Orthopedic auxiliary examination method and system based on image processing

Abstract

The invention discloses an orthopedic auxiliary examination method and system based on image processing, and belongs to the technical field of medical image processing, and the method specifically comprises the steps: extracting an implant contour and stable anatomical points of a first frame image, and building a reference feature set; performing global spatial transformation on all the images to construct a registration image; expanding along the outline of the implant to generate an annular analysis area, and dividing the annular analysis area into a rigid area and an elastic skeleton area; by comparing image texture features of the elastic skeleton region and the reference corresponding region, generating a spatial transformation field and performing local fine registration; identifying a pixel set of which the gray value changes in a one-way trend along with time and which is spatially adjacent, and marking the pixel set as an abnormal region; according to the method, by means of layered registration and time sequence analysis, the sensitivity of early prosthesis looseness sign detection and the objectivity of quantitative evaluation are improved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to an orthopedic auxiliary examination method and system based on image processing. Background Technology

[0002] Long-term follow-up after total joint replacement surgery relies on comparative analysis of a series of medical images. Among these, the appearance and continuous widening of a radiolucent line at the prosthesis-bone interface is a key radiographic marker for diagnosing aseptic loosening of the prosthesis. Achieving objective and accurate quantitative assessment of this marker is of great significance for early clinical intervention.

[0003] To improve evaluation efficiency, existing technologies generally employ feature-point-based image registration methods. This method extracts and matches the prosthesis contour and stable bony anatomical landmarks in images from different time points, and calculates a global spatial transformation (usually a rigid or affine transformation). This aims to correct the overall image displacement, rotation, and scaling caused by differences in shooting position, and to establish a unified coordinate benchmark for subsequent image comparison.

[0004] However, such registration methods based on global spatial transformation have inherent limitations. Their technical premise is the assumption that the target region (including the prosthesis and surrounding bone) maintains its geometric shape during follow-up. But in reality, postoperative bone is not a rigid body; it undergoes subtle and non-uniform elastic deformation due to physiological or pathological processes such as stress remodeling and local bone metabolism. This real, localized morphological change is ignored or incorrectly corrected by the existing global rigid transformation model. The direct consequence is that in the core analysis area of ​​the prosthesis-bone interface, images from different time periods, after so-called "registration," do not achieve accurate alignment of the bone's microscopic edges and texture at the microscopic scale. This residual, non-rigid local misalignment creates significant structural noise in subsequent differential analyses aimed at detecting subtle grayscale changes, severely interfering with the identification of early radiolucent lines and potentially completely masking true pathological changes, thus limiting the sensitivity and reliability of existing technologies in early diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide an orthopedic auxiliary examination method and system based on image processing, and to solve the following technical problems: Existing image registration methods based on global rigidity transformation ignore the local elastic deformation of the bone after surgery, resulting in inaccurate alignment of the prosthesis-bone interface area at the microscale, which in turn interferes with the subsequent detection of early pathological changes.

[0006] The objective of this invention can be achieved through the following technical solutions: An image processing-based orthopedic auxiliary examination method includes the following steps: S1. Obtain the orthopedic image sequence of the patient during the postoperative recovery period, arranged in the order of acquisition time, extract the implant contour data and stable bony anatomical point data of the first orthopedic image in the sequence, and establish a baseline image feature set. S2, Based on the reference image feature set, perform spatial transformation on each orthopedic image in the orthopedic image sequence to construct a globally spatially aligned registered image set; S3, for each registered image in the registered image set, based on the plant contour data of the registered image, extend to the image region outside the contour to generate a ring-shaped analysis region, and divide the ring-shaped analysis region into a rigid region defined by the plant contour and an elastic skeletal region outside the contour. S4. Extract the image texture features of the elastic bone region of each registered image and compare them with the image texture features of the corresponding bone region of the first orthopedic image. Generate a spatial transformation field based on the comparison results and apply the spatial transformation field to the elastic bone region of the corresponding registered image to form a local fine-registered image set. S5. Within the annular analysis area of ​​each local fine-registered image, identify pixel clusters whose gray values ​​show a unidirectional trend change along the time sequence and are spatially adjacent, and mark each pixel cluster as an abnormal region. S6 measures the scale of each anomalous region in the direction perpendicular to the inner plant outline, associates its positional information along the inner plant outline, and outputs a structured quantitative report of the anomalous region.

[0007] As a further aspect of the present invention: the specific process of establishing the reference image feature set in S1 is as follows: In the orthopedic image sequence, the orthopedic image acquired first in the time dimension is defined as the reference image. Pixel regions in the reference image with gray values ​​higher than a preset gray threshold are extracted as candidate regions. Morphological closing operations are applied to the candidate regions to obtain the endophyte image portion. Boundary pixel tracking is performed on the inner plant image portion to obtain its outer closed boundary, and the outer closed boundary is encoded as inner plant contour data. The inner plant contour data represents a closed planar contour in the form of an ordered coordinate sequence. In the reference image, the pixel region corresponding to the inner plant image portion is excluded. In the remaining pixel region, the gray-level gradient magnitude of each pixel is calculated. Pixels with gradient magnitudes exceeding a preset gradient threshold are marked as candidate feature points. Among the candidate feature points, points whose spatial coordinates correspond to specific bony structures in a predefined anatomical coordinate system are selected as stable bony anatomical points. The coordinate information of the stable bony anatomical points is recorded to generate stable bony anatomical point data. The implant contour data and the stable bony anatomical point data are stored in the same data set to obtain the reference image feature set.

[0008] As a further aspect of the present invention: in step S2, the specific process of constructing a globally spatially aligned registered image set is as follows: Read the implant contour data and stable bone anatomical point data corresponding to each orthopedic image located after the reference image from the orthopedic imaging sequence; The implant contour data and stable bony anatomical point data stored in the reference image feature set are defined as a reference coordinate set, and the implant contour data and stable bony anatomical point data corresponding to any orthopedic image are defined as a target coordinate set. Based on the reference coordinate set and the target coordinate set, a spatial linear transformation model is constructed. For each pixel location in the orthopedic image, coordinate mapping is performed according to the spatial linear transformation model, and the pixel grayscale value of the orthopedic image is resampled according to the mapped coordinates to obtain a registered image; the registered images corresponding to all orthopedic images are obtained and arranged according to the acquisition time sequence to obtain a registered image set.

[0009] As a further aspect of the present invention: the specific process for generating the annular analysis region in step S3 is as follows: Acquire the inner plant contour data of any frame of the registered image, call the inner plant contour data of the frame of the registered image, the inner plant contour data represents a closed planar contour in the form of an ordered coordinate sequence; based on the ordered coordinate sequence, determine the unit normal vector at each discrete vertex on the planar contour by calculating the difference vector of adjacent coordinate points, the direction of the unit normal vector is constrained to the external space of the planar contour. Starting from each vertex on the planar contour, a spatial offset is made along the direction of its corresponding unit normal vector by a preset Euclidean distance to obtain a corresponding extension vertex. All the extension vertices are connected in sequence to obtain a closed planar contour. The two-dimensional region enclosed by the planar contour and the closed planar contour is designated as the annular analysis region. The region within the annular analysis region that is within the area enclosed by the planar contour is designated as the rigid region. The region between the planar contour and the closed planar contour in the annular analysis region is designated as the elastic skeleton region.

[0010] As a further aspect of the present invention: in S4, the specific generation process of the spatial transformation field is as follows: Obtain the first pixel matrix of the elastic skeleton region of the registered image and the second pixel matrix of the corresponding skeleton region of the first orthopedic image; the first pixel grayscale matrix and the second pixel grayscale matrix have the same dimension; For each pixel in the first pixel matrix, the sum of squared grayscale differences between the local image block centered on it and the corresponding image block at the same coordinate position in the second pixel matrix is ​​calculated to obtain a local difference value. Using the local coordinate displacements of all pixels in the first pixel matrix as the variables to be solved, and the sum of all local differences as the objective function, an optimization equation that minimizes the objective function is established and solved to obtain the coordinate displacement value of each pixel in the first pixel matrix. The coordinate displacement values ​​of each pixel are then arranged according to their original positions to generate a spatial transformation field.

[0011] As a further aspect of the present invention: in step S5, the specific process for obtaining the abnormal region is as follows: From the local fine-registered image set, obtain the pixel grayscale data corresponding to the annular analysis region at multiple acquisition time points, and arrange the pixel grayscale data in chronological order to form a grayscale value time series for each pixel; For each pixel, the gray value time series is trend-fitted, the slope of the gray value change over time is calculated, and pixels with a slope greater than a preset positive threshold are marked as potential increasing pixels, and pixels with a slope less than a preset negative threshold are marked as potential decreasing pixels. Spatial connectivity analysis is performed on the potential increasing pixels and potential decreasing pixels respectively, and pixels that are spatially adjacent and belong to the same category are merged into a candidate pixel set; Calculate the average slope of all pixel changes in each candidate pixel set, mark the candidate pixel set with the average slope greater than a preset positive threshold as a positive abnormal change region, and mark the candidate pixel set with the average slope less than a preset negative threshold as a negative abnormal change region; For each marked abnormal change region, perform morphological closing operation to fill the voids inside the region and smooth the region boundary to obtain the abnormal region.

[0012] As a further aspect of the present invention: in step S6, the specific process for generating the structured abnormal region quantification report is as follows: Acquire any abnormal region and its associated inner plant contour data, wherein the inner plant contour data consists of an ordered sequence of contour points; in the ordered sequence of contour points, locate the contour point closest to the geometric center of the abnormal region as a measurement reference point; based on the measurement reference point and the coordinates of its preceding and following adjacent contour points in the sequence, calculate the contour tangent direction at the measurement reference point, and define the direction perpendicular to the contour tangent direction as the normal measurement direction. Two boundary points passing through the abnormal region are determined along the normal measurement direction. The Euclidean distance between the two boundary points is calculated and recorded as scale data. A one-dimensional arc length parameterized curve is generated based on the inner plant contour data. The coordinates of the abnormal region are projected onto the one-dimensional arc length parameterized curve to obtain the corresponding arc length coordinate set. The minimum continuous coverage interval of the arc length coordinate set is determined. The start and end values ​​of the minimum continuous coverage interval are recorded as extension range information. The scale data and extension range information are bound to the unique identifier of the abnormal region to obtain the quantitative data record of the abnormal region; the quantitative data records of all abnormal regions are summarized to obtain the abnormal region quantitative report.

[0013] As a further aspect of the present invention: S6 further includes comparing the scale data of each abnormal region in the quantitative report based on preset clinical diagnostic standard parameters; when the scale data exceeds a preset first-level threshold, the abnormal region is marked as a first-level warning region; when the scale data exceeds a second-level threshold higher than the first-level threshold, it is marked as a second-level warning region; according to the warning level, the corresponding abnormal region data in the quantitative report are identified and prioritized, and the region data of different warning levels are presented differently in the final output report.

[0014] The present invention also includes an image processing-based orthopedic auxiliary examination system for the above-mentioned image processing-based orthopedic auxiliary examination method, comprising: The image acquisition module is used to acquire orthopedic image sequences of patients during the postoperative recovery period, arranged in the order of acquisition time, extract the implant contour data and stable bony anatomical point data of the first orthopedic image in the sequence, and establish a baseline image feature set. The image registration module is used to perform spatial transformation on each orthopedic image in the orthopedic image sequence based on the reference image feature set, and construct a globally spatially aligned registered image set. The region division module is used to extend the inner plant contour data of each registered image in the registered image set to the image region outside the contour to generate a ring-shaped analysis region, and divide the ring-shaped analysis region into a rigid region defined by the inner plant contour and an elastic skeleton region outside the contour. The data processing module is used to extract the image texture features of the elastic bone region of each registered image and compare them with the image texture features of the corresponding bone region of the first orthopedic image. Based on the comparison results, a spatial transformation field is generated and applied to the elastic bone region of the corresponding registered image to form a local fine-registered image set. The data optimization module is used to identify spatially adjacent pixel clusters whose gray values ​​show a unidirectional trend change along the time sequence within the annular analysis area of ​​each local fine-registered image, and to mark each pixel cluster as an abnormal region. The results generation module measures the scale of each anomalous region in the direction perpendicular to the inner plant outline, associates its positional information along the inner plant outline, and outputs a structured quantitative report of the anomalous region.

[0015] The beneficial effects of this invention are: 1) By establishing a hierarchical registration mechanism of "first global rigid alignment, then local elastic deformation," the microscopic interference caused by the non-rigid physiological changes of the bone itself to image comparison is fundamentally eliminated. This invention first utilizes the implant contour and stable bone points to achieve rigid alignment of the entire image, correcting macroscopic positional differences. Subsequently, an innovative annular analysis region is defined around the implant, divided into an indeformable rigid part (prosthesis) and a deformable elastic bone part. Within the elastic bone part, by calculating the difference in texture features between the current image and the reference image, and solving for a pixel-level displacement field that minimizes this difference, non-rigid deformation adjustment is applied only to this local area. This step is equivalent to simulating and compensating for the microscopic deformation of the bone caused by remodeling or absorption at the pixel level, ensuring that at the key interface for subsequent change detection, images at different time points achieve anatomically accurate pixel-level alignment, providing a geometrically unbiased comparison benchmark for identifying real pathological changes.

[0016] 2) By implementing a change detection strategy that combines temporal trend analysis with spatial morphological constraints, high sensitivity and specificity in identifying early and occult lesions are achieved. Within the annular analysis region with high-precision spatiotemporal registration, the system no longer relies on the absolute grayscale threshold at a single time point. Instead, it constructs a sequence model of the grayscale value change over time for each pixel, detecting and extracting pixels whose grayscale values ​​exhibit a unidirectional, continuous evolution trend (such as continuous darkening). This pattern is a highly specific characteristic of pathological processes such as progressive bone resorption. Subsequently, the system applies spatial connectivity analysis and morphological optimization to pixels that conform to the trend, aggregating discrete pixel signals into abnormal regions with continuous and complete boundaries. This effectively distinguishes stable signals representing true bone resorption or hyperplasia from transient imaging artifacts or random noise, thereby achieving high sensitivity detection and high reliability confirmation of early lesions.

[0017] 3) By performing "precise quantification based on anatomical landmarks" and "structured output directly correlated with clinical diagnostic criteria" on detected abnormal areas, objective and quantifiable key imaging evidence is provided for the early diagnosis of aseptic loosening of prostheses. By precisely correlating identified abnormal signal areas representing potential bone resorption or hyperplasia with the anatomical geometry of the implant contour, for each abnormal area, a unique measurement normal direction perpendicular to the prosthesis-bone interface is automatically determined based on the contour tangent at its location. The maximum width of the area is then calculated along this direction, directly corresponding to the clinically relevant "luminosity line" scale. Simultaneously, all abnormal areas are mapped to a one-dimensional coordinate system based on the prosthesis contour, calibrating their precise longitudinal position and extent along the interface. Finally, these core quantitative indicators directly characterizing the "width," "position," and "range" of the luminosity line are integrated to generate a structured diagnostic data report. This process directly transforms abnormal signals on images into quantitative imaging standards that conform to clinical diagnostic norms and are used to assess the risk of loosening. This allows doctors to base their diagnostic decisions on unified and objective measurement data, significantly improving the consistency and reliability of diagnostic results. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a schematic diagram of an orthopedic auxiliary examination method based on image processing according to the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of an orthopedic auxiliary examination system based on image processing according to the present invention. Detailed Implementation

[0021] 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.

[0022] Please see Figure 1 As shown, the present invention is an orthopedic auxiliary examination method based on image processing, comprising the following steps: S1. Obtain the orthopedic image sequence of the patient during the postoperative recovery period, arranged in the order of acquisition time, extract the implant contour data and stable bony anatomical point data of the first orthopedic image in the sequence, and establish a baseline image feature set. First, the system connects to the Picture Archiving and Communication System (PACS) and retrieves a series of follow-up images after total joint replacement surgery, such as a standard anteroposterior X-ray of the hip joint, according to the patient's identification and examination time. These images are automatically sorted by acquisition date to form a time series. To eliminate potential temporary positional biases in the initial operating room fluoroscopy images, this method defaults to selecting the first outpatient follow-up image (usually taken 4-6 weeks post-surgery) as the "first orthopedic image," i.e., the baseline image. This image reflects the stable state of the prosthesis after the patient begins normal weight-bearing and is more valuable for follow-up comparison.

[0023] The processing of reference images involves two core operations: Implant contour data extraction: Due to the significant high grayscale features of metallic implants in X-ray images, the system employs an adaptive threshold segmentation algorithm to initially extract connected regions with grayscale values ​​higher than a preset threshold as candidate implant regions. Considering that the prosthesis edges may appear slightly blurred due to partial volume effects during imaging, morphological closing operations are then applied to the segmentation results to fill small voids and smooth the boundaries. Finally, an edge tracking algorithm (such as the Moore-Neighbor boundary tracking method) is performed on the processed binary image to obtain an ordered sequence of pixel coordinates describing the outer edge of the prosthesis (such as the femoral stem or acetabular cup), which is encoded as "implant contour data".

[0024] Extraction of stable bony anatomical point data: To provide additional, prosthesis-independent, rigid references for subsequent registration, stable feature points on the bone need to be selected. First, the extracted implant area is masked from the baseline image to avoid interference. In the remaining bony region, the grayscale gradient amplitude of each pixel is calculated, and points with local gradient maxima are marked as candidate feature points. These points typically correspond to cortical bone edges, bony ridges, or specific anatomical landmarks (such as the tip of the lesser trochanter of the femur or a specific angle of the ischial tuberosity). Crucially, not all candidate points are stable. By comparing a small number of validation images taken of the same patient within very short time intervals (e.g., at different angles during surgery), points with minimal spatial coordinate variation across different images are selected. These points are considered "stable bony anatomical points" least affected by soft tissue peristalsis and slight positional changes. Finally, the two-dimensional pixel coordinates of at least two such points are recorded to generate "stable bony anatomical point data." The above contour data is associated with and stored with the anatomical point data, thus forming the "baseline image feature set" used for analysis throughout the follow-up period.

[0025] S2, Based on the reference image feature set, perform spatial transformation on each orthopedic image in the orthopedic image sequence to construct a globally spatially aligned registered image set; First, the extraction process in S1 is repeated on the image to obtain the "current implant contour data" and "current stable bony anatomical point data" of the current image. Next, a registration model is established using the reference image feature set as a fixed reference. The two sets of data (contour key points and anatomical points) from the reference and the current image are used as corresponding point sets, and a least-squares fitting algorithm is used to solve for an optimal affine transformation matrix. This matrix can describe rotation, translation, scaling, and shearing, sufficient to correct for macroscopic positional differences. During calculation, since the number of implant contour points is much greater than that of anatomical points, their fitting weight is higher, ensuring a high degree of overlap in the prosthesis shape after registration. Then, the solved affine transformation matrix is ​​uniformly applied to the entire current image. By resampling the original image pixels (e.g., using bilinear interpolation), a new image is generated whose internal prosthesis and bony landmark positions are almost perfectly aligned with the reference image; this is called the "registered image." This process is performed sequentially on all subsequent images in the sequence, ultimately resulting in a "globally spatially aligned registered image set" where the prosthesis positions in all images are aligned. At this point, images from different time points provide the basis for pixel-level comparison. S3, for each registered image in the registered image set, based on the plant contour data of the registered image, extend to the image region outside the contour to generate a ring-shaped analysis region, and divide the ring-shaped analysis region into a rigid region defined by the plant contour and an elastic skeletal region outside the contour. Generating the annular analysis region: The implant contour data from the image is read. For each point on this closed contour, the unit normal vector at that point is calculated, pointing outwards (i.e., towards the bone side). Then, along the normal vector direction at each point, a fixed distance (e.g., 5 mm) is extended outwards. Connecting all the extended endpoints forms a new, similar to but wider than the original closed contour. The annular band between the original contour and this new contour is defined as the "annular analysis region." This region acts like an "observation corridor," tightly enclosing the prosthesis-bone interface.

[0026] Region Division: Within the annular analysis area, logical division is performed using the original implant outline as the boundary: The area within the outline corresponds to the metal prosthesis itself, which will not deform during the follow-up period, and is therefore defined as the "rigid region". The area outside the outline to the outer boundary corresponds to the host bone adjacent to the prosthesis, which may undergo changes in density and micromorphology due to stress shielding, bone resorption, or hyperplasia, and is therefore defined as the "elastic bone region".

[0027] S4. Extract the image texture features of the elastic bone region of each registered image and compare them with the image texture features of the corresponding bone region of the first orthopedic image. Generate a spatial transformation field based on the comparison results and apply the spatial transformation field to the elastic bone region of the corresponding registered image to form a local fine-registered image set. Despite global registration via S2, the trabecular texture of images at different time points may exhibit microscopic displacement due to bone remodeling within the "elastic skeleton region." For a subsequent follow-up image, local texture features are extracted from its "elastic skeleton region" and the corresponding skeleton region of the baseline image, respectively. One feasible implementation is to divide each region into multiple overlapping small image patches and calculate the Local Binary Pattern (LBP) histogram or Gray-Level Co-occurrence Matrix (GLCM) features for each image patch. These features are insensitive to absolute grayscale changes but sensitive to texture structure.

[0028] Using the texture features of the reference image as a reference, a dense displacement vector field, or "spatial transformation field," is calculated using optimization algorithms (such as the Demons algorithm or a B-spline-based elastic registration algorithm). This field defines how far (vector) each pixel in the current image's "elastic skeleton region" needs to move to achieve the best match between its local texture pattern and the reference image. This spatial transformation field is applied only to the "elastic skeleton region" of the current image, performing non-rigid resampling and repositioning of pixels within that region. Meanwhile, the "rigid region" (the prosthesis portion) remains unchanged. After processing, the bone texture of the current image in the critical prosthesis-bone interface region achieves sub-pixel-level fine alignment with the reference image. This process is repeated for each follow-up image to obtain a "locally finely registered image set." Thus, the temporal images achieve truly precise alignment within the target region, eliminating anatomical deformation interference.

[0029] S5. Within the annular analysis area of ​​each local fine-registered image, identify pixel clusters whose gray values ​​show a unidirectional trend change along the time sequence and are spatially adjacent, and mark each pixel cluster as an abnormal region. Within the annular analysis region, for the same spatial location (pixel coordinates), the grayscale values ​​at all time points (reference image and each locally finely registered image) are extracted to form a time series. Linear regression or a non-parametric trend test (such as the Mann-Kendall test) is performed on the time series of each pixel to calculate the slope of its grayscale change over time. Pixels with significantly negative slopes (continuous darkening, indicating bone resorption) or significantly positive slopes (continuous brightening, indicating bone hyperplasia) are marked as "trend pixels".

[0030] In two-dimensional space, adjacent "trend pixels" with the same trend direction are subjected to connected component analysis and aggregated into initially selected pixel clusters. Subsequently, morphological operations (such as closing operations) are applied to these clusters to fill internal micro-voids, smooth irregular boundaries, and filter out clusters with excessively small areas (which may be noise). Each resulting continuous, smooth pixel cluster is then labeled as an "abnormal region." This process effectively distinguishes between genuine, spatially continuous pathological changes and scattered imaging noise.

[0031] S6 measures the scale of each anomalous region in the direction perpendicular to the inner plant outline, associates its positional information along the inner plant outline, and outputs a structured quantitative report of the anomalous region.

[0032] For an "abnormal region," first locate the nearest implant outline segment and calculate the tangent direction of the line along that segment to the point closest to the geometric center of the region. The perpendicular direction of this tangent is the "normal measurement direction." Traverse the region along this direction and find the farthest points of the region's boundary in both directions. Calculate the Euclidean distance between these two points, which serves as the width of the region in the direction perpendicular to the prosthesis surface. This is a core clinical indicator for assessing radiolucent lines or the thickness of hyperplastic bone.

[0033] The implant outline is unfolded into a continuous one-dimensional parametric coordinate system from the start point to the end point. All pixels of the anomalous region are projected onto this coordinate line to obtain the arc length it covers. The coordinates of the start point, end point, and midpoint of this range are recorded. Simultaneously, this coordinate system is mapped to standard anatomical zones (such as the Gruen and DeLee zones for the hip joint) to form an anatomical location description such as "located in Gruen Zone 2 of the femoral stem".

[0034] Information such as the width, anatomical location, arc length range, trend of change (absorption / proliferation), and area of ​​each abnormal region is integrated into a structured data record. Data from all abnormal regions is then aggregated to automatically generate a quantitative report. The report can be presented in a combination of tables and diagrams, highlighting abnormal regions on the prosthesis outline diagram and providing detailed measurement data. This offers physicians clear, objective evidence that can be directly used for clinical diagnosis and follow-up comparisons.

[0035] In a preferred embodiment of the present invention, the specific process of establishing the reference image feature set in S1 is as follows: In the orthopedic image sequence, the orthopedic image acquired first in the time dimension is defined as the reference image, and the pixel regions in the reference image with gray values ​​higher than a preset threshold are extracted as candidate regions. Morphological closing operations are applied to the candidate regions to obtain the endophyte image portion. Boundary pixel tracking is performed on the inner plant image portion to obtain its outer closed boundary. The outer closed boundary is encoded as inner plant contour data. The pixel region corresponding to the inner plant image portion is excluded in the reference image. The gray-level gradient magnitude of each pixel point is calculated in the remaining pixel region. Pixel points with gradient magnitude exceeding a preset gradient threshold are marked as candidate feature points. Among the candidate feature points, points whose spatial coordinates correspond to specific bony structures in a predefined anatomical coordinate system are selected as stable bony anatomical points. The coordinate information of the stable bony anatomical points is recorded to generate stable bony anatomical point data. The implant contour data and the stable bony anatomical point data are stored in the same data set to obtain the reference image feature set.

[0036] The earliest image was chosen as the starting point for the entire follow-up timeline. This usually corresponds to the image taken when the patient returns to the hospital for the first time after surgery. For example, a pelvic X-ray taken six weeks after a total hip replacement surgery. This image was chosen as the baseline because the acute tissue edema caused by the surgery has basically subsided, the prosthesis position has stabilized, and there has been no significant long-term bone remodeling. Therefore, this image is most likely to represent an "initial and stable" anatomical state, providing a reliable starting point for all subsequent comparisons.

[0037] Subsequently, two core elements need to be precisely located on this baseline image: the implanted prosthesis and the patient's own inherent, stable bony landmarks. For prosthesis localization, the physical property of metal absorbing a large amount of radiation and exhibiting extremely high brightness in X-ray images is utilized. By setting a threshold higher than the grayscale of bone and soft tissue, all potentially bright pixel groups belonging to the metal prosthesis in the entire image are initially screened out. These pixel groups constitute candidate regions. However, simple threshold segmentation may result in small holes or edge burrs in the extracted regions due to image noise or blurred prosthesis edges. Therefore, morphological closing operations are then applied to these candidate regions. The geometric principle of this operation is to first expand the region boundary outward through a dilation operation to close small cracks, and then shrink the boundary back to an approximate original position through an erosion operation. The effect is to smooth the contour and fill the black holes inside the region caused by imaging noise, thereby obtaining a complete and continuous "implant image portion", such as a connected white region representing the femoral stem and acetabular cup of the hip joint.

[0038] Next, boundary pixel tracking is performed. The algorithm starts from any pixel on the edge of the region, searches for and connects to the next adjacent edge pixel according to a preset search direction, and repeats this process until it returns to the starting point. This results in a sequentially arranged ring of pixel coordinates, which encodes the precise spatial path of the boundary between the prosthesis and surrounding tissues—the "implant contour data." Next, the system needs to find stable anatomical reference points that are independent of the prosthesis and belong to the patient. To eliminate the strong interference of the metal prosthesis on image analysis, the pixel regions corresponding to the extracted implant image portion in the reference image are temporarily zeroed out or masked, allowing the system to focus on analyzing the remaining skeletal region. In these skeletal regions, the grayscale gradient magnitude of a point is evaluated by calculating the grayscale difference between each pixel and its neighboring pixels (up, down, left, and right). Physically, this measures the intensity of the brightness change at that point in the image. Because bone edges, ridges, or specific anatomical protrusions (such as the tip of the lesser trochanter of the femur) appear as rapid transitions from light to dark on X-rays, the gradient magnitudes at these locations are significantly higher than in flat cortical areas or the interior of cancellous bone. By setting an appropriate gradient threshold, these high-gradient points can be labeled as "candidate feature points." However, not all high-gradient points are suitable as stable references for long-term follow-up. For example, some points may be located on the cancellous bone texture, and their positions may shift with small changes in the shooting angle. Therefore, the final step is to filter these candidate feature points based on predefined anatomical knowledge. For example, in a standard lateral anatomical coordinate system of the hip joint, only those points whose coordinate positions match those of recognized bony structures that are not easily displaced by changes in posture (such as the apex of the most anterior and inferior aspect of the lesser trochanter of the femur, or a specific bony prominence at the junction of the ischial tuberosity and the lower edge of the acetabulum) will be ultimately adopted. The pixel coordinates of these points are recorded to generate "stable bony anatomical point data." Finally, the contour data describing the shape of the prosthesis and the coordinate point data describing the inherent anatomical position are stored together in a dataset, which constitutes the "baseline image feature set" used to govern the entire analysis process.

[0039] On the one hand, by extracting implant contour data, the system accurately grasps the geometric shape and spatial positioning of the prosthesis in its baseline state. This provides a crucial basis for correcting prosthesis projection deformation caused by different shooting angles in subsequent steps, ensuring pixel-level alignment of the prosthesis in images from different periods. On the other hand, by additionally extracting stable anatomical point data independent of the prosthesis and derived from the patient's own skeleton, it is equivalent to setting up a "built-in coordinate system" hidden within the skeleton in the analysis. This coordinate system will not fail due to possible minor displacement of the prosthesis or local remodeling of the surrounding bone, thus providing redundant verification and reinforcement for the entire registration process. These two types of data complement each other, together forming a robust baseline framework. Ultimately, the successful execution of this step lays the foundation for fair comparison of all subsequent analyses, enabling the system to accurately distinguish between genuine physiological or pathological changes in the skeleton over time (such as bone resorption around the prosthesis) and differences in image appearance caused by differences in imaging techniques or local tissue elastic deformation, thereby directly targeting the ultimate clinical goal of "early detection of signs of prosthesis loosening."

[0040] In another preferred embodiment of the present invention, the specific process of constructing a globally spatially aligned registered image set in step S2 is as follows: Read the implant contour data and stable bone anatomical point data corresponding to each orthopedic image located after the reference image from the orthopedic imaging sequence; The implant contour data and stable bony anatomical point data stored in the reference image feature set are defined as a reference coordinate set, and the implant contour data and stable bony anatomical point data corresponding to any orthopedic image are defined as a target coordinate set. Based on the reference coordinate set and the target coordinate set, a spatial linear transformation model is constructed. For each pixel location in the orthopedic image, coordinate mapping is performed according to the spatial linear transformation model, and the pixel grayscale value of the orthopedic image is resampled according to the mapped coordinates to obtain a registered image; the registered images corresponding to all orthopedic images are obtained and arranged according to the acquisition time sequence to obtain a registered image set.

[0041] In another preferred embodiment of the present invention, the specific process for generating the ring-shaped analysis region in step S3 is as follows: Acquire the inner plant contour data of any frame of registered image, the inner plant contour data representing a closed planar contour in the form of an ordered coordinate sequence; based on the ordered coordinate sequence, determine the unit normal vector at each discrete vertex on the planar contour by calculating the difference vector of adjacent coordinate points, the direction of the unit normal vector being constrained to the external space of the planar contour; Starting from each vertex on the planar contour, a spatial offset is made along the direction of its corresponding unit normal vector by a preset Euclidean distance to obtain a corresponding extension vertex. All the extension vertices are connected in sequence to obtain a closed planar contour. The two-dimensional region enclosed by the planar contour and the closed planar contour is designated as the annular analysis region. The region within the annular analysis region that is within the area enclosed by the planar contour is designated as the rigid region. The region between the planar contour and the closed planar contour in the annular analysis region is designated as the elastic skeleton region.

[0042] After prosthesis implantation, all clinically significant biological responses—whether indicating bone resorption or reflecting stable osseointegration—are concentrated at the narrow interface between the prosthesis surface and the host bone in direct contact or adjacent areas. However, the entire image contains a large amount of irrelevant information, such as distant bones, soft tissues, and background. Analyzing this global scope would not only be computationally burdensome, but more importantly, irrelevant image changes far from the interface (such as changes in soft tissue thickness or deformations in other areas caused by differences in projection angles) would act as significant noise, drowning out the truly critical, weak signals. Therefore, the primary objective of this step is to implement "precise focusing" by generating a constant-width annular band closely following the outer edge of the prosthesis contour. This strictly limits the analytical field of view for all subsequent complex calculations to the "bullseye" region where problems are most likely to be discovered, greatly improving the signal-to-noise ratio and efficiency of the analysis. Further dividing this annular band into "rigid regions" and "elastic bone regions" is a "differentiated processing" strategy based on a precise understanding of the physical properties of different parts within the region. The internal region defined by the implant contour corresponds to the metal prosthesis itself, which is absolutely rigid within the human body and will not undergo any deformation or physiological changes in grayscale. Marking it as a rigid region means that in subsequent analysis (especially during local registration), this region will be regarded as a stable and unchanging reference anchor point, prohibiting any displacement adjustment. This ensures the strict invariance of the prosthesis's own geometry as a reference coordinate system. The annular region outside the contour corresponds to the living skeleton, which may undergo subtle density changes and geometric deformations due to stress remodeling, inflammation, or dissolution. Therefore, it is marked as an elastic skeleton region, granting it the "authority" for subsequent local elastic deformation adjustments. The essence of this division is to establish a physically accurate model at the image processing level: a fixed core (prosthesis) is encased in a shell (interface bone) that may experience slight peristalsis. Its greatest advantage is that it provides clear physical constraints and operational objects for the next step (S4) of implementing local non-rigid registration—registration (i.e., alignment) only needs to be performed within the "elastic skeleton region," using the "rigid region" as an immovable reference frame. This approach allows the system to compensate for microscopic deformations of the bone caused by physiology or early pathology, while preventing excessive deformation or drift of the entire structure, thus ensuring the uniqueness and accuracy of the interpretation.

[0043] In another preferred embodiment of the present invention, the specific generation process of the spatial transformation field in step S4 is as follows: Obtain the first pixel matrix of the elastic skeleton region of the registered image and the second pixel matrix of the corresponding skeleton region of the first orthopedic image; the first pixel grayscale matrix and the second pixel grayscale matrix have the same dimension; For each pixel in the first pixel matrix, the sum of squared grayscale differences between the local image block centered on it and the corresponding image block at the same coordinate position in the second pixel matrix is ​​calculated to obtain a local difference value. Using the local coordinate displacements of all pixels in the first pixel matrix as the variables to be solved, and the sum of all local differences as the objective function, an optimization equation that minimizes the objective function is established and solved to obtain the coordinate displacement value of each pixel in the first pixel matrix. The coordinate displacement values ​​of each pixel are then arranged according to their original positions to generate a spatial transformation field.

[0044] One image patch is a region of elastic bone in the currently processed, globally registered follow-up image, and the other is a region of bone in the initially established reference image that corresponds exactly to its anatomical location. Since the previous global registration step aligned the overall spatial coordinate systems of the two images, these two image patches are identical in the number of rows and columns of their matrices. This is analogous to comparing plots of land within the same coordinate range cut from two overlaid maps. If the two image patches represent the exact same bone tissue without any deformation, then the texture of each local region should be almost identical. However, if the bone undergoes slight elastic deformation, this manifests as a shift in the position of the local texture pattern. To quantify this local shift, this invention employs a point-by-point comparison method: for each pixel in the follow-up image patch (first pixel matrix), a small square window, such as a 7-pixel by 7-pixel local image patch, is cut out centered on that pixel. This window contains the texture information surrounding that pixel. Simultaneously, in the reference image patch (second pixel matrix), a local image patch of the same size is cut out centered on the exact same row and column coordinates. Next, the difference in grayscale values ​​of corresponding pixels in these two small image patches is calculated, and the squares of all differences are summed to obtain a single value, the "local difference value". The significance of this sum-of-squares operation is that it not only considers the magnitude of the grayscale difference but also amplifies the impact of significant differences through squaring, thus more sensitively reflecting the degree of dissimilarity between two local regions. If the textures of the two local image patches are completely identical, this value will be close to zero; if there is texture misalignment, the value will increase. After performing the above operation on every pixel in the first pixel matrix, a set of countless local difference values ​​is obtained.

[0045] By treating the assumed displacement of each pixel in the follow-up image patch (i.e., how far it needs to move in the row and column directions) as unknown variables, and summing the new local difference values ​​obtained after recalculating all pixel positions, a function of these displacement variables is constructed, called the objective function. The value of this objective function intuitively represents the overall dissimilarity of the two image patches after displacement adjustment. The system then automatically finds the specific set of displacement variable values ​​that minimizes the value of this objective function through iterative optimization mathematical methods, i.e., finding a pixel-level displacement scheme that best matches the two image patches as a whole. The final calculated displacements in the row and column directions corresponding to each pixel in the first pixel matrix, arranged according to the original matrix positions of the pixels, form a two-dimensional, dense displacement vector field, which is the "spatial transformation field" to be generated in this step. This field is essentially a detailed map that precisely indicates the direction and distance that each pixel in the elastic skeleton region of the follow-up image needs to move to perfectly align with the reference image.

[0046] After artificial joint replacement surgery, the living bone surrounding the prosthesis is not a monolithic structure. It undergoes slow adaptive remodeling due to changes in the mechanical environment. This remodeling manifests in imaging as minute compressions, stretching, or twisting of local trabecular structures—a non-uniform, non-rigid geometric change. Traditional global rotation and translation transformations are completely ineffective in this regard; forcibly applying them only results in microscopic misalignment of key interface areas. The advantage of this approach is that it no longer treats the bone region as a whole for coarse processing. Instead, it acknowledges its internal deformability and adopts a "divide and conquer" strategy. Building upon the already largely global alignment, it focuses on finding the optimal alignment within smaller local windows. By calculating the grayscale differences in the neighborhood of each pixel and seeking a displacement scheme that minimizes the overall difference, the system is essentially simulating a local, adaptive "elastic stretch," allowing the bone texture in the follow-up images to be subtly adjusted, like an elastic net, to conform to the texture of the reference image. This allows for pixel-level, ultra-high-precision anatomical alignment at the prosthesis-bone interface—the most critical and potentially disease-prone area—surpassing overall rigidity. Only with this level of alignment can subsequent steps of detecting grayscale changes over time be truly reliable, as any residual geometric misalignment will be mistaken for a change in grayscale (i.e., bone density). Therefore, this step is the technological cornerstone of this approach's high-sensitivity detection of early signs of loosening. By compensating for interference from physiological deformation, it ensures that the "abnormal change signal" ultimately captured by the system is highly likely to be genuine pathological bone resorption or hyperplasia, rather than artifacts caused by poor registration, thus significantly improving the accuracy and reliability of the entire method in early diagnosis.

[0047] In another preferred embodiment of the present invention, the specific process of obtaining the abnormal region in step S5 is as follows: From the local fine-registered image set, obtain the pixel grayscale data corresponding to the annular analysis region at multiple acquisition time points, and arrange the pixel grayscale data in chronological order to form a grayscale value time series for each pixel; For each pixel, the gray value time series is trend-fitted, the slope of the gray value change over time is calculated, and pixels with a slope greater than a preset positive threshold are marked as potential increasing pixels, and pixels with a slope less than a preset negative threshold are marked as potential decreasing pixels. Spatial connectivity analysis is performed on the potential increasing pixels and potential decreasing pixels respectively, and pixels that are spatially adjacent and belong to the same category are merged into a candidate pixel set; Calculate the average slope of all pixel changes in each candidate pixel set, mark the candidate pixel set with the average slope greater than a preset positive threshold as a positive abnormal change region, and mark the candidate pixel set with the average slope less than a preset negative threshold as a negative abnormal change region; For each marked abnormal change region, perform morphological closing operation to fill the voids inside the region and smooth the region boundary to obtain the abnormal region.

[0048] The first step involves constructing a grayscale history for each tiny image unit (pixel) across multiple follow-ups, essentially creating a "behavioral profile" for each point. The slope of this change is calculated through trend fitting, essentially determining whether the density of that point is continuously increasing or decreasing. Pixels with slopes exceeding positive or negative thresholds are marked, completing the initial screening. This step is based on the principle that random noise fluctuations do not form stable trends, while true bone resorption or hyperplasia is a slow but directional cumulative process. However, temporal trends alone are insufficient, as scattered, isolated trend pixels may still be some kind of noise or extremely localized, clinically insignificant minor variations. Therefore, the next step introduces spatial connectivity analysis, requiring the marked pixels to be physically connected into patches. This is based on an important clinical fact: meaningful pathological changes (such as radiolucent lines) are usually not a few isolated points, but rather regions with a certain degree of spatial continuity. Aggregating adjacent pixels with consistent trends into a "candidate pixel set" means that only signals exhibiting coherent consistency in both time and space deserve further attention. Further calculation of the average slope of change of pixels within each set and secondary threshold judgment confirm the overall trend strength of the region, aiming to eliminate false regions that, although spatially connected, have weak average trends and may be formed by edge noise sets. Finally, morphological closing operations are performed on the selected regions to fill in small holes caused by image noise or insufficient local contrast, and to smooth any jagged irregular protrusions at their boundaries, thus obtaining "abnormal regions" with clear boundaries and complete shapes. The ultimate goal of this entire process is to achieve highly specific identification of early, subtle lesions. Its greatest advantage lies in its strict distinction between "change" and "abnormal change": it is understood that this invention does not respond to all grayscale differences, but only to those organized change patterns that exhibit a clear unidirectional development trend and form continuous regions in anatomical space. This pattern highly specifically corresponds to key pathological processes such as "progressive bone resorption" or "progressive bone hyperplasia."

[0049] In another preferred embodiment of the present invention, the specific process of generating the structured anomaly region quantification report in step S6 is as follows: Acquire any abnormal region and its associated inner plant contour data, wherein the inner plant contour data consists of an ordered sequence of contour points; in the ordered sequence of contour points, locate the contour point closest to the geometric center of the abnormal region as a measurement reference point; based on the measurement reference point and the coordinates of its preceding and following adjacent contour points in the sequence, calculate the contour tangent direction at the measurement reference point, and define the direction perpendicular to the contour tangent direction as the normal measurement direction. Two boundary points passing through the abnormal region are determined along the normal measurement direction. The Euclidean distance between the two boundary points is calculated and recorded as scale data. A one-dimensional arc length parameterized curve is generated based on the inner plant contour data. The coordinates of the abnormal region are projected onto the one-dimensional arc length parameterized curve to obtain the corresponding arc length coordinate set. The minimum continuous coverage interval of the arc length coordinate set is determined. The start and end values ​​of the minimum continuous coverage interval are recorded as extension range information. The scale data and extension range information are bound to the unique identifier of the abnormal region to obtain the quantitative data record of the abnormal region; the quantitative data records of all abnormal regions are summarized to obtain the abnormal region quantitative report.

[0050] By finding the prosthesis contour point closest to the abnormal area and calculating the tangent direction at that point, it was ensured that subsequent width measurements were performed in the anatomically correct direction perpendicular to the prosthesis-bone interface. This ensured that the measured "scale data" directly corresponded to the clinically relevant radiolucent lines or the true thickness of the hyperplastic bone, avoiding errors caused by arbitrary measurement directions and guaranteeing the clinical interpretability of the data. By converting the prosthesis contour into arc-length coordinates and projecting them onto the abnormal area, the three-dimensional, curved interface problem was reduced to a standard one-dimensional linear coordinate system. The resulting "extension range information" not only accurately recorded the length and specific segments of the abnormality distributed along the prosthesis surface, but more importantly, it achieved a standardized description of the location. This allows doctors to uniquely and accurately reproduce the location of the abnormality on the prosthesis model based on arc-length coordinates, regardless of when or where they view the report, achieving unambiguous localization and the possibility of follow-up comparison. Finally, binding and summarizing scale, location, and other data with region identifiers means that each suspicious pixel cluster is transformed into a structured data object containing core quantitative features. The entire report is transformed from an image requiring subjective interpretation into a list of objective data items that can be directly used for analysis and judgment. This approach is crucial to the ultimate goal of the program. It moves from "seeing" suspicious changes to "measuring" and "locating" those changes, transforming subtle signs of early loosening from mere shadows on a screen that require repeated comparison and speculation into report items with clear numerical values ​​and coordinates. This greatly improves the objectivity, repeatability, and accuracy of the diagnosis, as well as the precision of comparisons between different time points, thereby truly achieving the core goal of accurately quantifying and assessing early pathological changes.

[0051] In another preferred embodiment of the present invention, step S6 further includes comparing the scale data of each abnormal region in the quantitative report based on preset clinical diagnostic standard parameters; when the scale data exceeds a preset first-level threshold, the abnormal region is marked as a first-level warning region; when the scale data exceeds a second-level threshold higher than the first-level threshold, it is marked as a second-level warning region; according to the warning level, the corresponding abnormal region data in the quantitative report are identified and prioritized, and the region data of different warning levels are presented differently in the final output report.

[0052] See Figure 2 The present invention also includes an image processing-based orthopedic auxiliary examination system for implementing the above-described image processing-based orthopedic auxiliary examination method, comprising: The image acquisition module is used to acquire orthopedic image sequences of patients during the postoperative recovery period, arranged in the order of acquisition time, extract the implant contour data and stable bony anatomical point data of the first orthopedic image in the sequence, and establish a baseline image feature set. The image registration module is used to perform spatial transformation on each orthopedic image in the orthopedic image sequence based on the reference image feature set, and construct a globally spatially aligned registered image set. The region division module is used to extend the inner plant contour data of each registered image in the registered image set to the image region outside the contour to generate a ring-shaped analysis region, and divide the ring-shaped analysis region into a rigid region defined by the inner plant contour and an elastic skeleton region outside the contour. The data processing module is used to extract the image texture features of the elastic bone region of each registered image and compare them with the image texture features of the corresponding bone region of the first orthopedic image. Based on the comparison results, a spatial transformation field is generated and applied to the elastic bone region of the corresponding registered image to form a local fine-registered image set. The data optimization module is used to identify spatially adjacent pixel clusters whose gray values ​​show a unidirectional trend change along the time sequence within the annular analysis area of ​​each local fine-registered image, and to mark each pixel cluster as an abnormal region. The results generation module measures the scale of each anomalous region in the direction perpendicular to the inner plant outline, associates its positional information along the inner plant outline, and outputs a structured quantitative report of the anomalous region.

[0053] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for orthopedic auxiliary examination based on image processing, characterized in that, Includes the following steps: S1. Obtain the orthopedic image sequence of the patient during the postoperative recovery period, arranged in the order of acquisition time, extract the implant contour data and stable bony anatomical point data of the first orthopedic image in the sequence, and establish a baseline image feature set. S2, Based on the reference image feature set, perform spatial transformation on each orthopedic image in the orthopedic image sequence to construct a globally spatially aligned registered image set; S3, for each registered image in the registered image set, based on the plant contour data of the registered image, extend to the image region outside the contour to generate a ring-shaped analysis region, and divide the ring-shaped analysis region into a rigid region defined by the plant contour and an elastic skeletal region outside the contour. S4. Extract the image texture features of the elastic bone region of each registered image and compare them with the image texture features of the corresponding bone region of the first orthopedic image. Generate a spatial transformation field based on the comparison results and apply the spatial transformation field to the elastic bone region of the corresponding registered image to form a local fine-registered image set. S5. Within the annular analysis area of ​​each local fine-registered image, identify pixel clusters whose gray values ​​show a unidirectional trend change along the time sequence and are spatially adjacent, and mark each pixel cluster as an abnormal area. S6 measures the scale of each anomalous region in the direction perpendicular to the inner plant outline, associates its positional information along the inner plant outline, and outputs a structured quantitative report of the anomalous region.

2. The orthopedic auxiliary examination method based on image processing according to claim 1, characterized in that, In S1, the specific process of establishing the reference image feature set is as follows: In the orthopedic image sequence, the orthopedic image acquired first in the time dimension is defined as the reference image. Pixel regions in the reference image with gray values ​​higher than a preset gray threshold are extracted as candidate regions. Morphological closing operations are applied to the candidate regions to obtain the endophyte image portion. Boundary pixel tracking is performed on the inner plant image portion to obtain its outer closed boundary, and the outer closed boundary is encoded as inner plant contour data. The inner plant contour data represents a closed planar contour in the form of an ordered coordinate sequence. In the reference image, the pixel region corresponding to the inner plant image portion is excluded. In the remaining pixel region, the gray-level gradient magnitude of each pixel is calculated. Pixels with gradient magnitudes exceeding a preset gradient threshold are marked as candidate feature points. Among the candidate feature points, points whose spatial coordinates correspond to specific bony structures in a predefined anatomical coordinate system are selected as stable bony anatomical points. The coordinate information of the stable bony anatomical points is recorded to generate stable bony anatomical point data. The implant contour data and the stable bony anatomical point data are stored in the same data set to obtain the reference image feature set.

3. The orthopedic auxiliary examination method based on image processing according to claim 2, characterized in that, In S2, the specific process of constructing a globally spatially aligned registered image set is as follows: Read the implant contour data and stable bone anatomical point data corresponding to each orthopedic image located after the reference image from the orthopedic imaging sequence; The implant contour data and stable bony anatomical point data stored in the reference image feature set are defined as a reference coordinate set, and the implant contour data and stable bony anatomical point data corresponding to any orthopedic image are defined as a target coordinate set. Based on the reference coordinate set and the target coordinate set, a spatial linear transformation model is constructed. For each pixel location in the orthopedic image, coordinate mapping is performed according to the spatial linear transformation model, and the pixel grayscale value of the orthopedic image is resampled according to the mapped coordinates to obtain a registered image; the registered images corresponding to all orthopedic images are obtained and arranged according to the acquisition time sequence to obtain a registered image set.

4. The orthopedic auxiliary examination method based on image processing according to claim 1, characterized in that, In S3, the specific process for generating the annular analysis region is as follows: Acquire the inner plant contour data of any frame of registered image, the inner plant contour data representing a closed planar contour in the form of an ordered coordinate sequence; based on the ordered coordinate sequence, determine the unit normal vector at each discrete vertex on the planar contour by calculating the difference vector of adjacent coordinate points, the direction of the unit normal vector being constrained to the external space of the planar contour; Starting from each vertex on the planar contour, a spatial offset is made along the direction of its corresponding unit normal vector by a preset Euclidean distance to obtain a corresponding extension vertex. All the extension vertices are connected in sequence to obtain a closed planar contour. The two-dimensional region enclosed by the planar contour and the closed planar contour is designated as the annular analysis region. The region within the annular analysis region that is within the area enclosed by the planar contour is designated as the rigid region. The region between the planar contour and the closed planar contour in the annular analysis region is designated as the elastic skeleton region.

5. The orthopedic auxiliary examination method based on image processing according to claim 1, characterized in that, In S4, the specific generation process of the spatial transformation field is as follows: Obtain the first pixel matrix of the elastic skeleton region of the registered image and the second pixel matrix of the corresponding skeleton region of the first orthopedic image; the first pixel grayscale matrix and the second pixel grayscale matrix have the same dimension; For each pixel in the first pixel matrix, the sum of squared grayscale differences between the local image block centered on it and the corresponding image block at the same coordinate position in the second pixel matrix is ​​calculated to obtain a local difference value. Using the local coordinate displacements of all pixels in the first pixel matrix as the variables to be solved, and the sum of all local differences as the objective function, an optimization equation that minimizes the objective function is established and solved to obtain the coordinate displacement value of each pixel in the first pixel matrix. The coordinate displacement values ​​of each pixel are then arranged according to their original positions to generate a spatial transformation field.

6. The orthopedic auxiliary examination method based on image processing according to claim 1, characterized in that, In step S5, the specific process for obtaining the abnormal region is as follows: From the local fine-registered image set, obtain the pixel grayscale data corresponding to the annular analysis region at multiple acquisition time points, and arrange the pixel grayscale data in chronological order to form a grayscale value time series for each pixel; For each pixel, the gray value time series is trend-fitted, the slope of the gray value change over time is calculated, and pixels with a slope greater than a preset positive threshold are marked as potential increasing pixels, and pixels with a slope less than a preset negative threshold are marked as potential decreasing pixels. Spatial connectivity analysis is performed on the potential increasing pixels and potential decreasing pixels respectively, and pixels that are spatially adjacent and belong to the same category are merged into a candidate pixel set; Calculate the average slope of all pixel changes in each candidate pixel set, mark the candidate pixel set with the average slope greater than a preset positive threshold as a positive abnormal change region, and mark the candidate pixel set with the average slope less than a preset negative threshold as a negative abnormal change region; For each marked abnormal change region, perform morphological closing operation to fill the voids inside the region and smooth the region boundary to obtain the abnormal region.

7. The orthopedic auxiliary examination method based on image processing according to claim 1, characterized in that, In step S6, the specific process for generating the structured anomaly region quantification report is as follows: Acquire any abnormal region and its associated inner plant contour data, wherein the inner plant contour data consists of an ordered sequence of contour points; in the ordered sequence of contour points, locate the contour point closest to the geometric center of the abnormal region as a measurement reference point; based on the measurement reference point and the coordinates of its preceding and following adjacent contour points in the sequence, calculate the contour tangent direction at the measurement reference point, and define the direction perpendicular to the contour tangent direction as the normal measurement direction. Determine two boundary points that pass through the anomaly region along the normal measurement direction, calculate the Euclidean distance between the two boundary points and record it as scale data; A one-dimensional arc length parameterized curve is generated based on the inner plant contour data. The coordinates of the abnormal region are projected onto the one-dimensional arc length parameterized curve to obtain the corresponding arc length coordinate set. The minimum continuous coverage interval of the arc length coordinate set is determined, and the start and end values ​​of the minimum continuous coverage interval are recorded as extension range information. The scale data and extension range information are bound to the unique identifier of the abnormal region to obtain the quantized data record of the abnormal region; By summarizing the quantitative data records of all abnormal areas, an abnormal area quantitative report is obtained.

8. The orthopedic auxiliary examination method based on image processing according to claim 1, characterized in that, S6 further includes comparing the scale data of each abnormal region in the quantitative report based on preset clinical diagnostic standard parameters; when the scale data exceeds the preset first-level threshold, the abnormal region is marked as a first-level warning region. When the scale data exceeds the second-level threshold, which is higher than the first-level threshold, it is marked as a second-level warning area; Based on the warning level, the corresponding abnormal area data in the quantitative report are identified and prioritized, and the area data of different warning levels are presented differently in the final output report.

9. An image processing-based orthopedic auxiliary examination system, used to implement the image processing-based orthopedic auxiliary examination method according to any one of claims 1-8, characterized in that, include: The image acquisition module is used to acquire orthopedic image sequences of patients during the postoperative recovery period, arranged in the order of acquisition time, extract the implant contour data and stable bony anatomical point data of the first orthopedic image in the sequence, and establish a baseline image feature set. The image registration module is used to perform spatial transformation on each orthopedic image in the orthopedic image sequence based on the reference image feature set, and construct a globally spatially aligned registered image set. The region division module is used to extend the inner plant contour data of each registered image in the registered image set to the image region outside the contour to generate a ring-shaped analysis region, and divide the ring-shaped analysis region into a rigid region defined by the inner plant contour and an elastic skeleton region outside the contour. The data processing module is used to extract the image texture features of the elastic bone region of each registered image and compare them with the image texture features of the corresponding bone region of the first orthopedic image. Based on the comparison results, a spatial transformation field is generated and applied to the elastic bone region of the corresponding registered image to form a local fine-registered image set. The data optimization module is used to identify spatially adjacent pixel clusters whose gray values ​​show a unidirectional trend change over time within the annular analysis area of ​​each local fine-registered image, and to mark each pixel cluster as an abnormal region. The results generation module measures the scale of each anomalous region in the direction perpendicular to the inner plant outline, associates its positional information along the inner plant outline, and outputs a structured quantitative report of the anomalous region.

Images