphology image is divided into several grid cells. Several image parameters are extracted from each grid cell and each grid cell is numbered to obtain a standard spinal morphology image. Several standard image parameters are extracted from each grid cell in the standard spinal morphology image. Using a structural similarity algorithm, the image parameters of each grid cell are compared with the standard image parameters to calculate the absolute difference of each image parameter. The weight of each image parameter is preset. The grid cell score is calculated by weighted summation of the absolute differences of each image parameter. A grid cell score threshold is preset. Grid cells that exceed the grid cell score threshold are marked as abnormal grids.

[0198] Determine image feature parameter units: Obtain historical usage records, count the number of times each grid unit is marked as an abnormal grid, calculate the abnormal frequency of each grid unit, preset an abnormal frequency threshold, and mark grid units exceeding the abnormal frequency threshold as feature grids. Collect historical usage records in which a feature grid appears as an abnormal grid at least once, obtain the image parameters of each feature grid, normalize all image parameters, perform eigenvalue decomposition on the normalized image parameter matrix using principal component analysis, calculate the contribution of each image parameter based on the variance contribution rate of each image parameter in the principal components, summarize the contributions of each image parameter, calculate the average contribution of each image parameter, preset the number of image feature parameters to be 'a', sort the image parameters from high to low according to the average contribution, select the top 'a' image parameters as image feature parameters, and generate feature grids and image feature parameter sets.

[0199] Feedback scoring module: Based on physiological information and posture images and videos in historical usage records, calculate the physiological score and posture score corresponding to each time window, and then calculate the user feedback score for each time window. By using the average user feedback score of all historical usage records, determine the range of the average user feedback score corresponding to each feedback level.

[0200] The feedback scoring module includes a physiological scoring unit and a posture scoring unit.

[0201] Physiological scoring unit: Obtain physiological information from historical usage records, extract physiological parameters from historical usage records, normalize each physiological parameter, preset time windows, construct time series analysis models for normalized physiological parameters using the time windows as units, calculate the rate of change and stability index of physiological parameters within each time window, preset the weights of the rate of change and stability index, perform weighted summation of the rate of change and stability index, calculate the weight of each physiological parameter in different time windows, and perform weighted summation of the normalized physiological parameters and the corresponding time window weights to calculate the physiological score for each time window;

[0202] The posture scoring unit acquires posture images and videos from historical usage records. Using visual human motion capture technology, it identifies and extracts consecutive image frames corresponding to each user's posture change. Posture features are extracted from each frame in the consecutive image frames, and the change value of each posture feature is obtained. The change value of each posture feature is normalized. Using a time window as the unit, a posture feature change vector is constructed from the normalized posture features. The posture change rate and smoothness index within the time window are calculated. Weights for the posture change rate and smoothness index are preset, and the posture change rate and smoothness index are weighted and summed to obtain the weight of each posture feature in different time windows. The normalized posture feature change value is weighted and summed with the corresponding weight of the time window to calculate the posture score for each time window.

[0203] Feedback rating prediction model module: Set the training period, calculate the grid cell score of each feature grid corresponding to the usage record during the training period based on the feature grid and image feature parameter set, and combine it with the average user feedback score to generate a feedback rating prediction model.

[0204] Initialization Model Module: Based on the range of average user feedback scores corresponding to each feedback level, determine the feedback level corresponding to each usage record, calculate the image score of each usage record, and combine it with the average user feedback score and the initialization control parameters of the spinal correction device to generate the initialization model for each feedback level.

[0205] Dynamic adjustment model module: At a certain feedback level, calculate the rate of change of physiological parameters, change value of posture characteristics, and adjustment value of control parameters for each time window within the record, and generate a dynamic adjustment model for each feedback level in different time windows;

[0206] Real-time adjustment module: Real-time acquisition of the user's original spinal morphology image, calculation of real-time initialization control parameters of the spinal correction device based on feedback scoring prediction model and initialization model, acquisition of real-time physiological parameter change rate and real-time posture feature change value within the time window, calculation of dynamic adjustment control parameters through dynamic adjustment model, and adjustment of the spinal correction device according to the dynamic adjustment control parameters.

[0207] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A data management method applied to a spinal column correction device, characterized in that, The methods include: Step S100: Through the spinal correction device platform, obtain the user's original spinal morphology image from the historical usage records, and perform similarity calculation with the standard spinal morphology image to generate a feature grid and a set of image feature parameters. Step S100 includes the following steps: Step S101: In the spinal correction device, an external physiological monitoring device is connected to monitor the user's physiological information during use. An image acquisition device is deployed on the spinal correction device to collect posture images and videos of the user during use. At the same time, the operating parameters of the spinal correction device during use are obtained. Based on time synchronization technology, the physiological information, posture images and videos, and operating parameters corresponding to the same timestamp are grouped into a timestamped information group. All information groups during use are summarized to generate a usage record, and the usage record is uploaded to the spinal correction device platform. Step S102: Extract the original spinal morphology image before the user's use. Divide the original spinal morphology image into several grid units through feature mapping of a convolutional neural network. Extract several image parameters from each grid unit and number each grid unit to obtain a standard spinal morphology image. Extract several standard image parameters from each grid unit in the standard spinal morphology image. Compare the image parameters of each grid unit with the standard image parameters through a structural similarity algorithm. Calculate the absolute difference of each image parameter. Preset the weight of each image parameter. Calculate the grid unit score by weighted summation of the absolute differences of each image parameter. Preset a grid unit score threshold. Mark grid units that exceed the grid unit score threshold as abnormal grids. Step S103: Obtain historical usage records, count the number of times each grid cell is marked as an abnormal grid, calculate the abnormal frequency of each grid cell, preset an abnormal frequency threshold, and mark grid cells that exceed the abnormal frequency threshold as feature grids. Step S104: Collect historical usage records where the feature grid is an abnormal grid at least once, obtain the image parameters of each feature grid, normalize all image parameters, perform feature decomposition on the normalized image parameter matrix using principal component analysis, calculate the contribution of each image parameter based on the variance contribution rate of each image parameter in the principal components, summarize the contributions of each image parameter, calculate the average contribution of each image parameter, preset the number of image feature parameters to be 'a', sort the image parameters from high to low according to the average contribution, select the first 'a' image parameters as image feature parameters, and generate the feature grid and image feature parameter set; Step S200: Based on the physiological information and posture images and videos in the historical usage records, calculate the physiological score and posture score corresponding to each time window, and then calculate the user feedback score for each time window. By using the average user feedback score of all historical usage records, determine the range of the average user feedback score corresponding to each feedback level. Step S300: Set the training period, calculate the grid cell score of each feature grid corresponding to the record during the training period based on the feature grid and the set of image feature parameters, and combine it with the average user feedback score to generate a feedback score prediction model. Step S400: Based on the range of the average user feedback score corresponding to each feedback level, determine the feedback level corresponding to each usage record, calculate the image score of each usage record, and combine it with the average user feedback score and the initialization control parameters of the spinal correction device to generate the initialization model for each feedback level. Step S500: In a certain feedback level, calculate the rate of change of physiological parameters, the change value of posture characteristics, and the adjustment value of control parameters for each time window in the record, and generate a dynamic adjustment model for each feedback level in different time windows; Step S500 includes the following steps: Step S501: In a set of usage records for a certain feedback level, collect the rate of change of physiological parameters and the change value of posture features for each time window in a certain usage record, and normalize the rate of change of physiological parameters and the change value of posture features. Step S502: Obtain the adjustment value of each control parameter corresponding to each time window in the usage record, and normalize the adjustment value of each control parameter. Combine the normalized physiological parameter change rate, posture feature change value, and control parameter adjustment value of all usage records in the feedback level under the same time window to form a sample and construct a dynamic adjustment training set. Step S503: Using the normalized physiological parameter change rate and posture feature change value as input, and the normalized control parameter adjustment value as output, a long short-term memory neural network is used for training to generate a dynamic adjustment model for each feedback level in different time windows. Step S600: Real-time acquisition of the user's original spinal morphology image; calculation of real-time initialization control parameters of the spinal correction device based on the feedback scoring prediction model and initialization model; acquisition of real-time physiological parameter change rate and real-time posture feature change value within the time window; calculation of dynamic adjustment control parameters through the dynamic adjustment model; and adjustment of the spinal correction device according to the dynamic adjustment control parameters.

2. The data management method for a spinal correction apparatus according to claim 1, wherein, Step S200 includes the following steps: Step S201: Obtain physiological information from historical usage records, extract physiological parameters from historical usage records, perform normalization calculations on each physiological parameter, preset time windows, construct time series analysis models for normalized physiological parameters using the time windows as units, calculate the rate of change and stability index of physiological parameters within each time window, preset the weights of the rate of change and stability index, perform weighted summation of the rate of change and stability index, calculate the weight of each physiological parameter in different time windows, perform weighted summation of the normalized physiological parameters and the weights of the corresponding time windows, and calculate the physiological score for each time window; Step S202: Acquire posture image videos from historical usage records. Using visual human motion capture technology, identify and extract continuous image frames corresponding to each posture change of the user. Extract posture features from each frame in the continuous image frames, obtain the change value of each posture feature, normalize the change value of each posture feature, construct a posture feature change vector for the normalized posture features in units of time windows, calculate the posture change rate and smoothness index within the time window, preset the weights of the posture change rate and smoothness index, and calculate the weight of each posture feature in different time windows by weighted summation. Calculate the posture score for each time window by weighted summation of the normalized posture feature change value and the weight of the corresponding time window. Step S203: Combine the physiological scores and posture scores of the same time window, preset the weights of the physiological scores and posture scores, calculate the user feedback score of a certain time window by weighted summation of the physiological scores and posture scores, summarize the user feedback scores corresponding to all time windows, and calculate the average user feedback score of historical usage records. Step S204: Summarize the average user feedback rating of all historical usage records, and calculate the variance of the average user feedback rating as b. If there are k feedback levels, generate the range of each feedback level as: {[D1, D1+1×b), [D1+1×b, D1+2×b), ..., [D1+(k-1)×b, D2]}, where D1 and D2 represent the minimum and maximum values ​​of the average user feedback rating, respectively.

3. A data management method for a spinal correction apparatus as recited in claim 2, wherein, Step S300 includes the following steps: Step S301: Select several consecutive days as the training period, obtain the usage records within the training period, obtain the original spinal morphology image in a certain usage record based on the feature grid and image feature parameter set, collect the image feature parameters of each feature grid, and calculate the grid cell score of each feature grid. Step S302: Collect physiological information and posture images and videos from the usage records, calculate the average user feedback score of the usage records, combine the grid cell score of each feature grid with the average user feedback score to generate an image training set, use the grid cell score of each feature grid as input and the average user feedback score as output, train the image training set through a linear regression model to generate a feedback score prediction model.

4. The data management method for a spinal correction apparatus according to claim 3, wherein, Step S400 includes the following steps: Step S401: Obtain the average user feedback rating of the usage records, and based on the range of feedback levels, determine the feedback level corresponding to each usage record, and extract the initialization control parameters of the spinal correction device corresponding to each usage record; Step S402: The grid cell scores of each feature grid in the usage record are weighted and summed to calculate the image score of the usage record. The usage records of the same feedback level are summarized. The image score and the average user feedback score are used as inputs, and the initialization control parameters are used as outputs. The multilayer perceptron regression model is used for training to generate the initialization model for each feedback level.

5. The data management method for a spinal correction apparatus of claim 1, wherein, Step S600 includes the following steps: Step S601: Real-time acquisition of the user's original spinal morphology image; calculation of the real-time grid cell score of each feature grid based on the feature grid and image feature parameter set; input into the feedback score prediction model to calculate the feedback score prediction value. Step S602: Based on the range of feedback levels, determine the real-time feedback level corresponding to the current user, calculate the real-time image score, and input the real-time image score and the predicted feedback score into the initialization model corresponding to the real-time feedback level to calculate the real-time initialization control parameters of the spinal correction device; Step S603: Monitor the real-time physiological parameter change rate and real-time posture feature change value within the current time window, and perform normalization calculation. Input the normalized real-time physiological parameter change rate and real-time posture feature change value into the dynamic adjustment model of the real-time feedback level in the current time window, calculate the dynamic adjustment control parameters, and upload them to the spinal correction device. Perform dynamic adjustment on the spinal correction device according to the dynamic adjustment control parameters.

6. A data management system for a spinal correction apparatus for implementing the data management method for a spinal correction apparatus according to any one of claims 1 to 5, characterized in that, The system includes an image analysis module, a feedback scoring module, a feedback scoring prediction model module, an initialization model module, a dynamic adjustment model module, and a real-time adjustment module. The image analysis module: acquires the user's original spinal morphology image from historical usage records through the spinal correction device platform, calculates the similarity with the standard spinal morphology image, and generates a feature grid and a set of image feature parameters. The feedback scoring module calculates the physiological score and posture score corresponding to each time window based on the physiological information and posture images and videos in the historical usage records, and then calculates the user feedback score for each time window. By using the average user feedback score of all historical usage records, the range of the average user feedback score corresponding to each feedback level is determined. The feedback rating prediction model module: sets a training period, calculates the grid cell rating of each feature grid corresponding to the record during the training period based on the feature grid and image feature parameter set, and combines it with the average user feedback rating to generate a feedback rating prediction model. The initialization model module: Based on the range of the average user feedback score corresponding to each feedback level, it determines the feedback level corresponding to each usage record, calculates the image score of each usage record, and combines it with the average user feedback score and the initialization control parameters of the spinal correction device to generate an initialization model for each feedback level. The dynamic adjustment model module: In a certain feedback level, it calculates the rate of change of physiological parameters, the change value of posture characteristics, and the adjustment value of control parameters for each time window within the record, and generates a dynamic adjustment model for each feedback level in different time windows; The real-time adjustment module: acquires the user's original spinal morphology image in real time, calculates the real-time initialization control parameters of the spinal correction device based on the feedback scoring prediction model and the initialization model, and acquires the real-time physiological parameter change rate and real-time posture feature change value within the time window. Through the dynamic adjustment model, it calculates the dynamic adjustment control parameters and adjusts the spinal correction device according to the dynamic adjustment control parameters.

7. A data management system for use with a spinal correction apparatus as defined in claim 6, wherein, The image analysis module includes a unit for identifying abnormal grid cells and a unit for determining image feature parameters. The process of identifying abnormal grid units involves: connecting external physiological monitoring equipment to the spinal correction device to monitor the user's physiological information during use; deploying image acquisition equipment on the spinal correction device to collect posture images and videos of the user during use; simultaneously acquiring the operating parameters of the spinal correction device during use; and, based on time synchronization technology, grouping physiological information, posture images / videos, and operating parameters corresponding to the same timestamp into timestamped information groups; summarizing all information groups during use to generate a usage record; and uploading the usage record to the spinal correction device platform; extracting the user's original spinal morphology image before use; and then performing feature mapping using a convolutional neural network. The original spinal morphology image is divided into several grid cells. Several image parameters are extracted from each grid cell and each grid cell is numbered to obtain a standard spinal morphology image. Several standard image parameters are extracted from each grid cell in the standard spinal morphology image. Using a structural similarity algorithm, the image parameters of each grid cell are compared with the standard image parameters to calculate the absolute difference of each image parameter. The weight of each image parameter is preset. The grid cell score is calculated by weighted summation of the absolute differences of each image parameter. A grid cell score threshold is preset. Grid cells that exceed the grid cell score threshold are marked as abnormal grids. The process of determining image feature parameters involves: acquiring historical usage records, counting the number of times each grid cell is marked as an abnormal grid, calculating the abnormal frequency of each grid cell, setting a preset abnormal frequency threshold, marking grid cells exceeding the threshold as feature grids, collecting historical usage records where a feature grid appears as an abnormal grid at least once, acquiring image parameters for each feature grid, normalizing all image parameters, performing eigenvalue decomposition on the normalized image parameter matrix using principal component analysis, calculating the contribution of each image parameter based on its variance contribution rate in the principal components, summing the contributions of each image parameter, calculating the average contribution of each image parameter, setting the preset number of image feature parameters to 'a', sorting the image parameters from highest to lowest average contribution, selecting the top 'a' image parameters as image feature parameters, and generating feature grids and a set of image feature parameters.

8. A data management system for use with a spinal correction apparatus as defined in claim 6, wherein, The feedback scoring module includes a physiological scoring unit and a posture scoring unit: The physiological scoring unit: acquires physiological information from historical usage records, extracts physiological parameters from historical usage records, performs normalization calculations on each physiological parameter, presets a time window, constructs a time series analysis model for the normalized physiological parameters using the time window as a unit, calculates the rate of change and stability index of physiological parameters within each time window, presets the weights of the rate of change and stability index, performs a weighted summation of the rate of change and stability index, calculates the weight of each physiological parameter in different time windows, and performs a weighted summation of the normalized physiological parameters and the weights of the corresponding time windows to calculate the physiological score for each time window; The posture scoring unit: acquires physiological information from historical usage records, extracts physiological parameters from historical usage records, performs normalization calculations on each physiological parameter, presets a time window, constructs a time series analysis model for the normalized physiological parameters using the time window as a unit, calculates the rate of change and stability index of physiological parameters within each time window, presets the weights of the rate of change and stability index, performs a weighted summation of the rate of change and stability index, calculates the weight of each physiological parameter in different time windows, and performs a weighted summation of the normalized physiological parameters with the weights of the corresponding time windows to calculate the physiological score for each time window.

Images