A bladder ultrasound 3D reconstruction method adapted to a portable device
By employing regional adaptive boundary extraction, dual verification and completion, anatomical feature point matching, and constrained Delaunay triangulation, the anti-interference and adaptability issues of bladder 3D reconstruction on portable devices are resolved, achieving real-time reconstruction with high precision and low computing power.
Patent Information
- Application Number
- CN202610752637.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-28
- Publication Date
- 2026-08-25
AI Technical Summary
Existing bladder 3D reconstruction technology has poor anti-interference ability on portable devices, poor adaptability to irregular bladder shapes, insufficient topological accuracy of reconstruction models, and high computing power requirements, making it difficult to run in real time on low computing power platforms.
A three-dimensional reconstruction of the bladder is achieved by employing regional adaptive boundary contour extraction, dual validity verification and intelligent completion of missing contours, fine registration based on anatomical feature points and constrained Delaunay triangulation algorithm, combined with lightweight geometric algorithms.
It significantly improves clinical anti-interference capabilities, adapts to more than 95% of irregular bladder shapes, has high consistency between the reconstructed model and anatomical structure, and can run in real time on ARM Cortex-M series embedded chips, reducing equipment costs.
Smart Images
Figure CN122636804A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical ultrasound image processing technology, specifically relating to a method for three-dimensional reconstruction of the bladder using portable ultrasound equipment, which is particularly suitable for clinical scenarios such as urinary retention, bladder lesion assessment, and postoperative morphological monitoring. Background Technology
[0002] In clinical practice, such as urology, rehabilitation medicine, and geriatrics, accurate three-dimensional reconstruction of the bladder is a crucial foundation for organ visualization, anatomical structure analysis, and dynamic functional assessment. While multi-slice ultrasound-based bladder 3D reconstruction solutions have achieved portable applications, their core algorithms still have significant shortcomings, making it difficult to simultaneously meet the core clinical requirements of interference resistance, high robustness, adaptation to irregular shapes, and operation on low-computing-power platforms.
[0003] Existing technologies are mainly divided into two categories:
[0004] The first type is a contour stitching reconstruction scheme based on multi-section ultrasound. This type of scheme typically acquires multiple bladder ultrasound sections by rotating the probe, extracts the boundaries, and then completes 3D reconstruction by sequentially triangulating adjacent sections point by point. However, this scheme has the following technical drawbacks:
[0005] 1. Lack of contour validity verification: In clinical practice, intestinal gas occlusion and ultrasound artifacts often lead to incomplete or abnormal extraction of bladder boundaries in single sections. Directly inputting such abnormal contours into the reconstruction process will cause reconstruction failure or distortion of the model's topology, resulting in poor anti-interference ability.
[0006] 2. Simple splicing method: It adopts sequential triangular splicing of adjacent cut surfaces point by point. When the bladder shape is irregular or the number of contour points of adjacent cut surfaces is inconsistent, problems such as self-intersection of triangular facets, surface holes, and topological errors are very likely to occur. It has extremely poor adaptability to irregular shapes such as prostatic hyperplasia compression and bladder diverticula.
[0007] The second category is bladder 3D reconstruction schemes based on deep learning. These schemes use neural networks to segment and extract features from a small number of ultrasound sections to generate a 3D model, but their core limitation restricts their application on portable devices.
[0008] 1. Strong data dependence and poor generalization: It requires massive amounts of labeled data for training and has insufficient generalization ability for rare bladder malformations or focal lesions.
[0009] 2. High computing power requirements: It requires high-performance computing power such as GPU / NPU, and cannot run in real time on the embedded chips or mobile platforms of portable devices, resulting in high device costs and making it difficult to popularize in grassroots and bedside scenarios.
[0010] Therefore, there is an urgent need for a bladder ultrasound 3D reconstruction method that can solve the above-mentioned pain points of existing technologies, and has high anti-interference ability, strong adaptability to irregular shapes, and is suitable for operation in portable devices with low computing power. Summary of the Invention
[0011] The present invention aims to provide a bladder ultrasound 3D reconstruction method adapted to portable devices, in order to solve the technical problems of poor anti-interference ability, poor adaptability to irregular bladder shape, insufficient topological accuracy of reconstruction model and high computing power requirements in the prior art.
[0012] To achieve the above objectives, the technical solution provided by the present invention is as follows:
[0013] A method for bladder ultrasound 3D reconstruction adapted to portable devices includes the following steps:
[0014] Step 1: Maximum Bladder Section Locking and Multi-Segment Ultrasound Acquisition
[0015] The ultrasound probe is placed above the patient's pubic symphysis, and the largest section of the bladder is locked as the rotation reference plane. The 3D rotating motor is controlled to rotate within a preset angle range at preset step angles, and the ultrasound cross-sectional image and rotation angle corresponding to each step angle are acquired and recorded simultaneously.
[0016] Step 2: Region-based adaptive bladder boundary contour extraction
[0017] The ultrasound images of each section are preprocessed; based on the differences in ultrasound imaging of different anatomical parts of the bladder, a differentiated strategy is adopted to extract the boundary contours: for the main body of the side wall, the Otsu method combined with the Canny operator is used to extract high-contrast boundaries; for the top and neck, the Chan-Vese level set algorithm is used to extract weak boundaries; finally, the closed two-dimensional bladder contours of each section are output.
[0018] Step 3: Contour validity verification and intelligent completion of missing contours
[0019] Perform double validity checks on each cross-sectional contour extracted in step 2:
[0020] Closure check: Calculate the pixel distance between the start and end points of the contour. If it is greater than the preset threshold, it is determined to be invalid.
[0021] Form rationality verification: Calculate the perimeter, convex hull area, and aspect ratio of the principal axis of the contour. If the difference between any parameter and the average value of the corresponding parameter of the adjacent valid contour exceeds the preset ratio, it is determined to be invalid.
[0022] For contours deemed invalid, a thin-plate spline interpolation algorithm is used for completion: for a single invalid contour, interpolation is performed based on the valid contours before and after it; for multiple consecutive invalid contours, multi-section joint interpolation is performed by combining the prior anatomical morphology of a bladder-like ellipsoid.
[0023] Step 4: Contour space coordinate transformation based on motor rotation angle
[0024] Based on the motor rotation angle corresponding to each cross-section recorded in step 1, all pixels of the two-dimensional closed contour output in step 3 are converted into three-dimensional spatial coordinates to obtain the three-dimensional point cloud data of each contour.
[0025] Step 5: Fine registration of adjacent contours based on feature point matching
[0026] For the 3D point cloud of each contour, the curvature value is calculated point by point, and points with absolute curvature values exceeding a preset threshold are extracted as anatomical feature points. For the contours of two adjacent cross-sections, coarse alignment based on centroid is first performed, and then the nearest neighbor matching algorithm is used to match based on the curvature similarity and spatial position of feature points. Finally, with the matched feature points as constraints, the fine registration of the two contours is completed through the iterative nearest point algorithm, and the adjacent contours are uniformly resampled to ensure consistent point count.
[0027] Step 6: Constraining Delaunay triangulation contour connection and constructing the bladder 3D model
[0028] The two adjacent cross-sectional contours processed in step 5 are defined as the upper contour point set and the lower contour point set, respectively. The connection between the anatomical feature points matched in step 5 is defined as the forced constraint edge. Based on the one-to-one correspondence of the adjacent contour matching, the constrained Delaunay triangulation algorithm is used to stitch the triangular facets together. The forced constraint edge is used as a hard constraint condition to ensure that all forced constraint edges are completely retained in the edge set of the final triangular mesh, thereby generating the bladder three-dimensional surface model.
[0029] Furthermore, in step 3, if the number of consecutive invalid contours exceeds a preset percentage (e.g., 40%) of the total number of captured cross-sections, a re-capture prompt is triggered.
[0030] Furthermore, in step 5, the curvature threshold is set to twice the average curvature of the current contour by default.
[0031] Further, in step 6, the set of forced constraint edges is represented as E = {E0, E1, ..., E...} m-1}, where m is the number of anatomical feature point pairs; the constrained Delaunay triangulation satisfies: ∀E k ∈E, E k ∈Edge(T), that is, each mandatory constraint edge E k Edge(T) must belong to the set of edges of the final triangular mesh T.
[0032] Furthermore, a bladder ultrasound 3D reconstruction system adapted for portable devices includes:
[0033] An ultrasound probe is used to acquire ultrasound images of the bladder.
[0034] A 3D rotary motor is used to drive the ultrasonic probe to rotate within a preset angle range at a preset step angle.
[0035] The processor executes a computer program to implement the steps of the above-described bladder ultrasound 3D reconstruction method adapted to portable devices.
[0036] Furthermore, a portable ultrasound device includes the aforementioned bladder ultrasound 3D reconstruction system adapted for portable devices.
[0037] Compared with the prior art, the present invention has the following advantages:
[0038] 1. Significantly improved clinical anti-interference capability: By setting up dual validity verification of contours and intelligent completion mechanism based on thin plate spline interpolation, it effectively addresses the problem of abnormal single-section boundary caused by intestinal gas and ultrasound artifacts, avoiding reconstruction failure or model topology distortion caused by single-section quality issues.
[0039] 2. Excellent adaptability to irregular bladder morphology and anatomical consistency: The innovative design of a fine registration and constraint Delaunay triangulation stitching method based on anatomical feature point matching, with stable anatomical feature points as hard constraints for topological connection, fundamentally solves problems such as self-intersection of triangular facets and surface holes, and can adapt to more than 95% of irregular bladder morphologies in clinical practice. The reconstructed model has high consistency with the real anatomical structure.
[0040] 3. Highly adaptable to engineering applications: The entire process uses lightweight geometric algorithms, without relying on deep learning models or high-performance computing power such as GPUs / NPUs. It can be directly adapted to existing portable ultrasound devices equipped with 3D rotating motors. It can run in real time on embedded chips such as ARM Cortex-M series or mobile platforms through software upgrades (the entire process takes less than 300ms), making it easy to popularize in grassroots, community, and bedside scenarios. Attached Figure Description
[0041] Figure 1 This is a flowchart illustrating the overall process of the bladder ultrasound 3D reconstruction method in this embodiment of the invention.
[0042] Figure 2 This is a top-view diagram of the 3D rotating motor in an embodiment of the present invention.
[0043] Figure 3 This is a schematic diagram of the original image of mutually perpendicular cross-sections of bladder ultrasound and the final 3D mesh model of the bladder in an embodiment of the present invention. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.
[0045] This invention provides a bladder ultrasound 3D reconstruction method adapted to portable devices to solve the technical problems of poor anti-interference ability, poor adaptability to irregular bladder shapes, insufficient topological accuracy of reconstruction models, and high computing power requirements in the prior art.
[0046] The embodiments of the present invention use a commercially available portable bladder ultrasound device equipped with a 3D rotating motor as a carrier. Its main control uses an ARM Cortex-M7 core embedded MCU and is not configured with a GPU / NPU.
[0047] like Figure 1 As shown, the specific implementation steps of the bladder ultrasound 3D reconstruction method adapted to portable devices according to the present invention are as follows:
[0048] Step 1: Maximum Bladder Section Locking and Multi-Segment Ultrasound Acquisition
[0049] The ultrasound probe is placed above the patient's pubic symphysis, and the largest section of the bladder is locked as the rotation reference plane. The 3D rotating motor is controlled to rotate within a preset angle range at preset step angles, and the ultrasound cross-sectional image and rotation angle corresponding to each step angle are acquired and recorded simultaneously.
[0050] Specifically, the operator places the probe above the patient's pubic symphysis, rotates and fine-tunes the probe to find the largest cross-section of the bladder, which serves as the 0° reference for the motor. The motor rotates within the 0°~180° range according to a preset step angle. The recommended step angle is 5°~15°, which can be adjusted according to clinical needs. The corresponding ultrasound cross-sectional image acquired at each step angle is simultaneously bound.
[0051] In this embodiment: the operator places the probe above the patient's pubic symphysis, rotates the probe to find the largest bladder section that can fully display the bladder, and locks it as the 0° mechanical reference plane for the motor. The system controls the 3D rotary motor to rotate within the range of 0° to 180° in 15° step angles, acquiring a total of 12 bladder ultrasound sections, such as... Figure 2 As shown, the real-time angle data of the motor corresponding to each cross-section (0°, 15°, 30°, ..., 165°) is recorded synchronously.
[0052] Step 2: Region-based adaptive bladder boundary contour extraction
[0053] The ultrasound images of each section are preprocessed; based on the differences in ultrasound imaging of different anatomical parts of the bladder, a differentiated strategy is adopted to extract the boundary contours: for the main body of the side wall, the Otsu method combined with the Canny operator is used to extract high-contrast boundaries; for the top and neck, the Chan-Vese level set algorithm is used to extract weak boundaries; finally, the closed two-dimensional bladder contours of each section are output.
[0054] Specifically, for each ultrasound section, a 3×3 adaptive median filter is first applied to denoise and preserve edge details. Based on the differences in ultrasound imaging of different anatomical locations of the bladder, the contours are extracted differentially: the main boundary of the bladder lateral wall has high imaging contrast, so the Otsu method combined with the Canny operator is used to extract the contours; the top and neck of the bladder are easily affected by tissue and gas interference and have weak boundary contrast, so the Chan-Vese level set algorithm is used to extract the weak boundaries; finally, the closed two-dimensional bladder contours of each section are output for subsequent 3D reconstruction.
[0055] In this embodiment: the system performs adaptive median filtering preprocessing on each of the 12 acquired cross-sectional images using a 3×3 window. Then, boundary extraction is performed based on the imaging differences of bladder anatomical locations: for areas with high imaging contrast, such as the lateral walls, the Otsu method combined with the Canny operator is used to extract clear boundaries; for areas with weak contrast, such as the top and neck, which are easily disturbed, the Chan-Vese level set algorithm is used to fit weak boundaries. Finally, 12 initial closed two-dimensional bladder contours are output.
[0056] Step 3: Contour validity verification and intelligent completion of missing contours
[0057] Perform double validity checks on each cross-sectional contour extracted in step 2:
[0058] Closure check: Calculate the pixel distance between the start and end points of the contour. If it is greater than the preset threshold, it is determined to be invalid.
[0059] Form rationality verification: Calculate the perimeter, convex hull area, and aspect ratio of the principal axis of the contour. If the difference between any parameter and the average value of the corresponding parameter of the adjacent valid contour exceeds the preset ratio, it is determined to be invalid.
[0060] For contours deemed invalid, a thin-plate spline interpolation algorithm is used for completion: for a single invalid contour, interpolation is performed based on the valid contours before and after it; for multiple consecutive invalid contours, multi-section joint interpolation is performed by combining the prior anatomical morphology of a bladder-like ellipsoid.
[0061] Specifically, to address the issue that intestinal gas and ultrasound artifacts often lead to abnormal contours in clinical practice, thereby causing the entire reconstruction process to fail, this step designs a dual verification and intelligent completion mechanism, which is implemented as follows:
[0062] First, perform double validity checks on each cross-sectional contour extracted in step 2:
[0063] 1. Closure check: Calculate the pixel distance between the start and end points of the contour. If the distance is greater than the preset threshold (the default value is 5 pixels, which can be adaptively adjusted according to the device image resolution), it is determined to be a non-closed invalid contour.
[0064] 2. Morphological rationality verification: Calculate three types of morphological feature parameters of the current contour: contour perimeter, contour convex hull area, and principal axis aspect ratio. If the difference between any parameter and the average value of the corresponding parameters of the two adjacent valid cross-sectional contours exceeds the preset ratio (the default value is 30%, which can be adaptively adjusted according to the clinical scenario), it is judged as an abnormal invalid contour.
[0065] For contours deemed invalid, they are not directly used in subsequent reconstruction, but are repaired using an intelligent completion algorithm based on thin-plate spline interpolation.
[0066] 1. For a single invalid contour: using the two valid cross-sectional contours adjacent to the invalid contour as a reference, and combining the morphological change trends of the two valid contours, the complete contour of the current cross-section is generated by the thin plate spline interpolation algorithm;
[0067] 2. For the extreme case of two or more consecutive invalid contours: combining the prior constraint of the bladder's anatomical shape as an ellipsoid, and using the valid contours before and after the consecutive invalid contours as the benchmark, multi-section joint interpolation completion is completed;
[0068] 3. If the number of consecutive invalid contours exceeds 40% of the total number of captured cross-sections, a re-capture prompt will be triggered to avoid serious distortion in the reconstructed model.
[0069] The above mechanism ensures that all subsequent reconstructed contours are valid, thus preventing reconstruction failures caused by single-section quality issues from the outset.
[0070] In this embodiment, the system performs dual checks on the closure and morphological rationality of 12 initial contours. One section, due to intestinal gas obstruction, resulted in a 12-pixel distance between the start and end points of the extracted contour, exceeding the preset 5-pixel threshold, and was therefore judged as a "non-closed invalid contour." The system then initiates a completion mechanism: using the two adjacent valid section contours as a reference, a thin plate spline interpolation algorithm is employed, combined with the morphological change trends of the two valid contours, to generate the completed contour of the current section. Ultimately, 12 continuous and complete valid contours are obtained.
[0071] Step 4: Contour space coordinate transformation based on motor rotation angle
[0072] Based on the motor rotation angle corresponding to each cross-section recorded in step 1, all pixels of the two-dimensional closed contour output in step 3 are converted into three-dimensional spatial coordinates to obtain the three-dimensional point cloud data of each contour, providing a spatial reference for subsequent contour registration.
[0073] In this embodiment: the system calls the motor rotation angles (0°, 15°, ..., 165°, with each 15° increment) corresponding to the 12 cross-sections recorded in step 1, and converts all pixels on the 12 valid two-dimensional closed contours output in step 3 into three-dimensional spatial coordinates one by one based on their angle information, finally obtaining 12 bladder contour three-dimensional point cloud data bound to the corresponding motor angles, providing a spatial reference for subsequent fine registration of adjacent contours.
[0074] Step 5: Fine registration of adjacent contours based on feature point matching
[0075] For the 3D point cloud of each contour, the curvature value is calculated point by point, and points with absolute curvature values exceeding a preset threshold are extracted as anatomical feature points. For the contours of two adjacent cross-sections, coarse alignment based on centroid is first performed, and then the nearest neighbor matching algorithm is used to match based on the curvature similarity and spatial position of feature points. Finally, with the matched feature points as constraints, the fine registration of the two contours is completed through the iterative nearest point algorithm, and the adjacent contours are uniformly resampled to ensure consistent point count.
[0076] Specifically, the correspondence between adjacent contours directly determines the stitching accuracy of the final model. This step uses anatomical feature points as a reference for fine registration, providing a misaligned correspondence reference for subsequent triangular stitching, as detailed below:
[0077] For each contour's 3D point cloud, the curvature value is calculated point by point, and points whose absolute curvature value exceeds a preset threshold are extracted as anatomical feature points. The default threshold is twice the average curvature of the current contour, which can be adaptively adjusted according to the bladder morphology. These feature points correspond to key morphological inflection points such as the bladder neck, top, diverticulum edge, and lesion boundary, and are not affected by bladder fullness or the number of sections, thus having stable anatomical significance.
[0078] For the contours of two adjacent cross-sections, coarse alignment based on centroid is first performed to eliminate the overall offset between the two contours. Then, the nearest neighbor matching algorithm is used to perform one-to-one matching based on the curvature feature similarity and spatial distance of feature points, eliminating mismatched point pairs and establishing a stable correspondence between feature points. Finally, with the matched feature points as constraints, the iterative nearest point algorithm (ICP) is used to complete the fine registration of the two contours, further eliminating the small local offsets between adjacent contours.
[0079] After the fine registration is completed, the two contours are uniformly resampled based on the matched feature points to ensure that the number of points of adjacent contours is consistent and the positions of the feature points correspond one-to-one, providing a misaligned reference for subsequent contour connection and adapting to the registration of adjacent contours with different advance angles and shapes.
[0080] In this embodiment: the system calculates the curvature value of each point in the 3D point cloud of each contour, and extracts points whose absolute curvature value exceeds twice the average curvature of the current contour as anatomical feature points (corresponding to key inflection points such as the bladder neck and top). For contours of two adjacent sections (such as 0° and 15° sections), the centroids of the two are first calculated for coarse alignment; then, a nearest neighbor matching algorithm is used to match feature points one by one based on the curvature feature similarity and spatial distance, eliminating mismatched point pairs; finally, using the matched feature points as constraints, the two contours are finely registered through an iterative nearest point algorithm to eliminate minor local offsets. After fine registration, adjacent contours are uniformly resampled to ensure that the number of points in the two contours is consistent and that the feature point positions correspond one-to-one.
[0081] Step 6: Constraining Delaunay triangulation contour connection and constructing the bladder 3D model
[0082] The two adjacent cross-sectional contours processed in step 5 are defined as the upper contour point set and the lower contour point set, respectively. The connection between the anatomical feature points matched in step 5 is defined as the forced constraint edge. Based on the one-to-one correspondence of the adjacent contour matching, the constrained Delaunay triangulation algorithm is used to stitch the triangular facets together. The forced constraint edge is used as a hard constraint condition to ensure that all forced constraint edges are completely retained in the edge set of the final triangular mesh, thereby generating the bladder three-dimensional surface model.
[0083] Specifically, this step is as follows:
[0084] 1. Pre-definition
[0085] Let the contours of two adjacent cross-sections after fine registration and uniform resampling in step 5 be (two bladder cross-sections with adjacent motor rotation angles):
[0086] 1) Complete set of contour points for two adjacent cross-sections (all contour points, one-to-one global correspondence)
[0087] Upper contour point set (the contour of the adjacent cross-section that is spatially forward / has a smaller angle): P u = { p u 0, p u 1,……, p u n-1}
[0088] Lower contour point set (the contour of the section that is spatially located further back / at a larger angle among adjacent sections): Pd = { p d 0, p d 1,……, p d n-1}
[0089] in:
[0090] n is the number of uniform points after contour resampling; p u i With p d i For the one-to-one correspondence points completed in step 5; p u i : Upper contour point set P u The i-th point (i=0,1,…,n-1) in the array p d i : Lower contour point set P d The i-th point in the array (i=0,1,…,n-1).
[0091] 2) Anatomical feature point set (key inflection points selected from ordinary points)
[0092] From the ordinary points of the above upper and lower contours, extract the anatomical feature points selected in step 5 (corresponding to key morphological inflection points such as the bladder neck, top, diverticulum edge, and lesion boundary, which are not affected by filling degree and have stable anatomical significance), and define the feature point set:
[0093] Upper contour feature point set: F u ={ f u 0, f u 1, ..., f u m-1} (for P) u (A subset, retaining only the key parts)
[0094] Lower contour feature point set: F d ={ f d 0, f d 1, ..., f d m-1} (for P) d (A subset, retaining only the key parts)
[0095] Total anatomical feature set: F=F u ∪F d
[0096] Where: m is the total number of extracted anatomical feature points;
[0097] (f u k ,f d kThe first-to-one correspondence of anatomical feature points is completed in step 5: that is, the kth key inflection point of the upper and lower contours is in a one-to-one correspondence.
[0098] f u k : Upper contour feature point set F u The k-th point in the array (k=0,1,…,m-1).
[0099] f d k : Lower contour feature point set F d The k-th point in the array (k=0,1,…,m-1).
[0100] 3) Forced constraint edge set (hard constraint connection of feature points)
[0101] For each pair of matched anatomical feature points (f) u k ,f d k The line connecting the two sides is defined as a forced constraint edge:
[0102] E k It is the connection of each pair of matching anatomical feature points f u k and f d k Three-dimensional space line segments.
[0103] E k =line segment(f) u k ,f d k ), k=0,1,…,m-1 constraint edge set E={E0,E1,…,E m-1 These constraint edges are hard constraints for subsequent triangulation: the algorithm must completely preserve all constraint edges, without breaking or misaligning them, thus locking the topological relationships of the key bladder anatomical structures from the root and avoiding model clipping and topological errors.
[0104] 2. Constrained Delaunay triangulation
[0105] Based on the one-to-one correspondence points obtained through adjacent contour matching, the constrained Delaunay triangulation algorithm is used to stitch triangular facets together the regions between adjacent contours to generate a 3D surface model of the bladder sidewall. The triangulation process uses anatomical feature points as the forced constraint boundary; the core constraint formula is:
[0106] ∀E k ∈E, E k ∈Edge(T)
[0107] Formula meaning: For any forced constraint edge E belonging to the constraint edge set E, k The edge must be completely preserved in the edge set of the final triangular mesh and must not be broken by partitioning.
[0108] T is the set of triangular meshes generated by the subdivision, and Edge(T) is the set of all edges of the triangular meshes.
[0109] Based on this, it can be ensured that the pre-paired forced connections between each group of bladder anatomical feature points must be preserved intact in the final 3D model, without being arbitrarily changed or misaligned.
[0110] This constraint is the core difference between this scheme and existing point-by-point sequential splicing and unconstrained Delaunay partitioning. It prioritizes locking the topological connection relationship of anatomical feature points, thereby avoiding self-intersection of triangular facets and surface clipping problems from the root. It has strong adaptability to irregular bladder shapes and non-equiangularly distributed cross-sections.
[0111] In this embodiment: the system defines the adjacent upper and lower contour point sets after step 5 as P. u and P d The line connecting each pair of anatomical feature points matched in step 5 is defined as a forced constraint edge E. Subsequently, based on P... u and P d For each corresponding point, the constrained Delaunay triangulation algorithm is used to stitch triangular faces together between adjacent contours. During the triangulation process, all forced constraint edges E are set as hard constraints to ensure that each edge E... k All elements are fully preserved in the final generated triangular mesh, without being interrupted or misaligned by the meshing algorithm. This ultimately generates a high-topological-accuracy, anatomically consistent 3D bladder surface model, outputting a final anatomically accurate bladder 3D reconstruction model, such as... Figure 3 As shown, the actual runtime of the entire process on the embedded platform is less than 300 milliseconds, achieving real-time reconstruction.
[0112] This invention also provides a bladder ultrasound 3D reconstruction system adapted for portable devices, comprising: an ultrasound probe, a 3D rotating motor, and a processor. The ultrasound probe is used to acquire bladder ultrasound images; the 3D rotating motor is used to drive the ultrasound probe to rotate within a preset angle range at preset step angles; and the processor executes a computer program to implement the steps of the bladder ultrasound 3D reconstruction method adapted for portable devices described in this invention.
[0113] The present invention also provides a portable ultrasound device, including the bladder ultrasound 3D reconstruction system adapted to portable devices described in the present invention.
[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art can make various improvements and modifications without departing from the spirit and principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for bladder ultrasound 3D reconstruction adapted to portable devices, characterized in that, Includes the following steps: Step 1: Place the ultrasound probe above the patient's pubic symphysis and lock the largest section of the bladder as the rotation reference plane; control the 3D rotating motor to rotate within the preset angle range at preset step angles, and simultaneously acquire and record the ultrasound cross-sectional image and rotation angle corresponding to each step angle. Step 2: Preprocess the ultrasound images of each section; extract the boundary contours using a differential strategy based on the differences in ultrasound imaging of different anatomical locations of the bladder; finally output the closed two-dimensional bladder contours of each section. Step 3: Perform dual validity checks on each cross-sectional contour extracted in Step 2, including closure check and morphological rationality check; for contours determined to be invalid, use an algorithm based on thin plate spline interpolation to complete them; Step 4: Based on the motor rotation angle corresponding to each cross-section recorded in Step 1, convert all pixels of the two-dimensional closed contour output in Step 3 into three-dimensional spatial coordinates to obtain the three-dimensional point cloud data of each contour. Step 5: For the 3D point cloud of each contour, calculate the curvature value point by point, and extract the points whose absolute curvature value exceeds the preset threshold as anatomical feature points; for the contours of two adjacent cross-sections, first perform coarse alignment based on centroid, then use the nearest neighbor matching algorithm to match based on the curvature similarity and spatial position of feature points, and finally use the matched feature points as constraints to complete the fine registration of the two contours through the iterative nearest point algorithm, and uniformly resample the adjacent contours to ensure the consistency of the number of points; Step 6: Define the two adjacent cross-sectional contours processed in Step 5 as the upper contour point set and the lower contour point set, respectively, and define the connection between the matched anatomical feature points in Step 5 as the forced constraint edge; based on the one-to-one correspondence of the adjacent contour matching points, use the constrained Delaunay triangulation algorithm to stitch the triangular facets, and use the forced constraint edge as a hard constraint condition to ensure that all forced constraint edges are completely retained in the edge set of the final triangular mesh, thereby generating the bladder three-dimensional surface model.
2. The bladder ultrasound 3D reconstruction method adapted to portable devices according to claim 1, characterized in that, The differential strategy for extracting boundary contours in step 2 specifically involves: using the Otsu method combined with the Canny operator to extract high-contrast boundaries for the sidewalls and main body; and using the Chan-Vese level set algorithm to extract weak boundaries for the top and neck.
3. The bladder ultrasound 3D reconstruction method adapted to portable devices according to claim 1, characterized in that, The dual validity check mentioned in step 3 specifically includes: Closure check: Calculate the pixel distance between the start and end points of the contour. If it is greater than the preset threshold, it is determined to be invalid. Shape rationality verification: Calculate the perimeter of the contour, the area of the convex hull, and the aspect ratio of the principal axis. If the difference between any parameter and the average value of the corresponding parameter of the adjacent valid contour exceeds the preset ratio, it is determined to be invalid.
4. The bladder ultrasound 3D reconstruction method adapted to portable devices according to claim 1, characterized in that, Step 3, which involves using the thin-plate spline interpolation algorithm for completion, specifically includes: For a single invalid contour, interpolation is performed based on the valid contours before and after it; for multiple consecutive invalid contours, multi-section joint interpolation is performed to complete them by combining the prior knowledge of the anatomical morphology of the bladder-like ellipsoid.
5. The bladder ultrasound 3D reconstruction method adapted to portable devices according to claim 1, characterized in that, In step 3, if the number of consecutive invalid contours exceeds a preset percentage of the total number of captured cross-sections, a re-capture prompt will be triggered.
6. The bladder ultrasound 3D reconstruction method adapted to portable devices according to claim 1, characterized in that, In step 5, the curvature threshold is twice the average curvature of the current contour.
7. The bladder ultrasound 3D reconstruction method adapted to portable devices according to claim 1, characterized in that, In step 6, the set of forced constraint edges is represented as E = {E0, E1, ..., E...} m-1 }, where m is the number of anatomical feature point pairs; the constrained Delaunay triangulation satisfies: ∀E k ∈E, E k ∈Edge(T), that is, each mandatory constraint edge E k Edge(T) must belong to the set of edges of the final triangular mesh T.
8. A bladder ultrasound 3D reconstruction system adapted to portable devices, characterized in that, include: An ultrasound probe is used to acquire ultrasound images of the bladder. A 3D rotary motor is used to drive the ultrasonic probe to rotate within a preset angle range at a preset step angle. A processor that executes a computer program to implement the steps of the bladder ultrasound 3D reconstruction method for a portable device as described in any one of claims 1 to 7.
9. A portable ultrasonic device, characterized in that, Including the bladder ultrasound 3D reconstruction system adapted for portable devices as described in claim 8.