A method and system for constructing a diabetic osteoporosis fracture risk prediction model

Abstract

This invention relates to the field of fracture risk prediction technology, and particularly to a method and system for constructing a fracture risk prediction model for diabetic osteoporosis. The method involves collecting bone metabolism data and hip and spine bone density computed tomography images from diabetic patients to construct a multi-dimensional three-dimensional model of bone tissue microstructure. Based on this model, blood glucose fluctuation-bone strength mapping is performed to calculate the trabecular fragility increment and identify the pre-gradient feedback characteristics of brittle mechanical signals to obtain brittle mechanical pre-gradient feedback data. Phase shift analysis of bone stress signals is then performed on the above data to extract phase shift feature data and construct a fracture risk prediction model for diabetic osteoporosis. Finally, dynamic assessment and remote monitoring of the patient's fracture risk are conducted. This method integrates three-dimensional bone microstructure modeling, blood glucose fluctuation impact analysis, and artificial intelligence prediction, improving the early warning accuracy of diabetic osteoporotic fractures.

Description

Technical Field

[0001] This invention relates to the field of fracture risk prediction technology, and in particular to a method and system for constructing a fracture risk prediction model for diabetic osteoporosis. Background Technology

[0002] With an aging population and changing lifestyles, the incidence of diabetes continues to rise globally, becoming a serious public health problem. Diabetes not only affects metabolic processes such as glucose, lipids, and proteins, but also causes significant damage to the skeletal system, leading to a significantly increased risk of diabetic osteoporosis. Although bone mineral density (BMD) values ​​in diabetic patients sometimes appear normal or even elevated, their fracture risk remains significantly higher than that of non-diabetic individuals, suggesting limitations in traditional BMD assessment methods for predicting fracture risk in diabetic patients. The root cause lies in factors such as chronic hyperglycemia, insulin deficiency or resistance, and the accumulation of advanced glycation end products (AGEs) caused by diabetes, which damage bone microstructure, reduce bone quality, weaken bone strength, and increase bone fragility, thus making fractures of the hip, spine, and other areas possible even under minor external forces.

[0003] Currently, clinical assessment of osteoporotic fracture risk mainly relies on bone mineral density (BMD) measurements using dual-energy X-ray absorptiometry (DXA) and clinical risk assessment tools such as FRAX. However, these methods cannot fully reflect the complex bone metabolic abnormalities and bone microstructural degeneration characteristics in diabetic patients. In recent years, with the development of medical imaging technology and biomechanical modeling, constructing three-dimensional microstructural models of bone tissue based on high-resolution CT or quantitative CT, combined with finite element analysis for bone strength simulation, has provided the possibility for more accurate assessment of bone quality. However, existing technologies still lack systematic modeling of the dynamic correlation between blood glucose fluctuations and the mechanical properties of bone microstructure, especially neglecting the gradual impact of long-term blood glucose fluctuations on the evolution of trabecular fragility and its precursory mechanical signal characteristics before fracture.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a method and system for constructing a risk prediction model for diabetic osteoporosis fractures. This aims to address the technical problem that existing methods for assessing the risk of osteoporosis fractures are unable to effectively reflect the dynamic impact of blood glucose fluctuations on bone microstructure and bone fragility in diabetic patients, and lack an early and accurate prediction model that integrates multimodal data and biomechanical characteristics.

[0006] To achieve the above objectives, the present invention provides a method for constructing a risk prediction model for diabetic osteoporosis fractures, the method comprising:

[0007] Step S1: Collect bone mineral density data and hip and spine bone mineral density images of diabetic patients through the medical and health information system to obtain bone metabolism data and bone mineral density tomographic images of diabetic patients; construct a multi-dimensional three-dimensional model of bone trabeculae based on the bone mineral density tomographic images of diabetic patients to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

[0008] Step S2: Perform blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data; calculate the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data to obtain trabecular fragility increment data; perform brittle mechanical signal pre-gradient feedback identification on the trabecular fragility increment data based on the blood glucose fluctuation-bone strength mapping data to obtain brittle mechanical pre-gradient feedback data.

[0009] Step S3: Perform phase shift analysis on the brittle mechanics pre-gradient feedback data to obtain the phase shift data of the brittle feedback bone stress signal; use the XGBoost algorithm to construct a risk prediction model for diabetic osteoporosis fractures based on the phase shift data of the brittle feedback bone stress signal, and obtain the risk prediction model for diabetic osteoporosis fractures.

[0010] Step S4: Design an automated monitoring firmware for the diabetic osteoporosis fracture risk prediction model to obtain the monitoring firmware for the diabetic osteoporosis fracture risk prediction model. Send the monitoring firmware for the diabetic osteoporosis fracture risk prediction model to the health management cloud platform to perform diabetic osteoporosis fracture risk prediction.

[0011] Optionally, step S1 includes the following steps:

[0012] Step S11: Collect bone mineral density data and hip and spine bone mineral density images of diabetic patients through the medical and health information system to obtain basic data of diabetic patients, including bone metabolism data and bone mineral density tomographic images of diabetic patients.

[0013] Step S12: Optimize the bone tissue noise in the bone density tomography images of diabetic patients to obtain bone density noise-optimized images;

[0014] Step S13: Perform multi-dimensional reconstruction analysis on the bone density tomographic images of diabetic patients based on the bone density noise optimization images to obtain multi-dimensional reconstruction data of bone density tomographic scans;

[0015] Step S14: Construct a multi-dimensional three-dimensional model of bone tissue microstructure based on the multi-dimensional reconstruction data of bone density tomography scan to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

[0016] Optionally, step S2 includes the following steps:

[0017] Step S21: Extract continuous blood glucose monitoring data from the bone metabolism signs data of diabetic patients to obtain continuous blood glucose data of diabetic patients;

[0018] Step S22: Based on the continuous blood glucose data of diabetic patients, perform blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data.

[0019] Step S23: Use the preset diabetic bone abnormality change recognition model to identify bone microstructure abnormalities in diabetic patients by analyzing blood glucose fluctuation-bone strength mapping data, and obtain bone microstructure abnormality data of diabetic patients.

[0020] Step S24: Calculate the trabecular fragility increment based on the blood glucose fluctuation-bone strength mapping data of diabetic patients to obtain the trabecular fragility increment data;

[0021] Step S25: Based on the blood glucose fluctuation-bone strength mapping data, perform brittle mechanical signal pre-gradient feedback identification on the trabecular brittleness increment data to obtain brittle mechanical pre-gradient feedback data.

[0022] Optionally, step S24 includes the following steps:

[0023] Step S241: Calculate the porosity of trabecular bone structure from the abnormal bone microstructure data of diabetic patients to obtain trabecular bone porosity data;

[0024] Step S242: Simulate bone elasticity decay based on trabecular porosity data and blood glucose fluctuation-bone strength mapping data to obtain bone elasticity decay simulation data;

[0025] Step S243: Perform stress increment direction decomposition calculation on the bone elasticity attenuation simulation data to obtain bone stress increment direction decomposition data;

[0026] Step S244: Calculate the trabecular brittleness increment based on the decomposition data of bone stress increment direction to obtain the trabecular brittleness increment data.

[0027] Optionally, the stress increment direction decomposition calculation of the bone elasticity attenuation simulation data includes the following steps:

[0028] The bone elasticity and mechanical decay simulation data were used to assess the decay of the trabecular load-bearing strength, and the bone load-bearing strength decay data were obtained.

[0029] Based on the bone load-bearing strength attenuation data, bone stress load strength was simulated to obtain bone stress load strength data;

[0030] Global stress tensor calculation was performed on the bone stress load strength data to obtain global stress tensor data;

[0031] Stress spatial distribution fluctuation analysis is performed based on global stress tensor data to obtain stress spatial distribution fluctuation data;

[0032] The stress potential energy relationship is quantified based on Hooke's law to obtain stress space potential energy relationship data.

[0033] Based on the stress-space potential energy relationship data and stress-space distribution fluctuation data, stress increment direction decomposition calculation is performed to obtain bone stress increment direction decomposition data.

[0034] Optionally, step S25 includes the following steps:

[0035] Step S251: Perform time-domain feature analysis on the blood glucose fluctuation-bone strength mapping data to obtain time-domain data of blood glucose fluctuation-bone strength;

[0036] Step S252: Perform a multifactorial correlation analysis on the trabecular fragility increment data based on the blood glucose fluctuation-bone strength time domain data to obtain the multifactorial correlation data on fragility.

[0037] Step S253: Perform multi-factor cluster analysis on the brittle multi-factor association data to obtain brittle multi-factor association cluster data;

[0038] Step S254: Based on the brittle multi-factor association clustering data, perform dynamic brittle response simulation on the incremental brittle data of bone trabeculae to obtain multi-factor dynamic brittle response data;

[0039] Step S255: Based on the multi-factor dynamic fragility response data, identify the stress response pre-gradient trend of the blood glucose fluctuation-bone strength mapping data to obtain the stress response pre-gradient trend data.

[0040] Step S256: Based on the stress response pre-gradient trend data and multi-factor dynamic brittle response data, perform brittle mechanical signal pre-gradient feedback identification to obtain brittle mechanical pre-gradient feedback data.

[0041] Optionally, step S3 includes the following steps:

[0042] Step S31: Perform phase shift analysis on the brittle mechanics pre-gradient feedback data to obtain the brittle feedback bone stress signal phase shift data;

[0043] Step S32: Perform spatial interpolation on the phase shift data of the brittle feedback bone stress signal to obtain stress shift spatial interpolation data;

[0044] Step S33: Perform signal compensation on the phase shift data of the brittle feedback bone stress signal based on the stress offset spatial interpolation data to obtain the brittle feedback signal phase shift compensation data;

[0045] Step S34: Use the XGBoost algorithm to construct a risk prediction model for diabetic osteoporosis fractures based on the phase shift compensation data of the brittle feedback signal, and obtain the risk prediction model for diabetic osteoporosis fractures.

[0046] Optionally, step S32 includes the following steps:

[0047] Step S321: Calculate the phase shift of the brittle bone stress signal phase shift data to obtain the brittle bone stress signal phase shift data;

[0048] Step S322: Perform nonlinear incremental calculation on the phase shift data of the brittle bone stress signal based on the phase shift data of the brittle bone stress signal to obtain the phase shift nonlinear incremental data;

[0049] Step S323: Perform spatial incremental trend analysis on the phase shift data of the brittle feedback bone stress signal based on the phase shift nonlinear incremental data to obtain the spatial incremental trend data of the phase shift;

[0050] Step S324: Based on the phase offset spatial increment trend data, perform spatial interpolation processing on the phase offset data of the brittle feedback bone stress signal to obtain stress offset spatial interpolation data.

[0051] Optionally, step S34 includes the following steps:

[0052] Step S341: Divide the brittle feedback signal phase offset compensation data into a dataset to obtain a brittle feedback signal phase offset compensation test set and a brittle feedback signal phase offset compensation training set.

[0053] Step S342: Select similar features of feedback signals from the training set for phase offset compensation of brittle feedback signals to obtain similar feature data of feedback signals;

[0054] Step S343: Using the XGBoost algorithm and based on the feedback signal similarity feature data, perform similar structure incremental learning on the brittle feedback signal phase offset compensation training set to obtain similar structure incremental data;

[0055] Step S344: Construct an initial diabetic osteoporotic fracture risk prediction model based on incremental data of similar structures to obtain the initial diabetic osteoporotic fracture risk prediction model.

[0056] Step S345: Test the initial diabetic osteoporosis fracture risk prediction model based on the brittle feedback signal phase offset compensation test set to obtain the diabetic osteoporosis fracture risk prediction model.

[0057] Furthermore, to achieve the above objectives, the present invention also provides a system for constructing a risk prediction model for diabetic osteoporosis fractures, the system comprising:

[0058] The data modeling module is used to collect bone mineral density data and hip joint and spine bone mineral density images of diabetic patients through the medical and health information system, respectively obtaining bone metabolism record data and bone mineral density tomographic images of diabetic patients; based on the bone mineral density tomographic images of diabetic patients, a multi-dimensional three-dimensional model of bone trabeculae is constructed to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

[0059] The fragility identification module is used to perform blood glucose fluctuation-bone strength mapping processing on a multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data; to calculate the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data to obtain trabecular fragility increment data; and to identify the brittle mechanical signal pre-gradient feedback based on the blood glucose fluctuation-bone strength mapping data and the trabecular fragility increment data to obtain brittle mechanical pre-gradient feedback data.

[0060] The model building module is used to perform phase shift analysis on the brittle mechanics pre-gradient feedback data of bone stress signal to obtain the phase shift data of brittle feedback bone stress signal; the XGBoost algorithm is used to build a risk prediction model for diabetic osteoporosis fracture based on the phase shift data of brittle feedback bone stress signal to obtain the risk prediction model for diabetic osteoporosis fracture.

[0061] The firmware deployment module is used to design automated monitoring firmware for the diabetic osteoporosis fracture risk prediction model, obtain the monitoring firmware for the diabetic osteoporosis fracture risk prediction model, and send the monitoring firmware for the diabetic osteoporosis fracture risk prediction model to the health management cloud platform to perform diabetic osteoporosis fracture risk prediction.

[0062] This invention provides a method for constructing a predictive model for fracture risk in diabetic osteoporosis. The method integrates bone metabolism data from diabetic patients with high-resolution bone density computed tomography (BMD) images to construct a multi-dimensional three-dimensional model of bone tissue microstructure, achieving refined characterization of the trabecular bone microstructure. Based on this, it innovatively introduces a blood glucose fluctuation-bone strength mapping mechanism to quantitatively analyze the impact of dynamic blood glucose changes on bone mechanical properties. Furthermore, through trabecular fragility increment calculation and pre-gradient feedback identification of fragility mechanical signals, it effectively captures early mechanical precursor signals of bone fragility evolution before fracture. Further, by combining bone stress signal phase shift analysis, sensitive temporal-spatial mechanical features are extracted, and a high-precision predictive model is constructed using the XGBoost algorithm, significantly improving the accuracy of predicting fracture risk in diabetic osteoporosis and enhancing early warning capabilities. Finally, through automated monitoring firmware design and integration into a health management cloud platform, intelligent deployment and long-term dynamic monitoring of the model are achieved, providing a scientific and efficient decision support tool for personalized prevention and clinical intervention of osteoporosis complications in diabetic patients. The holistic approach integrates medical imaging, biomechanical modeling, and artificial intelligence technologies, overcoming the limitations of traditional bone mineral density assessment in diabetic patients, and has promising clinical application prospects and promotional value. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating an embodiment of the method for constructing a risk prediction model for diabetic osteoporosis fractures according to the present invention.

[0064] Figure 2 This is a structural block diagram of an embodiment of the diabetic osteoporosis fracture risk prediction model construction system of the present invention.

[0065] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0066] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0067] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the method for constructing a risk prediction model for diabetic osteoporosis fractures according to the present invention.

[0068] In one embodiment, the method for constructing the diabetic osteoporosis fracture risk prediction model includes:

[0069] Step S1: Collect bone mineral density (BMD) data and hip and spine BMD images of diabetic patients through a medical health information system to obtain BMD records and BMD computed tomographic images. Based on the BMD computed tomographic images, construct a multi-dimensional three-dimensional model of bone trabeculae to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

[0070] The bone metabolism data of diabetic patients can be a set of clinical biochemical indicators and physiological parameters reflecting the bone metabolism status of diabetic patients. This data can provide metabolic background information for bone microstructure modeling and mechanical property analysis, and help explain the biochemical mechanisms of abnormal bone quality. In this embodiment, the bone metabolism data of diabetic patients can be extracted from electronic medical records, laboratory test reports, or wearable devices through a medical health information system, including bone formation markers (such as P1NP), bone resorption markers (such as CTX), insulin levels, and glycated hemoglobin. The bone density computed tomography (CT) images of diabetic patients can be cross-sectional bone tissue images obtained from the hip joint or spine region of diabetic patients using high-resolution CT or quantitative CT (QCT). This data can be used as the original spatial data source for constructing a three-dimensional microstructure model of bone trabeculae. Furthermore, the bone density computed tomography images of diabetic patients can be obtained by performing tomographic scanning on the target bone region using medical imaging equipment to generate a voxel-level grayscale image sequence. In an exemplary embodiment, the bone density computed tomography images of diabetic patients can use high-resolution peripheral quantitative CT images, micro-CT images, clinical QCT images, etc.

[0071] A multi-dimensional three-dimensional model of bone tissue microstructure can be a digital three-dimensional structural model reconstructed from bone density tomographic images, containing the geometric morphology, connectivity, porosity, and spatial distribution characteristics of trabecular bone. This model can be used to achieve a refined characterization of the microstructure of trabecular bone, providing a geometric and topological basis for subsequent biomechanical analysis. In one specific embodiment, the multi-dimensional three-dimensional model of bone tissue microstructure can extract the trabecular bone network from tomographic images using image segmentation, skeleton extraction, and surface reconstruction algorithms, and assign material properties to support mechanical simulation. Furthermore, the multi-dimensional three-dimensional model of bone tissue microstructure can be the input object for blood glucose fluctuation-bone strength mapping processing, and its structural features directly affect the accuracy of the mapping results.

[0072] Collecting bone mineral density (BMD) data and hip and spine BMD images from diabetic patients through healthcare information systems can be achieved by calling interfaces of Hospital Information System (HIS), Laboratory Information System (LIS), and Picture Archiving and Communication System (PACS) to simultaneously acquire structured vital sign data and DICOM format images. Furthermore, this operation can be implemented by pulling the latest patient test results and image data in real time via API, or by periodically synchronizing historical data to a local data lake using batch ETL tools. This achieves standardized integration of multi-source heterogeneous health data, providing a complete input for subsequent modeling. Constructing a multi-dimensional 3D model of bone trabeculae based on bone mineral density tomographic images of diabetic patients can be achieved by preprocessing, thresholding, binarizing, and reconstructing the tomographic images to generate a finite element mesh model with material properties. Further, this operation can be achieved by using the MarchingCubes algorithm for surface reconstruction followed by importing into ANSYS for mesh generation, or by using a deep learning semantic segmentation model (such as U-Net) to extract bone trabeculae and directly generating a voxelized finite element model. This achieves the technical effect of transforming 2D images into a 3D microstructure digital twin suitable for mechanical simulation.

[0073] Step S2: Perform blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data; calculate the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data to obtain trabecular fragility increment data; perform brittle mechanical signal pre-gradient feedback identification on the trabecular fragility increment data based on the blood glucose fluctuation-bone strength mapping data to obtain brittle mechanical pre-gradient feedback data.

[0074] The blood glucose fluctuation-bone strength mapping data can be a dataset describing the quantitative correlation between dynamic changes in blood glucose (such as fluctuation amplitude, frequency, and duration) and the local mechanical strength of bone tissue. It can be used to reveal the progressive weakening effect of chronic hyperglycemia on the mechanical properties of bone microstructure, providing input for brittleness evolution analysis. In this embodiment, the blood glucose fluctuation-bone strength mapping data can be modeled by coupling blood glucose time-series data with the mechanical response of each region in the bone microstructure model, establishing a mapping function or lookup table. Furthermore, the blood glucose fluctuation-bone strength mapping data can be jointly generated from a multi-dimensional three-dimensional model of bone tissue microstructure and blood glucose time-series data, and used as direct input for calculating the trabecular brittleness increment. Performing blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure can be achieved by coupling the patient's blood glucose time-series data with the mechanical response of each unit in the model, establishing a functional relationship between blood glucose parameters and local bone strength. Furthermore, this operation can derive the bone matrix stiffness attenuation coefficient based on the AGEs accumulation dynamic model and then map it to the finite element material properties, or use historical cohort data to train a blood glucose-strength regression model and assign strength values ​​to the current model units. This can achieve the technical effect of quantitatively characterizing the impact of dynamic blood glucose changes on bone mechanical properties.

[0075] Trabecular fragility increment data can be the incremental change in the mechanical fragility of trabecular bone caused by blood glucose fluctuations within a specific time period. This data can be used to quantify the rate of bone fragility evolution over time and to identify accelerated degeneration stages. In this embodiment, the trabecular fragility increment data can be calculated based on blood glucose fluctuation-bone strength mapping data, calculating the difference between current and historical bone strength and converting it into an incremental value of fragility indicators. Calculating the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data can involve comparing the current and baseline bone strength distributions and calculating the change in fragility indicators (such as the reciprocal of fracture toughness) per unit time. Furthermore, this operation can be achieved by calculating the increment of local failure probability based on Weibull's statistical theory of fragility, or by deriving the fragility increment through the rate of change of energy release rate, thereby achieving the technical effect of quantifying the rate of bone fragility evolution with blood glucose fluctuations.

[0076] Brittle mechanics pre-gradient feedback data can be a set of signals reflecting the gradient change trend of trabecular brittleness increment in spatial and temporal dimensions, used to identify mechanical abnormalities preceding fractures. In this embodiment, brittle mechanics pre-gradient feedback data can perform spatiotemporal gradient calculation on trabecular brittleness increment data to identify abnormally steep increase regions and their propagation directions. For example, brittle mechanics pre-gradient feedback data can include, but is not limited to, one or more of the following: brittle spatial gradient vector field, brittle time derivative sequence, brittle mutation heatmap, etc. Based on blood glucose fluctuation-bone strength mapping data, brittle mechanics signal pre-gradient feedback identification of trabecular brittleness increment data can be achieved by performing gradient calculation on the brittleness increment data in the spatiotemporal domain to identify abnormal growth regions and their propagation patterns. Furthermore, this operation can be implemented by using the Sobel operator to calculate the spatial gradient of the three-dimensional brittle field, combined with temporal difference to identify mutation points, or by using a convolutional neural network to automatically extract brittle gradient anomaly feature maps, thereby achieving the technical effect of early identification of mechanical instability precursors in high-risk fracture areas.

[0077] Step S3: Perform phase shift analysis on the brittle mechanics pre-gradient feedback data to obtain the phase shift data of the brittle feedback bone stress signal. Use the XGBoost algorithm to construct a risk prediction model for diabetic osteoporosis fractures based on the phase shift data of the brittle feedback bone stress signal, thus obtaining the risk prediction model for diabetic osteoporosis fractures.

[0078] The brittle feedback bone stress signal phase shift data can be the phase delay or advance of the internal stress response signal of bone tissue relative to standard healthy bone under loading simulation conditions, corrected by brittle feedback. This data is used to extract temporal-spatial mechanical features sensitive to bone microstructural degeneration, enhancing the discriminative ability of the machine learning model. In this embodiment, the brittle feedback bone stress signal phase shift data can be obtained by applying periodic loads through finite element simulation, extracting the temporal stress signals of key nodes, and calculating the shift by comparing them with a reference phase. The XGBoost algorithm is an ensemble machine learning algorithm based on gradient boosting decision trees, which can be used to build high-precision classification models and learn fracture risk discrimination rules from complex mechanical features. The diabetic osteoporosis fracture risk prediction model can be a machine learning model trained based on the brittle feedback bone stress signal phase shift data to predict the probability of individual fractures. It can be used to achieve high-precision prediction and risk stratification of future fracture events in diabetic patients. In this embodiment, the diabetic osteoporosis fracture risk prediction model can use the XGBoost algorithm as a framework, inputting multidimensional mechanical features and outputting a fracture risk score or classification label.

[0079] Phase shift analysis of bone stress signals from brittle mechanics pre-gradient feedback data can be performed by simulating bone tissue stress response under virtual loading conditions, extracting key node stress time-series signals, and comparing their phase difference with a healthy reference signal. Further, this operation can be achieved by applying sinusoidal loads and analyzing the dominant frequency phase shift using FFT, or by simulating falls using impact loads and analyzing the arrival time shift of transient responses. This achieves the technical effect of obtaining dynamic mechanical characteristics highly sensitive to microstructural degradation. Constructing a diabetic osteoporosis fracture risk prediction model using the XGBoost algorithm on brittle feedback bone stress signal phase shift data can be achieved by using the phase shift feature vector as input and the presence or absence of fracture as a label to train an XGBoost classifier. Further, this operation can be achieved by using five-fold cross-validation to optimize hyperparameters before training the final model, or by combining SHAP values ​​for feature selection before training a lightweight model. This achieves the technical effect of constructing a fracture risk prediction model with high generalization ability.

[0080] Step S4: Design an automated monitoring firmware for the diabetic osteoporosis fracture risk prediction model to obtain the monitoring firmware for the diabetic osteoporosis fracture risk prediction model. Send the monitoring firmware for the diabetic osteoporosis fracture risk prediction model to the health management cloud platform to perform diabetic osteoporosis fracture risk prediction.

[0081] The monitoring firmware for the diabetic osteoporosis fracture risk prediction model can be an automated monitoring program module that encapsulates the prediction model and can run on edge devices or embedded systems. This module can support long-term, automated fracture risk monitoring in out-of-hospital or community settings. In this embodiment, the monitoring firmware for the diabetic osteoporosis fracture risk prediction model can enable the model to perform inference on low-power hardware through model compression, interface encapsulation, and driver adaptation. The health management cloud platform can be a remote information platform integrating health data storage, analysis, and service delivery. It can be used to receive and run the monitoring firmware, enabling model deployment, data aggregation, and the delivery of clinical early warning information.

[0082] The design of automated monitoring firmware for a diabetic osteoporosis fracture risk prediction model involves converting the trained model into a format that can run on embedded devices and encapsulating data input, preprocessing, and result output interfaces. Furthermore, this operation can be deployed on smart bracelets or home health terminals to periodically trigger risk assessments, or integrated into hospital follow-up systems to automatically call new data for risk reassessment, thereby achieving the technical effect of enabling the model to operate autonomously in non-clinical environments. Sending the diabetic osteoporosis fracture risk prediction model monitoring firmware to a health management cloud platform can be achieved by uploading the firmware package to the cloud platform via a secure communication protocol and registering it as a schedulable service module. Further, this operation can be pushed to user terminal devices via OTA or containerized and deployed on the cloud platform for multiple institutions to call API services, thereby achieving the technical effects of centralized model management, remote updates, and large-scale application.

[0083] Taking community-based bone health management for diabetic patients as an example, the method for constructing a diabetic osteoporosis fracture risk prediction model in this embodiment can be as follows: A 65-year-old type 2 diabetic patient regularly undergoes QCT scans and blood tests at a community hospital. The system automatically collects bone metabolism signs and spinal images to construct a three-dimensional model of trabecular bone. Combined with continuous glucose monitoring (CGM) data, the system calculates the mapping relationship between blood glucose fluctuations and bone strength, and finds that the fragility increment of the lumbar trabecular bone has increased significantly in the past 3 months. Further analysis shows that the fragility gradient has abnormally accumulated in the anterior column region of the L2 vertebral body, and the stress phase shift has reached the threshold. The XGBoost model outputs a high fracture risk score, and the monitoring firmware pushes the warning information to the health management cloud platform. After receiving the reminder, the family doctor arranges bisphosphonate intervention and fall prevention education.

[0084] In one embodiment, step S1 includes the following steps:

[0085] Step S11: Collect bone mineral density data and hip and spine bone mineral density images of diabetic patients through the medical and health information system to obtain basic data of diabetic patients, including bone metabolism data and bone mineral density tomographic images of diabetic patients.

[0086] Among them, the basic data of diabetic patients can be a collection of bone metabolism signs records and bone density tomography images of diabetic patients, which can be used to provide complete clinical and imaging basic information for bone microstructure modeling.

[0087] Step S12: Optimize the bone tissue noise in the bone density tomography images of diabetic patients to obtain bone density noise-optimized images;

[0088] The bone density noise-optimized image can be a bone density tomographic image that has undergone targeted denoising processing. This preserves the true structure of the trabecular bone and suppresses imaging noise, improving the edge sharpness and signal-to-noise ratio of the internal structure of the trabecular bone, and reducing topological errors in subsequent 3D reconstruction. For example, the bone density noise-optimized image can be one or more of the following: denoised images based on non-local means, denoised images based on convolutional autoencoders, and images optimized by anisotropic diffusion filtering. Furthermore, optimizing bone tissue noise in bone density tomographic images of diabetic patients can involve applying image denoising algorithms specifically designed for bone tissue to process the original tomographic image, suppressing unstructured noise while preserving fine trabecular bone. In an exemplary embodiment, this operation can be achieved by using a deep learning model with a U-Net architecture trained on labeled micro-CT data to denoise clinical QCT images, or by employing an improved anisotropic diffusion filter to suppress smoothing in bone edge regions with large gradients and enhance details in low-contrast regions. This can improve the discriminability of the trabecular bone structure in the image and reduce segmentation and reconstruction errors caused by noise.

[0089] Step S13: Perform multi-dimensional reconstruction analysis on the bone density tomographic images of diabetic patients based on the bone density noise optimization images to obtain multi-dimensional reconstruction data of bone density tomographic scans;

[0090] The multi-dimensional reconstruction data from bone density tomography can be enhanced tomographic data generated from noise-optimized images, containing multi-dimensional features such as spatial geometry, gray-level gradient, and texture directionality. This data can provide a more complete and feature-rich intermediate representation for 3D model construction, supporting high-fidelity microstructure modeling. In one specific embodiment, the multi-dimensional reconstruction data from bone density tomography can be one or more of the following: local structural tensor reconstruction data, multi-scale curvature enhancement data, and directional gray-level gradient field data. Furthermore, multi-dimensional reconstruction analysis of bone density tomographic images of diabetic patients based on noise-optimized bone density images can be achieved by extracting and fusing information from multiple structural dimensions based on the noise-optimized images to generate an enhanced tomographic representation. For example, this operation can be achieved by using a Hessian matrix to analyze local structural features and constructing a multi-scale tubular enhanced response map for reconstruction, or by extracting the responses of bone trabeculae in different orientations using a directional Gabor filter group and synthesizing multi-channel reconstruction data. This can restore the spatial continuity and microstructural anisotropy of bone trabeculae obscured by noise, improving the realism of subsequent modeling.

[0091] Step S14: Construct a multi-dimensional three-dimensional model of bone tissue microstructure based on the multi-dimensional reconstruction data of bone density tomography scan to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

[0092] Furthermore, constructing a multi-dimensional 3D model of bone tissue microstructure based on multi-dimensional reconstruction data from bone density tomography can be achieved by performing 3D segmentation and surface / voxel reconstruction on the multi-dimensional reconstruction data to generate a digital model of bone trabeculae containing geometric and topological features. In an exemplary embodiment, this operation can be achieved by segmenting using multi-threshold region growth combined with skeletal constraints, then generating a surface mesh using MarchingCubes, or by directly inputting multi-channel reconstruction data into 3DU-Net for end-to-end semantic segmentation of bone trabeculae and outputting a voxelized finite element model. This allows for obtaining a more accurate and detailed 3D model of bone microstructure, providing high-fidelity input for biomechanical simulation.

[0093] Taking precise modeling under high-noise clinical QCT images as an example, the method for constructing a risk prediction model for diabetic osteoporosis fractures in this embodiment can be based on an elderly diabetic patient undergoing a clinical QCT scan, where significant noise is present in the spinal images due to respiratory motion and low-dose settings. The system first collects bone metabolism markers and the original images as basic data; then, it performs deep learning-based bone tissue noise optimization on the images, significantly improving the visibility of vertebral trabeculae; next, it performs multi-dimensional reconstruction analysis to restore the connection paths of trabeculae in the coronal, sagittal, and axial directions; finally, the constructed three-dimensional model accurately presents the characteristics of sparse trabeculae but clear residual structural orientation, providing a reliable geometric basis for subsequent blood glucose fluctuation-bone strength mapping, and avoiding overestimation of fracture risk due to noise misjudgment as widespread fracture.

[0094] In one embodiment, step S2 includes the following steps:

[0095] Step S21: Extract continuous blood glucose monitoring data from the bone metabolism signs data of diabetic patients to obtain continuous blood glucose data of diabetic patients;

[0096] Continuous blood glucose data for diabetic patients can be high-frequency time-series data reflecting the dynamic changes in blood glucose levels over time, acquired through continuous glucose monitoring (CGM) devices. This data can provide high-temporal-resolution characteristics of blood glucose fluctuations, accurately characterizing the dynamic impact of blood glucose variability on the mechanical properties of bone microstructure. In one exemplary embodiment, continuous blood glucose data for diabetic patients can extract minute-level blood glucose value sequences collected and stored by subcutaneous glucose sensors from bone metabolism-related vital signs (BMR) records. Extraction of CGM data from BMR records for diabetic patients can involve screening and parsing time-series blood glucose values ​​from the CGM device within structured BMR records. Furthermore, extraction of CGM data from BMR records for diabetic patients can be achieved by extracting CGM data streams from electronic health records via the FHIR standard interface, or by using natural language processing to identify and structure CGM reports from unstructured medical order notes. This provides high-density, continuous dynamic blood glucose information, replacing traditional single-point blood glucose indicators and improving the accuracy of blood glucose fluctuation characterization.

[0097] Step S22: Based on the continuous blood glucose data of diabetic patients, perform blood glucose fluctuation bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation bone strength mapping data;

[0098] The blood glucose fluctuation bone strength mapping data can be a dataset describing the quantitative relationship between dynamic changes in blood glucose and local bone strength, generated by coupling continuous blood glucose data with a multi-dimensional three-dimensional model of bone tissue microstructure. This dataset can be used to achieve a refined correlation between blood glucose fluctuation patterns and bone mechanical performance degradation, improving the physiological realism and timeliness of the mapping mechanism. In one specific embodiment, the blood glucose fluctuation bone strength mapping data can map the time window characteristics (such as mean, coefficient of variation, and area under the curve) of continuous blood glucose data to the material properties of each element in the bone model, generating a spatiotemporally corresponding bone strength distribution. Furthermore, the blood glucose fluctuation bone strength mapping data can be used as input to a diabetic bone abnormality identification model and provide basic data for subsequent fragility increment calculations. Performing blood glucose fluctuation bone strength mapping processing on a multi-dimensional three-dimensional model of bone tissue microstructure based on continuous blood glucose data from diabetic patients can involve spatially and temporally coupling the temporal statistical characteristics of continuous blood glucose data with each voxel or finite element element in the bone model, endowing it with dynamic bone strength properties. Furthermore, this mapping process can use a sliding time window to calculate blood glucose variability indices and linearly map them to local bone Young's modulus, or simulate the accumulation rate of AGEs in the bone matrix based on a pharmacokinetic model and indirectly derive the strength decay function. This allows the establishment of a time-dimensional blood glucose-bone strength mapping relationship, reflecting the cumulative and instantaneous dual effects of blood glucose fluctuations on bone mechanical properties.

[0099] Step S23: Use the preset diabetic bone abnormality change recognition model to identify bone microstructure abnormalities in diabetic patients by analyzing blood glucose fluctuation bone strength mapping data, and obtain bone microstructure abnormality data of diabetic patients.

[0100] The pre-defined diabetic bone abnormality identification model can be a pre-trained machine learning or rule-driven model used to automatically identify diabetes-specific bone microstructural abnormalities. It can be used to accurately locate the trabecular bone regions most significantly affected by pathological factors such as chronic hyperglycemia and AGEs deposition, improving the targeting of subsequent fragility analysis. In an exemplary embodiment, the pre-defined diabetic bone abnormality identification model can be a classification or segmentation model constructed through supervised learning based on bone microstructural images and clinical outcome data from a cohort of diabetic patients. For example, the pre-defined diabetic bone abnormality identification model can include, but is not limited to, a U-Net-based bone abnormality segmentation model, a graph neural network-based bone connectivity abnormality detection model, and a clustering-based bone microstructural degradation pattern recognition model. The diabetic patient bone microstructural abnormality data can be data output after processing by the diabetic bone abnormality identification model, identifying spatial regions in bone tissue with diabetes-related microstructural degeneration and their degree of abnormality. This data can be used to provide spatial weights or regional constraints for calculating trabecular fragility increments, focusing on pathologically active areas for biomechanical evolution analysis. In one specific embodiment, bone microstructural abnormality data of diabetic patients can be used by an identification model to extract and discriminate features from blood glucose fluctuation-based bone strength mapping data, outputting abnormal region masks and confidence scores. For example, bone microstructural abnormality data of diabetic patients may include, but is not limited to, markers of trabecular sparsity regions, probability maps of high-incidence microcrack areas, and heat maps of bone-matrix interface degeneration. Using a pre-set diabetic bone abnormality alteration identification model to identify bone microstructural abnormalities in blood glucose fluctuation-based bone strength mapping data can involve using the mapped spatial distribution of bone strength as input and identifying abnormal patterns consistent with diabetic bone disease characteristics through a pre-trained model. Furthermore, this identification can be achieved through pixel-level abnormality segmentation of the three-dimensional bone strength map using a convolutional neural network, or node-level classification of trabecular network topological abnormalities based on graph embedding methods, thereby automatically distinguishing diabetic-specific bone microstructural degeneration from bone quality decline caused by other reasons, enhancing pathological specificity.

[0101] Step S24: Calculate the trabecular fragility increment based on the bone microstructure abnormality data of diabetic patients and the bone strength mapping data of blood glucose fluctuations to obtain the trabecular fragility increment data;

[0102] Based on bone microstructural abnormalities in diabetic patients, the calculation of trabecular fragility increments using blood glucose fluctuation-mapped bone strength data can be performed. This can be done by calculating the fragility index increment corresponding to the difference between current and historical bone strength within identified abnormal regions. Furthermore, this calculation can be performed only within the regions covered by the abnormal mask, calculating the temporal increment of the inverse of the Weibull modulus, or assigning higher weights to abnormal regions and using a weighted average to calculate the overall fragility increment. This allows for focusing on key pathological regions to quantify the fragility evolution rate and avoids signal dilution caused by averaging across the entire bone.

[0103] Step S25: Based on the blood glucose fluctuation bone strength mapping data, perform brittle mechanical signal pre-gradient feedback identification on the trabecular brittleness increment data to obtain brittle mechanical pre-gradient feedback data.

[0104] Based on blood glucose fluctuation bone strength mapping data, pre-gradient feedback identification of brittle mechanical signals in trabecular bone fragility increment data can be performed. This can be achieved by combining the spatiotemporal background provided by the blood glucose fluctuation bone strength mapping data with gradient field analysis of the brittle increment data to identify precursor signals. Furthermore, this identification can calculate the spatial gradient of the brittle increment within a specific delay window after the blood glucose peak, or construct a blood glucose-brittleness response transfer function to identify anomalous feedback regions with leading phases. This allows for the identification of the spatial propagation trend of accelerated brittleness evolution driven by blood glucose fluctuations, improving early warning capabilities.

[0105] Taking the follow-up assessment at a diabetes specialist outpatient clinic as an example, the method for constructing a risk prediction model for diabetic osteoporosis fractures in this embodiment can be as follows: A female patient with type 2 diabetes for 10 years wears a CGM device for two weeks, and her data is integrated into the bone metabolism-related signs record; the system extracts continuous blood glucose data from it, combines it with the lumbar spine QCT images to construct a three-dimensional model of bone microstructure, and performs blood glucose fluctuation bone strength mapping; subsequently, the preset diabetic bone abnormality change recognition model identifies a significant microstructural degeneration area below the vertebral endplate; the system calculates the fragility increment based on the bone strength mapping data in this area and finds that the fragility increase rate has reached a threshold in the past month; further gradient feedback analysis shows that the fragility in this area is spreading to the adjacent pedicle; the system determines this as a high fracture risk and triggers clinical intervention recommendations.

[0106] In one embodiment, step S24 includes the following steps:

[0107] Step S241: Calculate the porosity of trabecular bone structure based on the abnormal bone microstructure data of diabetic patients to obtain trabecular bone porosity data.

[0108] The trabecular porosity data can be the proportion of trabecular void volume to the total volume and its spatial distribution characteristics calculated within the abnormal bone microstructure region of diabetic patients. This data can be used to quantify the degree of trabecular sparsity caused by diabetes, providing structural parameters for modeling mechanical performance degradation. In this embodiment, the trabecular porosity data can be obtained by calculating local porosity based on the regions identified by the abnormal bone microstructure data of diabetic patients using voxel counting or morphological analysis. Calculating the trabecular structure porosity of the abnormal bone microstructure data of diabetic patients can be achieved by statistically analyzing the proportion of non-bone voxels or estimating the effective porosity using morphological opening operations within the abnormal regions output by the diabetic bone abnormality change identification model. Furthermore, this operation can be achieved by calculating a three-dimensional porosity heatmap using a local sliding window or by calculating a joint index of topological porosity and geometric porosity based on the skeletonization results, thereby converting qualitative abnormality markers into quantitative structural parameters to support subsequent mechanical modeling.

[0109] Step S242: Simulate bone elasticity decay based on trabecular porosity data and blood glucose fluctuation bone strength mapping data to obtain bone elasticity decay simulation data.

[0110] The bone elasticity decay simulation data can be obtained by combining trabecular porosity data and blood glucose fluctuation bone strength mapping data, and simulating the spatial decay distribution of bone tissue elastic modulus or stiffness through a physical model. This data can be used to reflect the stiffness degradation of bone materials under the combined effects of hyperglycemia and microstructural damage, improving the realism of mechanical simulation. In an exemplary embodiment, the bone elasticity decay simulation data can be generated by using porosity as an input parameter, substituting it into empirical formulas or finite element material assignment rules, and combining it with blood glucose-induced matrix degradation factors. For example, the bone elasticity decay simulation data can include, but is not limited to, Young's modulus spatial decay map, shear modulus degradation field, Poisson's ratio anomaly distribution data, etc. Simulating bone elasticity decay based on trabecular porosity data and blood glucose fluctuation bone strength mapping data can be achieved by introducing porosity as a spatial variable into the elastic modulus decay function and superimposing the strength baseline obtained from blood glucose fluctuation mapping to generate a comprehensive mechanical property field. Furthermore, this operation can be achieved by mapping porosity to elastic modulus using a power-law relationship and multiplying it by a blood glucose-related decay coefficient, or by assigning different material cards to each element according to porosity in finite element software. This allows for the integration of both structural sparsity and biochemical degradation factors, resulting in more realistic bone material performance modeling.

[0111] Step S243: Perform stress increment direction decomposition calculation on the bone elasticity attenuation simulation data to obtain bone stress increment direction decomposition data.

[0112] The bone stress increment direction decomposition data can be obtained by performing stress increment analysis on a model after bone elasticity decay under simulated loading conditions, and decomposing the stress components according to principal directions (such as axial, transverse, and shear). This data can be used to reveal the local fragility of bone trabeculae under different mechanical paths and identify the principal stress directions that are prone to microcrack propagation. In a specific embodiment, the bone stress increment direction decomposition data can be obtained by applying a standard load step in a virtual mechanical simulation, calculating the stress increment tensor of each element, and obtaining the directional components through eigenvalue decomposition or coordinate projection. The stress increment direction decomposition calculation of the bone elasticity decay simulation data can be performed by running a finite element simulation under a standard load (such as axial compression), extracting the stress increment tensor, and decomposing it into principal direction components or coordinate system projection components. Furthermore, this operation can be achieved by performing eigenvalue decomposition on the stress increment tensor of each element to extract the direction and amplitude of the maximum principal stress, or by decomposing the global load into compression, bending, and torsional components and calculating the corresponding stress responses, thereby obtaining direction-sensitive mechanical response characteristics and identifying potential paths for microstructural failure.

[0113] Step S244: Calculate the trabecular brittleness increment based on the decomposition data of bone stress increment direction to obtain the trabecular brittleness increment data.

[0114] The calculation of trabecular brittleness increments based on the decomposition data of bone stress increment directions can be achieved by combining the stress increment amplitudes in each direction with historical baselines to calculate the direction-weighted changes in brittleness indices. Furthermore, this operation can be implemented by fusing multiple brittleness increment components according to the principal stress direction weights to generate a comprehensive brittleness evolution rate, or by preferentially triggering brittleness threshold determination logic in high shear stress directions. This allows brittleness assessment to consider mechanical path dependence, improving predictive adaptability to complex load scenarios such as actual falls.

[0115] For example, in the context of studying the mechanical mechanisms of bone microstructure, the method for constructing a risk prediction model for diabetic osteoporosis fractures in this embodiment can be as follows: Researchers analyze the lumbar QCT images of a diabetic patient. First, a diabetic bone abnormality change identification model is used to locate microstructural abnormalities in the anterior and middle part of the L3 vertebral body. Then, it is calculated that the trabecular porosity in this area is 45%, which is significantly higher than the surrounding area. Based on this, the system performs elastic decay simulation on the bone strength mapping data of blood glucose fluctuations, showing that the Young's modulus in this area decreases by 60%. Under the simulated standing-fall combined load, the stress increment direction decomposition shows that the maximum shear stress is concentrated at the horizontal trabecular connection. Finally, the brittleness increment calculated based on this directional stress increment indicates that this area is in a rapid brittleness stage, and although there is no fracture, it already has high-risk mechanical characteristics.

[0116] In one embodiment, the stress increment direction decomposition calculation of bone elasticity attenuation simulation data includes the following steps:

[0117] The bone elasticity and mechanical decay simulation data were used to assess the decay of the trabecular load-bearing strength, and the bone load-bearing strength decay data were obtained.

[0118] The bone load-bearing strength attenuation data can be quantitative data that further assesses the degree of decline in the local load-bearing capacity of trabeculae under physiological loads, based on bone elasticity attenuation simulation. This data can reflect the degradation of trabecular bone's resistance to failure under actual stress caused by diabetes, providing strength boundary conditions for load simulation. In an exemplary embodiment, the bone load-bearing strength attenuation data can be calculated based on the spatial distribution of bone elastic modulus, combined with yield criteria (such as VonMises or Drucker-Prager), to determine the load-bearing strength threshold for each region, and compared with a healthy baseline to obtain the attenuation ratio. Assessing the trabecular load-bearing strength attenuation based on bone elasticity attenuation simulation data can be done by applying material failure criteria to calculate the local load-bearing strength based on the attenuated elastic modulus field, and comparing it with a healthy reference value to determine the degree of attenuation. For example, this operation can be achieved by calculating the equivalent strength attenuation rate using the VonMises yield criterion, or by directly mapping the load-bearing strength based on an experimentally fitted bone strength-porosity-AGEs ternary relationship model, thereby converting material stiffness degradation into structural load-bearing capacity loss and providing physical constraints for subsequent load simulations.

[0119] Based on the bone load-bearing strength attenuation data, bone stress load strength was simulated to obtain bone stress load strength data;

[0120] The bone stress load intensity data can be the equivalent load intensity distribution at various points within bone tissue under simulated physiological loads (such as standing, walking, and fall impact). This data can be used to reconstruct the stress state of bone microstructures under realistic mechanical conditions and identify high-risk loading areas. In one specific embodiment, the bone stress load intensity data can use bone load-bearing strength decay data as a material failure threshold. Standard biomechanical loads are applied in the finite element model to calculate the equivalent stress or load intensity on each element. Simulating bone stress load intensity based on bone load-bearing strength decay data can be achieved by applying standardized physiological loads (such as axial compression of 1.5 times body weight) in the finite element model, using the load-bearing strength decay data as the upper limit of the material load, and calculating the actual stress distribution. For example, this operation can be achieved by simulating the hip load path when standing on one leg, or by applying transient impact loads to simulate a fall event, thereby generating an internal load response that closely resembles a real biomechanical scenario and avoiding biases from idealized assumptions.

[0121] Global stress tensor calculation was performed on the bone stress load strength data to obtain global stress tensor data;

[0122] The global stress tensor data can be a complete second-order stress tensor representation at each voxel or finite element node in three-dimensional space, containing three normal stress components and three shear stress components. It can be used to provide a complete description of the local mechanical state, supporting principal direction analysis and failure mode identification. In this embodiment, the global stress tensor data can be used to solve the mechanical field of bone load intensity data using a finite element solver, outputting the stress state in full tensor form. Calculating the global stress tensor of bone load intensity data can be achieved by outputting a complete 6-component stress tensor for each computational unit using a finite element solver. For example, this operation can be implemented using nonlinear statics solutions with finite element analysis software, or by using a GPU-accelerated lattice Boltzmann method for large-scale parallel stress field calculations, thereby obtaining a complete mechanical state description that can be used for directional decomposition and failure analysis.

[0123] Stress spatial distribution fluctuation analysis is performed based on global stress tensor data to obtain stress spatial distribution fluctuation data;

[0124] The stress spatial distribution fluctuation data can be the gradient, abrupt change, or non-uniformity characteristics of the global stress tensor in the spatial domain, reflecting stress concentration and transmission anomalies. It can be used to identify stress transmission interruptions or localized concentration areas caused by microstructural defects, which are precursors to microcrack initiation. In a specific embodiment, the stress spatial distribution fluctuation data can be used to perform spatial derivative calculations or local variance analysis on the global stress tensor data to extract spatial fluctuation indices of the stress field. Stress spatial distribution fluctuation analysis based on global stress tensor data can involve calculating the spatial gradient, standard deviation, or Laplacian operator response of the stress tensor in the neighborhood to quantify its spatial non-uniformity. For example, this operation can be achieved by using the Sobel or Scharr operator to calculate the three-dimensional stress gradient amplitude, or by calculating the local stress variation coefficient through a sliding window, thereby identifying regions of abnormal stress transmission and locating potential micro-damage initiation points.

[0125] The stress potential energy relationship is quantified based on Hooke's law to obtain stress space potential energy relationship data.

[0126] The stress-space potential energy relationship data can be derived from Hooke's law by converting stress spatial distribution fluctuations into strain energy density per unit volume and its spatial distribution relationship. This data can be used to characterize changes in the ability of bone tissue to store deformation energy from an energy perspective, revealing the thermodynamic mechanism of increased brittleness. In this embodiment, the stress-space potential energy relationship data can utilize linear elastic constitutive relations to convert the stress field into a strain energy density field and analyze its coupling relationship with stress fluctuations. Quantifying the stress-space distribution fluctuation data according to Hooke's law can be achieved by using linear elastic constitutive relations to convert the stress field into a strain energy density field and analyzing its spatial coupling mode with stress fluctuations. For example, this operation can be achieved by directly integrating the stress-strain product to obtain the local potential energy, or by constructing a regression mapping function between the stress fluctuation amplitude and the potential energy density. This allows for the introduction of an energy perspective to explain the physical nature of increased bone brittleness, enhancing the physical interpretability of the characteristics.

[0127] Based on the stress-space potential energy relationship data and stress-space distribution fluctuation data, stress increment direction decomposition calculation is performed to obtain bone stress increment direction decomposition data.

[0128] The stress increment direction decomposition calculation is performed based on stress-space potential energy relationship data and stress-space distribution fluctuation data. This can be achieved by fusing the potential energy gradient direction and the principal axis of stress fluctuation to perform directional decomposition of stress changes under incremental load. In an exemplary embodiment, this operation can be achieved by using the potential energy gradient as a weight to perform weighted decomposition of the principal stress directions, or by preferentially extracting the direction of maximum shear stress increment in the high potential energy gradient region. This allows for the acquisition of direction-sensitive stress increment characteristics that combine energy-driven and structural response features, thereby improving the accuracy of brittleness prediction.

[0129] Taking the study of bone mechanics precursor signals as an example, the method for constructing a risk prediction model for diabetic osteoporosis fractures in this embodiment can be as follows: Researchers analyze the femoral neck QCT model of a diabetic patient: First, based on the simulation data of bone elasticity decay, it is assessed that the trabecular load-bearing strength decays by 40%; then, a single-leg standing load is simulated to obtain bone stress load strength data; the finite element solution outputs the global stress tensor, showing that there is a high shear stress zone in the lateral subcortical region; further fluctuation analysis reveals a sudden change in the stress gradient in this region; the strain energy density at this location is significantly reduced but the gradient increases sharply, calculated using Hooke's law; finally, the system integrates the potential energy relationship and stress fluctuation, decomposing the shear stress increment along the horizontal trabecular direction as the dominant failure mode, indicating that this region is a high-risk microcrack initiation point, even though the bone marrow diameter (BMD) is normal.

[0130] In one embodiment, step S25 includes the following steps:

[0131] Step S251: Perform time-domain feature analysis on the blood glucose fluctuation bone strength mapping data to obtain time-domain data of blood glucose fluctuation bone strength;

[0132] The time-domain data of bone strength during blood glucose fluctuations can be a dynamic feature sequence reflecting the change of bone strength over time, including temporal attributes such as fluctuation patterns, response delays, and recovery rates. This data can be used to characterize the dynamic response of bone strength to blood glucose fluctuations, providing a temporal input basis for multi-factor correlation analysis. In this embodiment, the time-domain data of bone strength during blood glucose fluctuations can be obtained by performing sliding window statistics, spectral analysis, or state segmentation on the time-dimensional mapping data of bone strength during blood glucose fluctuations, extracting features such as peak response time, strength decay rate, and oscillation period. Temporal feature analysis of the mapping data of bone strength during blood glucose fluctuations can involve feature extraction on the time-dimensional mapping data, including statistical calculation, frequency domain transformation, or state segmentation. Furthermore, this operation can be achieved by using wavelet transform to extract multi-scale time-frequency features of bone strength response, or by using a hidden Markov model to identify bone strength state transition points, thereby transforming static mapping data into a dynamic response representation and revealing the temporal dependence of bone strength on blood glucose fluctuations.

[0133] Step S252: Perform multifactorial correlation analysis on the trabecular fragility increment data based on the time-domain data of blood glucose fluctuation bone strength to obtain multifactorial correlation data of fragility;

[0134] Multifactorial fragility association data can be datasets describing the statistical or mechanistic associations between trabecular bone fragility increment and various diabetes-related pathological factors (such as blood glucose variability, AGEs levels, insulin sensitivity, etc.). This data can be used to reveal the multifactorial driving mechanism of fragility evolution, avoiding biases caused by single-factor modeling. In an exemplary embodiment, multifactorial fragility association data can be constructed based on time-domain data of blood glucose fluctuation and bone strength, combined with clinical metabolic indicators, using multiple regression, causal inference, or graphical models to establish the joint dependence of fragility on multiple factors. Multifactorial fragility association analysis of trabecular bone fragility increment data based on time-domain data of blood glucose fluctuation and bone strength can involve multivariate modeling of time-domain features and fragility increment, quantifying the contribution weights and interaction effects of each pathological factor on fragility evolution. In a specific embodiment, this operation can be achieved by constructing a structural equation model to analyze the path coefficients between blood glucose, AGEs, and fragility, or by using SHAP values ​​to decompose the marginal contributions of each time-domain feature to fragility increment in the XGBoost model. This allows for the establishment of a multifactorial explanatory framework for fragility evolution, improving the interpretability of the mechanism and the robustness of prediction.

[0135] Step S253: Perform multi-factor cluster analysis on the brittle multi-factor association data to obtain brittle multi-factor association cluster data;

[0136] The fragility multifactor association clustering data can be grouped to represent different fragility evolution patterns, and can be used to achieve heterogeneous subtyping of diabetic osteoporosis, supporting personalized risk assessment and intervention strategy development. In this embodiment, the fragility multifactor association clustering data can be clustered using K-means, spectral clustering, or Gaussian mixture models to identify patient or bone region categories with similar pathomechanical response patterns.

[0137] Multifactor clustering analysis of fragile multifactor association data can be performed by unsupervised clustering of multidimensional association feature vectors to classify subgroups with similar fragile driving mechanisms. Furthermore, this operation can be achieved by using t-SNE dimensionality reduction followed by DBSCAN clustering to discover non-spherical clusters, or by combining clinical outcome labels with semi-supervised clustering to optimize subtype discrimination boundaries. This can identify potential clinical subtypes of diabetic osteoporosis and support precise stratified management.

[0138] Step S254: Based on the brittle multi-factor association clustering data, perform dynamic brittle response simulation on the incremental brittle data of bone trabeculae to obtain multi-factor dynamic brittle response data;

[0139] The multi-factor dynamic brittle response data can be the dynamic process data of the evolution of trabecular brittleness over time under different pathological combinations, simulated based on the results of multi-factor brittleness association clustering. This data can be used to generate pathology-specific brittleness evolution trajectories to identify high-risk dynamic patterns. In an exemplary embodiment, the multi-factor dynamic brittle response data can be introduced into a finite element model with the corresponding multi-factor parameter combinations for clustering, performing time-progressive mechanical simulations, and outputting dynamic trajectories of brittleness indices (such as fracture energy and crack propagation rate). Dynamic brittleness response simulation of incremental trabecular brittleness data based on multi-factor brittleness association clustering data can be performed by setting corresponding pathological parameter combinations for each cluster subtype, conducting time-series simulations in a biomechanical model, and generating brittleness dynamic evolution trajectories. In a specific embodiment, this operation can be achieved by embedding a user-defined material subroutine in Abaqus to simulate AGEs-dependent stiffness degradation, or by using a surrogate model to accelerate multi-scenario brittleness evolution simulations, thereby generating subtype-specific brittleness evolution predictions and enhancing the model's adaptability to individual differences.

[0140] Step S255: Based on the multi-factor dynamic brittleness response data, perform stress response pre-gradient trend identification on the blood glucose fluctuation bone strength mapping data to obtain stress response pre-gradient trend data;

[0141] Among them, the stress response pre-fracture gradient trend data can be the abnormal gradient change trend data of the local stress field of bone tissue before fracture occurs under the driving force of multi-factor dynamic brittle response. It can be used to capture early signals of mechanical instability that have not yet formed structural damage before fracture. In this embodiment, the stress response pre-fracture gradient trend data can be used to update the material properties of the bone microstructure model based on multi-factor dynamic brittle response data, calculate the spatiotemporal gradient of the stress field after applying a standard load, and identify the premature stress concentration or distribution distortion trend.

[0142] Based on multi-factor dynamic fragility response data, stress response gradient trend identification can be performed on blood glucose fluctuation bone strength mapping data. This involves feeding the dynamic fragility response results back to the bone microstructure model, recalculating the stress field, and analyzing the abnormal evolution trend of its gradient over time. Furthermore, this operation can identify abrupt inflection points by calculating the time derivative of the principal stress gradient in the anterior column region of the L2 vertebral body, or by constructing a stress gradient propagation velocity field to detect abnormally accelerated regions. This allows for the identification of early disturbance signals in the stress field before macroscopic failure, improving the early warning lead time.

[0143] Step S256: Based on the stress response pre-gradient trend data and multi-factor dynamic brittle response data, perform brittle mechanical signal pre-gradient feedback identification to obtain brittle mechanical pre-gradient feedback data.

[0144] Brittle mechanical signal precursor feedback identification can be performed based on stress response precursor gradient trend data and multi-factor dynamic brittle response data. This can be achieved by fusing stress gradient trends and brittle dynamic response data, and identifying mechanical precursor signals with high predictive value through multimodal feature fusion. In a specific embodiment, this operation can be achieved by using an attention mechanism to weightedly fuse stress gradient and brittle trajectory features, or by constructing a two-stream neural network to process the two types of data separately and then performing decision-level fusion. This can achieve high-confidence identification of brittle mechanical precursor signals and reduce false positive warnings.

[0145] For example, in the scenario of early intervention guided by diabetic bone disease subtypes, the method for constructing a risk prediction model for diabetic osteoporosis fractures in this embodiment can be as follows: After continuous blood glucose monitoring and QCT scan, the system extracts the time-domain data of bone strength under blood glucose fluctuations from a 68-year-old male diabetic patient. It is found that his bone strength shows a significant delayed recovery after postprandial hyperglycemia; multivariate association analysis shows that his fragility increment is mainly driven by AGEs and blood glucose variability; cluster analysis classifies him into the "high AGEs-high volatility" subtype; dynamic simulation shows that his hip trabecular fragility will enter an accelerated increase phase within 6 months; stress gradient trend identifies a stress concentration front on the medial side of the femoral neck; based on this, the system determines it to be high risk and recommends AGEs inhibitors combined with anti-bone resorption therapy, rather than just calcium supplementation.

[0146] In one embodiment, step S3 includes the following steps:

[0147] Step S31: Perform phase shift analysis on the brittle mechanics pre-gradient feedback data to obtain the brittle feedback bone stress signal phase shift data;

[0148] The brittle feedback bone stress signal phase shift data can be a set of signals reflecting the temporal differences in local stress transmission in bone tissue under load, obtained by analyzing the phase response characteristics of the brittle mechanical pre-gradient feedback data. This data is used to characterize the hysteresis characteristics of the mechanical response of bone microstructure in diabetic states, providing the original basis for subsequent spatial interpolation and compensation.

[0149] Step S32: Perform spatial interpolation on the phase shift data of the brittle feedback bone stress signal to obtain stress shift spatial interpolation data;

[0150] The stress offset spatial interpolation data can be generated by interpolating the phase offset data of brittle feedback bone stress signal in the spatial domain. This data possesses higher spatial resolution and continuity, and can be used to compensate for stress field discontinuities caused by image resolution limitations or finite element mesh coarsening, thereby improving the spatial integrity of mechanical features. In this embodiment, the stress offset spatial interpolation data can be based on the spatial distribution of the original phase offset data. In missing or sparse regions, an interpolation algorithm is used to calculate the phase offset value of intermediate points, forming a dense mesh.

[0151] Spatial interpolation of the phase shift data of brittle feedback bone stress signals can be performed on discretely distributed phase shift node data in a 3D bone model coordinate system to generate a high-density phase field that continuously covers the entire region of interest. Furthermore, spatial interpolation of the brittle feedback bone stress signal phase shift data can be achieved by using radial basis functions for global interpolation on an unstructured mesh, or by using an inverse distance weighting method for fast interpolation in a local neighborhood. This enhances the spatial continuity and detail representation of the stress phase signal, mitigating feature loss caused by insufficient modeling resolution.

[0152] Step S33: Perform signal compensation on the phase shift data of the brittle feedback bone stress signal based on the stress offset spatial interpolation data to obtain the brittle feedback signal phase shift compensation data;

[0153] The brittle feedback signal phase shift compensation data can be an optimized set of phase shift signals obtained by correcting the original brittle feedback bone stress signal phase shift data using stress shift spatial interpolation data. This optimized set eliminates local distortion and boundary effects and can be used to improve the physical consistency and noise resistance of the phase shift features, providing more reliable input for machine learning models. In an exemplary embodiment, the brittle feedback signal phase shift compensation data can use interpolation data as a reference benchmark to perform weighted correction or residual compensation on low-confidence regions (such as edges and near pores) in the original phase shift data. Compensating the brittle feedback bone stress signal phase shift data based on stress shift spatial interpolation data can involve comparing the original phase shift data with the interpolation data, identifying deviation regions, and generating a compensated signal through weighted fusion or residual correction. In one specific embodiment, signal compensation for the phase shift data of brittle feedback bone stress signal based on stress shift spatial interpolation data can be achieved by linear weighted fusion of interpolation data as the main component and original data as the auxiliary component in the low confidence region, or by calculating the residual field between the original and interpolation data and superimposing the smoothed residual onto the original data to achieve compensation. This corrects the phase distortion caused by finite element mesh coarseness, simplified boundary conditions, or numerical errors, thereby improving feature fidelity.

[0154] Step S34: Use the XGBoost algorithm to construct a risk prediction model for diabetic osteoporosis fractures based on the phase shift compensation data of the brittle feedback signal, and obtain the risk prediction model for diabetic osteoporosis fractures.

[0155] For example, in the scenario of early intervention for high-risk diabetic patients, the method for constructing a fracture risk prediction model for diabetic osteoporosis in this embodiment can be as follows: A QCT scan of a 70-year-old subject shows that the bone mineral density of the lumbar spine is at the lower limit of normal. However, after the system performs blood glucose-strength mapping on the trabecular bone model, it finds that the fragility of the anterior column of the L1 vertebra is significantly increased. Further phase shift analysis shows that there is obvious stress response hysteresis in this area. However, due to the coarse finite element mesh, some microstructural areas have no effective signal. The system then performs Kriging space interpolation on the phase shift data to generate a continuous stress shift field, and performs boundary compensation on the original signal accordingly. The compensated phase data is input into the XGBoost model, which outputs an increased fracture risk warning, indicating that there may be an increased risk of fracture.

[0156] In one embodiment, step S32 includes the following steps:

[0157] Step S321: Calculate the phase shift of the brittle bone stress signal phase shift data to obtain the brittle bone stress signal phase shift data;

[0158] The brittle bone stress signal phase offset data can be a set of quantified values ​​extracted from the brittle feedback bone stress signal phase offset data, characterizing the lag or advance of the stress response at each spatial location relative to the reference phase. This data can be used to convert the original phase offset signal into a computable scalar field, providing a basis for subsequent nonlinear incremental analysis. In an exemplary embodiment, the brittle bone stress signal phase offset data can be obtained by calculating the phase angle difference between the stress time-series signal at each finite element node or voxel and the healthy reference signal. Calculating the phase offset of the brittle feedback bone stress signal phase offset data can be achieved by comparing the stress time-series signal at each spatial point with the reference phase of standard healthy bone and calculating the phase angle difference. Exemplarily, this operation can be implemented by extracting the instantaneous phase using Hilbert transform and then calculating the offset, or by calculating the steady-state offset using the FFT main peak phase difference in the frequency domain. This transforms the original complex signal form of the phase offset into a quantifiable scalar field, facilitating subsequent mathematical operations.

[0159] Step S322: Perform nonlinear incremental calculation on the phase shift data of the brittle bone stress signal based on the phase shift data of the brittle bone stress signal to obtain the phase shift nonlinear incremental data;

[0160] The phase shift nonlinear incremental data can be calculated based on phase shift data, reflecting the nonlinear change of phase shift over time or pathological progression. It can be used to reveal the nonlinear evolution of abnormal stress response during diabetic bone microstructural degeneration, avoiding misjudgments of complex mechanical behaviors by linear models. In a specific embodiment, the phase shift nonlinear incremental data can employ nonlinear differential operators or higher-order difference methods to capture the acceleration or abrupt change trend of phase shift in local regions, rather than simple linear differences.

[0161] Nonlinear increment calculations are performed on the phase shift data of brittle bone stress signals based on the phase shift data of brittle bone stress signals. This can be done by applying nonlinear difference or differential operators to the phase shift over time or in the dimension of pathological progression to calculate the nonlinear characteristics of its rate of change. In an exemplary embodiment, this operation can be achieved by approximating the nonlinear acceleration term using a second-order central difference or by using a local polynomial fit followed by differentiation to estimate the nonlinear increment. This allows for the capture of abrupt changes in stress response or accelerated deterioration caused by AGEs deposition, microcrack aggregation, etc.

[0162] Step S323: Perform spatial incremental trend analysis on the phase shift data of the brittle feedback bone stress signal based on the phase shift nonlinear incremental data to obtain the spatial incremental trend data of the phase shift;

[0163] The phase shift spatial increment trend data can describe the spatial distribution directionality and gradient change pattern of the phase shift nonlinear increment in the three-dimensional bone structure. It can be used to identify the spatial propagation path of brittle accumulation in the trabecular bone network, providing physical guidance priors for interpolation. In this embodiment, the phase shift spatial increment trend data can be used to perform spatial gradient tensor analysis or manifold learning on the phase shift nonlinear increment data to extract the dominant propagation direction and hotspot regions. Based on the phase shift nonlinear increment data, spatial increment trend analysis of the brittle feedback bone stress signal phase shift data can be performed on the nonlinear increment field using spatial gradient, directional derivative, or principal component analysis to identify its dominant change direction and aggregation pattern in three-dimensional space. Furthermore, this operation can be achieved by calculating the gradient tensor of the nonlinear increment field and extracting the largest eigenvector as the propagation direction, or by using DBSCAN clustering to identify high increment regions and fitting their spatial distribution trend surface. This reveals the spatial propagation law of brittle evolution in the bone microstructure, providing structural constraints for interpolation.

[0164] Step S324: Based on the phase offset spatial increment trend data, perform spatial interpolation processing on the phase offset data of the brittle feedback bone stress signal to obtain stress offset spatial interpolation data.

[0165] Spatial interpolation of phase shift data of brittle feedback bone stress signals based on spatial incremental trend data of phase shift can be performed by using the spatial incremental trend as the directionality or weight prior of the interpolation algorithm to guide the generation of phase values ​​that conform to pathomechanical laws in sparse regions. In a specific embodiment, this operation can be achieved by introducing the trend direction as an anisotropic kernel parameter in radial basis function interpolation or by using trend-guided Kriging interpolation, incorporating the incremental gradient as a drift term into the model. This improves the physical rationality and pathological sensitivity of the interpolation results and avoids mechanical distortion caused by unconstrained interpolation.

[0166] Taking early monitoring of bone quality deterioration in diabetes as an example, the method for constructing a fracture risk prediction model for diabetic osteoporosis in this embodiment can be as follows: A 62-year-old type 2 diabetic patient's QCT scan showed normal bone mineral density, but after the system calculated the phase shift of his trabecular bone model, it was found that there was a significant lag on the right side of the L3 vertebral body; further nonlinear incremental analysis showed that the phase shift in this region increased exponentially, and spatial trend analysis revealed that it spread along the principal axis of the trabecular bone; based on this, the system performed trend-guided interpolation in the low-resolution region to reconstruct a continuous high-fidelity potential field; the compensated data was input into the XGBoost model, indicating an increased fracture risk earlier than the changes in traditional BMD indicators.

[0167] In one embodiment, step S34 includes the following steps:

[0168] Step S341: Divide the brittle feedback signal phase offset compensation data into a dataset to obtain a brittle feedback signal phase offset compensation test set and a brittle feedback signal phase offset compensation training set.

[0169] The fragile feedback signal phase shift compensation training set can be a subset of the fragile feedback signal phase shift compensation data used for model training. It includes samples labeled with fracture outcomes and their corresponding compensated phase shift features. This subset provides a training basis for selecting similar features of the feedback signal and incremental learning, ensuring that the model learning process covers various bone microstructure degeneration patterns. In this embodiment, the fragile feedback signal phase shift compensation training set can extract training samples from the original compensation data at a preset ratio (e.g., 8:2) using random stratified sampling or time-series segmentation. The fragile feedback signal phase shift compensation test set can be an independent subset of the fragile feedback signal phase shift compensation data used for model validation and performance evaluation. This subset can be used to objectively evaluate the generalization ability of the initial model, avoiding inflated performance due to overfitting. For example, the fragile feedback signal phase shift compensation test set can retain samples not used in training during the dataset partitioning process, maintaining their original labels and feature integrity.

[0170] Dataset partitioning for brittle feedback signal phase shift compensation data can be achieved by dividing the compensated complete dataset into mutually exclusive training and testing sets according to a preset ratio or strategy. Furthermore, dataset partitioning for brittle feedback signal phase shift compensation data can be achieved by using hierarchical random partitioning to maintain a consistent ratio of fracture / non-fracture samples, or by partitioning by patient ID to avoid the same patient's data appearing in both the training and testing sets. This ensures the independence of model training and evaluation, preventing performance overestimation due to data leakage.

[0171] Step S342: Select similar features of feedback signals from the training set for phase offset compensation of brittle feedback signals to obtain similar feature data of feedback signals;

[0172] The feedback signal similarity feature data can be a subset of mechanical response features selected from the training set that are highly similar in phase shift spatiotemporal patterns. This subset can be used to focus on biomechanical indicators strongly correlated with the evolution of diabetic bone fragility, reducing feature dimensionality and improving model interpretability. In one specific embodiment, the feedback signal similarity feature data can be used to identify sample features with common degradation paths based on clustering (e.g., K-means, DBSCAN) or similarity measures (e.g., Dynamic Time Warping (DTW), Cosine Similarity). Selecting feedback signal similarity features for the fragility feedback signal phase shift compensation training set can involve calculating the similarity of phase shift signals among training samples and selecting a subset of features with common mechanical response patterns. For example, selecting feedback signal similarity features for the fragility feedback signal phase shift compensation training set can be achieved by using Dynamic Time Warping (DTW) to measure the similarity of temporal phase signals followed by clustering, or by selecting nearest neighbor features based on Euclidean distance in the principal component space. This allows focusing on common biomechanical patterns strongly correlated with fracture risk, improving feature effectiveness and model robustness.

[0173] Step S343: Using the XGBoost algorithm and based on the feedback signal similarity feature data, perform similar structure incremental learning on the brittle feedback signal phase offset compensation training set to obtain similar structure incremental data;

[0174] The incremental data of similar structures can be the learning weight update amount or structural split information accumulated gradually based on the similar feature data of feedback signals during XGBoost training. This can be used to enhance the model's ability to model the progressive path of diabetes-specific bone micro-damage and improve its sensitivity to hidden degradation. In this embodiment, the incremental data of similar structures can be used in each iteration to preferentially expand the tree structure within the similar feature subspace and record the incremental nodes that significantly contribute to the prediction. Using the XGBoost algorithm and performing incremental learning of similar structures on the brittle feedback signal phase offset compensation training set based on the similar feature data of feedback signals, the tree split can be guided to preferentially occur in the similar feature subspace during XGBoost training, accumulating structured learning increments. For example, this operation can be achieved by introducing a similar feature weighting term into the loss function to increase the gradient contribution of common patterns, or by using two-stage training: pre-training on a subset of similar features and then fine-tuning across all features. This can enhance the model's ability to model the degenerative path of diabetes-specific bone microstructures and improve its sensitivity to early mechanical abnormalities.

[0175] Step S344: Construct an initial diabetic osteoporotic fracture risk prediction model based on incremental data of similar structures to obtain the initial diabetic osteoporotic fracture risk prediction model.

[0176] The initial diabetic osteoporosis fracture risk prediction model can be a preliminary XGBoost prediction model built based on incremental data of similar structures, which has not been validated on a test set. This model can serve as a starting point for model optimization and must be evaluated on an independent test set before deployment. In an exemplary embodiment, the initial diabetic osteoporosis fracture risk prediction model can use incremental data of similar structures as training targets to complete the integrated construction of XGBoost base learners. Building the initial diabetic osteoporosis fracture risk prediction model based on incremental data of similar structures can involve integrating all incremental learning results to generate a complete XGBoost ensemble model structure. For example, this operation can be achieved by directly outputting an XGBoostBooster object or exporting a model parameter file for subsequent loading and testing, thereby forming a model instance with preliminary prediction capabilities for subsequent validation.

[0177] Step S345: Test the initial diabetic osteoporosis fracture risk prediction model based on the brittle feedback signal phase offset compensation test set to obtain the diabetic osteoporosis fracture risk prediction model.

[0178] Testing an initial diabetic osteoporotic fracture risk prediction model based on a brittle feedback signal phase shift compensation test set can be achieved by inputting the test set into the initial model and calculating consistency indices (such as AUC and F1-score) between the prediction results and the true labels. Further, this process can be accomplished by calculating the area under the ROC curve and calibration curve to assess discrimination and calibration capabilities, or by performing bootstrap resampling to estimate the performance confidence interval. This allows for an objective evaluation of the model's generalization performance and confirmation of its clinical usability.

[0179] Taking multi-center model validation and deployment as an example, the method for constructing a diabetic osteoporotic fracture risk prediction model in this embodiment can be as follows: the research team collects compensated phase shift data from 200 diabetic patients from three hospitals and divides the training / test sets in a 7:3 ratio; in the training set, three typical fragility evolution patterns are identified through DTW clustering, and similar features are selected accordingly; XGBoost performs incremental learning in this subspace to generate the initial model; the AUC reaches 0.92 on the independent test set, which is significantly better than the traditional FRAX tool; finally, the model is packaged into monitoring firmware and deployed to the regional health management cloud platform.

[0180] In addition, refer to Figure 2 To achieve the above objectives, the present invention also provides a system for constructing a risk prediction model for diabetic osteoporosis fractures, the system comprising:

[0181] Data modeling module 10 is used to collect bone mineral density data and hip joint and spine bone mineral density images of diabetic patients through the medical and health information system, respectively obtaining bone metabolism record data and bone mineral density tomographic images of diabetic patients; and constructing a multi-dimensional three-dimensional model of bone trabeculae based on the bone mineral density tomographic images of diabetic patients to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

[0182] The fragility identification module 20 is used to perform blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data; to calculate the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data to obtain trabecular fragility increment data; and to perform brittle mechanical signal pre-gradient feedback identification on the trabecular fragility increment data based on the blood glucose fluctuation-bone strength mapping data to obtain brittle mechanical pre-gradient feedback data.

[0183] The model building module 30 is used to perform phase shift analysis of bone stress signal on the brittle mechanics pre-gradient feedback data to obtain brittle feedback bone stress signal phase shift data; the XGBoost algorithm is used to build a diabetic osteoporosis fracture risk prediction model on the brittle feedback bone stress signal phase shift data to obtain the diabetic osteoporosis fracture risk prediction model.

[0184] Firmware deployment module 40 is used to design automated monitoring firmware for the diabetic osteoporosis fracture risk prediction model, obtain the diabetic osteoporosis fracture risk prediction model monitoring firmware, and send the diabetic osteoporosis fracture risk prediction model monitoring firmware to the health management cloud platform to perform diabetic osteoporosis fracture risk prediction.

[0185] Other embodiments or specific implementations of the diabetic osteoporosis fracture risk prediction model construction system described in this invention can be referred to the above-mentioned method embodiments, and will not be repeated here.

[0186] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for constructing a risk prediction model for fractures caused by diabetic osteoporosis, characterized in that, The method includes: Step S1: Collect bone mineral density data and hip and spine bone mineral density images of diabetic patients through the medical and health information system to obtain bone metabolism data and bone mineral density tomographic images of diabetic patients; construct a multi-dimensional three-dimensional model of bone trabeculae based on the bone mineral density tomographic images of diabetic patients to obtain a multi-dimensional three-dimensional model of bone tissue microstructure. Step S2: Perform blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data; calculate the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data to obtain trabecular fragility increment data; perform brittle mechanical signal pre-gradient feedback identification on the trabecular fragility increment data based on the blood glucose fluctuation-bone strength mapping data to obtain brittle mechanical pre-gradient feedback data. Step S3: Perform phase shift analysis on the brittle mechanics pre-gradient feedback data to obtain the phase shift data of the brittle feedback bone stress signal; use the XGBoost algorithm to construct a risk prediction model for diabetic osteoporosis fractures based on the phase shift data of the brittle feedback bone stress signal, and obtain the risk prediction model for diabetic osteoporosis fractures. Step S4: Design an automated monitoring firmware for the diabetic osteoporosis fracture risk prediction model to obtain the monitoring firmware for the diabetic osteoporosis fracture risk prediction model. Send the monitoring firmware for the diabetic osteoporosis fracture risk prediction model to the health management cloud platform to perform diabetic osteoporosis fracture risk prediction.

2. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 1, characterized in that, Step S1 includes the following steps: Step S11: Collect bone mineral density data and hip and spine bone mineral density images of diabetic patients through the medical and health information system to obtain basic data of diabetic patients, including bone metabolism data and bone mineral density tomographic images of diabetic patients. Step S12: Optimize the bone tissue noise in the bone density tomography images of diabetic patients to obtain bone density noise-optimized images; Step S13: Perform multi-dimensional reconstruction analysis on the bone density tomographic images of diabetic patients based on the bone density noise optimization images to obtain multi-dimensional reconstruction data of bone density tomographic scans; Step S14: Construct a multi-dimensional three-dimensional model of bone tissue microstructure based on the multi-dimensional reconstruction data of bone density tomography scan to obtain a multi-dimensional three-dimensional model of bone tissue microstructure.

3. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 1, characterized in that, Step S2 includes the following steps: Step S21: Extract continuous blood glucose monitoring data from the bone metabolism signs data of diabetic patients to obtain continuous blood glucose data of diabetic patients; Step S22: Based on the continuous blood glucose data of diabetic patients, perform blood glucose fluctuation-bone strength mapping processing on the multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data. Step S23: Use the preset diabetic bone abnormality change recognition model to identify bone microstructure abnormalities in diabetic patients by analyzing blood glucose fluctuation-bone strength mapping data, and obtain bone microstructure abnormality data of diabetic patients. Step S24: Calculate the trabecular fragility increment based on the blood glucose fluctuation-bone strength mapping data of diabetic patients to obtain the trabecular fragility increment data; Step S25: Based on the blood glucose fluctuation-bone strength mapping data, perform brittle mechanical signal pre-gradient feedback identification on the trabecular brittleness increment data to obtain brittle mechanical pre-gradient feedback data.

4. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 3, characterized in that, Step S24 includes the following steps: Step S241: Calculate the porosity of trabecular bone structure from the abnormal bone microstructure data of diabetic patients to obtain trabecular bone porosity data; Step S242: Simulate bone elasticity decay based on trabecular porosity data and blood glucose fluctuation-bone strength mapping data to obtain bone elasticity decay simulation data; Step S243: Perform stress increment direction decomposition calculation on the bone elasticity attenuation simulation data to obtain bone stress increment direction decomposition data; Step S244: Calculate the trabecular brittleness increment based on the decomposition data of bone stress increment direction to obtain the trabecular brittleness increment data.

5. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 4, characterized in that, The stress increment directional decomposition calculation of the bone elasticity attenuation simulation data includes the following steps: The bone elasticity and mechanical decay simulation data were used to assess the decay of the trabecular load-bearing strength, and the bone load-bearing strength decay data were obtained. Based on the bone load-bearing strength attenuation data, bone stress load strength was simulated to obtain bone stress load strength data; Global stress tensor calculation was performed on the bone stress load strength data to obtain global stress tensor data; Stress spatial distribution fluctuation analysis is performed based on global stress tensor data to obtain stress spatial distribution fluctuation data; The stress potential energy relationship is quantified based on Hooke's law to obtain stress space potential energy relationship data. Based on the stress-space potential energy relationship data and stress-space distribution fluctuation data, stress increment direction decomposition calculation is performed to obtain bone stress increment direction decomposition data.

6. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 3, characterized in that, Step S25 includes the following steps: Step S251: Perform time-domain feature analysis on the blood glucose fluctuation-bone strength mapping data to obtain time-domain data of blood glucose fluctuation-bone strength; Step S252: Perform a multifactorial correlation analysis on the trabecular fragility increment data based on the blood glucose fluctuation-bone strength time domain data to obtain the multifactorial correlation data on fragility. Step S253: Perform multi-factor cluster analysis on the brittle multi-factor association data to obtain brittle multi-factor association cluster data; Step S254: Based on the brittle multi-factor association clustering data, perform dynamic brittle response simulation on the incremental brittle data of bone trabeculae to obtain multi-factor dynamic brittle response data; Step S255: Based on the multi-factor dynamic fragility response data, identify the stress response pre-gradient trend of the blood glucose fluctuation-bone strength mapping data to obtain the stress response pre-gradient trend data. Step S256: Based on the stress response pre-gradient trend data and multi-factor dynamic brittle response data, perform brittle mechanical signal pre-gradient feedback identification to obtain brittle mechanical pre-gradient feedback data.

7. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform phase shift analysis on the brittle mechanics pre-gradient feedback data to obtain the brittle feedback bone stress signal phase shift data; Step S32: Perform spatial interpolation on the phase shift data of the brittle feedback bone stress signal to obtain stress shift spatial interpolation data; Step S33: Perform signal compensation on the phase shift data of the brittle feedback bone stress signal based on the stress offset spatial interpolation data to obtain the brittle feedback signal phase shift compensation data; Step S34: Use the XGBoost algorithm to construct a risk prediction model for diabetic osteoporosis fractures based on the phase shift compensation data of the brittle feedback signal, and obtain the risk prediction model for diabetic osteoporosis fractures.

8. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 7, characterized in that, Step S32 includes the following steps: Step S321: Calculate the phase shift of the brittle bone stress signal phase shift data to obtain the brittle bone stress signal phase shift data; Step S322: Perform nonlinear incremental calculation on the phase shift data of the brittle bone stress signal based on the phase shift data of the brittle bone stress signal to obtain the phase shift nonlinear incremental data; Step S323: Perform spatial incremental trend analysis on the phase shift data of the brittle feedback bone stress signal based on the phase shift nonlinear incremental data to obtain the spatial incremental trend data of the phase shift; Step S324: Based on the phase offset spatial increment trend data, perform spatial interpolation processing on the phase offset data of the brittle feedback bone stress signal to obtain stress offset spatial interpolation data.

9. The method for constructing a risk prediction model for diabetic osteoporosis fractures as described in claim 7, characterized in that, Step S34 includes the following steps: Step S341: Divide the brittle feedback signal phase offset compensation data into a dataset to obtain a brittle feedback signal phase offset compensation test set and a brittle feedback signal phase offset compensation training set. Step S342: Select similar features of feedback signals from the training set for phase offset compensation of brittle feedback signals to obtain similar feature data of feedback signals; Step S343: Using the XGBoost algorithm and based on the feedback signal similarity feature data, perform similar structure incremental learning on the brittle feedback signal phase offset compensation training set to obtain similar structure incremental data; Step S344: Construct an initial diabetic osteoporotic fracture risk prediction model based on incremental data of similar structures to obtain the initial diabetic osteoporotic fracture risk prediction model. Step S345: Test the initial diabetic osteoporosis fracture risk prediction model based on the brittle feedback signal phase offset compensation test set to obtain the diabetic osteoporosis fracture risk prediction model.

10. A system for constructing a model to predict the risk of fractures in diabetic osteoporosis, characterized in that, The system includes: The data modeling module is used to collect bone mineral density data and hip joint and spine bone mineral density images of diabetic patients through the medical and health information system, respectively obtaining bone metabolism record data and bone mineral density tomographic images of diabetic patients; based on the bone mineral density tomographic images of diabetic patients, a multi-dimensional three-dimensional model of bone trabeculae is constructed to obtain a multi-dimensional three-dimensional model of bone tissue microstructure. The fragility identification module is used to perform blood glucose fluctuation-bone strength mapping processing on a multi-dimensional three-dimensional model of bone tissue microstructure to obtain blood glucose fluctuation-bone strength mapping data; to calculate the trabecular fragility increment on the blood glucose fluctuation-bone strength mapping data to obtain trabecular fragility increment data; and to identify the brittle mechanical signal pre-gradient feedback based on the blood glucose fluctuation-bone strength mapping data and the trabecular fragility increment data to obtain brittle mechanical pre-gradient feedback data. The model building module is used to perform phase shift analysis on the brittle mechanics pre-gradient feedback data of bone stress signal to obtain the phase shift data of brittle feedback bone stress signal; the XGBoost algorithm is used to build a risk prediction model for diabetic osteoporosis fracture based on the phase shift data of brittle feedback bone stress signal to obtain the risk prediction model for diabetic osteoporosis fracture. The firmware deployment module is used to design automated monitoring firmware for the diabetic osteoporosis fracture risk prediction model, obtain the monitoring firmware for the diabetic osteoporosis fracture risk prediction model, and send the monitoring firmware for the diabetic osteoporosis fracture risk prediction model to the health management cloud platform to perform diabetic osteoporosis fracture risk prediction.

Images