Data processing method for orthopedic spinal imaging

Abstract

The present application relates to the technical field of healthcare informatics, and discloses a data processing method for orthopedic spinal imaging, comprising: converting spinal tomography image data into a voxel grayscale matrix, counting local spatial variance to adjust an anatomical boundary threshold, removing noise pseudo-edge points and constructing a structured seed set; driving a topological state sequence to step along the seed set to map semantic labels and calibrate a center point, calculating a direction vector to generate a dynamic search axis; using a servo spatial coordinate transformation matrix to correct the search window of the next vertebral segment, correcting geometric distortion and outputting multi-dimensional anatomical parameter geometric atlas data, the present application adjusts the boundary threshold to suppress noise interference, uses a servo coordinate transformation to correct the search window to correct geometric distortion, and blocks connected branches generated by osteophyte fusion to eliminate semantic counting misplacement divergence, so that atlas data sequence counting is smoothly convergent.

Description

Technical Field

[0001] This invention belongs to the field of healthcare informatics technology, and in particular relates to a data processing method for orthopedic spinal imaging. Background Technology

[0002] Currently, acquiring skeletal anatomical features using computed tomography is a common process. The processing program acquires voxel grayscale matrices based on spatial density differences and marks discrete vertebral body boundaries based on the abrupt change law of grayscale gradient field. However, when processing distorted images of severe osteophyte formation, joint ankylosis, or narrow intervertebral space, the physiological and anatomical gaps between adjacent vertebrae disappear due to osteophyte fusion and filling by hyperplastic tissue. This results in a completely degenerate state of grayscale features in the adhesion lesions, and the underlying pixel clustering logic cannot capture discrete gradient extrema in the fused lesion area, leading to computational disorder in the image segmentation process. Traditional improvement approaches usually attempt to deepen the number of neural network feature extraction layers or allow on-site physicians to manually define the segmentation planes. However, increasing the number of network layers consumes system computing resources exponentially and produces false edge misjudgments in adhesion areas with uniform grayscale. While introducing manual intervention can compensate for the lack of image boundary information to some extent, it interrupts the automated closed loop of data flow, increases the data conversion cycle, and is difficult to adapt to the high-timeliness digital management needs of clinical practice.

[0003] Optimization of hardware-dependent scanning slice thickness and radiation dose is limited by the physical boundaries of imaging equipment. Even with complex anatomical distortions, it remains difficult to eliminate grayscale degeneracy at tissue adhesions. Besides hardware limitations, control software and diagnostic algorithms also face technical bottlenecks when processing such variant images. For example, Chinese invention patent application CN114049955A discloses a computed tomography-assisted diagnostic system for vertebral fractures. This system utilizes a parallel and weighted backbone network to extract multi-segment vertebral features and introduces consistency comparison constraints by calculating the cosine correlation matrix between features. This scheme is based on the assumption that each vertebral segment... While pixel space is clearly separable and the distribution of features among normal vertebrae is consistent, when encountering severe osteophyte formation accompanied by joint ankylosis and fusion, the physical adhesion of adjacent vertebrae leads to degeneracy of pixel-level grayscale features. The parallel backbone network extracts confused connected abnormal features in the fusion region, causing the consistency comparison mechanism to fail fundamentally due to distortion of the basic data source. Because there is no dynamic following correction path for the curvature of the spatial topology, and no discrete conditional control flow to strongly constrain the memory coordinates of the variant segments, the discrimination error is cascaded along the spinal axis, causing the anatomical label count to diverge.

[0004] Therefore, the technical problem to be solved by this invention is how to utilize the processor in the image processing system to get rid of the single dependence on the original pixel features of the bottom layer when grayscale mixing occurs due to hyperplasia and adhesion at the anatomical boundary. This is achieved by introducing discrete conditional control flow and spatial topological state constraints with anatomical causal characteristics, and automatically triggering deviation from the calibration path to rewrite the memory coordinates of the variant segment at the moment of local feature annihilation, so as to ensure that multiple constraints achieve hierarchical coupling and thus enable the smooth convergence of the spinal anatomical semantic label sequence counting under complex conditions. Summary of the Invention

[0005] This invention aims to solve the problem of misalignment and divergence in the counting of spinal anatomical semantic label sequences under conditions of bone tissue anatomical boundary hyperplasia and adhesion.

[0006] In this technical solution, a data processing method for orthopedic spinal imaging includes the following steps:

[0007] Step S1: Obtain orthopedic spine three-dimensional tomographic image data containing low-dose scanning noise and vertebral deformity areas, and convert it into a spine voxel gray matrix; calculate the local spatial variance of the spine voxel gray matrix to extract the local noise variance weight; add the nominal anatomical boundary threshold to the product of the local noise variance weight and the preset adjustment coefficient in a stepwise adjustment to remove pseudo-edge extreme points in low-dose scanning noise, construct a structured seed set containing discrete nodes, and complete the initialization of the anatomical structure topology state sequence.

[0008] Step S2: Drive the anatomical structure topological state sequence to iteratively step along the structured seed set to map anatomical semantic identifiers and mark discrete anatomical center points section by section; calculate the direction vectors of multiple consecutive discrete anatomical center points determined in the previous mapping to generate dynamic search axes;

[0009] Step S3: The original three-dimensional spatial coordinates are projected in situ onto the following coordinate system that is curved along the anatomical trajectory line of the spine center using the dynamic anatomical spatial coordinate transformation matrix. The local search cone window range of the next vertebral segment is corrected by spatial curvature deformation, the geometric distortion caused by the vertebral body distortion area is corrected, and the search path of the anatomical structure topology state sequence on the dynamic search axis is updated. The topological connection relationship between each discrete node carrying anatomical semantic label is encapsulated into multidimensional anatomical parameter geometric atlas data output.

[0010] Preferably, step S1 includes the following sub-steps: Step S11, calculate the gray variance of each voxel in the three-dimensional neighborhood of the target voxel in the spinal voxel gray matrix, and map the gray variance to the local noise variance weight of the target voxel position; Step S12, when there is low-dose scanning noise in the spinal voxel gray matrix, extract the local noise variance weight and the preset adjustment coefficient, and change the nominal anatomical boundary threshold to the sum of the products of the nominal anatomical boundary threshold, the local noise variance weight and the preset adjustment coefficient, and complete the step adjustment of the nominal anatomical boundary threshold.

[0011] Preferably, step S2 includes the following sub-steps: Step S21, loading a preset topological structure adjacency constraint matrix, which contains the spatial mutual exclusion boundaries and relative connection order data of adjacent vertebrae in normal spinal anatomy; Step S22, during the evolution and flow of the anatomical structure topological state sequence, comparing and verifying the current jump search node with the topological structure adjacency constraint matrix, and assigning a corresponding anatomical semantic identifier label to the jump search node when the jump search node conforms to the anatomical semantic boundary of the topological structure adjacency constraint matrix, thereby completing the constraint control of the flow path of the anatomical structure topological state sequence.

[0012] Preferably, step S3 includes the following sub-steps: Step S31, in the preceding mapping, calculate in real time a plurality of determined discrete anatomical center points. The three-dimensional curvature center is determined, and a dynamic search axis is constructed along the tangent of the three-dimensional curvature center; in step S32, a conical local search cone window range is constructed with the dynamic search axis as the center; when the vertebral body deformity area is identified, the three-dimensional nonlinear spatial deformation distortion local search cone window range is called by the following anatomical space coordinate transformation matrix, so that the corrected local search cone window range is kept to coincide with the nonlinear curved spinal axis.

[0013] Preferably, the step of encapsulating the topological connections between discrete nodes carrying anatomical semantic labels into multidimensional anatomical parameter geometric atlas data output includes the following sub-steps: Step S33, combining the three-dimensional spatial coordinates of each discrete node with the corresponding anatomical semantic labels into structured semantic node data; determining the directed topological edges between each anatomical semantic label based on the topological adjacency constraint matrix, connecting the corresponding structured semantic node data using the directed topological edges, and generating a structured three-dimensional spinal anatomy network atlas; calculating the nominal center distance of the vertebral body, the three-dimensional radius of curvature of scoliosis, and the topological volume parameters of osteophytes on the vertebral margins between adjacent structured semantic node data, and associating the calculated parameters as primitive attributes with the corresponding nodes and edges of the structured three-dimensional spinal anatomy network atlas to complete the construction of multidimensional anatomical parameter geometric atlas data.

[0014] Preferably, step S2 further includes the following sub-steps: Step S23, when the anatomical topological state sequence flows through the vertebral body tomographic imaging region of severe degenerative osteophyte accompanied by joint ankylosis and fusion, the maximum single-step spatial search span of the internal state jump of the anatomical topological state sequence is limited to a closed numerical range of 12mm to 18mm, and the deformation coefficient of the following anatomical spatial coordinate transformation matrix is ​​dynamically adjusted according to the calculated topological adjacency state divergence residual factor, so that the divergence residual factor of the evolution of the anatomical topological state sequence is stabilized within a preset range not exceeding 0.05.

[0015] Preferably, the step between step S2 and step S3 further includes the following steps: Step S101, capturing the nonlinear following delay error of the anatomical structure topological state sequence during the state switching process in real time; calculating the three-dimensional spatial dynamic lag compensation factor based on the nonlinear following delay error; adjusting the transformation operator of the following anatomical spatial coordinate transformation matrix using the feedforward gain of the three-dimensional spatial dynamic lag compensation factor, correcting the spatial deformation correction amount of the local search cone window range online, and compensating for the spatial search axis response lag caused by the state jump.

[0016] Preferably, after step S3, the following steps are also included: Step S102, the multidimensional anatomical parameter geometric atlas data generated by similarity matching is compared with the known standard spinal topology atlas model that conforms to prior clinical anatomical knowledge, and a matching and co-tuned structure is output. When the topological distortion parameter extracted within the matching and co-tuned structure is less than the known compliance judgment threshold, it is determined that the multidimensional anatomical parameter geometric atlas data has no logical breaks, and the multidimensional anatomical parameter geometric atlas data is transmitted to the downstream medical information system network receiving end in a structured medical image stream format.

[0017] Preferably, step S102 includes the following sub-steps: Step S1021, non-destructive information compression of multidimensional anatomical parameter geometric atlas data to encapsulate it into a data packet conforming to the network protocol of medical image archiving and communication system; transmit the data packet to the input end of the clinical decision support system through a known data transmission interface for digital anatomical quantitative measurement of the spine and three-dimensional surgical planning in the clinical decision support system.

[0018] Compared with existing technologies, the data processing method for orthopedic spinal imaging of the present invention has the following advantages:

[0019] 1. In the data processing of orthopedic spinal images, the starting reference is anchored by the geometric centroid of the candidate boundary pixel node set in the spatial three-dimensional voxel gray matrix. The monotonic flow of the finite state machine is controlled to decouple the spatial constraints of bone tissue segment by segment. When the state machine traverses the anatomical candidate center points, the absolute difference between the spatial center distance vector and the nominal distance is extracted. When the absolute difference exceeds the preset step tolerance threshold, the deviation calibration operator inversely retrieves the spatial curvature evolution features of the normal vertebral segment to linearly extrapolate the theoretical coordinates and replace the memory spatial coordinates of the current variant segment. This blocks the abnormal connected branches generated by osteophyte fusion and eliminates the semantic counting cascade misalignment divergence caused by physical connectivity. It achieves smooth convergence of the sequence counting of structured atlas data, thereby directly blocking the logical disorder of the electronic data entry of patient orthopedic medical records in the medical management system caused by anatomical variations. This comprehensively improves the stability and automated interaction efficiency of the medical information system in processing high-noise complex medical images.

[0020] 2. Before constructing the candidate boundary pixel node set, the adaptive noise gating module is started to calculate the transient noise variance parameter by statistically analyzing the local neighborhood variance of the voxel gray matrix. When high-frequency quantum interference is introduced into the tomographic image during low-dose scanning, the adaptive noise gating module adjusts the preset boundary threshold in situ step by step according to the value of the transient noise variance parameter. This filters out pseudo-edge extreme points caused by quantum perturbation from the source, prevents false noise from causing an abnormal surge in the number of nodes during the initialization stage of the state machine, and coordinates the maintenance of the necessary scale of discrete nodes inside the finite state machine to ensure the stable operation of subsequent conditional control flow branching.

[0021] 3. In the monotonic flow of the finite state machine control flow, a follower spatial coordinate transformation matrix is ​​introduced. For the nonlinear torsion of the spinal axis caused by anatomical distortion, the tangent direction vectors of multiple confirmed continuous anatomical center points are extracted in real time during the preceding label generation stage as the latest dynamic forward search axis. The follower spatial coordinate transformation matrix is ​​used to perform three-dimensional nonlinear spatial deformation correction on the local search cone window range of the next vertebral segment to offset geometric distortion and prevent illegal interruption of the traversal path due to lateral curvature. This not only broadens the system's coverage of extremely complex cases, but also enables the state machine to maintain the continuity of data flow under severe deformation conditions. Attached Figure Description

[0022] Figure 1 This is a flowchart of the overall data processing of orthopedic spinal images according to the present invention;

[0023] Figure 2 This is a topological state sequence diagram of orthopedic spinal imaging according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0025] A data processing method for orthopedic spinal imaging includes the following steps:

[0026] Step S1: Obtain orthopedic spine three-dimensional tomographic image data containing low-dose scanning noise and vertebral deformity areas, and convert it into a spine voxel gray matrix; calculate the local spatial variance of the spine voxel gray matrix to extract the local noise variance weight; add the nominal anatomical boundary threshold to the product of the local noise variance weight and the preset adjustment coefficient in a stepwise adjustment to remove pseudo-edge extreme points in low-dose scanning noise, construct a structured seed set containing discrete nodes, and complete the initialization of the anatomical structure topology state sequence.

[0027] Step S2: Drive the anatomical structure topological state sequence to iteratively step along the structured seed set to map anatomical semantic identifiers and mark discrete anatomical center points section by section; calculate the direction vectors of multiple consecutive discrete anatomical center points determined in the previous mapping to generate dynamic search axes;

[0028] Step S3: The original three-dimensional spatial coordinates are projected in situ onto the following coordinate system that is curved along the anatomical trajectory line of the spine center using the dynamic anatomical spatial coordinate transformation matrix. The local search cone window range of the next vertebral segment is corrected by spatial curvature deformation, the geometric distortion caused by the vertebral body distortion area is corrected, and the search path of the anatomical structure topology state sequence on the dynamic search axis is updated. The topological connection relationship between each discrete node carrying anatomical semantic label is encapsulated into multidimensional anatomical parameter geometric atlas data output.

[0029] Preferably, step S1 includes the following sub-steps: Step S11, calculate the gray variance of each voxel in the three-dimensional neighborhood of the target voxel in the spinal voxel gray matrix, and map the gray variance to the local noise variance weight of the target voxel position; Step S12, when there is low-dose scanning noise in the spinal voxel gray matrix, extract the local noise variance weight and the preset adjustment coefficient, and change the nominal anatomical boundary threshold to the sum of the products of the nominal anatomical boundary threshold, the local noise variance weight and the preset adjustment coefficient, and complete the step adjustment of the nominal anatomical boundary threshold.

[0030] Preferably, step S2 includes the following sub-steps: Step S21, loading a preset topological structure adjacency constraint matrix, which contains the spatial mutual exclusion boundaries and relative connection order data of adjacent vertebrae in normal spinal anatomy; Step S22, during the evolution and flow of the anatomical structure topological state sequence, comparing and verifying the current jump search node with the topological structure adjacency constraint matrix, and assigning a corresponding anatomical semantic identifier label to the jump search node when the jump search node conforms to the anatomical semantic boundary of the topological structure adjacency constraint matrix, thereby completing the constraint control of the flow path of the anatomical structure topological state sequence.

[0031] Preferably, step S3 includes the following sub-steps: Step S31, in the preceding mapping, calculate in real time a plurality of determined discrete anatomical center points. The three-dimensional curvature center is determined, and a dynamic search axis is constructed along the tangent of the three-dimensional curvature center; in step S32, a conical local search cone window range is constructed with the dynamic search axis as the center; when the vertebral body deformity area is identified, the three-dimensional nonlinear spatial deformation distortion local search cone window range is called by the following anatomical space coordinate transformation matrix, so that the corrected local search cone window range is kept to coincide with the nonlinear curved spinal axis.

[0032] Preferably, the step of encapsulating the topological connections between discrete nodes carrying anatomical semantic labels into multidimensional anatomical parameter geometric atlas data output includes the following sub-steps: Step S33, combining the three-dimensional spatial coordinates of each discrete node with the corresponding anatomical semantic labels into structured semantic node data; determining the directed topological edges between each anatomical semantic label based on the topological adjacency constraint matrix, connecting the corresponding structured semantic node data using the directed topological edges, and generating a structured three-dimensional spinal anatomy network atlas; calculating the nominal center distance of the vertebral body, the three-dimensional radius of curvature of scoliosis, and the topological volume parameters of osteophytes on the vertebral margins between adjacent structured semantic node data, and associating the calculated parameters as primitive attributes with the corresponding nodes and edges of the structured three-dimensional spinal anatomy network atlas to complete the construction of multidimensional anatomical parameter geometric atlas data.

[0033] Preferably, step S2 further includes the following sub-steps: Step S23, when the anatomical topological state sequence flows through the vertebral body tomographic imaging region of severe degenerative osteophyte accompanied by joint ankylosis and fusion, the maximum single-step spatial search span of the internal state jump of the anatomical topological state sequence is limited to a closed numerical range of 12mm to 18mm, and the deformation coefficient of the following anatomical spatial coordinate transformation matrix is ​​dynamically adjusted according to the calculated topological adjacency state divergence residual factor, so that the divergence residual factor of the evolution of the anatomical topological state sequence is stabilized within a preset range not exceeding 0.05.

[0034] Preferably, the step between step S2 and step S3 further includes the following steps: Step S101, capturing the nonlinear following delay error of the anatomical structure topological state sequence during the state switching process in real time; calculating the three-dimensional spatial dynamic lag compensation factor based on the nonlinear following delay error; adjusting the transformation operator of the following anatomical spatial coordinate transformation matrix using the feedforward gain of the three-dimensional spatial dynamic lag compensation factor, correcting the spatial deformation correction amount of the local search cone window range online, and compensating for the spatial search axis response lag caused by the state jump.

[0035] Preferably, after step S3, the following steps are also included: Step S102, the multidimensional anatomical parameter geometric atlas data generated by similarity matching is compared with the known standard spinal topology atlas model that conforms to prior clinical anatomical knowledge, and a matching and co-tuned structure is output. When the topological distortion parameter extracted within the matching and co-tuned structure is less than the known compliance judgment threshold, it is determined that the multidimensional anatomical parameter geometric atlas data has no logical breaks, and the multidimensional anatomical parameter geometric atlas data is transmitted to the downstream medical information system network receiving end in a structured medical image stream format.

[0036] Preferably, step S102 includes the following sub-steps: Step S1021, non-destructive information compression of multidimensional anatomical parameter geometric atlas data to encapsulate it into a data packet conforming to the network protocol of medical image archiving and communication system; transmit the data packet to the input end of the clinical decision support system through a known data transmission interface for digital anatomical quantitative measurement of the spine and three-dimensional surgical planning in the clinical decision support system.

[0037] Example 1: In the data analysis scenario of lumbar spine tomographic images containing degenerative osteophyte ankylosis, automated identification of spinal segment geometry is crucial for surgical planning. The system receives three-dimensional tomographic image data acquired by CT equipment. The image data includes multiple vertebral bodies and fused areas of osteophytes and ankylosis between adjacent vertebral bodies. The processor maps the three-dimensional tomographic image data into a spatial three-dimensional voxel grayscale matrix. The processor processes the matrix Perform differential operations to calculate the three-dimensional gradient vector field for each voxel, and filter those whose absolute gradient values ​​are greater than a preset boundary threshold. voxel points, construct candidate boundary pixel node set The system loads the preset adjacency matrix of the spinal anatomy. Topological adjacency matrix Pre-defined linear cascade topological relationships between segments of the cervical, thoracic, and lumbar vertebrae, as well as the nominal spacing between corresponding segments. The processor uses candidate boundary pixel node sets The spatial geometric centroid serves as the initial anchor point, driving the state machine logic flow in accordance with the longitudinal anatomical axis of the spine.

[0038] The state machine logic flow traverses and dissects candidate center points At that time, calculate the current candidate center point of the anatomy. The anatomical center point has been marked in the previous step. The spatial center distance vector between The processor calculates the spatial center distance vector. The modulus and the nominal spacing of the corresponding segment The absolute value of the difference, and real-time monitoring of the topological adjacency state divergence residual factor. , The calculation formula is as follows: ,in, The residual factor for the divergence of topological adjacency states. The magnitude of the spatial center distance vector. Nominal spacing, These are the standard spatial variance calibration coefficients for the current spinal segment. The values ​​are derived from statistical normalization processing of a large-scale clinical dataset of three-dimensional spinal images of healthy individuals. In practice, for different anatomical locations such as the cervical, thoracic, and lumbar vertebrae, the values ​​of healthy vertebrae are pre-calculated. The mean standard deviation of the voxel distribution within the neighborhood, for example, when dealing with lumbar segments. The value is typically between 0.8 mm and 1.2 mm. This coefficient serves as the normalized denominator for calculating the divergence residual factor of the topological adjacency state. Its physical meaning lies in providing a standardized metric for measuring the degree of anatomical structural variation, ensuring that the degree of divergence under different bone density backgrounds can be compared laterally on the same dimensionless numerical scale.

[0039] If the processor performs a branch decision, and the absolute value of the difference is less than or equal to the preset step tolerance threshold... And set The processor will measure the current candidate center point of the anatomy at 2.5mm. If a structure is determined to conform to standard anatomical structures, and based on its corresponding anatomical semantic tags, the absolute value of the difference exceeds a preset step tolerance threshold, then... The system enters an abnormal branch. At this point, the deviation calibration operator is triggered. The system extracts the spatial curvature evolution gradient of the three consecutive anatomical center points that have been previously calibrated, calculates the coordinates of the theoretical anatomical center point of the current segment through linear extrapolation, and sets the current candidate anatomical center point as the... The spatial coordinates are forcibly rewritten to the coordinates of the theoretical anatomical center point to achieve convergence of anatomical semantic labels. After processing all anatomical segments, the system encapsulates the connection relationships of each node carrying converged anatomical semantic labels into structured multidimensional parametric geometric atlas data of the spine and transmits the data to the clinical surgical planning terminal. The above closed-loop processing prevents anatomical segment counting errors caused by local pixel adhesion due to osteophyte ankylosis, ensuring the structural integrity of the output atlas data.

[0040] Example 2: This example aims to verify the structural reconstruction capability of the present invention when processing images of degenerative spinal diseases through experiments. The experimental platform integrates a 1.2GHz CPU, 4GB cache, and an image processing unit. By simulating the computing load environment of a clinical imaging workstation, anatomical segmental structure analysis is performed on lumbar spine CT images with osteophyte ankylosis. The experimental sample consists of a dataset of 32 vertebral body tomographic images of degenerative osteophyte formation accompanied by joint ankylosis and fusion. During processing, the system first calculates the local neighborhood variance of the voxel gray matrix and extracts the local noise variance weight to dynamically correct the nominal anatomical boundary threshold. This filters out spurious edge extrema points generated by low-dose scanning noise. When determining the boundary threshold, the experiment sets... The value is 182.4. Real-time monitoring of the noise variance ensures... The system automatically compensates for fluctuations in background noise levels in the image, eliminating noise-induced deviations in anatomical node identification. The system drives the state machine logic flow to step along the topological state sequence of the anatomical structure, calibrating the anatomical center point and generating a dynamic search axis. The processor calculates the center distance vector. Nominal spacing The differences are used to monitor the divergence residual factor of the topological adjacency state. , The calculation formula is as follows: ,in, The residual factor for the divergence of topological adjacency states. The magnitude of the spatial center distance vector. This represents the nominal spacing of the corresponding segments. Set the spatial variance calibration coefficients for the current segment. It is 2.5mm, if If the corresponding value exceeds the threshold, the deviation calibration operator uses the spatial curvature evolution gradient linear extrapolation theoretical coordinates of the preceding three normal vertebral segments to correct the geometric distortion of the current osteophyte fusion segment.

[0041] To verify the effectiveness of the technology, an experiment was conducted with the present invention sample group, control sample group A, and control sample group B for performance comparison. The present invention sample group adopted the complete steps described above; control sample group A removed the adaptive noise gating module and directly performed boundary screening on the original matrix; control sample group B... The test results, set at 5.0 mm, show that the sample group of this invention, when traversing the ankylosis region of 5 segments, The maximum value is 0.04, the average positioning error is 0.45mm, there is no counting misalignment, and the control sample group A has invalid points mixed into the candidate boundary pixel node set due to noise interference, resulting in a count value. The maximum value rose to 0.18, with 9 samples showing missed or miscounted counts. Control group B, due to... Exceeding the reasonable range led to misjudgment of osteophyte edges, increasing the average positioning error to 2.15 mm, and anatomical semantic label drift occurred in 11 samples; experimental data records: the average positioning error of the sample group of this invention was 0.45 mm, with a counting misalignment rate of 0%; the average positioning error of control sample group A was 1.25 mm, with a counting misalignment rate of 28.1%; and the average positioning error of control sample group B was 2.15 mm, with a counting misalignment rate of 34.4%. The data shows that adjusting the boundary threshold by local noise variance weighting effectively suppressed pseudo-edge interference; and further... Constrained within a physical range of 2.5 mm, and in conjunction with the extrapolation mechanism of the deviation calibration operator, effective correction of geometric distortion of anatomical structures in orthopedic spinal images was achieved. The performance indicators of the control group deteriorated, objectively verifying the synergistic effect of the various technical features of this invention in processing complex orthopedic images and the critical effectiveness of the numerical constraints.

[0042] Example 3: Current orthopedic surgical planning scenarios involve image reconstruction tasks for degenerative scoliosis accompanied by vertebral osteophyte lesions. Such lesions often exhibit nonlinear fusion of anatomical boundaries in multi-slice tomographic images, causing conventional edge detection algorithms to fail and resulting in uncontrollable divergence in the sequence count of anatomical semantic labels. This example constructs an adaptive defensive reconstruction method based on spatial anatomical constraints to solve the problem of segment identification accuracy in osteophyte fusion regions.

[0043] The processor receives three-dimensional orthopedic spinal images and converts them into voxel grayscale matrices. To characterize the complexity of the anatomical features of the lesion region, the system quantifies the degree of morphological nonlinearity of the lesion region by calculating the second-order spatial derivative of the voxel gray matrix, and constructs defense boundary parameters. , The calculation formula is as follows: ,in, To dissect the boundary gradient defense threshold, The second derivative response coefficient is 0.85. Let be the second-order rate of change of the three-dimensional voxel grayscale matrix along the spatial axes. The grayscale baseline offset correction value for the lesion area is set to 1.25. When processing image sequences containing segmental fusion, conventional boundary screening algorithms cannot distinguish the true anatomical interface due to the high overlap in gradient distribution between lesion osteophytes and normal vertebral margins. The system enables anatomical constraint defense logic, mapping the gradient features of the candidate node set to the defense boundary parameters. The processor monitors the changing trends of morphological features of candidate nodes in real time. When the value of a local morphological deviation feature changes abruptly and crosses... When the threshold is set, if the recognition result is determined to be a lesion with geometrical distortion, the system will automatically switch to the extrapolation mode of the deviation calibration operator.

[0044] The deviation calibration operator extracts the spatial curvature evolution features of the three preceding normal anatomical segments unaffected by lesions. It then calculates the coordinates of the theoretical anatomical center point of the current segment via linear extrapolation. The system rewrites the spatial coordinates of the current candidate anatomical center point as the coordinates of the theoretical anatomical center point and reconstructs the anatomical semantic label of the segment using this pose. Clinical data validation shows that, on a test dataset including segment fusion, after employing the deviation calibration defense mechanism, the divergence residual factor of the spinal anatomical semantic label count is significantly reduced. Maintaining a value below 0.05, the average positioning error decreased from 3.8 mm in the control group to 0.75 mm, blocking the anatomical reconstruction distortion caused by diseased osteophytes and achieving stable output of spinal geometric atlas data under high deformation conditions.

[0045] Example 4: In clinical practice of digital processing of spinal images, it is necessary to process complex data where the vertebral anatomical features are discontinuous due to degenerative changes. The system receives three-dimensional tomographic image data and converts the voxel grayscale matrix... The data is stored in the system's high-speed memory space, and the initial calibration of the anatomical semantic labels of the spinal segments is performed based on the prior geometric constraints of the spinal anatomy. The system uses an adaptive spatial geometric constraint model to initialize the anatomical semantic label flow of the anatomical structure. The processor initializes the flow of anatomical semantic labels based on the preset average height of the spinal segments. Before processing image data, the system determines the longitudinal search step length benchmark for the spine based on preset baseline environmental parameters. An initial response deviation correction library for the deviation calibration operator was established. This benchmark library was obtained by preprocessing 200 spinal tomographic images from a clinical sample set to ensure that the system has stability for different bone density distributions from the initial deployment stage.

[0046] When processing spinal computed tomography images containing lesions, the processor calculates the anatomical center point of the current segment. With respect to the theoretical anatomical center point Instantaneous spatial deviation value and will As a benchmark parameter for evaluating reconstruction accuracy, the processor determines... Is it within the normal response range preset by the calibration operator? Within the range, among which, To ensure safe dissection of geometric deformation thresholds, the processor monitors the geometric deformation characteristics generated in the lesion area. consecutively exceeding At that time, the system automatically switches the anatomical search window to the pose correction path based on the compensation operators pre-stored in the deviation correction library, correcting the data deviation caused by the divergence of the anatomical semantic label sequence count, thereby achieving stable convergence of the anatomical semantic label sequence. Before outputting the multidimensional anatomical parameter geometric atlas data, the system performs de-identification processing on the image data. The processor performs hash mapping processing on the patient identification information to ensure that the output geometric parameter atlas only contains skeletal anatomical geometric morphology indicators, thereby achieving secure isolation of patient privacy data while ensuring the availability of clinical surgical planning data.

[0047] Example 5: Before performing anatomical reconstruction, current spinal computed tomography (CT) image processing systems require a standardized offline calibration process to eliminate geometric resolution biases caused by differences in imaging hardware and environmental disturbances. The system constructs an anatomical baseline database containing 500 normal bone morphology samples to define the physical baseline of the anatomical geometry of spinal segments. The system reads the scanning parameters of the current clinical imaging equipment, including slice thickness. With scanning dose coefficient Through a preset physical mapping function The scanning parameters are normalized to device-independent standardized physical dimensions. The calculation formula is as follows: ,in, For standardized physical conversion factors, The response constant of the imaging system is calibrated to a value of 1.25. This is the actual scanned layer thickness. It is the product of tube current and dose. This is the environmental noise compensation constant, with a value of 0.45.

[0048] processor based The pixel grayscale matrix of the spinal tomographic images is adjusted to map the grayscale distribution from different devices to a standard anatomical control space. The system establishes an initial anatomical search axis, starting from the centroid coordinates of the first lumbar vertebra and extending along the spatial geometric curvature evolution characteristics. The processor calculates the candidate anatomical center point for the current segment. With respect to the theoretical anatomical center point Spatial deviation vector between If the magnitude of the vector exceeds the preset decoupling threshold... The system enters the abnormal branch calibration process, and the processor calculates the projected area of ​​osteophytes in the lesion region. And use this as the parameter weight for the deviation correction logic: ,in, This represents the total area of ​​projected pixels in the osteophyte region. It is the set of spatial pixel neighborhoods of the locally fused segments; The voxel gray value at the coordinate position; The calibrated anatomical boundary segmentation threshold.

[0049] processor based The deviation difference from the preset segment decoupling criterion is used to update the offset correction amount in real time. The system will The pose transformation matrix acting on the anatomical search axis, through coordinate translation and rotation correction, converges the anatomical center point of the lesion segment to the theoretical trajectory. Under the action of this calibration procedure, when the system processed images of 30 osteophyte fusion lesions, the segment positioning error was reduced from 2.1 mm in the original scan to 0.35 mm. The above procedure constructs a physical alignment link for imaging parameters and eliminates the uncertainty of geometric reconstruction caused by environmental factors.

[0050] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A data processing method for orthopedic spinal imaging, characterized in that, Includes the following steps: Step S1: Obtain orthopedic spine three-dimensional tomographic image data containing low-dose scanning noise and vertebral deformity areas, and convert it into a spinal voxel grayscale matrix. The local spatial variance of the spinal voxel gray matrix is ​​statistically analyzed to extract the local noise variance weights. The nominal anatomical boundary threshold is accumulated by multiplying the local noise variance weight and the preset adjustment coefficient in steps to remove false edge extreme points in low-dose scanning noise, construct a structured seed set containing discrete nodes, and complete the initialization of the flow of the anatomical structure topology state sequence. Step S2: Drive the anatomical structure topological state sequence to iteratively step along the structured seed set to map anatomical semantic identifiers and mark discrete anatomical center points section by section; calculate the direction vectors of multiple consecutive discrete anatomical center points determined in the previous mapping to generate dynamic search axes; Step S3: The original three-dimensional spatial coordinates are projected in situ onto the following coordinate system that is curved along the anatomical trajectory line of the spine center using the following anatomical spatial coordinate transformation matrix. The local search cone window range of the next vertebral segment is corrected by spatial curvature deformation, the geometric distortion caused by the vertebral body deformity area is corrected, and the search path of the anatomical structure topology state sequence on the dynamic search axis is updated. The topological connections between discrete nodes carrying anatomical semantic labels are encapsulated into multidimensional anatomical parametric geometric atlas data output.

2. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11, calculate the gray variance of each voxel in the three-dimensional neighborhood of the target voxel in the spinal voxel gray matrix, and map the gray variance to the local noise variance weight of the target voxel position. Step S12: When low-dose scanning noise exists in the spinal voxel grayscale matrix, extract the local noise variance weight and the preset adjustment coefficient, and change the nominal anatomical boundary threshold to the sum of the products of the nominal anatomical boundary threshold, the local noise variance weight, and the preset adjustment coefficient, thus completing the step adjustment of the nominal anatomical boundary threshold.

3. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21, load the preset topology adjacency constraint matrix, which contains the spatial mutual exclusion boundaries and relative connection order data of the upper and lower adjacent vertebral segments of the normal spinal anatomy. Step S22: During the evolution and flow of the anatomical structure topology state sequence, the current jump search node is compared and verified with the topology adjacency constraint matrix. When the jump search node conforms to the anatomical semantic boundary of the topology adjacency constraint matrix, the corresponding anatomical semantic identifier label is assigned to the jump search node to complete the constraint control of the anatomical structure topology state sequence flow path.

4. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, Step S3 includes the following sub-steps: Step S31, in the preceding mapping, calculate in real time multiple consecutive discrete anatomical center points that have been determined. The three-dimensional curvature center is determined, and a dynamic search axis is constructed along the tangent of the three-dimensional curvature center; in step S32, a conical local search cone window range is constructed with the dynamic search axis as the center; when the vertebral body deformity area is identified, the three-dimensional nonlinear spatial deformation distortion local search cone window range is called by the following anatomical space coordinate transformation matrix, so that the corrected local search cone window range is kept to coincide with the nonlinear curved spinal axis.

5. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, The steps for encapsulating the topological connections between discrete nodes carrying anatomical semantic labels into multidimensional anatomical parameter geometric atlas data output include the following sub-steps: Step S33, combining the three-dimensional spatial coordinates of each discrete node with the corresponding anatomical semantic labels into structured semantic node data; determining the directed topological edges between each anatomical semantic label based on the topological adjacency constraint matrix, and using the directed topological edges to connect the corresponding structured semantic node data to generate a structured three-dimensional spinal anatomy network atlas; calculating the nominal center distance of the vertebral body, the three-dimensional radius of curvature of scoliosis, and the topological volume parameters of vertebral marginal osteophytes between adjacent structured semantic node data, and associating the calculated parameters as primitive attributes with the corresponding nodes and edges of the structured three-dimensional spinal anatomy network atlas to complete the construction of multidimensional anatomical parameter geometric atlas data.

6. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, Step S2 also includes the following sub-steps: Step S23, when the anatomical topological state sequence flows through the vertebral body tomographic imaging region of severe degenerative osteophyte formation accompanied by joint ankylosis and fusion, the maximum single-step spatial search span of the internal state jump of the anatomical topological state sequence is limited to a closed numerical range of 12mm to 18mm. Based on the calculated topological adjacency state divergence residual factor, the deformation coefficient of the following anatomical spatial coordinate transformation matrix is ​​dynamically adjusted so that the divergence residual factor of the evolution of the anatomical topological state sequence is stabilized within a preset range not exceeding 0.

05.

7. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, Between steps S2 and S3, the following steps are also included: Step S101, capturing the nonlinear following delay error of the anatomical structure topology state sequence during state switching in real time; calculating the three-dimensional spatial dynamic lag compensation factor based on the nonlinear following delay error; adjusting the transformation operator of the following anatomical spatial coordinate transformation matrix using the feedforward gain of the three-dimensional spatial dynamic lag compensation factor, correcting the spatial deformation correction amount of the local search cone window range online, and compensating for the spatial search axis response lag caused by state jump.

8. The data processing method for orthopedic spinal imaging according to claim 1, characterized in that, Following step S3, the following steps are also included: Step S102, the multidimensional anatomical parameter geometric atlas data generated by similarity matching is compared with a known standard spinal topology atlas model that conforms to prior clinical anatomical knowledge, and a matching and co-tuned structure is output. When the topological distortion parameter extracted within the matching and co-tuned structure is less than the known compliance judgment threshold, it is determined that the multidimensional anatomical parameter geometric atlas data has no logical breaks, and the multidimensional anatomical parameter geometric atlas data is transmitted to the downstream medical information system network receiving end in a structured medical image stream format.

9. A data processing method for orthopedic spinal imaging according to claim 8, characterized in that, Step S102 includes the following sub-steps: Step S1021, non-destructive information compression of multidimensional anatomical parameter geometric atlas data to encapsulate it into a data packet conforming to the network protocol of medical image archiving and communication system; transmit the data packet to the input end of the clinical decision support system through a known data transmission interface for digital anatomical quantitative measurement of the spine and three-dimensional surgical planning in the clinical decision support system.

Images