Centrum stability assessment method and system for senile osteoporosis centrum compression fracture

Abstract

The invention relates to a vertebral stability assessment method and system for senile osteoporosis vertebral compression fracture. The method comprises the following steps: collecting dynamic stress data and cone motion data of a target cone in real time, and calculating a stress concentration coefficient, an energy dissipation rate and a motion coupling degree; performing weighted calculation processing on the stress concentration coefficient, the motion coupling degree and the energy dissipation rate to obtain an initial stability index, and performing neural network verification processing on the initial stability index and the bone mineral density value to obtain a dynamic stability index; and according to a preset grading standard, carrying out grading and strategy matching processing on the dynamic stability index to obtain a vertebral stability grade and a corresponding treatment strategy. According to the method, dynamic stress and motion data are collected in real time, multi-parameter calculation is fused, a dynamic stability index is obtained through neural network verification, precise evaluation and personalized treatment strategy matching of the vertebral stability are achieved, and evaluation comprehensiveness and treatment pertinence are improved.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and in particular relates to a method and system for assessing the stability of vertebral compression fractures in elderly patients with osteoporosis. Background Technology

[0002] Osteoporotic vertebral compression fractures are a common condition in the elderly, and assessing vertebral stability is crucial for treatment selection. Currently, clinical practice primarily relies on static imaging examinations such as X-rays and CT scans to measure vertebral height loss and fracture morphology. However, while these methods provide static information such as vertebral morphology and bone mineral density, they cannot reflect the real-time mechanical stability of the vertebra under dynamic loads, nor can they comprehensively assess the synergistic effect of multi-dimensional mechanical parameters and bone mineral density. For example, X-rays and CT scans can only display fracture morphology and bone mineral density values, but cannot quantify the stress distribution and motion coupling characteristics of the vertebra during positional changes or axial rotation. In recent years, with the rapid development of computer technology, dynamic mechanical assessment techniques have gradually gained attention. Although some studies have attempted to predict vertebral stress distribution through finite element analysis, their calculations rely on static CT data, ignore real-time motion loads, and require complex modeling processes with high radiation doses. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for assessing vertebral stability in osteoporotic vertebral compression fractures of the elderly, in order to overcome the above-mentioned technical problems and achieve accurate classification of stability levels and generation of personalized treatment strategies.

[0004] Firstly, this application provides a method for assessing the stability of vertebral bodies in osteoporotic vertebral compression fractures in the elderly, including:

[0005] Real-time acquisition of dynamic stress data of the target vertebral body to obtain the raw stress dataset, and acquisition of vertebral body motion data to obtain the raw motion dataset;

[0006] The stress concentration factor and energy dissipation rate are calculated based on the original stress dataset, and the motion coupling degree is calculated based on the original motion dataset.

[0007] The stress concentration factor, motion coupling degree, and energy dissipation rate are weighted and calculated to obtain the initial stability index. The initial stability index and bone density value are then verified by a neural network to obtain the dynamic stability index.

[0008] Based on the preset grading criteria, the dynamic stability index is classified and matched with strategies to obtain the vertebral body stability level and the corresponding treatment strategy.

[0009] In one embodiment, dynamic stress data of the target vertebra is acquired in real time to obtain a raw stress dataset, and vertebral motion data is acquired to obtain a raw motion dataset, including:

[0010] The sensor array deployment distribution is obtained by deploying a flexible piezoelectric sensor array in the target vertebral region of the patient's back;

[0011] Based on the optically traceable markers on the skin projection areas corresponding to the target vertebra and adjacent vertebrae, the marker deployment distribution is obtained;

[0012] Acquire patient's motion execution data by instructing the patient to perform standard motion protocols that include positional changes and axial rotation;

[0013] Based on the sensor array deployment distribution and motion execution data, multi-axis dynamic stress waveforms are collected to obtain the raw stress dataset;

[0014] Based on the distribution of marker points and action execution data, three-dimensional motion trajectories are collected to obtain the raw motion dataset.

[0015] In one embodiment, the stress concentration factor and energy dissipation rate are calculated based on the original stress dataset, including:

[0016] The original stress dataset was subjected to stress peak analysis to obtain the stress peak values ​​at the anterior and posterior edges of the target vertebral body.

[0017] Calculate the ratio of the peak stress at the leading edge to the peak stress at the trailing edge to obtain the stress ratio.

[0018] Logarithmic normalization of the stress ratio yields the stress concentration factor.

[0019] The original stress dataset is subjected to frequency domain transformation to obtain the power spectral density distribution, and the power spectral density distribution is subjected to frequency band energy integration to obtain the low-frequency energy integral and the high-frequency energy integral.

[0020] The energy dissipation rate is obtained by proportionally calculating the low-frequency energy integral and the high-frequency energy integral.

[0021] In one embodiment, the motion coupling degree is calculated based on the original motion dataset, including:

[0022] The original motion dataset is processed by angle curve extraction to obtain buckling angle time series curves and rotation angle time series curves.

[0023] Based on a preset threshold, threshold detection processing is performed on the buckling angle time series curve to obtain the time point in the buckling angle time series curve where the buckling angle reaches the preset threshold.

[0024] Based on the time series curves of time points and rotation angles, the absolute value of the rotation angle is extracted to obtain the motion coupling degree.

[0025] In one embodiment, the initial stability index is obtained using the following formula:

[0026] ISI = w1·C s +w2·C m +w3·D e

[0027] Among them, C s C is the stress concentration factor. m For the degree of motion coupling, D e Let w1, w2, and w3 be the energy dissipation rate, where w1, w2, and w3 are all preset weighting coefficients, and w1 + w2 + w3 = 1.

[0028] In one embodiment, the dynamic stability index is classified and matched with a strategy according to a preset grading standard to obtain the vertebral body stability level and the corresponding treatment strategy, including:

[0029] The dynamic stability index is compared with the first and second grade thresholds, and the comparison results show that the first grade threshold is smaller than the second grade threshold.

[0030] When the dynamic stability index is less than the first grade threshold, the vertebral body stability level is grade one.

[0031] When the comparison result shows that the dynamic stability index is greater than the first grade threshold and less than the second grade threshold, the vertebral body stability level is level two.

[0032] When the comparison result shows that the dynamic stability index is greater than the second grade threshold, the vertebral body stability level is grade three.

[0033] When the vertebral body stability level is level 2 or level 3, the original stress dataset is processed by region extraction to select the set of spatial coordinates with stress concentration coefficient greater than the stress risk threshold.

[0034] Based on a pre-defined strategy database, a treatment strategy is generated by processing the strategy according to the stability level and the set of spatial coordinates.

[0035] In one embodiment, the initial stability index and bone mineral density value are subjected to neural network verification processing to obtain a dynamic stability index, including:

[0036] The initial stability index and bone mineral density value are concatenated to obtain the input feature vector.

[0037] Input the input feature vector into the pre-trained 3D-ResNet neural network model to obtain the original output value of the neural network;

[0038] The original output value of the neural network is normalized using the Sigmoid function to obtain the dynamic stability index;

[0039] The pre-trained 3D-ResNet neural network model is obtained through the following steps:

[0040] The dynamic mechanical dataset and animal bone mineral density values ​​of an animal spinal compression fracture model were obtained. The dynamic mechanical dataset includes stress waveforms, motion trajectories and in vitro mechanical test results, including vertebral stiffness values.

[0041] Calculate the reference stability index based on the dynamic mechanics dataset;

[0042] A training set was constructed based on the reference stability index, animal bone mineral density values, and preset stability grading labels.

[0043] The 3D-ResNet neural network model is trained based on the training set, and the network weights are optimized through the backpropagation algorithm to obtain the pre-trained 3D-ResNet neural network model.

[0044] Secondly, this application also provides a vertebral stability assessment system for osteoporotic vertebral compression fractures in the elderly, comprising:

[0045] The data acquisition module is used to collect dynamic stress data of the target vertebra in real time to obtain the raw stress dataset, and to acquire vertebral motion data to obtain the raw motion dataset.

[0046] The parameter calculation module is used to calculate the stress concentration factor and energy dissipation rate based on the original stress dataset, and to calculate the motion coupling degree based on the original motion dataset.

[0047] The stability index generation module is used to perform weighted calculations on stress concentration factor, motion coupling degree and energy dissipation rate to obtain initial stability index. The initial stability index and bone density value are then verified by a neural network to obtain dynamic stability index.

[0048] The grading and strategy matching module is used to grading and matching the dynamic stability index according to preset grading standards to obtain the vertebral body stability level and the corresponding treatment strategy.

[0049] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the first aspect.

[0050] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the first aspect.

[0051] The aforementioned method and system for assessing vertebral stability in osteoporotic vertebral compression fractures in the elderly provides foundational data for capturing the mechanical response of the vertebra under dynamic loads by real-time acquisition of dynamic stress and motion data of the target vertebra. Secondly, based on this dataset, stress concentration coefficient, energy dissipation rate, and motion coupling degree are calculated, transforming complex dynamic mechanical information into quantifiable key parameters and providing multi-dimensional indicators for subsequent stability assessment. Furthermore, an initial stability index is obtained through weighted calculation, and a dynamic stability index is derived by combining it with bone mineral density values ​​through neural network verification. This achieves deep integration of dynamic mechanical parameters and bone mineral density characteristics, improving the accuracy of stability assessment. Finally, the dynamic stability index is graded and matched with strategies according to preset grading standards to obtain the vertebral stability level and corresponding treatment strategy, thereby providing personalized and precise treatment recommendations for vertebrae with different stability states.

[0052] Compared with traditional assessment methods based on static imaging or single parameters, this method reflects the mechanical stability characteristics of the vertebral body through dynamic data acquisition, multi-parameter fusion, and intelligent verification mechanisms. It overcomes the limitations of static assessment and single-parameter assessment, and provides a more scientific and reliable basis for clinical treatment decisions. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart of a method for assessing vertebral stability in osteoporotic vertebral compression fractures in the elderly, provided as an exemplary embodiment of the present invention;

[0055] Figure 2 A flowchart illustrating a method for calculating stress concentration factor and energy dissipation rate based on an original stress dataset, as provided in an exemplary embodiment of the present invention;

[0056] Figure 3 A schematic diagram of the structure of a vertebral stability assessment system for osteoporotic vertebral compression fractures in the elderly, provided as an exemplary embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] In one embodiment, such as Figure 1 As shown, a method for assessing vertebral stability in osteoporotic vertebral compression fractures of the elderly is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] S101: Real-time acquisition of dynamic stress data of the target vertebral body to obtain the raw stress dataset, and acquisition of vertebral motion data to obtain the raw motion dataset.

[0060] Specifically, dynamic stress data of the target vertebra can be acquired using a flexible piezoelectric sensor array deployed on the skin projection area corresponding to the target vertebra on the patient's back. This array, composed of multiple distributed piezoelectric sensing units, can closely adhere to the skin surface, sensing the stress transmission and distribution changes of the vertebra under different movement states in real time. The acquired raw stress data is output as a voltage signal, which can be processed through signal filtering and amplification to convert it into a digital signal and stored, forming a raw stress dataset. This dataset can include multi-dimensional information such as timestamps, sensor numbers, and corresponding stress values. Vertebral motion data can be obtained by using techniques such as optical tracking and inertial measurement to record the vertebra's trajectory, velocity, acceleration, and other kinematic parameters in three-dimensional space, resulting in a raw motion dataset. This raw motion dataset reflects the dynamic behavior characteristics of the vertebra under different activity states. Combined with the stress data, it provides a more comprehensive understanding of the vertebra's mechanical state.

[0061] S102: Calculate the stress concentration factor and energy dissipation rate based on the original stress dataset, and calculate the motion coupling degree based on the original motion dataset.

[0062] Specifically, the stress concentration factor is calculated based on the non-uniformity of stress distribution. Illustratively, the stress concentration factor can be calculated by analyzing the peak stress regions and gradient changes in stress distribution within the original stress dataset. This factor quantitatively reflects the degree of stress concentration at a specific location on the vertebral body, and a higher stress concentration factor indicates that the vertebral structure in that region is more susceptible to damage. Furthermore, when the vertebral body is subjected to dynamic stress, some energy is dissipated, primarily related to changes in the vertebral body's internal microstructure. Based on characteristic parameters of stress fluctuations, the energy dissipation rate can be calculated. This energy dissipation rate reflects the vertebral body's ability to absorb and release energy during dynamic stress; a lower energy dissipation rate indicates weaker buffering capacity and poorer stability. The motion coupling degree is related to the correlation between motion parameters of the vertebral body in different motion directions. By calculating the correlation coefficients between motion parameters such as displacement, velocity, and acceleration in different directions, the motion coupling degree can be obtained. This motion coupling degree reflects the coordination and consistency of the vertebral body during multi-directional motion; a higher motion coupling degree indicates that the vertebral body can better maintain overall mechanical balance during multi-directional motion, resulting in relatively better stability.

[0063] S103: The stress concentration factor, motion coupling degree, and energy dissipation rate are weighted and calculated to obtain the initial stability index. The initial stability index and bone density value are then verified by a neural network to obtain the dynamic stability index.

[0064] Specifically, the weighting coefficients in the weighting process are set according to the importance of each parameter to vertebral stability. By analyzing a large amount of clinical and experimental data, the weighting coefficients of each parameter can be determined, allowing the obtained initial stability index to more accurately reflect the stability state of the vertebral body. Bone mineral density (BMD) values, obtained through conventional bone mineral density testing methods such as dual-energy X-ray absorptiometry (DXA), reflect the density of vertebral bone and are an important indicator for assessing the degree of osteoporosis. Subsequently, a pre-trained neural network model can be used to perform neural network validation on the initial stability index and BMD values. This neural network model can be constructed based on neural network structures such as multilayer perceptrons and long short-term memory networks. It is trained using a large amount of known vertebral stability data and corresponding initial stability index and BMD values ​​as training samples, thereby learning the complex nonlinear relationship between the initial stability index, BMD values, and actual vertebral stability. Through neural network validation, the initial stability index can be corrected and calibrated to obtain a more accurate dynamic stability index.

[0065] S104: Based on the preset grading standards, the dynamic stability index is classified and matched with strategies to obtain the vertebral body stability level and the corresponding treatment strategy.

[0066] Specifically, the pre-defined grading criteria are developed based on clinical experience and practical needs in vertebral stability assessment. The dynamic stability index can be divided into different grade ranges, each corresponding to a different vertebral stability state. For example, the dynamic stability index can be divided into high stability, moderate stability, and low stability levels. Subsequently, appropriate treatment strategies can be matched based on the grade range to which the dynamic stability index belongs. These strategies can include conservative treatment (such as medication and rehabilitation training), minimally invasive surgery, and open surgery, thus providing clinicians with targeted treatment recommendations.

[0067] The aforementioned method, by acquiring real-time dynamic stress and motion data of the target vertebral body, can capture the mechanical characteristics of the vertebral body under dynamic conditions from multiple dimensions such as stress change trends, motion amplitude, and trajectory, providing a data basis for subsequent stability assessment. Secondly, based on the original stress dataset, the stress concentration coefficient and energy dissipation rate are calculated, and based on the original motion dataset, the motion coupling degree is calculated, realizing the quantification of local stress concentration, energy consumption characteristics, and motion coordination of the vertebral body, improving the reliability and specificity of key assessment parameters. Finally, an initial stability index is obtained through weighted calculation, and a dynamic stability index is obtained by combining bone mineral density values ​​with neural network verification, realizing the fusion analysis of in vivo dynamic response and baseline bone condition, further improving the physiological relevance of stability quantification. Furthermore, by classifying levels according to preset grading standards and matching treatment strategies, the scientific rigor of stability assessment and the adaptability of treatment plans are enhanced, improving the accuracy of assessment results and the effectiveness of clinical treatment.

[0068] In one embodiment, dynamic stress data of the target vertebra is acquired in real time to obtain a raw stress dataset, and vertebral motion data is acquired to obtain a raw motion dataset, including:

[0069] The sensor array deployment distribution is obtained by deploying a flexible piezoelectric sensor array in the target vertebral region of the patient's back;

[0070] Based on the optically traceable markers on the skin projection areas corresponding to the target vertebra and adjacent vertebrae, the marker deployment distribution is obtained;

[0071] Acquire patient's motion execution data by instructing the patient to perform standard motion protocols that include positional changes and axial rotation;

[0072] Based on the sensor array deployment distribution and motion execution data, multi-axis dynamic stress waveforms are collected to obtain the raw stress dataset;

[0073] Based on the distribution of marker points and action execution data, three-dimensional motion trajectories are collected to obtain the raw motion dataset.

[0074] Specifically, the flexible piezoelectric sensor array can adopt a serpentine topology arrangement, consisting of several piezoelectric sensing units connected in series on a flexible substrate such as a polyimide film. The size of a single sensing unit can be set to 5mm × 5mm, with a spacing of 3mm between adjacent units. This serpentine topology allows the array to adaptively bend with the deformation of the skin surface during patient movement, avoiding sensor unit detachment or signal distortion caused by changes in body position, and ensuring that each unit can independently sense stress changes in its corresponding area. Illustratively, the spatial coordinates of the sensor array deployment can be recorded using a pre-calibrated optical positioning system. This coordinate system establishes a three-dimensional coordinate system with the center of the target vertebra as the origin. By acquiring the positional information of each sensing unit, a dataset of the sensor array deployment can be obtained.

[0075] When obtaining the marker deployment distribution based on optically traceable markers on the skin projection areas corresponding to the target vertebra and adjacent vertebrae, these optically traceable markers can be passive infrared reflective spheres with a diameter of 3mm, their surfaces covered with a high-reflectivity coating to reflect infrared light of a specific wavelength. One marker can be pasted at the center of the skin projection area of ​​the target vertebra as a reference point, one marker at the center of the skin projection areas of the adjacent upper and lower vertebrae, and one marker 1.5cm to each side of the target vertebra, forming a distribution structure containing 5 markers. This distribution structure can eliminate errors caused by the displacement of a single marker through multi-point collaborative positioning, improving the stability of motion trajectory calculation. After the marker deployment is completed, the markers can be spatially located using an infrared optical tracking module and calibrated using a 3D calibration target to complete spatial coordinate calibration. Subsequently, the initial positions of each marker in the 3D coordinate system are recorded and stored to obtain the marker deployment distribution dataset.

[0076] Indicatively, when acquiring patient movement execution data, the standard movement protocol may include three core movement sequences, each executed at a preset speed and amplitude. For example, the first sequence is a postural transition movement, including slowly flexing forward from an upright position to a 30° flexion, holding for 2 seconds, and then slowly returning to the starting position, repeated 3 times. The second sequence is an axial rotation movement, including rotating 30° to the left from a neutral position, holding for 2 seconds, returning to the neutral position, then rotating 30° to the right, holding for 2 seconds, and returning to the starting position, repeated twice. The third sequence is a compound movement, involving simultaneous axial rotations of 20° to the left and right from a 30° flexion position, repeated twice. During movement execution, voice prompts and visual guidance can be used to ensure the patient's movements conform to the standards, and medical staff can monitor the process to prevent deviations due to pain or limited mobility. Movement execution data is recorded as timestamp-linked movement stage tags.

[0077] Specifically, during the process of deploying and distributing sensor arrays and acquiring raw stress datasets based on motion execution data, each sensing unit of the flexible piezoelectric sensor array can convert mechanical stress into changes in electrical charge. Furthermore, during the acquisition process, stress signals at different motion stages can be segmented and labeled according to the timestamps in the motion execution data; for example, the stress waveform of a flexion motion can be labeled as "flexion-stress-1" (the number represents the number of repetitions). For the acquisition of multi-axis dynamic stress, this can be achieved using sensing units arranged in different directions within a serpentine topology array. Units arranged along the Y-axis primarily sense compressive stress in the vertical direction, while units arranged along the X-axis primarily sense shear stress. By synthesizing the signals from each unit, three-dimensional stress components can be obtained. Finally, the raw stress dataset can be stored in data frames, with each frame containing a timestamp, sensor unit number, three-dimensional stress component values, and corresponding motion stage labels.

[0078] Furthermore, when collecting 3D motion trajectories based on the distribution of marker points and the data from motion execution, the infrared optical tracking module can capture the 3D spatial coordinates of each marker point per frame and calculate the real-time displacement of each marker point based on its initial position in the marker point distribution. A rigid body transformation-based algorithm is then used to calculate the 3D motion trajectory. For example, treating five marker points as a rigid whole, the spatial distribution of the marker points is fitted using the least squares method to eliminate non-rigid deformation errors caused by skin slippage. The translational parameters of the target vertebra are calculated based on the relative displacement between the reference point and adjacent vertebral marker points. Rotational parameters such as flexion angle, lateral flexion angle, and rotation angle are calculated based on the angular changes of the marker points in the horizontal and sagittal planes. The flexion angle is the angle between the target vertebral axis and the vertical line in the sagittal plane, and the rotational angle is the angle between the target vertebral axis and the midline in the horizontal plane. Similar to stress data, the 3D motion trajectory data can also be segmented and labeled according to the timestamps of the motion execution data, ultimately yielding the original motion dataset. Each frame of this dataset contains a timestamp, the 3D coordinates of each marker point, vertebral translational parameters, rotational parameters, and motion stage labels.

[0079] In one embodiment, such as Figure 2 As shown, the stress concentration factor and energy dissipation rate are calculated based on the original stress dataset through the following steps:

[0080] S201: Perform stress peak analysis on the original stress dataset to obtain the stress peak value at the anterior and posterior edges of the target vertebral body;

[0081] S202: Calculate the ratio of the peak stress at the leading edge to the peak stress at the trailing edge to obtain the stress ratio;

[0082] S203: Logarithmically normalize the stress ratio to obtain the stress concentration factor;

[0083] S204: Perform frequency domain transformation on the original stress dataset to obtain the power spectral density distribution, and perform frequency band energy integration on the power spectral density distribution to obtain the low-frequency energy integral and the high-frequency energy integral.

[0084] S205: The energy dissipation rate is obtained by proportionally calculating the low-frequency energy integral and the high-frequency energy integral.

[0085] Specifically, when performing stress peak analysis on the original stress dataset, the three-dimensional stress components in the dataset can first be filtered by direction, focusing on extracting the stress components along the vertebral axis, i.e., the Z-axis direction perpendicular to the vertebral endplate. This directional stress directly reflects the compressive load borne by the vertebra and is most closely related to the stability of compression fractures. A sliding window peak detection algorithm can be used, setting the window size and step size, and determining the local maximum value as a candidate peak by comparing the stress component values ​​of each sampling point within the window. Subsequently, threshold filtering is applied to the candidate peaks to remove noise peaks smaller than a preset noise threshold, and false peaks are removed by verifying the continuity of peaks in adjacent windows, such as ensuring the time interval between adjacent peaks is not less than 100ms. In addition, based on the spatial coordinates of the sensing units deployed in the sensor array, the peak value collected by the sensing unit located in the anterior projection area of ​​the target vertebra (positive X-axis direction, 1-2 cm from the midline) can be defined as the anterior stress peak value, and the peak value collected by the sensing unit located in the posterior projection area (negative X-axis direction, 1-2 cm from the midline) can be defined as the posterior stress peak value. Finally, three maximum peak values ​​are extracted from the stress waveform of each action stage and used as the anterior stress peak value and posterior stress peak value of that stage, respectively.

[0086] To illustrate, when calculating and logarithmically normalizing the stress ratio, the peak anterior stress value can be compared to the peak posterior stress value during the same movement phase. Since the stress distribution in a normal vertebra is relatively uniform, the stress ratio is typically in the range of 0.8-1.2. However, in vertebrae with compromised stability, this value may increase significantly due to anterior compression. To map the stress ratio to a uniform quantization range, natural logarithmic normalization can be used to convert the stress ratio into a dimensionless coefficient between 0 and 1. When the stress ratio is 1, the stress concentration coefficient is 0, indicating no stress concentration. When the stress ratio is the maximum stress ratio observed in clinical data, the stress concentration coefficient is 1, indicating severe stress concentration, thus achieving standardized characterization of stress ratios of different magnitudes. Furthermore, when performing frequency domain transformation on the original stress dataset, stress waveform segments from the stable loading period (such as the maintenance phase in flexion) of each movement phase can be selected and windowed using the Hanning window function to reduce spectral leakage. The time-domain signal was then converted to the frequency-domain signal using a Fast Fourier Transform (FFT) to obtain a power spectral density distribution in the frequency range of 0-250 Hz. The power spectral density value represents the power per unit frequency, reflecting the energy proportion of different frequency components. Based on the biomechanical characteristics of the vertebral body, the frequency band can be divided into a low-frequency band (0-5 Hz) and a high-frequency band (5-20 Hz). The low-frequency band corresponds to stress fluctuations caused by overall vertebral deformation, while the high-frequency band corresponds to the stress response caused by vibrations of vertebral microstructures such as trabeculae. Furthermore, when integrating the power spectral density distribution across frequency bands, the trapezoidal integral method can be used to calculate the area under the power spectral density curves for the low-frequency and high-frequency bands, respectively, yielding the low-frequency energy integral and the high-frequency energy integral. The boundary frequencies of the integration intervals can be determined through spectral characteristic analysis of clinical samples, ensuring coverage of the main energy distribution range of the stress signal.

[0087] Furthermore, the sum of the low-frequency energy integral and the high-frequency energy integral can be calculated, and the energy dissipation rate can be obtained by calculating the ratio of the low-frequency energy integral to the sum. Illustratively, a stable vertebral structure under dynamic load can absorb low-frequency energy through elastic deformation and dissipate high-frequency energy through the damping effect of its microstructure. Therefore, the energy dissipation rate can intuitively reflect the energy absorption and dissipation characteristics of the vertebral structure, providing a quantitative basis for subsequent stability assessment.

[0088] In one embodiment, calculating motion coupling based on the original motion dataset includes:

[0089] The original motion dataset is processed by angle curve extraction to obtain buckling angle time series curves and rotation angle time series curves.

[0090] Based on a preset threshold, threshold detection processing is performed on the buckling angle time series curve to obtain the time point in the buckling angle time series curve where the buckling angle reaches the preset threshold.

[0091] Based on the time series curves of time points and rotation angles, the absolute value of the rotation angle is extracted to obtain the motion coupling degree.

[0092] Specifically, when extracting angle curves from the original motion dataset, rotational parameters related to the target vertebral motion can be selected first. The flexion angle is the rotation angle of the target vertebral body around the X-axis in the sagittal plane, with flexion being positive and extension being negative. The rotation angle is the rotation angle of the target vertebral body around the Z-axis in the horizontal plane, with left rotation being positive and right rotation being negative. Based on the timestamps and rotational parameters in the original motion dataset, linear interpolation can be used to interpolate the discrete data points, adjusting the sampling frequency to match the sampling frequency of the infrared optical tracking module to eliminate potential time interval inconsistencies during data acquisition. Subsequently, moving average filtering can be used to smooth the interpolated angle data, removing high-frequency jitter noise and generating continuous flexion and rotation angle time-series curves.

[0093] Furthermore, the preset threshold can be set based on clinical kinematic research data. For example, 30° can be selected as the key threshold for the flexion angle. This angle is within the range of movements that most elderly patients can safely complete, while effectively stimulating the motion coupling characteristics of the vertebral body under moderate load. The threshold detection process can employ an adaptive threshold crossing algorithm, which involves scanning the flexion angle time-series curve point by point. When the curve first rises from the region below 30° and crosses 30°, the time point at that moment is recorded. If multiple crossings occur in the same movement sequence, such as during repetitive movements, the time point of the first crossing is taken as valid data to avoid errors caused by repeated calculations. During the detection process, stage labels such as "flexion stage" in the movement execution data can also be used for verification to ensure that the time point falls within the preset movement stage, eliminating false detections caused by patient movement deviations. If the detection result does not match the movement stage, the curve can be smoothed again and the detection repeated until a time point that conforms to the movement logic is obtained.

[0094] In illustrative terms, when extracting the absolute value of rotation angle based on time-series curves of rotation angles, the angle value corresponding to the time point can be located in the rotation angle time-series curve. This value reflects the rotation angle accompanying the target vertebra at the instant the flexion angle reaches 30°. Since rotation angles have directional differences, and motion coupling degree needs to reflect the magnitude of coupling rather than direction, the absolute value of this angle can be taken. To improve the reliability of the results, the above processing can be performed on three repeated executions of the same action sequence, obtaining three absolute values. The arithmetic mean of these three values ​​is then used as the final motion coupling degree. Because the rotation angle accompanying normal vertebral movements during flexion is relatively small (usually less than 5°), while vertebrae with decreased stability exhibit significant motion coupling due to impaired structural integrity, the magnitude of motion coupling degree directly reflects the quality of vertebral motion coordination, providing crucial kinematic evidence for assessing its stability.

[0095] In one embodiment, based on the stress concentration factor, motion coupling degree, and energy dissipation rate calculated above, the initial stability index can be obtained using the following formula:

[0096] ISI = w1·C s +w2·C m +w3·D e

[0097] Among them, C s C is the stress concentration factor. m For the degree of motion coupling, D e Let w1, w2, and w3 be the energy dissipation rate, where w1, w2, and w3 are all preset weighting coefficients, and w1 + w2 + w3 = 1.

[0098] In one embodiment, the dynamic stability index is classified and matched with a strategy according to a preset grading standard to obtain the vertebral body stability level and the corresponding treatment strategy, including:

[0099] The dynamic stability index is compared with the first and second grade thresholds, and the comparison results show that the first grade threshold is smaller than the second grade threshold.

[0100] When the dynamic stability index is less than the first grade threshold, the vertebral body stability level is grade one.

[0101] When the comparison result shows that the dynamic stability index is greater than the first grade threshold and less than the second grade threshold, the vertebral body stability level is level two.

[0102] When the comparison result shows that the dynamic stability index is greater than the second grade threshold, the vertebral body stability level is grade three.

[0103] When the vertebral body stability level is level 2 or level 3, the original stress dataset is processed by region extraction to select the set of spatial coordinates with stress concentration coefficient greater than the stress risk threshold.

[0104] Based on a pre-defined strategy database, a treatment strategy is generated by processing the strategy according to the stability level and the set of spatial coordinates.

[0105] Specifically, the dynamic stability index ranges from [0,1], and its value is negatively correlated with vertebral stability; that is, the larger the value, the worse the vertebral stability. The first and second grading thresholds are set based on statistical analysis of clinical case data, and the optimal segmentation point is determined using the receiver operating characteristic curve. The first grading threshold can be set to 0.3, and the second grading threshold can be set to 0.6. The comparison process can be implemented using a numerical comparison algorithm, comparing the dynamic stability index one by one with the two thresholds to generate a comparison report, providing a direct basis for subsequent grading.

[0106] For example, a dynamic stability index less than 0.3 is classified as Level 1 stability, corresponding to an intact vertebral structure, uniform stress distribution under dynamic load, and low motion coupling, indicating good vertebral stability and a low risk of fracture. A dynamic stability index greater than 0.3 and less than 0.6 is classified as Level 2 stability, corresponding to mild vertebral structural damage, insignificant stress concentration, but slightly decreased motion coordination, suggesting moderate stability with a potential risk of progression. A dynamic stability index greater than 0.6 is classified as Level 3 stability, corresponding to significant vertebral structural damage, significant stress concentration, and high motion coupling, indicating poor stability and a high risk of fracture progression or recurrence. For illustrative purposes, the classification results are stored as text labels and associated with the specific values ​​of the dynamic stability index, facilitating clinical physicians' ability to trace the assessment basis.

[0107] Furthermore, when the vertebral stability level is grade II or III, the original stress dataset can be processed for region extraction to screen for stress concentration risk areas. For example, firstly, the spatial coordinates and corresponding stress concentration coefficients of all sensor units are extracted from the original stress dataset, establishing a mapping table between spatial coordinates and stress concentration coefficients. This threshold can be determined by analyzing the difference in stress concentration coefficient distribution between fractured and normal vertebrae. A stress risk threshold of 0.7 can be set, corresponding to the upper limit of the 95% confidence interval. Subsequently, by traversing the mapping table, the coordinates of all sensor units with stress concentration coefficients greater than 0.7 are screened to form a spatial coordinate set. This set can be stored in the form of a three-dimensional coordinate array, with each element containing coordinate values. Through the mapping relationship with the anatomical divisions of the target vertebral body, such as the anterior and posterior edges, the specific anatomical region of stress concentration (such as the upper-middle part of the anterior edge of the vertebral body) can be intuitively located, providing a spatial positioning basis for the formulation of subsequent treatment strategies.

[0108] Furthermore, the pre-defined strategy database is a structured data storage module containing three levels of strategy templates: a basic strategy template for Level 1 stability, a reinforcement strategy template for Level 2 stability, and an intervention strategy template for Level 3 stability. Each template includes core content such as conservative treatment plans, rehabilitation training plans, and clinical monitoring cycles. When the stability level is Level 1, the basic strategy template can be directly invoked to output a treatment strategy, such as one primarily based on anti-osteoporosis drug therapy combined with core muscle training three times a week, with a monitoring cycle set at three months. This strategy generation process can be performed through database queries and conditional judgment algorithms. The final output treatment strategy is presented in the form of a structured report, including treatment methods, specific operational parameters, execution cycles, and risk warnings, ensuring the operability and accuracy of clinical applications.

[0109] In one embodiment, the initial stability index and bone mineral density value are subjected to neural network verification processing to obtain the dynamic stability index, including:

[0110] The initial stability index and bone mineral density value are concatenated to obtain the input feature vector.

[0111] Input the input feature vector into the pre-trained 3D-ResNet neural network model to obtain the original output value of the neural network;

[0112] The original output value of the neural network is normalized using the Sigmoid function to obtain the dynamic stability index;

[0113] The pre-trained 3D-ResNet neural network model is obtained through the following steps:

[0114] The dynamic mechanical dataset and animal bone mineral density values ​​of an animal spinal compression fracture model were obtained. The dynamic mechanical dataset includes stress waveforms, motion trajectories and in vitro mechanical test results, including vertebral stiffness values.

[0115] Calculate the reference stability index based on the dynamic mechanics dataset;

[0116] A training set was constructed based on the reference stability index, animal bone mineral density values, and preset stability grading labels.

[0117] The 3D-ResNet neural network model is trained based on the training set, and the network weights are optimized through the backpropagation algorithm to obtain the pre-trained 3D-ResNet neural network model.

[0118] Specifically, when performing vector concatenation on the initial stability index and bone mineral density (BMD) value, the initial stability index is a dimensionless value obtained through weighted calculation, ranging from [0,1], while the BMD value is the bone mineral density of the target vertebral region measured using dual-energy X-ray absorptiometry (DXA). The initial stability index can then be considered as a one-dimensional feature vector [S], and the BMD value as another-dimensional feature vector [BMD]. A feature concatenation algorithm combines these two into a two-dimensional input feature vector. The concatenation process must ensure sample correlation between the two, meaning each input feature vector corresponds to synchronous measurement data from the same patient. Furthermore, data standardization maps the BMD value to the [0,1] interval, eliminating dimensional differences and providing consistent input for the neural network's feature learning.

[0119] Furthermore, the 3D-ResNet neural network model is based on the 3D-ResNet-18 architecture, which consists of five convolutional stages. The first stage is the input layer, which receives a two-dimensional input feature vector and transforms it into a 3D feature tensor with a size of 1×1×2 through dimensional expansion. The second to fourth stages are residual blocks, each consisting of two 3D convolutional layers (3×3×3 kernel size, 1×1×1 stride), a batch normalization layer, and a ReLU activation function. Skip connections are used to alleviate the gradient vanishing problem and gradually extract higher-order nonlinear correlations of features. The fifth stage consists of a global average pooling layer and a fully connected layer, which compresses the high-dimensional feature tensor into a 1×1×64 feature vector, and then outputs the original value, i.e., the original output value of the neural network, through the fully connected layer. The model's pre-trained parameters are stored in a weight file. During inference, a forward propagation algorithm is used. After the input feature vector is input into the model and processed through each layer, the original output value with an unconstrained range is obtained. Normalizing the original output value of the neural network using the Sigmoid function maps the original output value to the [0,1] interval. The result of the mapping is the dynamic stability index.

[0120] In a schematic manner, during the training of the 3D-ResNet neural network model, when acquiring the dynamic mechanical dataset and animal bone mineral density values ​​for an animal spinal compression fracture model, a pre-defined animal can be used to construct an osteoporosis model, and vertebral compression devices can be used to establish compression fracture models of different degrees, such as 10%, 20%, and 30% of the vertebral height. The dynamic mechanical dataset can be acquired by deploying miniature piezoelectric sensors and fluorescent markers on rat spinal specimens, including stress waveforms at different loading rates, three-dimensional motion trajectories, and out-of-body mechanical test results measured by a materials mechanics testing machine. The vertebral stiffness value in this result can be calculated using the slope of the linear segment of the load-displacement curve. The animal bone mineral density value can be calculated using micro-CT scanning to obtain the bone mineral density of the vertebral region. Subsequently, when calculating the reference stability index based on the dynamic mechanical dataset, the calculation method for the reference stability index is consistent with the initial stability index.

[0121] Specifically, when constructing the training set based on the reference stability index, animal bone density values, and preset stability grading labels, the preset stability grading labels can be determined based on the results of in vitro mechanical testing. For example, when the vertebral stiffness value is greater than 80 N / mm, the label is stable, corresponding to a low dynamic stability index; when the stiffness value is between 40 and 80 N / mm, the label is moderately unstable; and when the stiffness value is less than 40 N / mm, the label is severely unstable. The training set can be expanded to 10,000 samples using data augmentation techniques such as adding ±5% Gaussian noise. Each sample contains the reference stability index, animal bone density value, and corresponding label, and is divided into training and validation sets in a 7:3 ratio. Based on the training set, a 3D-ResNet neural network model can be trained using the Adam optimizer with the cross-entropy loss function. The network weights are updated through backpropagation. For example, each iteration inputs 32 samples, calculates the loss value between the predicted output and the label, calculates the partial derivative of the loss function with respect to the weights of each layer using the chain rule, and adjusts the weight values ​​according to the gradient descent direction. The training process can last for 200 epochs, with model accuracy evaluated on the validation set every 10 epochs. Training stops when the validation set accuracy shows no improvement for 20 consecutive epochs, and the network weights at this point can be saved as a pre-trained model. This model has learned the non-linear relationship between the reference stability index, bone mineral density value, and actual vertebral stability, and can be directly used for dynamic stability index verification of human data.

[0122] like Figure 3As shown, based on the same inventive concept, this application also provides a vertebral stability assessment system 300 for vertebral compression fractures in osteoporotic patients, as described above. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the vertebral stability assessment system for osteoporotic vertebral compression fractures in osteoporotic patients provided below can be found in the limitations of the vertebral stability assessment method for osteoporotic vertebral compression fractures in osteoporotic patients described above, and will not be repeated here. The system includes:

[0123] The data acquisition module 301 is used to collect dynamic stress data of the target vertebra in real time to obtain the original stress dataset, and to acquire vertebral motion data to obtain the original motion dataset.

[0124] The parameter calculation module 302 is used to calculate the stress concentration factor and energy dissipation rate based on the original stress dataset, and to calculate the motion coupling degree based on the original motion dataset.

[0125] The stability index generation module 303 is used to perform weighted calculations on stress concentration factor, motion coupling degree and energy dissipation rate to obtain initial stability index, and to perform neural network verification on initial stability index and bone density value to obtain dynamic stability index.

[0126] The grading and strategy matching module 304 is used to grade and match the dynamic stability index according to the preset grading standard to obtain the vertebral body stability level and the corresponding treatment strategy.

[0127] In one exemplary embodiment, the present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method for assessing vertebral stability in osteoporotic vertebral compression fractures of the present application. A multi-core processor is preferred to improve the system's parallel processing capabilities. The memory provides sufficient temporary storage space to support program execution and data processing. The memory capacity should be large enough to accommodate a large amount of information and computational tasks.

[0128] In one exemplary embodiment, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for assessing the stability of vertebral compression fractures in patients with osteoporosis.

[0129] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for assessing vertebral stability in osteoporotic vertebral compression fractures of the elderly, characterized in that, The method includes: Real-time acquisition of dynamic stress data of the target vertebral body to obtain the raw stress dataset, and acquisition of vertebral body motion data to obtain the raw motion dataset; The stress concentration factor and energy dissipation rate are calculated based on the original stress dataset, and the motion coupling degree is calculated based on the original motion dataset; The stress concentration factor, the motion coupling degree, and the energy dissipation rate are weighted and calculated to obtain the initial stability index. The initial stability index and the bone density value are then verified by a neural network to obtain the dynamic stability index. Based on the preset grading criteria, the dynamic stability index is classified and matched with strategies to obtain the vertebral body stability level and the corresponding treatment strategy.

2. The method according to claim 1, characterized in that, The real-time acquisition of dynamic stress data of the target vertebral body is used to obtain a raw stress dataset, and vertebral body motion data is acquired to obtain a raw motion dataset, including: The sensor array deployment distribution is obtained by deploying a flexible piezoelectric sensor array in the target vertebral region of the patient's back; Based on the optically traceable markers on the skin projection areas corresponding to the target vertebra and adjacent vertebrae, the marker deployment distribution is obtained; The patient's action execution data is acquired by instructing the patient to perform a standard action protocol that includes positional changes and axial rotation. Based on the sensor array deployment and the action execution data, multi-axis dynamic stress waveforms are collected to obtain the original stress dataset; Based on the distribution of the marker points and the action execution data, a three-dimensional motion trajectory is collected to obtain the original motion dataset.

3. The method according to claim 1, characterized in that, The calculation of stress concentration factor and energy dissipation rate based on the original stress dataset includes: The original stress dataset is subjected to stress peak analysis to obtain the stress peak value at the anterior edge and the stress peak value at the posterior edge of the target vertebral body. Calculate the ratio of the peak stress at the leading edge to the peak stress at the trailing edge to obtain the stress ratio. The stress ratio is logarithmically normalized to obtain the stress concentration factor; The original stress dataset is subjected to frequency domain transformation to obtain the power spectral density distribution, and the power spectral density distribution is subjected to frequency band energy integration to obtain low-frequency energy integral and high-frequency energy integral. The energy dissipation rate is obtained by proportionally calculating the low-frequency energy integral and the high-frequency energy integral.

4. The method according to claim 1, characterized in that, The calculation of motion coupling degree based on the original motion dataset includes: The original motion dataset is processed by angle curve extraction to obtain buckling angle time-series curves and rotation angle time-series curves; According to a preset threshold, the buckling angle time series curve is subjected to threshold detection processing to obtain the time point in the buckling angle time series curve where the buckling angle reaches the preset threshold. Based on the time point and the time series curve of the rotation angle, the absolute value of the rotation angle is extracted to obtain the motion coupling degree.

5. The method according to any one of claims 1 to 4, characterized in that, The initial stability index is obtained by the following formula: ISI=w1·C s +w2·C m +w3·D e Among them, C s C is the stress concentration factor. m Let D be the degree of motion coupling. e The energy dissipation rate is defined as w1, w2, and w3, which are all preset weighting coefficients and satisfy w1 + w2 + w3 = 1.

6. The method according to claim 1, characterized in that, The process of classifying and matching the dynamic stability index according to a preset grading standard to obtain the vertebral body stability level and corresponding treatment strategy includes: The dynamic stability index is compared with the first grading threshold and the second grading threshold, and the comparison result is that the first grading threshold is less than the second grading threshold. When the comparison result shows that the dynamic stability index is less than the first grading threshold, the vertebral body stability level is Level 1 stability; When the comparison result shows that the dynamic stability index is greater than the first grading threshold and less than the second grading threshold, the vertebral body stability level is level two. When the comparison result shows that the dynamic stability index is greater than the second grading threshold, the vertebral body stability level is level three. When the stability level of the vertebral body is level two or level three, the original stress dataset is subjected to region extraction processing to filter out the set of spatial coordinates whose stress concentration coefficient is greater than the stress risk threshold. Based on a preset strategy database, the treatment strategy is obtained by performing strategy generation processing according to the stability level and the set of spatial coordinates.

7. The method according to claim 1, characterized in that, The process of performing neural network verification on the initial stability index and bone mineral density value to obtain the dynamic stability index includes: The initial stability index and the bone density value are concatenated to obtain the input feature vector; The input feature vector is input into a pre-trained 3D-ResNet neural network model to obtain the original output value of the neural network; The original output value of the neural network is normalized using the Sigmoid function to obtain the dynamic stability index. The pre-trained 3D-ResNet neural network model is obtained through the following steps: A dynamic mechanical dataset and animal bone mineral density values ​​of an animal spinal compression fracture model are obtained. The dynamic mechanical dataset includes stress waveforms, motion trajectories, and in vitro mechanical test results, including vertebral stiffness values. Calculate the reference stability index based on the dynamic mechanics dataset; A training set is constructed based on the reference stability index, the animal bone density value, and the preset stability grading label; The 3D-ResNet neural network model is trained based on the training set, and the network weights are optimized through backpropagation algorithm to obtain the pre-trained 3D-ResNet neural network model.

8. A system for assessing vertebral stability in osteoporotic vertebral compression fractures in the elderly, characterized in that, The system includes: The data acquisition module is used to collect dynamic stress data of the target vertebra in real time to obtain the raw stress dataset, and to acquire vertebral motion data to obtain the raw motion dataset. The parameter calculation module is used to calculate the stress concentration factor and energy dissipation rate based on the original stress dataset, and to calculate the motion coupling degree based on the original motion dataset. The stability index generation module is used to perform weighted calculation processing on the stress concentration coefficient, the motion coupling degree, and the energy dissipation rate to obtain the initial stability index, and to perform neural network verification processing on the initial stability index and the bone density value to obtain the dynamic stability index. The grading and strategy matching module is used to grading and matching the dynamic stability index according to a preset grading standard to obtain the vertebral stability level and the corresponding treatment strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images