Endometrial dynamic change tracking analysis method and system based on image registration

Abstract

The application relates to the technical field of image data processing, and discloses an endometrial dynamic change tracking analysis method and system based on image registration, which comprises the following steps: extracting boundary pixel coordinates of a target region in a first image frame and a second image frame, and constructing a discrete coordinate chain with a neighborhood index relationship; constructing a spatial distance field representing pixel space displacement constraints based on the target region of the second image frame; extracting gradient vectors of the discrete coordinate chain in the spatial distance field, and decoupling displacement bias into a normal driving component and a tangential homogenization component by using second-order central difference; generating a spatial deformation field through vector superposition, and performing coordinate resampling on the first image frame; the application realizes the decoupling of spatial mapping logic and pixel features by using an orthogonal decoupling mechanism, effectively constrains pixel distribution density in the deformation process, solves the problems of spatial folding and topological tearing under large-scale nonlinear deformation, and improves image tracking stability.

Description

Technical Field

[0001] This invention relates to a method and system for tracking and analyzing dynamic changes in the endometrium based on image registration, belonging to the field of image data processing technology. Background Technology

[0002] Current image registration techniques form the basis for multi-temporal analysis and dynamic evolution tracking. Conventional techniques typically follow the assumption of pixel intensity consistency and achieve spatial alignment between two frames by solving for the minimum value of the energy functional of the deformation field. When the target to be registered is a sequence with weak texture such as endometrial images and accompanied by large-scale nonlinear deformation, the local gray-level gradient distribution exhibits high uniformity. Under such conditions, the optimization process of the energy function loses its clear direction, resulting in non-physical spatial overlap of the deformation field during the solution process. This causes non-physical tearing of the image topology. Due to the lack of obvious contrast features, matching algorithms that rely solely on gray-level information produce ambiguity in the pixel mapping process, causing the spatial topological relationship after deformation to deviate from the constraints of normal physiological structure.

[0003] While increasing the control grid density or introducing higher-order smoothing constraints can suppress deformation fluctuations to some extent, they still cannot solve the mapping deviation caused by the lack of discriminative feature support. Existing technologies, when dealing with the coordination between large-scale deformation and topology preservation, lack effective modeling of the spatial correlation characteristics between pixels, making it difficult to guarantee the global consistency of coordinate mapping after resampling. This limitation manifests as localized grid accumulation or vacuum when faced with drastic distortion of the target object's geometry. Besides the registration bottleneck caused by missing organizational features, the displacement field solution control logic also has limitations. For example, [the following is an example of a patent announcement: CN1] Chinese invention patent 12949723B discloses a method for classifying pathological images of endometrium. It uses an improved U-Net network to perform segmentation and classification of endometrial tissue images. The logic is anchored to static feature extraction and probability mapping. However, when faced with large-scale nonlinear deformation caused by physiological peristalsis, this classification method ignores the physical topological constraints of continuous evolution between tissues and lacks a positive control mechanism of displacement field Jacobian determinant. When processing dynamic sequences with large inter-frame displacement fluctuations, it produces pixel mapping spatial folding or local stacking of computational grids. It cannot maintain the geometric coherence of the endometrial boundary movement process and is difficult to support thickness evolution monitoring and area change rate analysis.

[0004] Therefore, how to construct an image spatial transformation mechanism with topology preservation capability to improve the mapping reliability of non-rigid registration under physiological deformation conditions has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: a method for tracking and analyzing dynamic changes in the endometrium based on image registration, comprising the following steps: Step S1: Obtain a first image frame and a second image frame containing the endometrial region to be registered; Step S2: Use the gradient operator to extract the boundary pixel coordinates of the target region in the first image frame and the second image frame, and obtain them as the first point set. With the second set The boundary pixel set, where the first point set The pixels in the graph form a discrete coordinate chain with neighborhood indexing relationships; Step S3, based on the second point set Perform a spatial distance transformation to construct a second point set representing the distances between each pixel. The spatial distance field of the shortest pixel distance; Step S4, extract the first point set The gradient vector in the spatial distance field is used to perform orthogonal component decomposition based on the second-order central difference of the discrete coordinate chain, resulting in the first point set. The displacement offset of the middle pixel is decoupled into a normal driving component along the discrete coordinate chain normal and a tangential homogenization component along the discrete coordinate chain tangential. Step S5: Extract the projection length of the gradient vector in the normal direction to generate the normal driving component, and calculate the second-order central difference between the coordinates of the current pixel and the coordinates of the adjacent previous and next pixels in the discrete coordinate chain to generate the tangential homogenization component. Then, vectorically superimpose the normal driving component and the tangential homogenization component to generate the corresponding first point set. Spatial deformation field; Step S6: Perform coordinate resampling on the first image frame according to the spatial deformation field to generate a registered tracking analysis image.

[0006] Preferably, in step S3, the process of constructing the spatial distance field includes: using the second point set The pixel coordinates are used to establish a distance mapping function, and the negative gradient direction of the distance mapping function is calculated to determine the displacement guide vector of each spatial grid point.

[0007] Preferably, in step S5, the tangential homogenization component... The calculation rules are as follows: ,in, This represents the coordinate vector of the current pixel in the discrete coordinate chain. and These are the coordinate vectors of the current pixel in the discrete coordinate chain, representing the previous and next pixels. λ is a preset weight greater than 0 used to adjust the pixel distribution density.

[0008] Preferably, when generating the spatial deformation field in step S5, the method further includes using local support basis functions to modify the first point set. The displacement vector is interpolated and diffused to map the boundary displacement to the non-boundary pixel positions of the first image frame.

[0009] Preferably, it further includes performing an adaptive adjustment of the interpolation radius based on the local curvature of the spatial distance field: calculating the second derivative of the spatial distance field to identify high curvature feature regions in the first image frame, and reducing the support radius of the local support basis function in the high curvature feature regions to improve the deformation mapping accuracy of local edge wrinkle features.

[0010] Preferably, it also includes performing temporal consistency calibration based on functional residuals: calculating the residual mapping distance of the resampled pixel coordinates in the spatial distance field; when the residual mapping distance exceeds a preset threshold, extracting the deformation field of the adjacent previous frame as the motion component reference, and performing weighted fusion on the current spatial deformation field.

[0011] Preferably, in step S2, when extracting the boundary pixel coordinates of the target region in the first image frame, an edge detection operator is used to identify gray-level abrupt change points at the junction of the endometrium and myometrium, and the gray-level abrupt change points are logically sorted according to the contour tracing order to establish a discrete coordinate chain.

[0012] Preferably, in step S4, the process of performing orthogonal component decomposition includes: constructing the projection matrix of each pixel position, projecting the gradient vector to the normal dimension orthogonal to the local tangent of the discrete coordinate chain, and simultaneously calculating the tangential component along the direction of the discrete coordinate chain. After step S6, the process also includes calculating the endometrial thickness evolution and area change rate between the first image frame and the second image frame based on the spatial deformation field, and generating dynamic spectral data reflecting the peristaltic characteristics of the endometrium.

[0013] Preferably, steps S1 to S6 are executed iteratively, updating the first point set in each iteration. The coordinates are determined and the mapping constraint state in the spatial distance field is recalculated until the value of the global energy functional reaches the preset convergence interval.

[0014] An image registration-based system for tracking and analyzing dynamic changes in the endometrium includes: The acquisition module is used to acquire a first image frame and a second image frame containing the endometrial region to be registered; The feature extraction module is used to extract the boundary pixel coordinates of the target region in the first image frame and the second image frame using gradient operators, thereby obtaining the first point set. With the second set And according to the first point set Construct a discrete coordinate chain with neighborhood index relationships; Field construction module, used for building upon the second point set Perform a spatial distance transformation to construct a second point set representing the distances between each pixel. The spatial distance field of the shortest pixel distance; The orthogonal decomposition module is used to extract the first point set. The gradient vector in the spatial distance field is decomposed into orthogonal components based on the second-order central difference of the discrete coordinate chain, and the displacement bias is decoupled into normal driving components and tangential homogenization components. The field generation module is used to calculate the vector superposition result of the normal driving component and the tangential homogenization component to generate the corresponding first point set. Spatial deformation field; The resampling module is used to perform coordinate resampling on the first image frame based on the spatial deformation field to generate a registered tracking analysis image.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the dynamic change tracking analysis of the endometrium, by constructing a globally covered topological potential field and combining it with an orthogonal decoupling mapping mechanism, the spatial folding and topological tearing problems of the image matrix during large-scale nonlinear deformation are solved. The topological potential field transforms the discrete boundary point set into a continuous spatial constraint energy field, providing a global coordinate reference for displacement calculation. On this basis, by decoupling the spatial gradient vector of the topological potential field into a normal anchor displacement component and a tangential repulsion displacement component, the sampling density of pixel points in the tangential direction of the equipotential line is constrained, ensuring the spatial distribution uniformity of the discrete deformation reference under extremely distorted conditions, so that the Jacobian determinant when generating the continuous deformation field always remains positive, and the generation of image overlap and anatomical structure voids is eliminated from the underlying data mapping logic.

[0016] 2. By utilizing the deformation field full propagation mechanism guided by the boundary topological skeleton, the problem of accurate registration in areas with weak texture or highly uniform pixel feature distribution is solved. By extracting the discrete point set of the boundary of the target region as the potential source, a geometric constraint field independent of pixel grayscale consistency is constructed, so that the calculation of deformation vector is transformed from traditional local pixel search to restricted potential field sliding. When the texture inside the image is missing, the gradient logic of the topological potential field can still drive the pixel to be matched to converge to the target topological structure. The boundary displacement deviation is linearly extended to all pixels in the image through a spatial interpolation algorithm based on local support basis functions. This mechanism decouples the spatial mapping logic from the pixel grayscale features, improving the applicability of the image tracking scheme in low signal-to-noise ratio image environments.

[0017] 3. By introducing a local stiffness adaptive modulation mechanism based on the second-order gradient of the potential field, the system accurately captures local high-curvature anatomical features. It uses the second-order derivative features of the topological potential field to identify geometrically complex regions in the image space and dynamically adjusts the support radius of the basis function in the interpolation algorithm accordingly. In edge fold regions with severe curvature fluctuations, the support radius is reduced to increase the local deformation stiffness, enabling the deformation field to closely fit the complex topological skeleton and suppressing the feature smoothing blurring phenomenon common in traditional algorithms. This differentiated processing method for the deformation features of heterogeneous tissues ensures that the system maintains global deformation smoothness while possessing a deep insight into microscale local evolution. Attached Figure Description

[0018] Figure 1 This is a flowchart of the spatial registration and tracking processing steps for endometrial images in this invention; Figure 2 This is a structural diagram of the multi-level endometrial dynamic evolution tracking and analysis system of the present invention.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] like Figure 1 As shown, this embodiment of the invention provides a process flow for spatial registration and tracking of endometrial images based on image registration. The method specifically includes the following steps: Step S1: Obtain a first image frame and a second image frame containing the endometrial region to be registered; Step S2: Use the gradient operator to extract the boundary pixel coordinates of the target region in the first image frame and the second image frame, and obtain them as the first point set. With the second set The boundary pixel set, where the first point set The pixels in the graph form a discrete coordinate chain with neighborhood indexing relationships; Step S3, based on the second point set Perform a spatial distance transformation to construct a second point set representing the distances between each pixel. The spatial distance field of the shortest pixel distance; Step S4, extract the first point set The gradient vector in the spatial distance field is used to perform orthogonal component decomposition based on the second-order central difference of the discrete coordinate chain, resulting in the first point set. The displacement offset of the middle pixel is decoupled into a normal driving component along the discrete coordinate chain normal and a tangential homogenization component along the discrete coordinate chain tangential. Step S5: Extract the projection length of the gradient vector in the normal direction to generate the normal driving component, and calculate the second-order central difference between the coordinates of the current pixel and the coordinates of the adjacent previous and next pixels in the discrete coordinate chain to generate the tangential homogenization component. Then, vectorically superimpose the normal driving component and the tangential homogenization component to generate the corresponding first point set. Spatial deformation field; Step S6: Perform coordinate resampling on the first image frame according to the spatial deformation field to generate a registered tracking analysis image.

[0022] Preferably, in step S3, the process of constructing the spatial distance field includes: using the second point set The pixel coordinates are used to establish a distance mapping function, and the negative gradient direction of the distance mapping function is calculated to determine the displacement guide vector of each spatial grid point.

[0023] Preferably, in step S5, the tangential homogenization component... The calculation rules are as follows: ,in, This represents the coordinate vector of the current pixel in the discrete coordinate chain. and These are the coordinate vectors of the current pixel in the discrete coordinate chain, representing the previous and next pixels. λ is a preset weight greater than 0 used to adjust the pixel distribution density.

[0024] Preferably, when generating the spatial deformation field in step S5, the method further includes using local support basis functions to modify the first point set. The displacement vector is interpolated and diffused to map the boundary displacement to the non-boundary pixel positions of the first image frame.

[0025] Preferably, it further includes performing an adaptive adjustment of the interpolation radius based on the local curvature of the spatial distance field: calculating the second derivative of the spatial distance field to identify high curvature feature regions in the first image frame, and reducing the support radius of the local support basis function in the high curvature feature regions to improve the deformation mapping accuracy of local edge wrinkle features.

[0026] Preferably, it also includes performing temporal consistency calibration based on functional residuals: calculating the residual mapping distance of the resampled pixel coordinates in the spatial distance field; when the residual mapping distance exceeds a preset threshold, extracting the deformation field of the adjacent previous frame as the motion component reference, and performing weighted fusion on the current spatial deformation field.

[0027] Preferably, in step S2, when extracting the boundary pixel coordinates of the target region in the first image frame, an edge detection operator is used to identify gray-level abrupt change points at the junction of the endometrium and myometrium, and the gray-level abrupt change points are logically sorted according to the contour tracing order to establish a discrete coordinate chain.

[0028] Preferably, in step S4, the process of performing orthogonal component decomposition includes: constructing the projection matrix of each pixel position, projecting the gradient vector to the normal dimension orthogonal to the local tangent of the discrete coordinate chain, and simultaneously calculating the tangential component along the direction of the discrete coordinate chain. After step S6, the process also includes calculating the endometrial thickness evolution and area change rate between the first image frame and the second image frame based on the spatial deformation field, and generating dynamic spectral data reflecting the peristaltic characteristics of the endometrium.

[0029] Preferably, steps S1 to S6 are executed iteratively, updating the first point set in each iteration. The coordinates are determined and the mapping constraint state in the spatial distance field is recalculated until the value of the global energy functional reaches the preset convergence interval.

[0030] like Figure 2 As shown, this embodiment of the invention also provides a multi-level endometrial dynamic evolution tracking and analysis system structure corresponding to the above method. This system specifically includes: The acquisition module is used to acquire a first image frame and a second image frame containing the endometrial region to be registered; The feature extraction module is used to extract the boundary pixel coordinates of the target region in the first image frame and the second image frame using gradient operators, thereby obtaining the first point set. With the second set And according to the first point set Construct a discrete coordinate chain with neighborhood index relationships; Field construction module, used for building upon the second point set Perform a spatial distance transformation to construct a second point set representing the distances between each pixel. The spatial distance field of the shortest pixel distance; The orthogonal decomposition module is used to extract the first point set. The gradient vector in the spatial distance field is decomposed into orthogonal components based on the second-order central difference of the discrete coordinate chain, and the displacement bias is decoupled into normal driving components and tangential homogenization components. The field generation module is used to calculate the vector superposition result of the normal driving component and the tangential homogenization component to generate the corresponding first point set. Spatial deformation field; The resampling module is used to perform coordinate resampling on the first image frame based on the spatial deformation field to generate a registered tracking analysis image.

[0031] Example 1: In the application scenario of tracking and analyzing the endometrial region in clinical high-frequency ultrasound sequences, the gray-level distribution of the endometrium and its adjacent myometrium is highly uniform and lacks characteristic texture. Traditional registration methods based on the assumption of pixel gray-level consistency often cause the energy functional to get trapped in local minima due to the lack of gray-level gradients. This limitation manifests in engineering practice as uncontrolled local accumulation of the pixels to be matched during deformation field calculation, leading to spatial folding and topological tearing of anatomical structures in the resampled image, causing subsequent dynamic change tracking and analysis to lose a reliable coordinate reference. To address the challenge of weak texture and large deformation coexisting, the processor acquires data containing the pixels to be registered. The first and second image frames of the endometrial region are used to extract the boundary pixel coordinates of the target region in the two frames using a gradient operator, resulting in a first point set and a second point set, respectively. The pixels in the first point set form a discrete coordinate chain with a neighborhood index relationship. A spatial distance transformation is performed based on the second point set to construct a spatial distance field representing the shortest pixel distance between each pixel position and the second point set. The gradient vector of the first point set in the spatial distance field is extracted, and orthogonal component decomposition based on the second-order central difference of the discrete coordinate chain is performed to decouple the displacement bias into a normal driving component along the normal direction of the discrete coordinate chain and a tangential homogenization component along the tangential direction of the discrete coordinate chain. The tangential homogenization component Ft follows the calculation rules. ,in This represents the coordinate vector of the current pixel in the discrete coordinate chain. and These are the coordinate vectors of the current pixel in the discrete coordinate chain, representing the previous and next pixels. λ is a preset weight greater than 0 used to adjust the pixel distribution density.

[0032] In orthogonal component decomposition, discrete coordinate chain pixels Local normal vector From adjacent points and The determined tangent vector is generated by rotating it 90 degrees, and the normal driving component is projected from the gradient vector. Direction obtained; global energy functional The calculation formula is ,in This is the global energy value. For pixels In the spatial distance field, the Euclidean distance value, β is the weighting constant with a value of 0.1, and J is the Jacobian determinant value at each point in the deformation field; the iteration involves five consecutive iterations. When the mean change rate is below 0.01%, the convergence interval is reached and coordinate updates cease. This orthogonal decoupling mapping mechanism changes the traditional point set independent optimization method in registration logic. By extracting the projection length of the gradient vector in the normal direction to generate the normal driving component, the target boundary is brought closer. At the same time, the tangential homogenization component based on the second-order central difference is used to apply geometric repulsion constraints in the tangential direction of the equipotential line. This synergistic effect ensures that the discrete deformation reference maintains a constant spatial sampling density as it approaches the target contour, resolving the contradiction between normal convergence and tangential stacking. This allows the Jacobian determinant of the generated deformation field to remain positive in the extreme distortion region. Thus, without relying on internal pixel texture features, the construction of image spatial transformation logic with topology preservation capability is realized, providing continuous and stable displacement field support for the calculation of endometrial thickness evolution and area change rate.

[0033] Example 2: To verify the effectiveness of the image registration-based endometrial dynamic change tracking and analysis method and system under nonlinear deformation conditions, an experimental platform was established using a computer workstation equipped with a multi-core processor with a main frequency of 3.0 GHz. The experimental data came from an anonymized uterine ultrasound image dataset, with a single frame resolution of 640×480 pixels, a pixel pitch of 0.2 mm, and a sampling frame rate of 30 Hz. To evaluate the system's anti-interference performance in a real acquisition environment, Gaussian white noise with a signal-to-noise ratio of 20.4 dB was superimposed on the raw image signal at the input end, and a random translational perturbation with an amplitude of 4 pixels was introduced to simulate probe squeezing. The spatial displacement bias caused by pressure or patient respiration is a key parameter in this experiment. The preset weight λ for adjusting pixel distribution density is used to balance the response speed of normal convergence and the constraint strength of tangential topology preservation. The processor performs gradient calibration experiments, using the maximum displacement gradient of the endometrial boundary in the image sequence as the core variable. The value of λ is changed in steps of 0.05 within the range of 0.05 to 0.5. Experimental results show that when the weight λ is in the range of 0.1 to 0.25, the Jacobian determinant of the calculated deformation field remains above 0.82, and the average boundary residual is stable below 0.14 mm. In this embodiment, the weight λ is selected as... The preset weight is set to 0.15. The engineering calibration process for this preset weight is as follows: During the initialization phase after the image acquisition terminal is powered on, 10 frames of idle static background images are continuously captured to establish a buffer of 640 by 480 pixels. The average variance of the grayscale of all pixels in the buffer is calculated as an indicator of the electronic noise level. The system has a preset mapping table: when the average variance is less than 5, the preset weight is set to 0.1; when the average variance is between 5 and 15, the preset weight increases linearly with the variance in increments of 0.01; when the average variance exceeds 20, the preset weight is fixed at 0.25. This dynamic gain adjustment based on the noise level ensures that the positioning jitter caused by grayscale gradient fluctuations is offset by enhancing the mutual repulsion between point sets in a low signal-to-noise ratio environment.

[0034] The experiment established four groups for performance benchmarking. The experimental group employed an orthogonal decoupling mapping mechanism based on the second-order central difference of a discrete coordinate chain; control group 1 used a single normal-driven scheme lacking tangential homogenization components; control group 2 used a traditional optical flow field registration algorithm; and control group 3 set the weight λ to 0.6 for out-of-range comparison. Key intermediate data for each group were recorded when processing the 25th frame image accompanied by endometrial peristalsis. The experimental group achieved a minimum Jacobian determinant of 0.86 for the deformation field and a variance of 0.11 pixels squared between adjacent pixels in the discrete coordinate chain. Control group 1 achieved a minimum Jacobian determinant of -0.18 for the deformation field at the same location. Control group 2 had a boundary matching residual of 2.12 mm, while control group 3... When the value deviates from the optimal window, the average residual at the boundary increases to 1.65 mm. Based on the above experimental data analysis, the experimental group achieved the extraction of endometrial thickness evolution on the basis of improving the signal-to-noise ratio to 28.2 dB. The average absolute error of the measured value was 0.11 mm, which proved that under the second-order central difference constraint of the discrete coordinate chain, the spatial mapping logic can maintain the coherence of the topological structure, solve the coordinate tracking drift problem of weak textured tissue under large deformation, and meet the requirements of high-precision medical image tracking system for deformation field stability.

[0035] Example 3: When the system faces a tracking scenario where the endometrial boundary has local edge folds, if the processor uses a fixed-radius interpolation diffusion strategy, it is prone to smoothing deviations in areas of abrupt curvature changes. This causes data fusion of details in the image during spatial imaging, leading to measurement errors in subsequent area change rate analysis. The processor acquires the first and second image frames to be registered, extracts the boundary pixels of the target region using an edge detection operator, and uses an 8-neighborhood chain code tracking algorithm to construct a discrete coordinate chain with neighborhood indexing relationships. It selects a starting anchor point from the boundary point set and searches for adjacent pixels in a clockwise direction, sequentially establishing a sequence containing pixel coordinate vectors. This provides the coordinates of the previous pixel for performing a second-order central difference operation based on the displacement vector. and the coordinates of the next pixel .

[0036] Based on the normal driving component and the tangential homogenization component, an adaptive adjustment of the interpolation radius based on the local curvature of the spatial distance field is performed. A Hessian matrix H is constructed using the second-order partial derivatives of the spatial distance field, and the trace of matrix H is extracted to determine the curvature index κ used to characterize the degree of local edge wrinkling. Where D is the distance value of the corresponding coordinate position in the spatial distance field; the system dynamically corrects the support radius R of the local support basis function according to the curvature index κ, which is determined as follows: in the smooth region where the curvature index κ is less than the preset curvature threshold T1, the support radius R is set to 5 pixels; when the curvature index κ is detected to exceed the preset curvature threshold T1, the support radius R converges as the curvature index κ increases, and the calculation formula is as follows: ;in, The base support radius is set to 5 pixels in this application example, and α is the attenuation coefficient, set to 0.25.

[0037] The local support basis function uses a compactly supported Gaussian kernel function, and the support radius R is determined by the local curvature κ of the spatial distance field. When κ exceeds the preset curvature threshold of 0.5 pixels per unit, the support radius R is reduced from 5 pixels to 2 pixels, improving the ability to capture the deformation features of the endometrial micro-folds. To adapt to ultrasound images of different resolutions, the distance transformation step size of the spatial distance field is 0.5 pixels, and the preset weight λ is calibrated according to the electronic noise level of the image. When the image grayscale variance exceeds 10, the value of λ is increased from 0.1 to 0.25, enhancing the tangential repulsion constraint strength to counteract the boundary positioning jitter induced by low signal-to-noise ratio. The intervention of the variable-scale interpolation mechanism limits the displacement constraint generated by the boundary point set to the spatial range of the target folds. When processing the edge of the endometrium with complex folds, the Jacobian determinant fluctuation range of the deformation field generated by this method at the fold peak is between 0.92 and 1.05, and the edge matching residual is reduced from 0.85 mm in the fixed radius scheme to 0.08 mm, thereby providing a physically consistent displacement field support for the dynamic tracking of images of weak textured tissues.

[0038] Example 4: When the system faces the challenge of deploying an image registration-based endometrial dynamic change tracking and analysis method on image acquisition terminals with different gain characteristics and noise floor, the processor implements a standardized benchmark calibration procedure to eliminate the interference of imaging equipment hardware differences on the accuracy of deformation field calculation. Static image data is acquired using a standard phantom containing simulated acoustic characteristics of uterine tissue, and the root mean square error of pixel grayscale values ​​within a preset constant region of the phantom is extracted to quantify the inherent electronic noise level of the current device. ,according to A linear function mapping relationship is used within the grayscale range of 5 to 25 pixels to determine the baseline value of the distance transformation step size when constructing the spatial distance field. ,Should The value of varies Increased by scaling the spatial derivative operation under low signal-to-noise ratio background, the generation of pseudo-gradient components is suppressed. In this application example, when detecting... When it is 12.4, the corresponding The unit is set to 1.2 pixels.

[0039] In on-site debugging scenarios before new equipment goes live, the processor's inherent electronic noise level is based on the calibration phase. Adaptive fine-tuning is performed on the preset weight λ for adjusting pixel distribution density by injecting a simulated vector field containing known displacement gradients into the deformation field solution module and monitoring the root mean square error of the boundary matching results in real time. , will make The weights that minimize the Jacobian determinant and keep it within the range of 0.8 to 1.2 are used as the operating parameters. Under a specific test environment of 12.4, λ converges to 0.158 after optimization calculation. The initial parameters determined by the pre-calibration process eliminate the calculation deviation of the algorithm when facing detector signal drift. Before the coordinate resampling begins in the first and second image frames, the processor adapts the geometric exclusion constraint strength of the system to the sampling accuracy boundary of the current physical terminal, so that the local deformation fidelity of the resampled image is maintained above 98.6%.

[0040] Example 5: In deployment scenarios involving dynamic changes in the physical resolution of ultrasound images, the processor determines the physical pixel spacing between the first and second image frames by reading the probe frequency and aperture parameters of the image acquisition terminal. and By implementing spatial scaling using a standardized acoustic phantom with preset millimeter-level scales, the discrete pixel displacements in the image coordinate system are converted into endometrial thickness evolution variables with physical dimensions, thus determining the minimum grid step size for constructing the spatial distance field. With a unit of 0.1 pixels, this pre-calibration process establishes the quantization mapping rules between discrete image features and tissue physical morphology before the system performs coordinate resampling.

[0041] When the system faces the risk of signal drift during long-term sequence tracking, the processor implements a time-consistency calibration procedure based on functional residuals to maintain the feedback loop of deformation calculation. After completing the coordinate resampling of the first image frame, the processor extracts the coordinates of the resampled boundary point set and calculates the residual mapping distance of the coordinates in the corresponding spatial distance field. Compare it with the preset motion consistency threshold. Implement quantitative comparisons, among which Determined based on 5% of the current average thickness of the endometrial lining; when detected Exceed At that time, the system extracts the deformation field of the adjacent previous frame as the reference value of the displacement component, and uses the fusion weight w to perform deformation field superposition correction. The weight w follows the calculation rules as follows: Where w is the fusion weight and e is the natural constant. For residual mapping distance, This is the motion consistency threshold; in this application example, when 1.25 pixels and When the unit is 0.8 pixels, the weight w is calculated to be 0.79, which enables the current deformation field and the reference value to be fused under the weight constraint. This filters out transient displacement noise induced by breathing motion or probe slippage, and maintains the spatial topological consistency of the resampled image at 98.4% level after continuous tracking for 180 frames.

[0042] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for tracking and analyzing dynamic changes in the endometrium based on image registration, characterized in that, Includes the following steps: Step S1: Obtain a first image frame and a second image frame containing the endometrial region to be registered; Step S2: Use the gradient operator to extract the boundary pixel coordinates of the target region in the first image frame and the second image frame, and obtain them as the first point set. With the second set The boundary pixel set, where the first point set The pixels in the graph form a discrete coordinate chain with neighborhood indexing relationships; In step S2, when extracting the boundary pixel coordinates of the target region in the first image frame, an edge detection operator is used to identify gray-level abrupt change points at the junction of the endometrium and myometrium, and the gray-level abrupt change points are logically sorted according to the contour line tracing order to establish a discrete coordinate chain. Step S3, based on the second point set Perform a spatial distance transformation to construct a second point set representing the distances between each pixel. The spatial distance field of the shortest pixel distance; Step S4, extract the first point set The gradient vector in the spatial distance field is used to perform orthogonal component decomposition based on the second-order central difference of the discrete coordinate chain, resulting in the first point set. The displacement offset of the middle pixel is decoupled into a normal driving component along the discrete coordinate chain normal and a tangential homogenization component along the discrete coordinate chain tangential. Specifically, the projection length of the gradient vector in the normal direction is extracted to generate the normal driving component, and the second-order central difference between the coordinates of the current pixel and the coordinates of the adjacent previous and next pixels in the discrete coordinate chain is calculated to generate the tangential homogenization component. The normal driving component and the tangential homogenization component are then vector-superimposed to generate the corresponding first point set. Spatial deformation field; Step S5: Perform coordinate resampling on the first image frame according to the spatial deformation field to generate a registered tracking analysis image.

2. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 1, characterized in that, In step S3, the process of constructing the spatial distance field includes: using the second point set The pixel coordinates are used to establish a distance mapping function, and the negative gradient direction of the distance mapping function is calculated to determine the displacement guide vector of each spatial grid point.

3. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 1, characterized in that, In step S4, the tangential homogenization component The calculation rules are as follows: ,in, This represents the coordinate vector of the current pixel in the discrete coordinate chain. and These are the coordinate vectors of the current pixel in the discrete coordinate chain, representing the previous and next pixels. λ is a preset weight greater than 0 used to adjust the pixel distribution density.

4. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 1, characterized in that, In step S4, when generating the spatial deformation field, the method also includes using local support basis functions on the first point set. The displacement vector is interpolated and diffused to map the boundary displacement to the non-boundary pixel positions of the first image frame.

5. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 4, characterized in that, It also includes performing adaptive adjustment of the interpolation radius based on the local curvature of the spatial distance field: calculating the second derivative of the spatial distance field to identify high curvature feature regions in the first image frame, and reducing the support radius of the local support basis function in the high curvature feature regions to improve the deformation mapping accuracy of local edge wrinkle features.

6. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 1, characterized in that, It also includes performing temporal consistency calibration based on functional residuals: calculating the residual mapping distance of the resampled pixel coordinates in the spatial distance field, and when the residual mapping distance exceeds a preset threshold, extracting the deformation field of the adjacent previous frame as the motion component reference, and performing weighted fusion on the current spatial deformation field.

7. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 1, characterized in that, In step S4, the process of performing orthogonal component decomposition includes: constructing the projection matrix of each pixel position, projecting the gradient vector to the normal dimension orthogonal to the local tangent of the discrete coordinate chain, and simultaneously calculating the tangential component along the direction of the discrete coordinate chain. After step S5, it also includes calculating the endometrial thickness evolution and area change rate between the first image frame and the second image frame based on the spatial deformation field, and generating dynamic spectral data reflecting the peristaltic characteristics of the endometrium.

8. The method for tracking and analyzing dynamic changes in the endometrium based on image registration according to claim 1, characterized in that, Steps S1 to S5 are executed iteratively, updating the first point set in each iteration. The coordinates are determined and the mapping constraint state in the spatial distance field is recalculated until the value of the global energy functional reaches the preset convergence interval.

9. An image-registration-based endometrial dynamic change tracking and analysis system, used to execute the image-registration-based endometrial dynamic change tracking and analysis method of claim 1, characterized in that, include: The acquisition module is used to acquire a first image frame and a second image frame containing the endometrial region to be registered; The feature extraction module is used to extract the boundary pixel coordinates of the target region in the first image frame and the second image frame using gradient operators, thereby obtaining the first point set. With the second set And according to the first point set Construct a discrete coordinate chain with neighborhood index relationships; Field construction module, used for building upon the second point set Perform a spatial distance transformation to construct a second point set representing the distances between each pixel. The spatial distance field of the shortest pixel distance; The orthogonal decomposition module is used to extract the first point set. The gradient vector in the spatial distance field is decomposed into orthogonal components based on the second-order central difference of the discrete coordinate chain, and the displacement bias is decoupled into normal driving components and tangential homogenization components. The field generation module is used to calculate the vector superposition result of the normal driving component and the tangential homogenization component to generate the corresponding first point set. Spatial deformation field; The resampling module is used to perform coordinate resampling on the first image frame based on the spatial deformation field to generate a registered tracking analysis image.

Images