Pelvic floor organ three-dimensional ultrasonic reconstruction and dynamic change analysis system

Abstract

The invention relates to the technical field of medical images, in particular to a pelvic floor organ three-dimensional ultrasonic reconstruction and dynamic change analysis system which comprises a pelvic floor volume probe, a three-dimensional pelvic floor organ reconstruction module, a relative position quantitative analysis module, a four-dimensional pelvic floor organ dynamic tracking module, a data storage module and a data display module. Three-dimensional volume data are collected through a pelvic floor volume probe, and the three-dimensional pelvic floor organ reconstruction module achieves accurate three-dimensional reconstruction of the bladder, the vagina and the rectum through the machine learning image segmentation and depth map reconstruction technology. The relative position quantitative analysis module is used for establishing a standardized pelvic floor coordinate system and quantitatively analyzing the relative position relationship between organs; the four-dimensional pelvic floor organ dynamic tracking module innovatively adopts multi-dimensional mark point dynamic tracking of topology perception and a four-dimensional space-time trajectory reconstruction technology based on manifold learning to realize dynamic function evaluation of pelvic floor organs, and provides static anatomical structure and dynamic function information at the same time.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical imaging technology, in particular to a pelvic floor organ three-dimensional ultrasound reconstruction and dynamic change analysis system, especially a medical imaging device applied to the diagnosis and evaluation of female pelvic floor dysfunction. BACKGROUND

[0002] Pelvic floor dysfunction is a common gynecological disease, including pelvic organ prolapse, stress urinary incontinence and defecation dysfunction, etc., which seriously affects the quality of life of patients. Accurate evaluation of the anatomical structure and functional state of the pelvic floor organ is crucial for the diagnosis and treatment of pelvic floor dysfunction.

[0003] The existing pelvic floor evaluation techniques mainly include conventional two-dimensional ultrasound, MRI and CT, etc., but these methods have obvious limitations. Conventional two-dimensional ultrasound can only provide single-plane anatomical information, and cannot fully understand the three-dimensional structure and spatial positional relationship of the pelvic floor organ; MRI can provide good soft tissue contrast, but the equipment is expensive, the examination time is long, and it cannot be evaluated in real time; CT has radiation risk, which is not suitable for pregnant women and patients who need repeated follow-up.

[0004] In recent years, three-dimensional ultrasound technology has made significant progress in clinical application, but most of the three-dimensional ultrasound systems on the market can only provide static three-dimensional reconstruction, and cannot effectively capture the dynamic change process of the pelvic floor organ. Pelvic floor dysfunction is a dynamic disease, and only static anatomical structure evaluation cannot fully understand the nature of the disease. In addition, the existing evaluation methods rely mainly on the subjective experience of doctors, lack of objective and quantitative evaluation criteria, resulting in large individual differences in diagnosis results.

[0005] Therefore, it is of great significance to develop a comprehensive evaluation system that can provide both static anatomical structure and dynamic functional information of the pelvic floor organ, in order to improve the diagnostic accuracy and treatment effect of pelvic floor dysfunction. SUMMARY

[0006] The purpose of the present application is to provide a pelvic floor organ three-dimensional ultrasound reconstruction and dynamic change analysis system, which aims to overcome the shortcomings of the prior art, realize accurate three-dimensional reconstruction and dynamic functional evaluation of the pelvic floor organ, and provide comprehensive and objective quantitative basis for the diagnosis and treatment of pelvic floor dysfunction.

[0007] The present application provides a pelvic floor organ three-dimensional ultrasound reconstruction and dynamic change analysis system, which comprises:

[0008] A pelvic floor volume probe is used to be placed directly above the pelvic floor organ and to collect three-dimensional volume data;

[0009] a three-dimensional pelvic floor organ reconstruction module connected with the pelvic floor volume probe, configured to receive the three-dimensional volume data and reconstruct a three-dimensional simulation model of the bladder, vagina and rectum;

[0010] a relative position quantification analysis module connected with the three-dimensional pelvic floor organ reconstruction module, configured to establish a standardized pelvic floor coordinate system and calculate anatomical position parameters;

[0011] a four-dimensional pelvic floor organ dynamic tracking module connected with the pelvic floor volume probe and the three-dimensional pelvic floor organ reconstruction module, configured to continuously collect dynamic change data of different states of the bladder, vagina and rectum, wherein the four-dimensional pelvic floor organ dynamic tracking module comprises a topologically-aware multi-dimensional marker point dynamic tracking system and a four-dimensional spatio-temporal trajectory reconstruction system based on manifold learning;

[0012] the topologically-aware multi-dimensional marker point dynamic tracking system is configured to collect position change information of the pelvic floor organs through heterogeneous marker points and perform topological feature preserving analysis;

[0013] the four-dimensional spatio-temporal trajectory reconstruction system based on manifold learning is configured to convert the marker point position change information into a continuous four-dimensional pelvic floor organ dynamic model;

[0014] a data storage module connected with the four-dimensional pelvic floor organ dynamic tracking module, configured to store the three-dimensional simulation model and dynamic change data;

[0015] a data display module connected with the data storage module, configured to dynamically display the three-dimensional simulation model and synchronously play back the change process of the pelvic floor organs in different states.

[0016] Preferably, the topologically-aware multi-dimensional marker point dynamic tracking system comprises:

[0017] a physical marker point subsystem configured to collect position change information of the vagina and pelvic floor muscles through micro-sensors;

[0018] a virtual marker point subsystem configured to identify and track bladder feature points in ultrasound images through image processing algorithms;

[0019] a marker point data processing unit configured to pre-process, noise-filter and multi-source data synchronously fuse the position information collected by the physical marker point subsystem and the virtual marker point subsystem;

[0020] a topologically-preserving marker point distribution strategy unit configured to ensure that the marker point distribution satisfies the topological homeomorphism condition based on the anatomical features of the pelvic floor region;

[0021] an adaptive sampling control unit configured to dynamically adjust the sampling frequency according to the motion state of the pelvic floor organs.

[0022] As preferred, the manifold learning based four-dimensional spatio-temporal trajectory reconstruction system comprises:

[0023] a manifold representation and dimension reduction unit for mapping the high-dimensional feature vectors of the labeled points to a low-dimensional manifold space;

[0024] a multi-scale manifold structure analysis unit for analyzing the pelvic floor dynamic behavior from three scales of micro, meso and macro;

[0025] a hierarchical reconstruction unit for realizing complete four-dimensional model reconstruction through three levels of local surface reconstruction, organ-level reconstruction and overall area reconstruction;

[0026] a spatio-temporal continuity constraint unit for ensuring the physiological feasibility of the reconstructed model through time continuity constraint, space continuity constraint and anatomical constraint;

[0027] a dynamic feature extraction unit for extracting key dynamic parameters from the four-dimensional model for functional evaluation.

[0028] As preferred, the micro-sensor in the physical labeled point subsystem comprises:

[0029] a three-axis accelerometer for measuring the acceleration change of the labeled point;

[0030] a three-axis gyroscope for measuring the angle change of the labeled point;

[0031] a wireless transmission module for transmitting the collected data to the labeled point data processing unit in real time;

[0032] an electromagnetic induction power supply system for providing passive power for the sensor through an external electromagnetic field.

[0033] As preferred, the virtual labeled point subsystem comprises:

[0034] a feature point recognition unit for automatically recognizing feature points based on tissue interfaces, reflection intensity features and texture features in the ultrasound image;

[0035] an inter-frame tracking unit for realizing inter-frame tracking of the feature points by combining hierarchical template matching with optical flow method;

[0036] a dynamic generation strategy unit for dynamically adjusting the number and distribution of the virtual labeled points according to the deformation of the pelvic floor tissue;

[0037] a confidence evaluation unit for assigning a confidence score to each virtual labeled point, and starting re-identification when the confidence is lower than a threshold.

[0038] As preferred, the hierarchical reconstruction unit comprises:

[0039] a local surface reconstruction subunit for reconstructing a local surface around the landmark points by radial basis function interpolation technique;

[0040] an organ-level reconstruction subunit for realizing complete reconstruction of a single organ in combination with shape priors and physical model constraints;

[0041] an overall region integration subunit for applying multi-body system theory to handle the inter-organ relationship and constraints;

[0042] a time series integration subunit for ensuring continuity in the time dimension by state space model.

[0043] As preferred, the relative position quantification analysis module is used for:

[0044] taking the center point of the three-dimensional pelvic floor ultrasound image as the position of the pelvic floor ischial spine;

[0045] establishing an xyz three-dimensional Cartesian coordinate system with the pelvic floor ischial spine as the origin of the coordinate system;

[0046] taking the direction of the pelvic floor ischial spine pointing to the cervical os as the z-axis, the direction perpendicular to the y-axis as the x-axis, and the direction perpendicular to the z-axis as the y-axis;

[0047] defining the bladder and vaginal organs as two objects, and dividing the relative position relationship into three categories: overlap, contact, and separation;

[0048] calculating the distance, angle, and overlapping area between the bladder and vaginal organs as the quantification parameters of the relative position.

[0049] As preferred, the three-dimensional pelvic floor organ reconstruction module comprises:

[0050] a machine learning image segmentation unit for learning the seed region by an active shape model algorithm, forming a pelvic floor organ segmentation template with active shape model parameters, and identifying and segmenting the pelvic floor organ transverse, sagittal, or coronal ultrasound images by using the template;

[0051] a depth map reconstruction unit for selecting a reference image from a series of pelvic floor organ ultrasound images, calculating a depth map, and registering other images with the reference image to realize depth reconstruction;

[0052] a surface fitting and fusion unit for meshing the three-dimensional reconstruction data, forming a fixed-size grid, establishing a boundary plane, and fusing the generated surface to complete the construction of the three-dimensional pelvic floor organ model.

[0053] As preferred, the dynamic feature extraction unit is used for extracting the following dynamic parameters from the four-dimensional model:

[0054] a curve of the position of the anatomical key point changing over time;

[0055] organ volume and shape parameter change curves;

[0056] inter-organ relative position parameters, including distance, angle, and overlap area time change functions;

[0057] tissue strain and stress distribution estimates;

[0058] motion range and velocity distribution;

[0059] motion coordination indicators;

[0060] The dynamic parameters are standardized and used for functional state evaluation.

[0061] As a preferred embodiment, the data display module comprises:

[0062] a three-dimensional visualization submodule for real-time display of three-dimensional simulation models of the bladder, vagina, and rectum;

[0063] a dynamic playback submodule for playback of the change process of the pelvic floor organs in different states according to stored dynamic change data;

[0064] a parameter curve display submodule for display of distance, overlap area, and relative angle time change curves between pelvic floor anatomical structures;

[0065] a functional evaluation result display submodule for display of pelvic floor functional state evaluation results and clinical diagnosis suggestions.

[0066] The present application has the following advantages:

[0067] 1. Simultaneous three-dimensional reconstruction of the bladder, vagina, rectum, and other pelvic floor organs is achieved, providing comprehensive anatomical structure information;

[0068] 2. By establishing a standardized pelvic floor coordinate system, objective quantification of pelvic floor anatomical position parameters is achieved, making the data between different patients comparable;

[0069] 3. Topology-aware multi-dimensional marker dynamic tracking technology and four-dimensional spatiotemporal trajectory reconstruction technology based on manifold learning are innovatively used to comprehensively capture and analyze the dynamic change process of the pelvic floor organs;

[0070] 4. Objective quantification of pelvic floor functional state evaluation indicators is provided, reducing the subjectivity of diagnosis and improving the accuracy of diagnosis;

[0071] 5. Through dynamic display and playback functions, the change process of the pelvic floor organs in different states is intuitively displayed, facilitating doctor understanding and patient education;

[0072] 6. The system adopts non-invasive ultrasound imaging technology, has no radiation risk, and is suitable for pregnant women and patients who need repeated follow-up.

[0073] In summary, the pelvic organ three-dimensional ultrasound reconstruction and dynamic change analysis system provided by the application can not only provide static anatomical structure information of the pelvic organ, but also capture and analyze the dynamic change process, thereby providing comprehensive technical support for diagnosis, treatment scheme formulation and efficacy evaluation of pelvic dysfunction. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 Fig. 1 is a schematic diagram of the overall architecture of the pelvic organ three-dimensional ultrasound reconstruction and dynamic change analysis system of the application;

[0075] Figure 2 Fig. 2 is a workflow diagram of the three-dimensional pelvic organ reconstruction module of the application;

[0076] Figure 3 Fig. 3 is a workflow diagram of the four-dimensional pelvic organ dynamic tracking module of the application;

[0077] Figure 4 Fig. 4 is a structural schematic diagram of the topologically aware multi-dimensional marker point dynamic tracking system of the application;

[0078] Figure 5 Fig. 5 is a workflow diagram of the four-dimensional spatiotemporal trajectory reconstruction system based on manifold learning of the application. DETAILED DESCRIPTION

[0079] Reference will now be made to Figures 1-5 , and specific embodiments will be described in detail below with reference to the accompanying drawings.

[0080] As shown in Figure 1 , the pelvic organ three-dimensional ultrasound reconstruction and dynamic change analysis system provided by the application includes a pelvic volume probe 1, a three-dimensional pelvic organ reconstruction module 2, a relative position quantification analysis module 3, a four-dimensional pelvic organ dynamic tracking module 4, a data storage module 5 and a data display module 6.

[0081] The pelvic volume probe 1 is used to be placed directly above the pelvic organ and to collect three-dimensional volume data. In an embodiment of the application, the pelvic volume probe 1 adopts a high-frequency ultrasound volume probe with a frequency range of 5-12 MHz, which can simultaneously obtain multi-angle and multi-slice ultrasound images to form complete three-dimensional volume data. Preferably, the scanning angle range of the probe is 70°-170°, and the scanning frame rate is 15-30 frames / second, so as to ensure that sufficient detailed information of the pelvic region is obtained.

[0082] The three-dimensional pelvic organ reconstruction module 2 is connected with the pelvic volume probe 1, and is used to receive three-dimensional volume data and reconstruct three-dimensional simulation models of the bladder, vagina and rectum. As shown inFigure 2 As shown, the three-dimensional pelvic floor organ reconstruction module 2 includes a machine learning image segmentation unit 21, a depth map reconstruction unit 22, and a surface fitting and fusion unit 23.

[0083] The machine learning image segmentation unit 21 learns the seed region through an active shape model algorithm (ASMs) to form a pelvic floor organ segmentation template with active shape model parameters. In an embodiment of the present application, 100-200 pelvic floor organ ultrasound images are first manually segmented by a professional physician as seed regions, which are input into the machine learning system for training. During the training process, the active shape model algorithm is implemented through the following steps:

[0084] 1) Mark each manually segmented contour point to form a point distribution model (PDM);

[0085] 2) Align the point distribution of all training samples;

[0086] 3) Perform principal component analysis (PCA) to extract the main mode of shape change;

[0087] 4) Establish a grayscale appearance model for subsequent matching process.

[0088] After training, the formed pelvic floor organ segmentation template can automatically identify and segment the pelvic floor organ region in new ultrasound images. Preferably, the accuracy rate of the segmentation template reaches more than 90%, which can effectively reduce the workload of manual segmentation.

[0089] The depth map reconstruction unit 22 selects the first image from a series of pelvic floor organ ultrasound images as a reference image, performs depth map calculation on the reference image as a benchmark for depth reconstruction; then registers other pelvic floor organ ultrasound images with the reference image, and transforms all other pelvic floor organ ultrasound images to the coordinate space of the reference image to complete the depth reconstruction. In an embodiment of the present application, the depth map calculation adopts a gradient-based method, and the calculation formula is as follows:

[0090] ,

[0091] wherein, is the depth value at point , with unit of millimeter (mm); is the gradient vector of the th frame image at point , representing the direction and amplitude of image intensity change; is the weight coefficient of the th frame image, ranging from 0 to 1, and satisfying is the number of image frames, with a typical value of 5-20. Image registration adopts a mutual information maximization method to ensure accurate alignment between different images.

[0092] The curved surface fitting and fusion unit 23 meshes the three-dimensional reconstruction data, forms a fixed-size grid, establishes a boundary plane, and fuses the generated curved surface to complete the construction of the three-dimensional pelvic organ model. Specifically, the meshing adopts a uniformly spaced hexahedral grid, and the grid size is 0.5 mm x 0.5 mm x 0.5 mm, which can better balance the reconstruction accuracy and computational efficiency. The curved surface fusion adopts a weighted average method to ensure the smoothness of the curved surface transition.

[0093] The relative position quantification analysis module 3 is connected with the three-dimensional pelvic organ reconstruction module 2, and is used to establish a standardized pelvic coordinate system and calculate anatomical position parameters. The relative position quantification analysis module 3 takes the center point of the three-dimensional pelvic ultrasound image as the position of the pelvic ischial spine, and establishes an xyz three-dimensional Cartesian coordinate system with the pelvic ischial spine as the origin of the coordinate system, wherein the direction of the pelvic ischial spine pointing to the cervical os is the z-axis, the direction perpendicular to the y-axis is the x-axis, and the direction perpendicular to the z-axis is the y-axis.

[0094] In the standardized coordinate system, the relative position quantification analysis module 3 defines the bladder and vaginal organs as two objects, divides the relative position relationship into three categories of overlap, contact and separation, and calculates the distance, angle and overlap area between the bladder and vaginal organs as the quantification parameters of the relative position. Specifically, the inter-organ distance is defined as the Euclidean distance between the center points of the two organs; the angle is defined as the angle between the main axes of the two organs; and the overlap area is obtained by calculating the intersection part of the volumes of the two organs. These quantification parameters provide quantitative basis for objective evaluation of pelvic floor dysfunction.

[0095] The four-dimensional pelvic organ dynamic tracking module 4 is connected with the pelvic volume probe 1 and the three-dimensional pelvic organ reconstruction module 2, and is used to continuously collect dynamic change data of different states of the bladder, vagina and rectum. As the core innovative point of the present application, as shown in Figure 3 The four-dimensional pelvic organ dynamic tracking module 4 includes a topologically aware multi-dimensional marker point dynamic tracking system 41 and a four-dimensional spatiotemporal trajectory reconstruction system 42 based on manifold learning.

[0096] As shown in Figure 4 The topologically aware multi-dimensional marker point dynamic tracking system 41 includes a physical marker point subsystem 411, a virtual marker point subsystem 412, a marker point data processing unit 413, a topologically preserved marker point distribution strategy unit 414 and an adaptive sampling control unit 415.

[0097] The physical marker point subsystem 411 is used to collect the position change information of the vagina and the pelvic floor muscles through micro sensors. In a preferred embodiment of the present application, the micro sensors in the physical marker point subsystem 411 include a three-axis accelerometer, a three-axis gyroscope, a wireless transmission module, and an electromagnetic induction power supply system. The three-axis accelerometer is used to measure the acceleration change of the marker point, with a measurement range of ±16g and a sampling frequency adjustable between 5-60Hz according to the motion state. The three-axis gyroscope is used to measure the angle change of the marker point, with a measurement range of ±2000° / s and an accuracy of 0.1°. The wireless transmission module uses low-power Bluetooth technology, with a transmission frequency dynamically adjusted within the range of 2.4GHz-2.485GHz and a transmission distance up to 10 meters, sufficient to cover the needs of clinical examination. The electromagnetic induction power supply system provides passive power for the sensor through an external electromagnetic field, avoiding the inconvenience and safety hazards of using a built-in battery.

[0098] The virtual marker point subsystem 412 is used to identify and track the feature points of the bladder in the ultrasound image through image processing algorithms. The virtual marker point subsystem 412 includes a feature point identification unit, an inter-frame tracking unit, a dynamic generation strategy unit, and a confidence evaluation unit. The feature point identification unit automatically identifies feature points based on the tissue interface, reflection intensity features, and texture features in the ultrasound image. Specifically, the tissue interface is detected by a gradient operator; the reflection intensity features are extracted by a local reflection intensity histogram feature; and the texture features are calculated by a gray level co-occurrence matrix (GLCM). The inter-frame tracking unit uses a hierarchical template matching combined with an optical flow method to achieve inter-frame tracking of the feature points. In an embodiment of the present application, the hierarchical template matching uses a three-layer pyramid structure to gradually refine the matching position from low resolution to high resolution; and the optical flow method uses the Lucas-Kanade algorithm to obtain the motion vector of the feature point by solving the optical flow equation. The dynamic generation strategy unit dynamically adjusts the number and distribution of virtual marker points according to the deformation of the pelvic floor tissue. When rapid tissue deformation is detected, the system automatically increases the marker point density to capture finer motion details; conversely, in areas where the tissue changes slowly, the system reduces the number of marker points to optimize computing resources. The confidence evaluation unit assigns a confidence score to each virtual marker point and initiates re-identification when the confidence is below a threshold. The confidence score is calculated based on the following factors:

[0099] ,

[0100] wherein, is the total confidence of the marker point , with a value range of 0-1, and a higher value indicating more reliable tracking; is the template matching confidence, reflecting the matching degree of the current marker point with the template, with a value range of 0-1; For motion coherence confidence, reflecting whether the marker point motion conforms to the physical law, the value range is 0-1; For topology consistency confidence, reflecting whether the marker point has maintained the topological relationship with the surrounding marker points, the value range is 0-1; 、 、 For weight coefficients, respectively representing the relative importance of the three confidence factors, usually set to 0.4, 0.3 and 0.3, and satisfying 1. When is lower than the preset threshold (usually 0.6), the system will start the re-identification process.

[0101] The marker point data processing unit 413 is used for pre-processing, noise filtering and multi-source data synchronous fusion of the position information collected by the physical marker point subsystem 411 and the virtual marker point subsystem 412. The pre-processing includes steps such as coordinate system conversion and data standardization; the noise filtering uses an adaptive Kalman filter, which can automatically adjust the filtering parameters according to the signal characteristics; the multi-source data synchronous fusion uses a time window technology to unify the data of different sampling frequencies to a standard time reference system.

[0102] The topology-maintained marker point distribution strategy unit 414 is used to ensure that the marker point distribution meets the topological homeomorphism condition based on the anatomical characteristics of the pelvic floor region. Specifically, the system first constructs a topological template of the pelvic floor region based on anatomical knowledge, identifies key functional points and anatomical structure interfaces, then applies Delaunay triangulation to construct an initial marker network, calculates the information gain of each candidate marker position, iteratively optimizes the marker point distribution until the coverage requirement is met, and finally applies topology verification to ensure that the marker network meets the homeomorphism condition.

[0103] The adaptive sampling control unit 415 is used to dynamically adjust the sampling frequency according to the motion state of the pelvic floor organ. When the motion speed is lower than the threshold A (usually set to 5 mm / s), low-frequency sampling (5-10 Hz) is used; when the motion speed is higher than the threshold B (usually set to 30 mm / s), high-frequency sampling (30-60 Hz) is used; in the intermediate state, the sampling frequency is calculated according to the following formula:

[0104] ,

[0105] Where, is the current sampling frequency, unit Hz, representing the number of samples per second; is the basic sampling frequency (usually 5-10 Hz), used for low-speed motion state; is the high-frequency sampling frequency (usually 30-60 Hz), used for high-speed motion state; is the current motion speed, unit mm / s; is a velocity threshold A (typically 5 mm / s), representing the upper limit of low-speed motion; is a velocity threshold B (typically 30 mm / s), representing the lower limit of high-speed motion. This adaptive sampling strategy significantly improves the system's response speed to violent motion, while optimizing data storage and processing efficiency.

[0106] The four-dimensional spatiotemporal trajectory reconstruction system 42 based on manifold learning is used to convert the landmark position change information into a continuous four-dimensional pelvic organ dynamic model. As shown in Figure 5 , the four-dimensional spatiotemporal trajectory reconstruction system 42 includes a manifold representation and dimension reduction unit 421, a multi-scale manifold structure analysis unit 422, a hierarchical reconstruction unit 423, a spatiotemporal continuity constraint unit 424, and a dynamic feature extraction unit 425.

[0107] The manifold representation and dimension reduction unit 421 is used to map the high-dimensional feature vector of the landmark point to a low-dimensional manifold space. In an embodiment of the present application, the state of each landmark point is represented by the following multi-dimensional feature vector:

[0108] ,

[0109] wherein, is the feature vector of the landmark point p; is the position coordinate of the landmark point in three-dimensional space, with units of mm; is the motion velocity of the landmark point in the direction of the three coordinate axes, with units of mm / s; is the motion acceleration of the landmark point in the direction of the three coordinate axes, with units of mm / s 2 ; is the local curvature of the surface on which the landmark point is located, with units of 1 / mm; is the relative distance matrix, with a size of , where n is the total number of landmark points, and each element represents the Euclidean distance from the current landmark point to other landmark points, with units of mm; is a topological relationship identifier, representing the topological connection relationship of the landmark point with surrounding landmark points; is the state confidence of the landmark point, with a value range of 0-1.

[0110] Through the local linear embedding (LLE) technique, the high-dimensional feature vector is mapped to a low-dimensional manifold space:

[0111] ,

[0112] wherein, is the mapped low-dimensional manifold representation, with a dimension of , is the number of data points; is the original high-dimensional feature matrix, with dimension , is the original feature dimension; is the number of neighbors (usually set to 12-20), representing the number of neighbor points considered when calculating the local geometry of each data point; is the target dimension after dimension reduction (usually set to 3-5), much smaller than the original dimension . LLE algorithm is implemented by the following steps: first, find nearest neighbors for each data point; then calculate the reconstruction weight, so that each point can be represented by a linear combination of its neighbors; finally, find the low-dimensional embedding that satisfies the reconstruction weight relationship by solving the eigenvalue problem. This dimension reduction method can maintain the local geometric structure of the data while significantly reducing the computational complexity.

[0113] The multi-scale manifold structure analysis unit 422 is used to analyze the dynamic behavior of the pelvic floor from three scales: micro, meso, and macro. The micro scale focuses on local deformation of the tissue, and differential geometry tools such as Gaussian curvature, mean curvature, etc. are used to analyze the deformation characteristics of the surface; the meso scale focuses on the movement of individual organs, and rigid body dynamics and elastic body mechanics are used to analyze the overall displacement and deformation of the organs; the macro scale focuses on the coordinated action of organ groups, and system dynamics and coordination theory are used to analyze the interaction and coordination mode between organs.

[0114] The hierarchical reconstruction unit 423 is used to realize complete four-dimensional model reconstruction through three levels of local surface reconstruction, organ-level reconstruction, and overall area reconstruction. The hierarchical reconstruction unit 423 includes a local surface reconstruction subunit, an organ-level reconstruction subunit, an overall area integration subunit, and a time sequence integration subunit. The local surface reconstruction subunit reconstructs the local surface around the marker point through radial basis function (RBF) interpolation technology, and the RBF interpolation function is defined as:

[0115] ,

[0116] wherein, is the interpolation result at point , representing the spatial position or other physical quantity of the point; is the radial basis function, which is a function of distance, and a multi-quadratic function is usually selected , is the smoothing parameter, which controls the smoothness of the surface, and the typical value is 1.0-5.0mm; is the weight coefficient of the th marker point, obtained by solving a linear equation system; represents the Euclidean distance from point to the th marker point , with the unit of mm; Low-order polynomial terms, typically linear or quadratic polynomials, are used to capture global trends; The number of landmark points involved in the interpolation calculation. RBF interpolation can handle irregularly distributed data points and has good smoothness, which is very suitable for reconstructing continuous surfaces from discrete landmark points.

[0117] The organ-level reconstruction sub-unit combines shape priors and physical model constraints to realize the complete reconstruction of a single organ. Shape priors are provided through statistical shape models (SSM), and physical model constraints include volume preservation, surface smoothness, and anatomical constraints. The overall region integration sub-unit applies multi-body system theory to handle the interrelationships and constraints between organs, ensuring that the reconstructed organs maintain the correct relative positional relationship. The time sequence integration sub-unit ensures continuity in the time dimension through a state space model, with the state equation and observation equation defined as:

[0118] ,

[0119] ,

[0120] where, is the state vector at time t, containing the position, shape, and motion parameters of the organ, with a dimension of is the number of state variables; is the control input vector, representing external forces or patient-initiated motion instructions, with a dimension of is the number of control variables; is the observation vector, containing variables directly observed from landmark data, with a dimension of is the number of observation variables; is the state transition matrix, with a dimension of , describing how the system state evolves over time; is the control matrix, with a dimension of , describing how control inputs affect the system state; is the observation matrix, with a dimension of , describing how state variables map to observation values; is the process noise vector, with a dimension of , assumed to be Gaussian white noise with a mean of zero, and the covariance matrix is is the observation noise vector, with a dimension of , assumed to be Gaussian white noise with a mean of zero, and the covariance matrix is . The state space model is solved through a Kalman filter or extended Kalman filter to achieve optimal estimation of the dynamic state of the pelvic floor organs.

[0121] The spatio-temporal continuity constraint unit 424 is used to ensure the physiological feasibility of the reconstructed model by time continuity constraint, spatial continuity constraint and anatomical constraint. The time continuity constraint is achieved by adding velocity and acceleration smoothness regularization terms; the spatial continuity constraint is achieved by physical model limiting the range of organ deformation; the anatomical constraint is based on expert knowledge, setting the feasible range and constraints of organ motion.

[0122] The dynamic feature extraction unit 425 is used to extract key dynamic parameters from the four-dimensional model for functional assessment. As described in claim 9, the dynamic feature extraction unit 425 extracts the following dynamic parameters from the four-dimensional model: the curve of the position of the anatomical key point changing with time; the curve of the volume and shape parameter change of the organ; the relative position parameters between organs, including the time-varying functions of distance, angle, and overlapping area; the distribution of tissue strain and stress estimation; the motion range and velocity distribution; the motion coordination index.

[0123] These parameters are standardized for functional status assessment. Standardization uses the Z-score method to convert each parameter to a standard score:

[0124]

[0125] wherein, is the standard score, dimensionless, indicating the number of standard deviations of the original parameter value from the average value of the reference population; is the original parameter value, the unit depends on the specific parameter; is the average value of the reference population, with the same unit as the original parameter; is the standard deviation of the reference population, with the same unit as the original parameter. By comparing with the normal population data, the system can objectively evaluate the normality of the pelvic floor function status, and provide quantitative basis for clinical diagnosis.

[0126] The data storage module 5 is connected with the four-dimensional pelvic floor organ dynamic tracking module 4, and is used to store the three-dimensional simulation model and dynamic change data. In an embodiment of the present application, the data storage module 5 adopts a hierarchical storage architecture, and stores the original data, intermediate results and final analysis results respectively, so as to optimize the storage space and data access efficiency. In addition, the data storage module 5 also realizes data backup and security protection function, ensuring the safety and privacy of patient data.

[0127] The data display module 6 is connected with the data storage module 5, and is used to dynamically display the three-dimensional simulation model and synchronously play back the change process of the pelvic floor organs under different states. As described in claim 10, the data display module 6 includes a three-dimensional visualization sub-module, a dynamic playback sub-module, a parameter curve display sub-module and a functional assessment result display sub-module.

[0128] ​The three-dimensional visualization submodule is used to display the three-dimensional simulation model of the bladder, vagina and rectum in real time. Preferably, the three-dimensional visualization adopts a volume rendering technique, which can display the internal structure and surface details of the organs. The dynamic playback submodule is used to playback the change process of the pelvic floor organs in different states according to the stored dynamic change data. The playback speed is adjustable, supporting slow playback to observe the micro changes and supporting fast playback to grasp the overall change trend. The parameter curve display submodule is used to display the distance, overlapping area and relative angle between the pelvic floor anatomical structures over time. These curves intuitively reflect the dynamic change characteristics of the pelvic floor organs, facilitating the doctor to perform functional evaluation. The functional evaluation result display submodule is used to display the pelvic floor functional state evaluation results and clinical diagnosis suggestions. The evaluation results are presented in the form of charts and texts, including functional normality evaluation, abnormal type determination and severity quantification.

[0129] In summary, the pelvic floor organ three-dimensional ultrasound reconstruction and dynamic change analysis system provided by the present application realizes comprehensive evaluation of the anatomical structure and functional state of the pelvic floor organs through the innovative four-dimensional pelvic floor organ dynamic tracking technology, and provides strong support for accurate diagnosis and personalized treatment of pelvic floor dysfunction.

[0130] The embodiments of the present application are not limited to the above examples, and various modifications and improvements made by those skilled in the art without departing from the concept of the present application should be considered as falling within the scope of protection of the present application.

Claims

1. A system for three-dimensional ultrasound reconstruction and dynamic change analysis of a pelvic floor organ, characterized in that, The application relates to a four-dimensional pelvic floor organ dynamic tracking system. The system comprises: a pelvic floor volume probe for being placed above the pelvic floor organs and collecting three-dimensional volume data; a three-dimensional pelvic floor organ reconstruction module connected with the pelvic floor volume probe, for receiving the three-dimensional volume data and reconstructing three-dimensional simulation models of the bladder, vagina and rectum; a relative position quantitative analysis module connected with the three-dimensional pelvic floor organ reconstruction module, for establishing a standardized pelvic floor coordinate system and calculating anatomical position parameters; a four-dimensional pelvic floor organ dynamic tracking module connected with the pelvic floor volume probe and the three-dimensional pelvic floor organ reconstruction module, for continuously collecting dynamic change data of different states of the bladder, vagina and rectum, wherein the four-dimensional pelvic floor organ dynamic tracking module comprises a topologically-aware multi-dimensional marker point dynamic tracking system and a four-dimensional spatiotemporal trajectory reconstruction system based on manifold learning; the topologically-aware multi-dimensional marker point dynamic tracking system is used for collecting position change information of the pelvic floor organs through heterogeneous marker points and performing topological feature preserving analysis; the four-dimensional spatiotemporal trajectory reconstruction system based on manifold learning is used for converting the marker point position change information into a continuous four-dimensional pelvic floor organ dynamic model; a data storage module connected with the four-dimensional pelvic floor organ dynamic tracking module, for storing the three-dimensional simulation models and dynamic change data; 2. The pelvic floor organ three-dimensional ultrasound reconstruction and dynamic change analysis system according to claim 1, characterized in that, a data display module connected with the data storage module, for dynamically displaying the three-dimensional simulation models and synchronously playing back the change process of the pelvic floor organs under different states. The topologically-aware multi-dimensional marker point dynamic tracking system comprises: a physical marker point subsystem for collecting position change information of the vagina and pelvic floor muscles through micro sensors; a virtual marker point subsystem for identifying and tracking bladder feature points in ultrasonic images through image processing algorithms; a marker point data processing unit for pre-processing, noise filtering and multi-source data synchronous fusion of the position information collected by the physical marker point subsystem and the virtual marker point subsystem; a topologically-preserving marker point distribution strategy unit for ensuring that the marker point distribution meets the topological homeomorphism condition based on the anatomical characteristics of the pelvic floor region; 3.The system of claim 1, wherein, an adaptive sampling control unit for dynamically adjusting the sampling frequency according to the motion state of the pelvic floor organs. The four-dimensional spatiotemporal trajectory reconstruction system based on manifold learning comprises: a manifold representation and dimension reduction unit for mapping the high-dimensional feature vectors of the marker points to a low-dimensional manifold space; a multi-scale manifold structure analysis unit for analyzing the pelvic floor dynamic behavior from three scales of micro, meso and macro; a hierarchical reconstruction unit for realizing complete four-dimensional model reconstruction through three levels of local surface reconstruction, organ-level reconstruction and overall region reconstruction; a spatiotemporal continuity constraint unit for ensuring the physiological feasibility of the reconstructed model through time continuity constraint, space continuity constraint and anatomical constraint; 4. The pelvic floor organ three-dimensional ultrasound reconstruction and dynamic change analysis system of claim 2, wherein, a dynamic feature extraction unit for extracting key dynamic parameters from the four-dimensional model for functional evaluation. The micro sensors in the physical marker point subsystem comprise: a three-axis accelerometer for measuring the acceleration change of the marker points; a three-axis gyroscope for measuring the angle change of the marker points; a wireless transmission module for transmitting the collected data to the marker point data processing unit in real time. An electromagnetic induction power supply system for providing passive power to a sensor through an external electromagnetic field.

5. The pelvic organ three-dimensional ultrasound reconstruction and dynamic change analysis system of claim 2, wherein, The virtual marker point subsystem comprises: A feature point recognition unit configured to automatically recognize feature points based on tissue interfaces, reflection intensity features, and texture features in the ultrasound images; An inter-frame tracking unit configured to realize inter-frame tracking of the feature points by using a hierarchical template matching method combined with an optical flow method; A dynamic generation strategy unit configured to dynamically adjust the number and distribution of the virtual marker points according to deformation of the pelvic floor tissue; A confidence assessment unit configured to assign a confidence score to each virtual marker point, and to initiate re-identification when the confidence score is lower than a threshold value. 6.The system of claim 3, wherein, The hierarchical reconstruction unit comprises: A local surface reconstruction subunit configured to reconstruct a local surface around the marker points by using a radial basis function interpolation technique; An organ-level reconstruction subunit configured to realize complete reconstruction of a single organ by combining shape priors and physical model constraints; An overall region integration subunit configured to apply a multi-body system theory to process mutual relationships and constraints between organs; A time series integration subunit configured to ensure continuity in the time dimension by using a state space model. 7.The system according to claim 1, wherein, The relative position quantitative analysis module is configured to: Take a center point of the three-dimensional pelvic floor ultrasound image as a position of the pelvic ischial spine; Establish an xyz three-dimensional Cartesian coordinate system with the pelvic ischial spine as an origin of the coordinate system; Take a direction in which the pelvic ischial spine points to the cervical os as a z-axis, a direction perpendicular to the y-axis as an x-axis, and a direction perpendicular to the z-axis as a y-axis; Define the bladder and the vaginal organ as two objects, and divide the relative position relationship into three categories: overlap, contact, and separation; Calculate a distance, an included angle, and an overlapping area between the bladder and the vaginal organ as quantitative parameters of the relative position. 8.The system of claim 1, wherein, The three-dimensional pelvic floor organ reconstruction module comprises: A machine learning image segmentation unit configured to learn a seed region by using an active shape model algorithm, form a pelvic floor organ segmentation template with an active shape model parameter, and identify and segment the pelvic floor organ transverse, sagittal, or coronal ultrasound images by using the template; A depth map reconstruction unit configured to select a reference image from a series of pelvic floor organ ultrasound images, calculate a depth map, and register other images with the reference image to realize depth reconstruction; A curved surface fitting and fusion unit configured to mesh three-dimensional reconstruction data, form a fixed-size grid, establish a boundary plane, fuse generated curved surfaces, and complete construction of a three-dimensional pelvic floor organ model. 9.The system of claim 3, wherein, The dynamic feature extraction unit is configured to extract the following dynamic parameters from the four-dimensional model: A curve of position changes of an anatomical key point over time; A curve of changes in organ volume and shape parameters; Relative position parameters between organs, including a distance, an included angle, and an overlapping area as a time-varying function; Tissue strain and stress distribution estimation; Motion range and velocity distribution; Motion coordination index; The dynamic parameters are used for functional state evaluation after standardization processing. 10.The system of claim 1, wherein, The data display module comprises: A three-dimensional visualization submodule configured to display three-dimensional simulation models of the bladder, the vaginal organ, and the rectum in real time; A dynamic playback submodule configured to play back changes in the pelvic floor organs in different states according to stored dynamic change data. The parameter curve display submodule is configured to display the distance, the overlapping area, and the relative angle between the anatomical structures of the pelvic floor changing over time. The functional evaluation result display submodule is configured to display the functional state evaluation result of the pelvic floor and the clinical diagnosis suggestion.

Images