Image data dynamic feature ai modeling method and system for spondylolisthesis diagnosis

Abstract

The application discloses an image data dynamic feature AI modeling method and system for spondylolysis diagnosis, and relates to the technical field of spondylus image detection. The image data dynamic feature AI modeling method and system for spondylolysis diagnosis, comprising S1, collecting and preprocessing geometric feature data and displacement feature data in spondylus images, and constructing a standardized spondylus state data set; S2, evaluating the coordination of the instantaneous coordination state of each spondylus, generating a local structure symmetry feature and a stability sample set; S3, performing trend analysis on the structure evolution trend of the spondylus, and identifying a difference-enhanced area; S4, performing risk assessment on the segment-level risk intensity of the spondylus, and dynamically adjusting the atlas structure presentation mode based on the risk assessment result. The problems that existing research focuses on spondylus static variation identification and cannot depict the dynamic evolution path of the spondylolysis state are solved.

Description

Technical Field

[0001] This invention relates to the field of vertebral image detection technology, specifically to an AI modeling method and system for dynamic features of image data for vertebral misalignment diagnosis. Background Technology

[0002] With the continuous improvement of resolution and dynamic acquisition capabilities of clinical imaging equipment, medical image analysis has been widely applied to the auxiliary diagnosis and preoperative assessment of various spinal structural abnormalities. Current mainstream diagnostic methods typically rely on manual identification of vertebral body boundaries and structural relationships in static image sequences, assessing changes in the physiological curvature of the spine and segmental misalignment through two-dimensional measurement, centerline fitting, and angle calculation. Building upon this, some auxiliary tools integrate image segmentation algorithms based on template matching and grayscale features to semi-automatically extract vertebral segment contours, center points, and spatial relative positions, improving positioning accuracy and reducing manual intervention.

[0003] For example, invention patent CN115330753B discloses a vertebral identification method, device, equipment, and storage medium to improve the identification speed and accuracy of vertebral identification. The method includes: retargeting and dividing spinal image data into image block grids to obtain spinal slice images and multiple image blocks; calculating the local maximum density projection of the spinal slice images based on the spinal bounding boxes to obtain a slice local maximum density projection map; generating multiple image block maximum density projection maps based on the slice local maximum density projection map; inputting these multiple maximum density projection maps into a vertebral identification and localization model for vertebral identification and localization to obtain a two-dimensional Gaussian heat map group; calculating a target three-dimensional Gaussian heat map group based on the two-dimensional Gaussian heat map group; calculating the vertebral body center of the target three-dimensional Gaussian heat map group to generate vertebral identification results, wherein the vertebral identification results include: the vertebral category and center coordinates of each vertebra.

[0004] For example, the invention patent with announcement number CN112381805B discloses a medical image processing method, including: acquiring rib points in 3D medical images; determining a center point (x0, y0), where x0 is associated with the x-coordinate of the rib point and y0 is associated with the y-coordinate of the rib point; acquiring vertebral information in any frame of 2D medical image, the vertebral information including: the direction of the straight line from the center point to each vertebral point in the frame of 2D medical image, the Z-coordinate of the vertebral point, and the vertebral point being located on a first line segment or on a second line segment parallel to the first line segment and at a preset distance from the first line segment, the first line segment passing through the center of the vertebral detection frame of the frame of 2D medical image and parallel to the X-axis; mapping the vertebral information from multiple frames of 2D medical images to a first coordinate system to obtain medical images corresponding to the multiple frames of 2D medical images. This invention improves mapping efficiency and accuracy while also facilitating doctors' image reading and diagnosis, thereby improving the efficiency and accuracy of diagnosis to a certain extent.

[0005] However, in dynamic image data analysis scenarios, the structural evolution of the spine during functional movement cycles exhibits significant temporal and stage-specific characteristics. Vertebral misalignment trends often manifest as complex evolutionary patterns such as gradual, periodic misalignment, and cross-segmental linkages. Traditional static image analysis methods struggle to capture their true motion trajectories and evolutionary paths. Existing dynamic image analysis methods mostly employ keyframe extraction and feature comparison to model structural changes. Some studies have introduced classic motion estimation algorithms such as optical flow and feature point tracking to obtain motion features such as the center position, rotation angle, and displacement direction of vertebral segments between consecutive frames. However, these methods generally suffer from problems such as single feature responses, weak tracking stability, and difficulty in cross-frame structural relocalization. Furthermore, they lack the ability to abstract high-order semantics of vertebral evolution trends, making it difficult to establish a complete model of misalignment evolution paths. Simultaneously, existing analysis processes largely rely on preset rules and manual screening criteria, lacking proactive identification of abnormal evolutionary patterns and a mechanism for quantifying risk levels. This prevents the early detection of potential structural instability trends, limiting their application in disease early warning and postoperative rehabilitation tracking.

[0006] To address the above issues, there is an urgent need for AI modeling methods and systems for dynamic features of imaging data in the diagnosis of vertebral misalignment. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an AI modeling method and system for dynamic features of imaging data for vertebral misalignment diagnosis, which solves the problem that existing research focuses on identifying static variations in vertebrae and cannot depict the dynamic evolution path of misalignment.

[0008] Technical solution To achieve the above objectives, this invention provides the following technical solution: an AI modeling method and system for dynamic features of image data for vertebral misalignment diagnosis, comprising: S1, collecting geometric and displacement feature data from vertebral images, preprocessing the collected geometric and displacement feature data to construct a standardized vertebral segment state dataset; S2, based on the standardized vertebral segment state dataset, evaluating the instantaneous collaborative state of each vertebral segment, and generating a local structural symmetry feature and stability sample set based on the collaborative evaluation results; S3, based on the standardized vertebral segment state dataset, performing trend analysis on the structural evolution trend of vertebral segments from the angle changes and spatial position offset states of each vertebral segment in multiple frames of images, and identifying regions of differential enhancement based on the trend analysis results; S4, combining the trend analysis results of vertebral segments to perform segmental-level risk assessment, and dynamically adjusting the atlas structure presentation method based on the risk assessment results.

[0009] Furthermore, the specific steps for acquiring geometric and displacement feature data from vertebral images are as follows: acquiring geometric feature data during the change of vertebral segment posture, including: rotation angle value and principal axis direction angle value of each vertebral segment in each frame image, and simultaneously recording the time index of each frame image; acquiring displacement feature data during the dynamic change of the vertebral segment center point, including: vertebral body midline longitudinal coordinate, structural center transverse coordinate value and structural center longitudinal coordinate value of each vertebral segment in each frame image, vertebral segment height, and center point coordinates.

[0010] Further, the specific steps for preprocessing the collected geometric feature data and displacement feature data to construct a standardized vertebral segment state dataset are as follows: For the geometric feature data during the change of vertebral segment posture, the rotation angle value and principal axis direction angle value of each vertebral segment in each frame image are obtained, and the time index of the corresponding frame is recorded synchronously. The geometric feature data is then sorted according to the time index. For the displacement feature data during the dynamic change of the vertebral segment center point, the ordinate of the vertebral body midline, the transverse coordinate value and longitudinal coordinate value of the structural center of each vertebral segment, the vertebral segment height, and the complete center point coordinates are obtained in each frame image. The absolute value of the difference between the longitudinal coordinate of the structural center and the midline is calculated based on the difference in the ordinate, and recorded as the horizontal offset value of the center point. Gray-level analysis is performed on the gray-level distribution of the left and right sides of the same vertebral segment in the image to obtain the gray-level difference between the left and right sides of the vertebral segment, which is recorded as the vertebral segment edge difference value. Based on the vertebral segment edge difference value of each vertebral segment, a vertebral segment edge gray-level change sequence is constructed. The standardized data is then processed. The processed geometric feature data and displacement feature data are normalized to construct a standardized vertebral segment state dataset. The center point coordinates of the target vertebra and the next vertebra in the same frame are obtained, and the displacement distance between adjacent vertebrae is obtained using the spatial Euclidean algorithm, which is recorded as the center point distance. The center point coordinates of the current vertebra in adjacent frames are obtained, and the path angle change value is calculated by the angle between the lines connecting the front and rear center points. The number of times the path angle change value of the same vertebra exceeds the angle change threshold during continuous posture changes is recorded as the vertebral active offset value. The center point coordinates of the same vertebra in all frames are obtained to construct a center point coordinate sequence. The center point distance between frames is calculated using the Euclidean algorithm and summed to obtain the total path length. The difference between the center point distances of the vertebra in the start and end frames is calculated to obtain the functional position distance. The vertebral height of the same vertebra in all frames is obtained to construct a vertebral height sequence. The maximum and minimum vertebral height values ​​in the vertebral height sequence are extracted, and the difference is calculated and recorded as the vertebral height difference value.

[0011] Furthermore, the specific steps for evaluating the synergistic state of each vertebra based on the standardized vertebral segment state dataset are as follows: Calculate the absolute value of the difference between the rotation angle value of the current vertebral segment and the rotation angle value of the next vertebral segment to obtain the rotation inconsistency value; calculate the difference between the transverse coordinate value of the structural center of the current vertebral segment and the transverse coordinate value of the center of the next vertebral segment, and take the square value to obtain the transverse offset value; calculate the difference between the longitudinal coordinate value of the structural center of the current vertebral segment and the longitudinal coordinate value of the center of the next vertebral segment, and take the square value to obtain the longitudinal offset value; add the transverse offset value and the longitudinal offset value and take the square root to obtain the relative position offset value; take the negative value of the rotation inconsistency value and substitute it into the exponential function to obtain the rotation inconsistency index; take the negative value of the relative position offset value and substitute it into the exponential function to obtain the relative position offset index; multiply the rotation inconsistency index and the relative position offset index to obtain the rotational lateral flexion synergistic value.

[0012] Further, the specific steps for generating a local structural symmetry feature and stability sample set based on the synergy assessment results are as follows: Real-time comparison of the rotational flexion synergy value and rotational flexion synergy threshold of the current vertebral segment: When the rotational flexion synergy value is less than or equal to the rotational flexion synergy threshold, the triangular annotation region of the current frame image of the vertebral segment is extracted, the ratio of the length of the three sides of the structure formed by the upper edge of the vertebral segment and the corner point of the pedicle is extracted, and the side length difference is compared with the corresponding regions of the upper and lower vertebral segments. At the same time, the attitude direction vector of the current frame image is superimposed, and the vector offset direction of the segment is extended to the upper and lower two vertebral segments to construct a multi-segment direction switching matrix for identifying the local rotation axis offset point; When the rotational flexion synergy value is greater than the rotational flexion synergy threshold, the extension region of the principal axis projection line of the current vertebral segment in the current frame and the previous frame image is read, the gray-scale change sequence of the vertebral segment edge in the horizontal region of the image is extracted, and the current vertebral segment region is cropped into equal-height rectangular blocks to generate a structural stability texture reference sample sequence.

[0013] Furthermore, the specific steps for trend analysis of the structural evolution trend of vertebral segments based on the standardized vertebral segment state dataset and the angular changes and spatial position shifts of each vertebral segment in multi-frame images are as follows: Calculate the absolute value of the difference between the principal axis angle values ​​of the current vertebral segment in the previous and next frames of the current frame, and divide this by the difference between the time index of the next frame and the time index of the previous frame to obtain the angle change rate; add one to the absolute value of the difference between the principal axis angle value of the current vertebral segment in the current frame and the principal axis angle value in the initial frame, and take the reciprocal. The angle offset suppression value is obtained; the difference between the center point distance in the previous frame and the center point distance in the next frame is calculated, divided by the difference between the time index of the next frame and the time index of the previous frame, and the absolute value is taken to obtain the position change rate; the center point distance in the current frame is subtracted from the center point distance in the initial frame, divided by the center point distance in the initial frame, and the absolute value is taken to obtain the relative offset ratio of the structure; the product of the angle change rate and the angle offset suppression value is added to the product of the position change rate and the relative offset ratio of the structure to obtain the staggered seam evolution trend value.

[0014] Further, the specific steps for identifying the differential enhancement region based on the trend analysis results are as follows: Construct a sequence of vertebral segment misalignment evolution trend values ​​based on the misalignment evolution trend values ​​of n consecutive frames, and calculate the local change slope and fluctuation amplitude of the misalignment evolution trend values; if the misalignment evolution trend value of the current frame is greater than the previous m frames and less than the following m frames, perform a bidirectional temporal comparison, and interpolate and reconstruct the vertebral segment principal axis angles in the previous and following m frames; if the misalignment evolution trend value of the current frame is the maximum value among the n consecutive frames, extract the displacement direction of the segment's center point in the current frame and... If the path angle change value is greater than the angle offset threshold, the vertebral segment boundary region is expanded along the main axis in the current frame and grayscale stretching is performed to enhance the clarity of the structural outline. If the current frame misalignment evolution trend value is the minimum value among n consecutive frames, the misalignment evolution trend value sequence of adjacent upper and lower vertebral segments is extracted with the current vertebral segment as the center, and the ranking of the misalignment evolution trend value sequence between vertebral segments is calculated. If the current vertebral segment has the lowest ranking, the complete vertebral segment structure block where the vertebral segment is located is cropped from the original image and recorded as a stable feature reference sample.

[0015] Furthermore, the specific steps for assessing the segmental risk intensity of vertebrae by combining the trend analysis results of vertebrae are as follows: The cumulative path offset value is obtained by taking the total path length of the current vertebrae as the numerator and the absolute value of the difference between the total path length of the vertebrae and the functional position distance plus one as the denominator; the vertebral height difference value of the current vertebrae is divided by the sum of the vertebral height of the initial frame and one, and then one is added to obtain the vertebral deformation amplification value; the active vertebral offset value is multiplied by the square of the horizontal offset value of the center point, divided by the sum of the vertebral edge difference value and one, and then multiplied by the misalignment evolution trend value to obtain the structural disturbance enhancement value; the sum of the path angle change value and the cumulative path offset value, the vertebral deformation amplification value, and the structural disturbance enhancement value are multiplied to obtain the dynamic map risk value.

[0016] Further, the specific steps for dynamically adjusting the presentation method of the atlas structure based on the risk assessment results are as follows: Real-time comparison of the current dynamic atlas risk value with the atlas risk threshold, which includes a first risk threshold and a second risk threshold: When the dynamic atlas risk value is less than or equal to the second risk threshold, mark the upper and lower edge coordinates of the vertebral segment in the current frame, construct an equal-width rectangular region and divide it into grayscale texture slices by column, compressing and storing it as a stable structural texture sample; When the dynamic atlas risk value is greater than the second risk threshold and less than or equal to the first risk threshold, perform piecewise interpolation fitting on the motion path of the vertebral segment's center point, and simultaneously extract the vertebral segment height from the starting frame to the current frame to construct a sequence of vertebral segment height difference changes; When the dynamic atlas risk value is greater than the first risk threshold, extract the extended neighborhood region of the vertebral segment based on its own vertebral height from the current frame, reconstruct the central principal axis vector and calculate the angle offset between the central principal axis vector and the midline of the spine, and input the cropped image of the extended neighborhood region as a high-risk structural fragment into the dynamic evolution risk atlas of the vertebral segment for thermal rendering.

[0017] The second aspect of this invention provides an AI modeling method for dynamic features of imaging data for vertebral misalignment diagnosis, comprising: a temporal image data acquisition module for acquiring geometric and displacement feature data from vertebral images, and preprocessing the acquired geometric and displacement feature data to construct a standardized vertebral segment state dataset; a vertebral body coordinated motion assessment module for evaluating the instantaneous coordinated state of each vertebra based on the standardized vertebral segment state dataset, and generating a local structural symmetry feature and stability sample set based on the coordinated assessment results; a misalignment evolution trend extraction module for analyzing the structural evolution trend of vertebral segments based on the angular changes and spatial position offsets of each vertebra in multiple frames of images, and identifying regions of differential enhancement based on the trend analysis results; and an evolution risk output module for assessing the segmental risk intensity of vertebral segments based on the trend analysis results, and dynamically adjusting the atlas structure presentation method based on the risk assessment results.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) The AI ​​modeling method and system for dynamic features of image data for vertebral misalignment diagnosis realizes continuous modeling of vertebral spatial displacement and posture changes by introducing a dynamic tracking mechanism of the center point of the vertebral segment and the structural boundary point in multiple frames of images.

[0019] (2) The AI ​​modeling method and system for dynamic features of imaging data for vertebral misalignment diagnosis improves the accuracy of automatic positioning of abnormal segments during motion by establishing a continuous inter-frame vertebral structure similarity analysis mechanism.

[0020] (3) The AI ​​modeling method and system for dynamic features of imaging data for vertebral misalignment diagnosis improves the ability to perceive and judge dynamic morphological abnormalities in the process of spinal disease diagnosis by uniformly processing structural features and motion trajectory information.

[0021] (4) The AI ​​modeling method and system for dynamic features of imaging data for vertebral misalignment diagnosis constructs a dynamic atlas representation of the vertebral motion evolution trend by integrating multivariate features such as rotation angle, path length and direction angle.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of the AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to the present invention. Figure 2 This is a structural diagram of the AI ​​modeling system for dynamic features of imaging data for vertebral misalignment diagnosis according to the present invention; Figure 3 This is a bar chart showing the rotational lateral bending synergy values ​​involved in this invention; Figure 4 This is a risk map of the dynamic evolution of vertebral segments involved in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-4This invention provides a technical solution: an AI modeling method and system for dynamic features of image data for vertebral misalignment diagnosis, comprising: S1, collecting geometric feature data and displacement feature data from vertebral images, and preprocessing the collected geometric feature data and displacement feature data to construct a standardized vertebral segment state dataset; S2, based on the standardized vertebral segment state dataset, performing a synergistic evaluation of the instantaneous synergistic state of each vertebral segment, and generating a local structural symmetry feature and stability sample set based on the synergistic evaluation results; S3, based on the standardized vertebral segment state dataset, performing trend analysis on the structural evolution trend of vertebral segments from the angle changes and spatial position offset states of each vertebral segment in multiple frames of images, and identifying regions of differential enhancement based on the trend analysis results; S4, combining the trend analysis results of vertebral segments to perform a risk assessment of the segmental risk intensity, and dynamically adjusting the atlas structure presentation method based on the risk assessment results.

[0026] Specifically, the steps for acquiring geometric and displacement feature data from vertebral images are as follows: During the dynamic acquisition of vertebral segment posture changes, geometric feature data is obtained by recording the motion state of the vertebral segments in the image sequence frame by frame. The time index of each frame is used as a temporal marker, so that the rotation angle value and the principal axis direction angle value form a continuous expression in the time dimension, thereby fully presenting the posture change trajectory of the vertebral segments during flexion, extension, rotation, and slight displacement. At the same time, in the acquisition of dynamic changes in the center point, the longitudinal coordinate of the vertebral body midline is used as a global reference. Combined with the transverse coordinate value of the structural center, the longitudinal coordinate value of the structural center, the height of the vertebral segment, and the coordinate of the center point of each vertebral segment in the continuous image frames, a temporal displacement description that can reflect the relative positional relationship, motion trend, and height change between different vertebral segments is formed. This allows the longitudinal position, transverse movement amplitude, segmental extension and contraction state, and center point offset characteristics of the vertebral segments in the overall arrangement to be accurately recorded over time, thereby constructing a continuous structural posture representation that reflects the dynamic behavior of the vertebral segments.

[0027] In this implementation scheme, by continuously acquiring the posture and displacement information of vertebral segments in the imaging sequence, a temporal data foundation that can realistically reflect the dynamic behavior characteristics of vertebral segments is established. This enables subsequent modules to conduct structural synergy assessment, evolutionary trend identification, and displacement risk determination on a stable and complete motion trajectory. By recording the rotation angle and principal axis angle values ​​of the vertebral segments, the posture change patterns of the vertebral segments during flexion-extension, lateral flexion, and slight torsion can be described. By recording the longitudinal coordinate of the vertebral body midline, the coordinates of the structural center point, and the height of the vertebral segment, the continuous displacement process of the vertebral segments in terms of longitudinal alignment, lateral displacement amplitude, and segmental morphological changes can be characterized. This ensures that the relative position, posture direction, and height changes between different vertebral segments are consistently expressed in the time dimension, thereby providing reliable basic data support for subsequent structural synergy analysis, local displacement identification, and segmental stability determination.

[0028] Specifically, the preprocessing of the collected geometric and displacement feature data to construct a standardized vertebral segment state dataset involves the following steps: In the geometric feature processing of vertebral segment posture changes, the rotation angle and principal axis angle values ​​of each vertebra are extracted one by one from consecutive image frames, enabling the posture information of each frame to form a clear and coherent change trajectory in the time dimension. Simultaneously, the time index of the corresponding frame is recorded, and the frame order is sorted according to the time index, ensuring the traceability and coherence of angle changes in the temporal direction, thus laying a stable foundation for subsequent posture evolution analysis. In the displacement feature processing of the dynamic changes of the vertebral segment center point, the longitudinal coordinates of the vertebral midline, the transverse coordinates of the structural center, the longitudinal coordinates, the vertebral segment height, and the complete center point coordinates are obtained frame by frame. Then, the absolute value of the difference between the longitudinal coordinate of the structural center and the midline is calculated based on the longitudinal coordinate difference. This value is defined as the center point horizontal offset value, used to characterize the lateral displacement amplitude of the vertebra, reflecting its subtle displacement performance in the arrangement relationship.

[0029] In expressing the lateral grayscale characteristics of images, grayscale analysis is performed on the grayscale distribution of the left and right sides of the same vertebral segment to obtain the grayscale difference between the left and right sides, which is then used as the edge difference value of the vertebral segment. Subsequently, based on the edge difference values ​​of multiple vertebral segments, a grayscale change sequence reflecting the trend of edge texture changes is constructed, so that the brightness changes of the vertebral segment at the boundary can be presented in the form of a time series. The standardized geometric feature data and displacement feature data are uniformly normalized to ensure that all indicators are expressed under the same dimension, thereby constructing a standardized vertebral segment state dataset to ensure the comparability of various values ​​in subsequent calculation steps.

[0030] When describing the spatial relationship between vertebral segments, the center point coordinates of the target vertebra and the next vertebra in the same frame are obtained. The spatial Euclidean algorithm is used to obtain the displacement distance between adjacent vertebral segments, which is recorded as the center point distance to characterize the instantaneous proximity between segments. When characterizing temporal motion changes, the center point coordinates of the current vertebra in adjacent frames are obtained, and the angle between the lines connecting the preceding and following center points is calculated to obtain the path angle change value. During continuous posture changes, the number of times the path angle change value exceeds the angle change threshold is counted and recorded as the vertebral active offset value to identify the activity level of the vertebra during posture changes.

[0031] To describe the overall change in the motion trajectory, a center point coordinate sequence is constructed by obtaining the center point coordinates of the same vertebral segment across all frames. The Euclidean algorithm is then used to calculate and sum the center point distances frame by frame to obtain the total path length. Simultaneously, the spatial distance between the center points of the start and end frames is extracted and used as the functional distance to reflect the displacement increment of the vertebral segment within the overall motion range. Finally, a vertebral segment height sequence is constructed by obtaining the vertebral segment heights across all frames. The maximum and minimum values ​​are then extracted from the sequence, and the difference is calculated. This difference is used as the vertebral segment height difference value to reflect the segment's variation in the vertical dimension.

[0032] In this implementation scheme, based on continuous image frames, key values ​​of vertebral segments during posture and displacement changes are recorded in a time-series manner, enabling accurate representation of rotational behavior, positional offset, trajectory morphology, and height changes within a unified data framework. By constructing geometric feature sequences, the posture change patterns of vertebral segments during rotation and orientation adjustments can be presented; by constructing displacement feature sequences, the movement trends of vertebral segments in the longitudinal and lateral directions can be shown; by continuously recording edge grayscale differences, subtle changes in vertebral segment edge texture in the temporal direction can be described; and by calculating path angle changes, active offsets, center point distances, total path lengths, functional position distances, and height differences, the motion amplitude and structural change characteristics of vertebral segments at short- and long-term scales can be further characterized. This step ultimately enables the geometric and displacement behaviors of vertebral segments in the image sequence to form a quantifiable, comparable, and correlated complete dynamic representation, providing a solid data foundation for subsequent collaborative analysis, trend identification, offset determination, and stability assessment.

[0033] Specifically, based on a standardized vertebral segment state dataset, the synergy assessment of the instantaneous synergy state of each vertebral segment involves the following steps: calculating the difference between the transverse coordinate value of the current vertebral segment's structural center and the transverse coordinate value of the next vertebral segment, and squaring the result to obtain the transverse offset value; calculating the difference between the longitudinal coordinate value of the current vertebral segment's structural center and the longitudinal coordinate value of the next vertebral segment, and squaring the result to obtain the longitudinal offset value; adding the transverse and longitudinal offset values ​​and taking the square root to obtain the relative position offset value; negatively substituting the rotational inconsistency value into an exponential function to obtain the rotational inconsistency index; negatively substituting the relative position offset value into an exponential function to obtain the relative position offset index; and multiplying the rotational inconsistency index and the relative position offset index to obtain the rotational lateral flexion synergy value.

[0034] The formula for calculating the rotational lateral bend synergy value is: In the formula, The rotation angle value of the kth vertebral segment in the image is used to quantify the degree of deflection between the main axis of the vertebral segment and the vertical reference axis. It is the core angle parameter for determining the change in the posture of the segment and is derived from the angle fitting calculation of the upper and lower edge structural lines of the vertebral segment in the registered image. The rotation angle value of the (k+1)th vertebral segment in the image is used to characterize the degree of axial displacement of the vertebral segment in the current posture and is the basic variable for collaborative analysis between adjacent segments. This represents the lateral coordinate value of the structural center of the k-th vertebra in the image. It is used to reflect the left and right offset position of the vertebra in the image and is the basic lateral quantity for constructing the relative displacement vector. It is derived from the extraction of the center point pixel coordinates after structural registration. This represents the lateral coordinate value of the structural center of the (k+1)th vertebral segment in the image, used to express the lateral relative relationship between adjacent vertebral segments; This represents the longitudinal coordinate value of the structural center of the k-th vertebra in the image, which reflects the vertical position of the vertebra in the image. It is the basic longitudinal quantity for constructing the relative displacement vector and is derived from the extraction of the center point pixel coordinates after structural registration. This represents the longitudinal coordinate value of the structural center of the (k+1)th vertebra in the image. It is used to express the longitudinal relative relationship between adjacent vertebrae and is the direct source for subsequent calculation of longitudinal offset.

[0035] In this implementation scheme, T1's Set to 2.1. Set to 2.4. Set to 10.0. Set to 10.5. Set to 20.0. Set to 20.3; T2 Set to 1.9. Set to 1.6. Set to 12.5. Set to 12.0. Set to 21.0. Set to 20.5; T3 Set to 2.3. Set to 2.0. Set to 11.0. Set to 11.5. Set to 19.5. Set to 19.0; T4 Set to 1.7. Set to 1.9. Set to 13.3. Set to 13.0. Set to 22.0. Set to 21.8; T5 Set to 2.0. Set to 1.8. Set to 12.0. Set to 12.5. Set to 20.5. Set to 20.0; T6 Set to 1.8. Set to 2.1. Set to 14.0. Set to 13.8. Set to 21.5. Set to 21.7; T7 Set to 2.2. Set to 2.0. Set to 13.5. Set to 13.0. Set to 22.2. Set the value to 22.5. Calculate the rotational lateral flexion synergy value for each vertebral segment. As shown in Table 1, the rotational lateral flexion synergy data is presented.

[0036] Table 1. Rotational Lateral Flexion Synergistic Value Data Table like Figure 3 As shown in Table 1, this is a table of rotational lateral bending synergy values ​​provided in this application example. Figure 3It can be seen that the rotational lateral flexion synergy value of segment T4 is the highest, indicating that the difference in rotational angle value between it and the adjacent segments is small. At the same time, the spatial offset of the horizontal and vertical coordinates is limited, making the overall state closer to a stable synergistic posture. This reflects that the posture consistency of this segment in the current sequence is high and the positional offset is controlled. It can be regarded as a segment with strong local structural stability and is suitable as a reference benchmark in subsequent time series analysis to enhance the reliability of local evolution trend judgment. The rotational lateral flexion synergy values ​​of segments T2 and T3 are at the lowest level. Although the angle difference is in the low to medium range, the spatial positional offset is relatively larger, which weakens the local synergy and shows a downward trend in overall stability. This will automatically reduce their reference weight in the structural matching and posture comparison process, avoid the introduction of oversensitive misjudgment due to amplified local offset, and thus improve the stability and robustness of overall trend recognition. The rotational flexion-rotational synergy value histogram can intuitively present the synergistic sensitivity distribution of each vertebral segment in the current sequence. The higher the rotational flexion-rotational synergy value, the closer the posture between vertebral segments and the lower the spatial deviation, making it more suitable as a stable reference segment. The lower the rotational flexion-rotational synergy value, the more likely the segment has a certain structural deviation tendency in the current frame sequence. Its weight should be reduced in subsequent dynamic analysis to avoid affecting the accuracy of overall trend judgment and risk identification.

[0037] Specifically, the steps for generating a sample set of local structural symmetry features and stability based on the synergy assessment results are as follows: The rotational flexion-rotational synergy value and the rotational flexion-rotational synergy threshold of the current vertebral segment are compared in real time, and the magnitude of the synergy value is used as the basis for judging the posture consistency and deviation risk of the current segment. When the rotational flexion-rotational synergy value is less than or equal to the rotational flexion-rotational synergy threshold, the vertebral segment located in the current frame image is taken as the analysis target, and a triangular labeled region is generated at this location. This region consists of three points: the midpoint of the upper edge and the left and right pedicle angles. The positions of these three points are obtained using an edge gradient combined with a geometric peak detection algorithm, and a closed triangular region formed by connecting the three points is used as a fixed analysis domain. Subsequently, the side lengths from the midpoint of the upper edge to the left pedicle angle, the midpoint of the upper edge to the right pedicle angle, and the base side length between the two pedicle angles are extracted within this region, and the ratio of these three sides is used to construct the structural three-side length ratio. This length ratio is compared one by one with the length ratios of the corresponding regions of adjacent vertebral segments, and the difference is used to identify local proportional deviations under posture instability. Based on this, the vertebral segment principal axis direction vector is obtained from the current frame image. This vector is used as the direction reference for offset propagation, and a continuous direction chain is generated by extending two vertebral segments upstream and downstream along the vector direction, centered on the current vertebra. Based on the direction deviation between each vector segment in the direction chain, a direction switching matrix is ​​constructed using a fixed angular step size. The number of rows and columns of the matrix is ​​determined by the number of vector segments and the angular step size. Each cell in the matrix records the direction interval distribution of the corresponding vector segment. By identifying the position with the highest cumulative value of direction jump in the matrix, this position is used as the axis offset point to determine the specific landing point where attitude instability occurs in the current segment.

[0038] When the rotational lateral flexion coordination value is greater than the rotational lateral flexion coordination threshold, it indicates that the current segment is in a highly consistent posture state. At this point, the region near the main axis projection line of the vertebral segment is extracted from the current and previous frames, and extended forward and backward along the main axis by a fixed distance, with a horizontal expansion width of several pixels, thus generating an extended region that covers the stable contour of the vertebral segment within a short time window. Subsequently, the edge grayscale change sequence in the horizontal direction is extracted from the extended region, allowing the texture brightness changes between consecutive frames to be expressed numerically. Based on this, the entire vertebral segment region is cropped into equal-height rectangular blocks along the height direction. The height of each rectangular block is a fixed segmental proportion of the vertebral segment height, and the step size is set to half the height of the rectangular block, achieving vertical local overlap and ensuring that each rectangular block maintains consistency at a local height scale. This generates a structural stability texture reference sample sequence, providing a reliable basis for subsequent stable segment identification and structural comparison.

[0039] In this implementation scheme, based on the real-time discrimination results of rotational lateral flexion synergy values, the structural performance of vertebral segments in unstable and stable states is extracted in layers, enabling clear distinction between posture displacement features and stable texture features within the same process. By generating triangular labeled regions and calculating the ratio of the three sides of the structure during the low synergy value stage, and then combining the direction chain and direction switching matrix to identify the axis offset point, the offset landing point and offset propagation direction can be accurately located when the vertebral segment exhibits posture inconsistency, allowing the spatial source and influence range of local instability to be directly quantified. By extracting the principal axis projection line extension region and constructing equal-height rectangular block texture samples during the high synergy value stage, the stable contour of the vertebral segment can be recorded when it maintains a highly consistent posture, allowing subsequent comparisons using stable textures as a benchmark, thereby improving the reliability of structural stability identification. This step ultimately achieves dynamic distinction between stable and unstable segments of the vertebral segment, providing a clear basis for subsequent trend analysis, offset detection, and risk assessment, and contributing to improved accuracy and continuity of structural determination.

[0040] Specifically, based on a standardized vertebral segment state dataset, the structural evolution trend of vertebral segments is analyzed from the angle changes and spatial position offsets of each segment in multi-frame images. The specific steps are as follows: Calculate the absolute value of the difference between the principal axis angle values ​​of the current vertebral segment in the previous and next frames of the current frame, and divide it by the difference between the time index of the next frame and the time index of the previous frame to obtain the angle change rate; add one to the absolute value of the difference between the principal axis angle value of the current vertebral segment in the current frame and the principal axis angle value in the initial frame, and take the reciprocal to obtain the angle offset suppression value; calculate the difference between the center point distance in the previous and next frames of the current frame, divide it by the difference between the time index of the next frame and the time index of the previous frame, and take the absolute value to obtain the position change rate; subtract the center point distance in the initial frame from the center point distance in the current frame, divide it by the center point distance in the initial frame, and take the absolute value to obtain the relative structural offset ratio; add the product of the angle change rate and the angle offset suppression value to the product of the position change rate and the relative structural offset ratio to obtain the misalignment evolution trend value.

[0041] The formula for calculating the staggered seam evolution trend value is: ; In the formula, It represents the principal axis direction angle value of the same vertebra in the next frame, used to characterize the rotation trend after the current frame, and is the endpoint reference value for calculating the rotation slope, which is extracted from the principal axis direction angle of the vertebral structure in frame b; It represents the principal axis direction angle value of the same vertebra in the previous frame, which is used to quantify the rotation state of the previous frame. It is the basic quantity for calculating the rate of angle change and is derived from the extraction of the direction angle of the line connecting the upper and lower edges of the vertebra in frame a. The time index of the next frame describes the time position of the next frame and is the time endpoint for calculating the trend slope. It is derived from the timestamp record. The time index of the previous frame describes the time position of the previous frame. It is the starting point of the denominator for angle and displacement change rate and comes from the timestamp record. This represents the angle value of the vertebral segment principal axis in the current frame. It is used to quantify the rotation state of the vertebral segment under the current posture and is an important variable for determining whether there is an angle deviation. It is derived from the angle between the line connecting the upper and lower edges of the vertebral segment and the vertical axis in this frame. It represents the principal axis direction angle value of the same vertebra in the initial frame. It is used to measure the cumulative deviation between the current frame and the standard posture. It is an important reference quantity for trend normalization and is derived from the principal axis angle measurement of the functional position motion start frame. It represents the distance between the center points of two adjacent vertebrae in the next frame, which is used to calculate the displacement change trend. It is the key data for calculating the trend growth rate and comes from the spatial distance measurement of the structural annotation points in frame b. This represents the distance between the center points of two adjacent vertebrae in the previous frame. It is used to calculate the starting position of the displacement slope and is derived from the calculation of the center point coordinate difference after structural registration in frame a. This represents the distance between the center point of the target vertebral segment and its next segment in the current frame. It is used to evaluate the spatial structural position of the current segment and is the core position parameter for misalignment identification. It is derived from the coordinate difference calculation of the center points of the two vertebral segments in the registered image. It represents the distance between the center points of two adjacent vertebrae in the initial frame, used to measure the cumulative offset of the current segment from the standard state. It is the benchmark value on which the evolution degree normalization process depends, and is derived from the structural point position extraction results in the functional bit start frame.

[0042] In this implementation scheme, the changes in the principal axis angle and the displacement of the center point in consecutive image frames are used as core quantities to construct a quantitative index that can characterize the rate of rotational displacement evolution of vertebral segments during temporal changes. This allows the rotational trend, movement trend, and structural displacement trend of vertebral segments between consecutive frames to be integrated with a unified dimension. The formula can output a misalignment evolution trend value that reflects the superimposed effect of three types of motion behaviors: rotational change, longitudinal displacement, and overall displacement. This provides a key basis for determining whether vertebral segments are experiencing accelerated rotation, amplified displacement, or deterioration, thereby supporting subsequent structural risk identification and dynamic behavior analysis.

[0043] Specifically, the steps for identifying regions of enhanced difference based on trend analysis results are as follows: A sequence of vertebral segment misalignment evolution trend values ​​is constructed based on the misalignment evolution trend values ​​of n consecutive frames, and the local slope and fluctuation amplitude of the misalignment evolution trend values ​​are calculated; where n represents the length of the continuous time window used for trend analysis, covering several preceding and following image frames, used to describe the rotational offset evolution law of the vertebral segment within a longer time interval. If the misalignment evolution trend value of the current frame is greater than the previous m frames and less than the following m frames, a bidirectional temporal comparison is performed, and the principal axis angle of the vertebral segment in the m frames before and after the current frame is interpolated and reconstructed; where m represents the length of the half-time window used to determine the local peak shape, used to identify the relative position of the current frame in the local trend. The interpolation reconstruction uses a spline method to ensure that the principal axis angle presents a continuous and smooth change trajectory within the local time range, thus avoiding unstable posture expression caused by discrete angle jumps.

[0044] If the current frame's seam evolution trend value is the maximum value among n consecutive frames, the displacement direction of the segment's center point and the change value of the path angle are extracted in the current frame. If the change value of the path angle is greater than the angle offset threshold, the segment boundary region is expanded along the main axis in the current frame and grayscale stretching is performed to make the segment outline present a clearer texture structure in the offset enhancement scene. Grayscale stretching uses a piecewise linear contrast function to improve the visibility of the boundary by enhancing the brightness gradient in the mid-grayscale area, so that the structural texture is more prominently expressed in the offset sensitive stage.

[0045] If the current frame's misalignment evolution trend value is the minimum among n consecutive frames, then the sequence of misalignment evolution trend values ​​for adjacent upper and lower vertebrae is extracted with the current vertebrae as the center, and ranking statistics are performed within this sequence. The ranking is based on the absolute sorting of trend values ​​for all segments within the window, which consists of the previous vertebrae, the current vertebrae, and the next vertebrae. If multiple segments have the same trend value, the final order is determined by the natural order of the vertebrae in the longitudinal arrangement. If the current vertebrae is at the lowest position in the ranking, then the complete structural block containing the current vertebrae is cropped from the original image and recorded as a stable feature reference sample for subsequent stable structure comparison and attitude benchmark generation.

[0046] In this implementation scheme, trend changes within a continuous time window are used to automatically distinguish between three states of the vertebral segment: the offset enhancement stage, the posture transition stage, and the structural stability stage. This allows key features in different states to be extracted specifically within a unified process. By constructing a trend value sequence within a window of length n and calculating the local slope, the rotational amplification trend and fluctuation level of the vertebral segment over a longer time range can be described. By identifying the relative position of the current frame in the bidirectional time series within a local window of length m and reconstructing the principal axis angle using splines, a smooth angle representation can be obtained when the posture change is in the transition stage. When the trend reaches the maximum value within the window, the boundary expression is enhanced by changes in the path angle and grayscale stretching, highlighting the detailed contours of the structure under the offset growth state, making the location of misalignment easier to identify in comparison. When the trend reaches the minimum value within the window, the most stable position is identified by cross-segment trend sorting, and a complete structural block is cut out at this position as a stable reference sample, providing a reliable benchmark for subsequent comparison processes. Overall, this step, through a segmented processing method driven by trend peak and valley features, enables the amplified, transitional, and stable states of the vertebral segment during dynamic use to be clearly separated, thereby enhancing the accuracy and continuity of subsequent offset identification, structural comparison, and risk assessment.

[0047] Specifically, the risk assessment of segmental risk intensity of vertebrae based on the trend analysis results of vertebrae is carried out in the following steps: The total path length of the current vertebrae is used as the numerator, and the absolute value of the difference between the total path length of the vertebrae and the functional position distance plus one is used as the denominator to obtain the cumulative path offset value; the vertebral height difference value of the current vertebrae is divided by the sum of the vertebral height of the initial frame and one, and then one is added to obtain the vertebral deformation amplification value; the active offset value of the vertebrae is multiplied by the square of the horizontal offset value of the center point, divided by the sum of the vertebral edge difference value and one, and then multiplied by the misalignment evolution trend value to obtain the structural disturbance enhancement value; the sum of the path angle change value and the cumulative path offset value, the vertebral deformation amplification value, and the structural disturbance enhancement value are multiplied to obtain the dynamic map risk value.

[0048] The formula for calculating the risk value of dynamic graphs is: ; In the formula: The value of the change in the path angle is used to quantify the magnitude of the sudden change in direction of the current segment in continuous motion. It is an important indicator for identifying turning structures in the motion trajectory and is derived from the calculation result of the direction angle of the line connecting the center points of the current frame and the frame before and after the current frame. It represents the total path length of the center point of the vertebral segment in consecutive frames. It is used to quantify the total spatial motion trajectory of the segment within the functional position motion cycle. It is the basic variable for judging the intensity of its motion participation and comes from the sum of the Euclidean distances between the center points of each frame. It represents the distance of the vertebral segment in its functional position, which is used to reflect the degree of deviation of the actual motion path from the theoretical straight trajectory. It is a reference quantity for judging whether the vertebral segment has a tendency to deviate from the track. It is derived from the straight distance calculation of the coordinates of the center point of the start and end frames. The value representing the difference in vertebral height is used to quantify the longitudinal deformation of the structure during motion. It is a characteristic indicator for identifying compression and stretching trends and is derived from the calculation of the vertical distance between the upper and lower edge structural points in each frame. The height of the vertebral segment in the initial frame is used to provide a reference for the deformation of the vertebral segment structure. It is the basic value for calculating the height normalization ratio and is derived from the length of the upper and lower edge line segments of the vertebral segment in the initial frame. This represents the staggered joint evolution trend value of the current frame, which is used to quantify the overall structural evolution intensity of the segment at the current time point. It is the core variable of the trend influence term in risk calculation and comes from the output result of the formula of the aforementioned trend extraction module. This represents the active offset value of the current vertebral segment, which is the number of frames in which the posture change exceeds a specified angle threshold during continuous motion. It is used to reflect the frequency of structural instability of this segment in the time sequence and is a statistical parameter of dynamic activity, derived from the statistical count of directional angle changes over multiple frames. This represents the current horizontal offset value of the vertebral segment center point, used to measure the dispersion of the lateral position of the vertebral segment. It is a spatial index for evaluating structural offset behavior and is derived from the horizontal projection distance between the center point and the vertebral body midline. This represents the edge difference value of the vertebral segment in the current frame, which is used to evaluate the edge symmetry of the image structure. It is an image-level feature variable for determining abnormal morphological features of the vertebral segment boundary, and is derived from the gray-level distribution analysis of the left and right symmetrical regions of the vertebral segment in the image.

[0049] In this implementation example, the multidimensional changes in vertebral segment behavior during dynamic motion are uniformly quantified into a dynamic atlas value that directly reflects the risk of displacement. This allows for a comprehensive displacement risk assessment result under the same expression framework, encompassing angle changes, path accumulation, spatial compression morphology, rotational trends, and intensity corrections. This formula integrates abrupt angle changes, path accumulation displacement, longitudinal compression morphology, temporal trend amplification effects, and spatial displacement intensity into a unified risk expression quantity. This enables the simultaneous presentation of rotational instability, path displacement accumulation, structural compression morphology deterioration, and trend acceleration during continuous vertebral movement. Consequently, it provides a quantitative conclusion on the overall displacement risk level of the vertebral segment in the current frame, offering a direct basis for subsequent risk atlas construction, key segment identification, and dynamic displacement monitoring.

[0050] Specifically, the steps for dynamically adjusting the presentation of the map structure based on the risk assessment results are as follows: compare the current dynamic map risk value with the map risk threshold in real time. The map risk threshold includes a first risk threshold and a second risk threshold. The relationship between the dynamic map risk value and the threshold is used as an important basis for judging the current segment structure status.

[0051] When the dynamic map risk value is less than or equal to the second risk threshold, the upper and lower edge coordinates of the vertebral segments in the current frame are marked. The actual width between the left and right boundaries of the vertebral body is used as the reference width, and a rectangular region of equal width is constructed according to a fixed ratio coefficient. The width of the rectangular region is the width of the vertebral body multiplied by a fixed ratio, so that the texture expression of different vertebral segments has a consistent scale in the lateral direction. Then, the rectangular region is divided into several grayscale texture slices by columns, with each column corresponding to a fixed pixel width, which is used to record the local changes in the lateral grayscale distribution. The grayscale texture slices are stored as stable structure texture samples by column order compression, providing a reference for subsequent comparison.

[0052] When the dynamic map risk value is greater than the second risk threshold and less than or equal to the first risk threshold, piecewise interpolation fitting is performed on the motion path of the center point of the vertebral segment. The interpolation adopts spline fitting to make the trajectory of the center point show continuous and smooth changes in the local interval. At the same time, the vertebral segment height sequence is extracted from the starting frame to the current frame, and the height difference between adjacent frames is calculated to construct the vertebral segment height difference change sequence, which is used to record the longitudinal compression and extension performance of the vertebral segment in the current stage.

[0053] When the dynamic atlas risk value exceeds the first risk threshold, an extended neighborhood region of the vertebral segment is extracted from the current frame, based on its own vertebral height. The vertical boundary of the extended neighborhood is set to the vertebral segment height multiplied by a fixed ratio, and the horizontal boundary is extended by a fixed pixel distance on both sides, so that the region can cover the complete texture outline of the vertebral segment in the high-risk stage. The central principal axis vector is reconstructed within the extended neighborhood, and then the angular offset between the principal axis vector and the midline of the spine is calculated to assess the degree of structural rotation offset. Finally, the cropped image of the extended neighborhood region is input into the dynamic atlas risk rendering process, and a thermal rendering is generated by linear mapping to a fixed color mark, so that the high offset area obtains more significant brightness and color response in the rendering result, which is used to indicate the location and range of the high-risk segment.

[0054] like Figure 4 The diagram shows a dynamic evolution risk map of vertebral segments provided in this application example. It visually displays the distribution of dynamic evolution risk for each vertebral segment across different frames during spinal movement. The horizontal axis represents the frame sequence, indicating the evolutionary process over time, while the vertical axis represents the vertebral segment number, corresponding to the spatial arrangement of the structure. The map presents the distribution of dynamic risk values ​​in a heatmap format, where each colored block represents the dynamic risk value of the corresponding vertebral segment in the current frame. The color increases progressively from black / red to yellow, indicating a rise in risk intensity. The color bars on the right side of the map show the range of dynamic risk values, supporting quantitative assessment of risk levels in local areas. The superimposed blue curve represents the trajectory of the vertebral body center point over time, reflecting the overall stability trend of the structure over time. Blue pentagrams mark the locations of abnormal evolution points, indicating key risk mutation areas identified during structural movement. This map integrates time-series image analysis results, combining the center point trajectory with changes in local risk peaks, providing a dynamic reference for subsequent vertebral segment structural anomaly assessment and functional evolution evaluation.

[0055] In this implementation plan, based on the classification results of the dynamic map risk values, the structural performance of vertebral segments in stable, transitional, and high-risk states is captured in a stratified manner, enabling the targeted extraction and accurate recording of structural information corresponding to different risk levels. This step, through a dynamic map risk value-driven hierarchical processing method, captures key features of vertebral segments in stable, transitional, and high-deviation stages, allowing the dynamic map to accurately express different risk states during structural evolution, providing a solid basis for subsequent trend identification and risk monitoring.

[0056] The second aspect of this invention provides an AI modeling system for dynamic features of image data for vertebral misalignment diagnosis, comprising: a temporal image data acquisition module that analyzes continuous vertebral images frame by frame to obtain the rotation angle, principal axis direction angle, structural center coordinates, longitudinal reference coordinates, and vertebral height of the vertebral segment at each time point; and then, combined with the horizontal offset of the center point and the grayscale difference of the edge, performs scale unification, temporal alignment, and feature normalization processing on all the original values, thereby forming a standardized vertebral segment state dataset that can reflect the posture and displacement behavior of the vertebral segment.

[0057] Based on the standardized vertebral segment state dataset of the vertebral body coordinated motion assessment module, the angular proximity, longitudinal and transverse distance proximity, and posture direction consistency of each vertebral segment and its adjacent segments in the same frame are comprehensively calculated for coordination. Based on this, the proportional relationship between the upper edge of the vertebral segment and the pedicle region, the feature expression of the texture stable segment are extracted, the local vertebral segment structure that maintains highly consistent motion characteristics in the current frame is extracted as structural symmetry features, and the texture fragments formed in the stable phase are used as the stability sample set.

[0058] The staggered joint evolution trend extraction module uses the rate of change of angle and spatial offset increment of continuous frames in the standardized vertebral segment state dataset to construct the evolution trend sequence of the vertebral segment. From the change slope, fluctuation amplitude and peak and valley positions expressed by the trend sequence, the rotation amplification area, displacement enhancement area and structural offset area are identified, so that the segment with the intensified trend can be clearly located in space.

[0059] Based on the trend analysis results, the evolution risk output module further performs segmental risk assessment of vertebrae by integrating the risk intensity value derived from angle mutation, path increment, longitudinal compression, and rotational offset. It also dynamically adjusts the presentation of the map using the risk intensity, so that stable regions, transition regions, and high-risk regions present differentiated appearances in the final map, thereby generating a risk map output that can reveal key positions in structural evolution.

[0060] In this implementation scheme, the role of the temporal image data acquisition module is to systematically extract the core values ​​of the vertebral segments in rotational behavior, displacement behavior and boundary texture changes by taking continuous image frames as input, and to form a standardized vertebral segment state dataset by unifying the scale and aligning the temporal sequence, so that all subsequent calculation steps are based on data that is consistent in scale, continuous in temporal sequence and complete in structure.

[0061] The role of the vertebral body coordinated motion assessment module is to determine the posture consistency of each vertebral segment at the same time point based on the standardized state dataset. It evaluates the instantaneous coordinated performance of the vertebral segments by the degree of angular proximity, the degree of center point proximity, and local proportional structural features, and generates symmetry features for verification and stability samples for comparison, providing a stable reference for subsequent offset detection and trend recognition.

[0062] The role of the staggered joint evolution trend extraction module is to identify the rotational amplification segment and displacement enhancement segment that occur in the vertebral segment during the movement based on the continuous frame angle change rate and spatial offset sequence. By characterizing the structural evolution speed and offset direction through trend slope, fluctuation amplitude and local peak and valley positions, the potential abnormal area can be clearly located in time sequence.

[0063] The role of the evolutionary risk output module is to combine trend values, path increments, compression morphology and rotation offset to construct segment-level risk intensity, and dynamically adjust the presentation of the dynamic map according to the risk intensity, so that normal segments, transition segments and high-risk segments form a clear hierarchical expression in the final map, thereby providing a visual risk basis for structural displacement monitoring and key segment identification.

[0064] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0065] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An AI modeling method for dynamic features of imaging data for vertebral misalignment diagnosis, characterized by: include: S1. Collect geometric and displacement feature data from vertebral images, and preprocess the collected geometric and displacement feature data to construct a standardized vertebral segment state dataset. S2, based on a standardized vertebral segment state dataset, evaluates the synergy of the instantaneous synergistic state of each vertebral segment, and generates a sample set of local structural symmetry features and stability based on the synergistic evaluation results; S3, based on a standardized vertebral segment state dataset, analyzes the structural evolution trend of vertebral segments from the angle changes and spatial position offsets of each vertebra in multi-frame images, and identifies regions of differential enhancement based on the trend analysis results. S4, combined with the trend analysis results of vertebral segments, conducts a risk assessment of the segmental risk intensity of vertebral segments, and dynamically adjusts the presentation of the atlas structure based on the risk assessment results.

2. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for acquiring geometric and displacement feature data from vertebral images are as follows: Geometric feature data of vertebral segment posture changes were collected. The geometric feature data included the rotation angle value and principal axis direction angle value of each vertebra in each frame image, and the time index of each frame image was recorded. Displacement feature data were collected during the dynamic change of the vertebral segment center point. The displacement feature data included: the longitudinal coordinate of the vertebral body midline, the transverse coordinate value of the structural center of each vertebral segment in each frame image, the longitudinal coordinate value of the structural center, the vertebral segment height, and the coordinate of the center point.

3. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for preprocessing the collected geometric feature data and displacement feature data to construct a standardized vertebral segment state dataset are as follows: For the geometric feature data during the posture change of vertebral segments, the rotation angle value and principal axis direction angle value of each vertebra in each frame image are obtained, and the time index of the corresponding frame is recorded synchronously. The geometric feature data is then sorted according to the time index. For the displacement feature data of the dynamic change of the vertebral segment center point, the vertical coordinate of the vertebral body midline, the horizontal coordinate value of the structural center of each vertebral segment, the vertical coordinate value, the height of the vertebral segment and the complete center point coordinates are obtained in each frame image. The absolute value of the difference between the vertical coordinate of the structural center and the midline is calculated based on the difference of the vertical coordinate and recorded as the horizontal offset value of the center point. Gray-level analysis was performed on the gray-level distribution of the left and right sides of the same vertebral segment in the image to obtain the gray-level difference between the left and right sides of the vertebral segment, which was recorded as the vertebral segment edge difference value. Based on the vertebral segment edge difference value of each vertebral segment, a gray-level change sequence of the vertebral segment edge was constructed. The geometric feature data and displacement feature data after standardization are normalized to construct a standardized vertebral segment state dataset; Obtain the center point coordinates of the target vertebra and the next vertebra in the same frame, and use the spatial Euclidean algorithm to obtain the displacement distance between adjacent vertebrae, which is denoted as the center point distance. Obtain the center point coordinates of the current vertebral segment in adjacent frames, calculate the path angle change value by the angle between the lines connecting the front and rear center points, and record the number of times the path angle change value of the same vertebral segment exceeds the angle change threshold during continuous posture changes, which is recorded as the active offset value of the vertebral segment. The center point coordinates of the same vertebra in all frames are obtained to construct a center point coordinate sequence. The center point distance between frames is calculated using the Euclidean algorithm and summed to obtain the total path length. The difference between the center point distances of the vertebra in the start and end frames is calculated to obtain the functional bit distance. The vertebral height of the same vertebral segment is obtained in all frames to construct a vertebral height sequence. The maximum and minimum vertebral height values ​​in the vertebral height sequence are extracted and the difference is calculated, which is recorded as the vertebral height difference value.

4. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for evaluating the instantaneous coordinated state of each vertebra based on the standardized vertebral segment state dataset are as follows: Calculate the absolute value of the difference between the rotation angle value of the current vertebral segment and the rotation angle value of the next vertebral segment to obtain the rotation inconsistency value; Calculate the difference between the horizontal coordinate of the current vertebral segment's structural center and the horizontal coordinate of the next vertebral segment, and square the result to obtain the horizontal offset value. Calculate the difference between the vertical coordinate of the current vertebral segment's structural center and the vertical coordinate of the next vertebral segment, and square the result to obtain the vertical offset value. Add the horizontal offset value and the vertical offset value together and take the square root to obtain the relative position offset value. The rotation inconsistency value is negative and then substituted into the exponential function to obtain the rotation inconsistency index. The relative position offset value is negative and then substituted into the exponential function to obtain the relative position offset index. The rotation inconsistency index and the relative position offset index are multiplied to obtain the rotational lateral buckling coordination value.

5. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for generating a sample set of local structural symmetry features and stability based on the collaborative evaluation results are as follows: Real-time comparison of the rotational lateral flexion coordination value and the rotational lateral flexion coordination threshold of the current vertebral segment: When the rotational lateral flexion coordination value is less than or equal to the rotational lateral flexion coordination threshold, the triangular annotation area of ​​the current frame image of the vertebral segment is extracted, the ratio of the length of the three sides of the structure formed by the upper edge of the vertebral segment and the corner point of the pedicle is extracted, and the side length difference is compared with the corresponding area of ​​the upper and lower vertebral segments. At the same time, the attitude direction vector of the current frame image is superimposed, and the vector offset direction of the segment is extended to the upper and lower two vertebral segments to construct a multi-segment direction switching matrix for identifying the local rotation axis offset point. When the rotational lateral flexion coordination value is greater than the rotational lateral flexion coordination threshold, the extension area of ​​the principal axis projection line of the current vertebral segment in the current frame and the previous frame image is read, the gray-scale change sequence of the vertebral segment edge in the horizontal region of the image is extracted, and the current vertebral segment region is cropped into equal-height rectangular blocks to generate a structural stability texture reference sample sequence.

6. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for trend analysis of the structural evolution trend of vertebral segments based on the standardized vertebral segment state dataset and the angular changes and spatial position offsets of each vertebral segment in multi-frame images are as follows: Calculate the absolute value of the difference between the principal axis angle values ​​of the current vertebral segment in the previous and next frames of the image, and divide it by the difference between the time index of the next frame and the time index of the previous frame to obtain the angle change rate. The absolute value of the difference between the current vertebral segment's principal axis direction angle value in the current frame and the principal axis direction angle value in the initial frame is increased by one and then the reciprocal is taken to obtain the angle offset suppression value. The position change rate is obtained by calculating the difference between the center point distance in the previous frame and the center point distance in the next frame, dividing it by the difference between the time index of the next frame and the time index of the previous frame, and then taking the absolute value. Subtract the center point distance in the initial frame from the center point distance in the current frame, divide by the center point distance in the initial frame, and take the absolute value to obtain the relative offset ratio of the structure. The value of the staggered joint evolution trend is obtained by adding the product of the angular change rate multiplied by the angular offset suppression value and the product of the position change rate multiplied by the relative offset ratio of the structure.

7. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for identifying regions with enhanced differences based on trend analysis results are as follows: Based on the misalignment evolution trend values ​​of n consecutive frames, a sequence of misalignment evolution trend values ​​for vertebral segments is constructed, and the local change slope and fluctuation amplitude of the misalignment evolution trend values ​​are calculated. If the current frame's misalignment evolution trend value is greater than the previous m frames and less than the next m frames, perform a bidirectional temporal comparison and interpolate and reconstruct the vertebral principal axis angles in the previous and next m frames of the current frame. If the current frame's misalignment evolution trend value is the maximum value among n consecutive frames, extract the displacement direction and path angle change value of the segment's center point in the current frame. If the path angle change value is greater than the angle offset threshold, expand the vertebral segment boundary region along the main axis direction in the current frame and perform grayscale stretching to enhance the clarity of the structural outline. If the current frame's misalignment evolution trend value is the minimum among n consecutive frames, then the misalignment evolution trend value sequence of adjacent upper and lower vertebrae is extracted with the current vertebrae as the center. The ranking of the misalignment evolution trend value sequence between vertebrae is calculated. If the current vertebrae has the lowest ranking, then the complete vertebral segment structure block where the vertebrae is located is cropped from the original image and recorded as a stable feature reference sample.

8. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for assessing the segmental risk intensity of vertebrae by combining the trend analysis results of vertebrae are as follows: The cumulative path offset value is obtained by taking the total path length of the current vertebral segment as the numerator and the absolute value of the difference between the total path length of the vertebral segment and the functional position distance plus one as the denominator. Divide the current vertebral segment height difference value by the sum of the vertebral segment height in the initial frame and one, and then add one to obtain the vertebral segment deformation amplification value. Multiply the active offset value of the vertebral segment by the square of the horizontal offset value of the center point, divide by the sum of the difference value of the vertebral segment edge and one, and then multiply by the misalignment evolution trend value to obtain the structural disturbance enhancement value. The dynamic map risk value is obtained by multiplying the sum of the path angle change value, the cumulative path offset value, the vertebral deformation amplification value, and the structural disturbance enhancement value.

9. The AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis according to claim 1, characterized in that: The specific steps for dynamically adjusting the presentation method of the map structure based on the risk assessment results are as follows: Real-time comparison of the current dynamic map risk value with the map risk threshold, which includes a first risk threshold and a second risk threshold: When the dynamic map risk value is less than or equal to the second risk threshold, mark the upper and lower edge coordinates of the vertebra in the current frame, construct an equal-width rectangular region and divide it into grayscale texture slices by column, and compress and store it as a stable structure texture sample. When the dynamic map risk value is greater than the second risk threshold and less than or equal to the first risk threshold, segmental interpolation fitting is performed on the motion path of the center point of the vertebral segment, and the vertebral segment height from the starting frame to the current frame is extracted to construct a sequence of vertebral segment height difference changes. When the risk value of the dynamic map is greater than the first risk threshold, the extended neighborhood region of the vertebral segment based on its own vertebral height is extracted from the current frame, the central principal axis vector is reconstructed, and the angle offset between the central principal axis vector and the midline of the spine is calculated. The cropped map of the extended neighborhood region is then input as a high-risk structural fragment into the dynamic evolution risk map of the vertebral segment for thermal rendering.

10. An AI modeling system for dynamic features of imaging data for vertebral misalignment diagnosis, employing the AI ​​modeling method for dynamic features of imaging data for vertebral misalignment diagnosis as described in any one of claims 1-9, characterized in that: include: The time-series image data acquisition module is used to acquire geometric feature data and displacement feature data in vertebral images, and to preprocess the acquired geometric feature data and displacement feature data to construct a standardized vertebral segment state dataset. The vertebral body coordinated motion assessment module is used to assess the synergy of the instantaneous coordinated state of each vertebra based on a standardized vertebral segment state dataset, and to generate a sample set of local structural symmetry features and stability based on the synergy assessment results. The staggered joint evolution trend extraction module is used to perform trend analysis on the structural evolution trend of each vertebra based on the angle change and spatial position offset of each vertebra in multiple frames of images, based on a standardized vertebral segment state dataset, and to identify regions of differential enhancement based on the trend analysis results. The evolution risk output module is used to assess the segmental risk intensity of vertebrae by combining the trend analysis results of vertebrae, and dynamically adjust the presentation of the atlas structure based on the risk assessment results.

Images