Minimally invasive sacral nerve modulation needle assisted positioning method and system based on three-dimensional reconstruction

Abstract

The present application relates to the field of medical devices and medical image processing technology, and particularly relates to a minimally invasive sacral nerve modulation acupuncture assisted positioning method and system based on three-dimensional reconstruction, which obtains a three-dimensional ultrasound image sequence, reconstructs a personalized sacrum navigation model containing a target nerve target point, establishes a precise spatial reference, dynamically registers a real-time body surface three-dimensional model, analyzes registration residuals and needle insertion parameters, dynamically compensates and corrects acupuncture depth, accurately calculates a second distance from a needle tip to the target point, and outputs positioning information. The present application realizes accurate mapping and real-time guidance of individualized anatomical structures, significantly improves the accuracy of minimally invasive sacral nerve modulation acupuncture assisted positioning results, and can be applied to sacral nerve modulation treatment.

Description

Technical Field

[0001] This invention relates to the fields of medical devices and medical image processing technology, specifically to a minimally invasive sacral nerve modulation acupuncture-assisted positioning method and system based on three-dimensional reconstruction. Background Technology

[0002] Sacral nerve modulation is a clinical therapy that uses physical stimulation of the sacral nerves to treat conditions such as urinary dysfunction. For example, this therapy is currently commonly used to treat refractory neurogenic bladder after pelvic surgery, a postoperative complication affecting urinary function. Its basic principle is to stimulate the target nerve, modulate its abnormal nerve reflexes, and thereby restore the normal function of target organs such as the bladder.

[0003] Currently, acupuncture is used clinically to modulate the sacral nerve. This involves percutaneously inserting acupuncture needles or specialized electrodes into the posterior sacral foramen and the vicinity of the target nerve to apply stimulation, thus treating urinary dysfunction. However, the following technical bottlenecks exist in practical application: First, the sacrum and the course of its internal nerves exhibit significant individual anatomical variations, making it difficult to achieve precise three-dimensional spatial mapping of deep nerve targets using traditional surface localization or two-dimensional image guidance. Second, acupuncture manipulation involves techniques such as insertion, lifting, and twisting, which can cause severe compression, displacement, and mechanical deformation of the skin and subcutaneous soft tissues.

[0004] Existing optical or depth camera tracking methods can only acquire static or superficial poses and cannot dynamically compensate for deep tissue deformation caused by the operation itself. This results in radial drift between the real-time positioning coordinates and the actual nerve target points in the patient's body, which seriously affects the accuracy of the needle insertion trajectory and the therapeutic effect of sacral nerve modulation. Summary of the Invention

[0005] To address the technical problem that existing acupuncture-assisted localization methods cannot overcome individual anatomical differences and dynamically compensate for tissue deformation induced by acupuncture, leading to inaccurate localization of deep sacral nerve targets, the present invention aims to provide a minimally invasive sacral nerve modulation acupuncture-assisted localization method and system based on three-dimensional reconstruction. The specific technical solution adopted is as follows: One embodiment of the present invention provides a minimally invasive sacral nerve modulation acupuncture-assisted localization method based on three-dimensional reconstruction, the method comprising the following steps: The patient's three-dimensional ultrasound transverse image sequence was acquired. Based on the image features, the bone and suspected nerve regions were distinguished. The spatial connectivity between adjacent transverse sections was combined to remove scanning artifacts and extract the distribution of target nerves. A three-dimensional model of the sacrum was reconstructed. The sacral 3D model is partitioned based on the bone morphology mutation features. The partitioned sacral 3D model is then registered with a preset standard reference model. Based on the spatial mapping relationship, the standard target points in the standard reference model are mapped to the sacral 3D model to obtain a sacral navigation model containing target nerve targets. The real-time three-dimensional model of the patient's body surface, the exposed length of the needle handles of several acupuncture needles, and the needle insertion posture parameters are obtained during acupuncture. The real-time three-dimensional model of the body surface is then registered with the sacral navigation model in real time to obtain the registration residual of the real-time registration. Calculate the first distance from the acupuncture point of each acupuncture needle to the corresponding target nerve point, and determine the current deviation index based on the temporal fluctuation of the real-time registration residual and the degree of abrupt change of the first distance; if the current deviation index meets the preset historical fluctuation approximation condition, then combine the change in the needle handle leakage length and the needle insertion posture parameters to correct and obtain the current acupuncture depth of each acupuncture needle. Based on the current acupuncture depth of each acupuncture needle, the second distance from the tip of each acupuncture needle to the corresponding target nerve point is determined, and the positioning information is output.

[0006] Another embodiment of the present invention provides a minimally invasive sacral nerve modulation acupuncture-assisted positioning system based on three-dimensional reconstruction, including a processor and a memory. The processor is used to process instructions stored in the memory to implement a minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction.

[0007] To address the positioning drift problem caused by deep tissue deformation in existing technologies, this invention achieves precise tracking of the needle tip position through multimodal data fusion and a dynamic compensation mechanism. Specific beneficial effects are as follows: This invention utilizes a three-dimensional ultrasound-reconstructed sacral model and combines it with zonal registration and standard target point mapping techniques to construct a navigation model containing the target nerve target, providing a precise spatial reference for subsequent localization. Through dynamic registration of the real-time surface model and the navigation model, combined with temporal analysis of registration residuals, localization deviations caused by tissue deformation are effectively identified. Furthermore, a depth compensation algorithm based on changes in needle handle protrusion length and needle insertion posture parameters corrects the acupuncture depth in real time, overcoming the technical shortcomings of traditional optical tracking that relies solely on static surface markers and cannot detect deep tissue displacement. Finally, by calculating the second distance from the needle tip to the target point and outputting visualized localization information, precise guidance of the needle tip's approach to the target nerve is achieved, significantly improving the accuracy and safety of sacral nerve modulation therapy. Attached Figure Description

[0008] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the steps of a minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to the present invention. Figure 2 This is an overview of the image processing algorithm of the minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction in this embodiment of the invention; Figure 3 This is a flowchart illustrating the implementation process of step S1 in an embodiment of the present invention. Figure 4 This is a schematic diagram of the three-dimensional section positioning and cross-sectional slice generation scheme in an embodiment of the present invention; Figure 5 This is a schematic diagram of the classification results of dual-threshold tissue regions in an embodiment of the present invention; Figure 6 This is a schematic diagram of the spatial connectivity analysis results of target neural region extraction in an embodiment of the present invention; Figure 7 This is a detailed six-view rendering of the sacrum STL 3D model in an embodiment of the present invention; Figure 8 This is a flowchart illustrating the implementation process of step S2 in an embodiment of the present invention. Figure 9 This is a schematic diagram of a model partitioning scheme based on morphological mutation features in an embodiment of the present invention; Figure 9 (a) is a sub-image of 3D feature points + partition cross-sections (12) in an embodiment of the present invention; Figure 9 (b) is a subplot of the morphological mutation density score (red > threshold 0.330) in an embodiment of the present invention; Figure 9 (c) is a subgraph of the model partitioning scheme (Gantt graph) in the embodiments of the present invention; Figure 10 This is a schematic diagram of the needle depth compensation mechanism in an embodiment of the present invention—a graph showing the relationship between registration residual, deviation index, and depth correction. Detailed Implementation

[0010] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the technical solution proposed according to the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0011] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0012] Furthermore, in all division and logarithmic operations involved in this invention, a protection mechanism is introduced to avoid computational crashes or invalid numerical values ​​caused by a zero denominator or a zero input value. This protection mechanism can be flexibly configured according to actual conditions. For example, when the denominator or the argument of the logarithmic function in a division operation is zero, a protection parameter with the same dimension as the denominator or the argument, or dimensionless, can be added. The value of this protection parameter can be set to a very small value greater than zero to ensure that the algorithm remains robust and feasible under extreme conditions. In addition, the normalization functions mentioned in this invention, unless otherwise specified, all employ the maximum-minimum normalization method to limit the normalization result to a minimum value. Within an interval or other continuous interval. The maximum and minimum values ​​used for maximum-minimum normalization can be obtained based on the actual application scenario. For example, when multiple values ​​can be obtained during implementation and it is necessary to compare the magnitudes of different values, the maximum and minimum values ​​can be determined by statistically analyzing multiple values; while when only a single value can be obtained during implementation or it is not necessary to compare the magnitudes of different values, the maximum and minimum values ​​can be statistically derived based on a large amount of historical experimental data or previously obtained prior data.

[0013] The application scenarios targeted by this invention can be: Because the target area for sacral nerve modulation using acupuncture is usually surrounded by the sacrum and adjacent to nerves and blood vessels, resulting in a complex anatomical structure, and because there are significant differences in individual skeletal morphology and nerve pathways among patients, traditional blind puncture or planar guided methods are prone to positioning errors. Therefore, this invention aims to provide a minimally invasive acupuncture-assisted positioning method based on three-dimensional reconstruction. This method first obtains the accurate three-dimensional distribution of the patient's bones and nerves through three-dimensional ultrasound scanning, and spatially registers it with a standard sacral model to accurately determine the coordinates of the patient's individualized acupuncture target area. Subsequently, in actual operation, the system uses visual capture to monitor and analyze in real time the soft tissue traction deformation caused by the patient's breathing or acupuncture techniques (such as lifting, thrusting, and twisting), and corrects the needle insertion depth according to the degree of dynamic deviation, thereby providing medical personnel with accurate and real-time dynamic auxiliary positioning information.

[0014] To address the aforementioned technical problem of poor accuracy in locating deep sacral nerve targets, specifically, one embodiment of the present invention provides a minimally invasive sacral nerve modulation acupuncture-assisted localization method based on three-dimensional reconstruction, such as... Figure 1 As shown, it includes the following steps: S1. Acquire the transverse image sequence of the patient's three-dimensional ultrasound, distinguish between bone and suspected nerve regions based on image features, and extract the target nerve distribution by removing scanning artifacts by combining the spatial connectivity between adjacent transverse sections, and reconstruct a three-dimensional model of the sacrum.

[0015] like Figure 2 As shown, the overall image processing algorithm of this invention includes, in sequence: edge detection (S101), feature intensity calculation (S102), tissue classification (S103), neural extraction (S104), 3D reconstruction (S105+S201), morphological mutation feature extraction (S201), and model partitioning (S202), ultimately outputting a sacral navigation model containing target neural points. This flowchart clearly illustrates the complete processing chain from the input of the original cross-sectional image to the output of the 3D navigation model.

[0016] The above steps are performed after the patient is fixed in a prone position and the body surface is marked. At this time, a personalized static "sacrum-nerve" standard 3D model of the patient is constructed by three-dimensional ultrasound scanning (including marker scanning). The model generated at this stage serves as the static geographical reference for the entire surgical process. In this embodiment, the patient analyzed has neurogenic bladder after pelvic surgery and is deemed by the doctor to require acupuncture treatment; furthermore, combined with the anatomical prior model, the pathway of the target nerve is deduced from the extracted bone edge morphology.

[0017] As an exemplary implementation, the process of step S1 described above is as follows: Figure 3 As shown, it includes: S101, perform edge detection on the cross-sectional images in the cross-sectional image sequence to obtain each candidate connected component.

[0018] As an exemplary implementation, the step of obtaining candidate connected components may include: The first step is to obtain a sequence of cross-sectional images.

[0019] like Figure 4As shown, during the 3D ultrasound scan, the probe moves at a constant speed along the Z-axis of the patient's sacrococcygeal region, acquiring a series of transverse slices. The figure shows the spatial distribution of the transverse slices on the 3D grid (a total of 8 equally spaced slices along the Z-axis, with an inter-slice spacing of approximately 34 mm, covering an anatomical range from Z=205 mm to Z=444 mm), as well as the 2D images (500×500 pixels) generated from each slice. Each transverse slice corresponds to a specific Z-axis spatial position, providing a precise spatial positioning reference for subsequent 3D reconstruction.

[0020] In this embodiment, under a specific prone position that restricts pelvic displacement and fully exposes the sacral region (e.g., by elevating the abdomen with abdominal support and keeping both lower limbs naturally extended to reduce muscle deformation interference), multimodal markers are placed on characteristic parts of the patient's body surface, and the multimodal markers are attached to the body surface positions corresponding to bony anatomical landmarks such as the posterior superior iliac spine. Subsequently, a three-dimensional ultrasound scan is performed on the patient's sacrococcygeal region to obtain anatomical structure image data of the sacrum and adjacent nerves, i.e., a sequence composed of cross-sectional images at various moments during a single scan, and the spatial position information of the markers in the current ultrasound image coordinate system is extracted simultaneously as a reference benchmark for subsequent spatial repositioning and registration.

[0021] The multimodal markers are visible in both ultrasound and binocular vision images. The number of multimodal markers can be greater than or equal to 3, and the implementer can set the number of multimodal markers according to the specific implementation situation.

[0022] The number of sampling frames and sampling frequency for cross-sectional images depend on the hardware parameters of the 3D ultrasound scanner and the probe translation speed. The specific number of frames is determined by the anatomical length of the scanning area and the preset slice spacing (e.g., 0.5 mm / frame). In specific sacral nerve modulation scenarios, a high sampling rate of 20 Hz to 40 Hz is recommended to ensure that more detailed nerve pathways can be captured during rapid translation or respiratory fluctuations.

[0023] Furthermore, to ensure the accuracy of subsequent 3D reconstruction based on image grayscale and edge features, the acquired original cross-sectional image sequence needs to be preprocessed. Specifically, the original cross-sectional image is first converted to grayscale, transforming it from an RGB three-channel image to a single-channel grayscale image. This removes interference from color measurement scales and UI characters exported by medical equipment, establishing a standardized grayscale data benchmark. Subsequently, Gaussian filtering is performed on the grayscale image to effectively smooth the image and suppress the speckle noise interference inherent in ultrasound imaging.

[0024] The second step is to perform edge detection on the cross-sectional images in the cross-sectional image sequence to obtain each candidate connected component.

[0025] In this embodiment, for each time step of the cross-sectional image sequence, the Canny edge detection algorithm is used to extract image edge features; the independent connected components formed in the edge detection results are marked as candidate connected components, which serve as the basic processing units for subsequent skeletal and neural region classification.

[0026] The specific implementation principle of the Canny edge detection algorithm is common knowledge in the field and will not be elaborated here. In other optional implementations, implementers can also use conventional edge detection algorithms such as the Sobel operator, Prewitt operator, or Roberts operator for equivalent replacement.

[0027] S102, for any candidate connected component, obtain the number of effective edge pixels, the average gray level, and the average gray level gradient magnitude of the candidate connected component, and perform data fusion processing to obtain the skeletal edge feature intensity of the candidate connected component.

[0028] Based on the differences in acoustic impedance between human tissues, bone tissue, due to its high density and significantly higher acoustic impedance than surrounding soft tissue, typically appears as bright with clear and continuous contours in ultrasound echo images. In contrast, neural tissue exhibits a smaller acoustic impedance difference compared to soft tissue, resulting in weaker echo signals and irregular textured edges. Utilizing these physical differences, effective separation of bone and neural regions can be achieved by extracting the grayscale, gradient, and edge geometric features of candidate connected regions.

[0029] Here, the intensity of the bone edge feature reflects the probability that the candidate connected region conforms to the boundary features of a high acoustic impedance object. The larger the value, the higher the probability that the candidate connected region is a bone edge; conversely, the candidate connected region is determined to be a suspected neural region or soft tissue background.

[0030] As an exemplary implementation, taking any candidate connected component as an example, the skeletal edge feature intensity of the candidate connected component is obtained, including: The first step is to obtain the number of valid edge pixels, the average gray level, and the average gray level gradient magnitude of the candidate connected components.

[0031] The first sub-step is to obtain the number of valid edge pixels of the candidate connected components.

[0032] Here, effective edge pixels are the points where the contour direction of the candidate connected region changes abruptly.

[0033] In this embodiment, the geometric feature values ​​of each region contour point in the candidate connected domain are extracted, specifically the 8-neighbor chain code values, and the first-order difference processing is performed on all the 8-neighbor chain code values ​​to obtain each difference value; the region contour points with non-zero difference values ​​are obtained as the effective edge pixels of the candidate connected domain, and the number of effective edge pixels corresponding to the candidate connected domain is counted.

[0034] Among them, the 8-neighbor chain code value can reflect the continuity and smoothness of the local direction of the edge. The first-order difference of the chain code sequence indicates that the edge direction has changed. The more contour points in the region where the difference value is not zero, the more broken the edge and the more irregular the shape.

[0035] The second sub-step involves first obtaining the grayscale values ​​of all pixels in the candidate connected region, and then calculating the average grayscale value of all pixels in the candidate connected region as the grayscale mean of the candidate connected region.

[0036] The third sub-step involves first obtaining the gray-level gradient magnitude of all contour points at the candidate connected region contour, and then calculating the average gray-level gradient magnitude of all contour points at the candidate connected region contour as the average gray-level gradient magnitude of the candidate connected region.

[0037] This process involves selecting a gradient operator, such as the Sobel operator, Prewitt operator, or Roberts operator, and then calculating the horizontal and vertical gradients of the contour points. Based on these horizontal and vertical gradients, the grayscale gradient magnitude of the contour points can be obtained. The implementation process of the gradient operator is existing technology and will not be elaborated here.

[0038] The second step is to determine the skeletal edge feature intensity of the candidate connected region based on the number of effective edge pixels, the average gray level, and the average gray level gradient magnitude of the candidate connected region.

[0039] Skeletal contours typically appear as continuous, smooth, arc-shaped, bright lines, while neural tissue edges are irregular and prone to artifacts. By calculating the first-order difference of the 8-neighborhood chain code, the number of effective edge pixels with non-zero difference values ​​characterizes the frequency of contour direction abrupt changes. The fewer the number, the smoother the contour, and the more it conforms to the geometric characteristics of bones; conversely, more of it indicates jagged edges caused by neural textures or artifacts. The average grayscale value of bones is significantly higher than that of soft tissues and nerves, allowing direct screening of high-echo cortical bone regions. The density abrupt change at the boundary between bones and soft tissues is dramatic; a larger average grayscale gradient amplitude indicates a sharper boundary, a unique characteristic of bones. In contrast, the boundary between nerves and soft tissues is blurred, with a significantly lower average gradient amplitude. Therefore, this embodiment integrates the average grayscale value, the average grayscale gradient amplitude, and the number of effective edge pixels to form a complementary relationship from three dimensions: brightness, contrast, and geometric shape, accurately determining the intensity of skeletal edge features used for screening skeletal regions.

[0040] As an example, the formula for calculating the bone edge feature intensity of the j-th candidate connected component can be: In the formula, represents the skeletal edge feature intensity of the j-th candidate connected component; norm represents the linear normalization function. If the maximum and minimum value normalization is used, the maximum and minimum values ​​in the maximum and minimum value normalization can be the maximum and minimum values ​​of the skeletal edge feature intensity of all candidate connected components. This represents the average gray value of the j-th candidate connected component. This represents the maximum mean gray level across all candidate connected components. This represents the average gray-level gradient magnitude of the j-th candidate connected component. This represents the mean of the maximum gray-level gradient magnitudes corresponding to all candidate connected components. This represents the total number of contour points of the j-th candidate connected region. This represents the number of valid edge pixels of the j-th candidate connected component. This represents a non-zero constant, used to avoid the possibility of the denominator of a fraction being zero. Its empirical value can be 0.01.

[0041] In the formula for calculating the intensity of bone edge features, and Used to eliminate and The dimensionless problem, and limit it to between 0 and 1; The reflection energy density of the candidate connected domains was characterized. The geometric smoothness factor characterizes the contour of the candidate connected region. The smoother the contour of the candidate connected region (i.e., the fewer effective edge pixels) and the brighter the echo (i.e., the higher the gray level and gradient), the greater the strength of the skeletal edge features.

[0042] S103, set the first and second feature intensity thresholds, and determine the skeletal region and suspected neural region in the cross-sectional image based on the comparison results of the bone edge feature intensity with the first and second feature intensity thresholds respectively.

[0043] like Figure 5 As shown, taking the 42nd frame cross-section as an example, the tissue classification results after dual-threshold grading are displayed. The left image shows the classification overlay effect, where the red area represents bone (bone edge feature intensity > 0.65), and the yellow area represents suspected nerves (bone edge feature intensity between 0.30 and 0.65); the right image shows the frame-by-frame classification statistics curve. There are a total of 0 bone regions and 6 suspected nerve regions in this frame, with a total of 119 suspected nerve regions extracted from the entire sequence, providing a candidate target set for subsequent spatial connectivity analysis.

[0044] In this embodiment, a first feature intensity threshold and a second feature intensity threshold are preset, wherein the first feature intensity threshold is greater than the second feature intensity threshold. As a preferred empirical reference value, the first feature intensity threshold can be set to 0.7, and the second feature intensity threshold can be set to 0.3. In specific implementation, for the bone edge feature intensity of any candidate connected region, the following hierarchical judgment is performed: 1) If the bone edge feature intensity is greater than the first feature intensity threshold (i.e., greater than 0.7), then the candidate connected region is determined to be a bone region; 2) If the bone edge feature intensity is between the second feature intensity threshold (excluding) and the first feature intensity threshold (including) (i.e., within) the threshold value, then the bone edge feature intensity is determined to be between the second feature intensity threshold (excluding) and the first feature intensity threshold (including) (i.e., within the threshold value). If the intensity of the bone edge feature is not greater than the second feature intensity threshold (i.e., less than or equal to 0.3), then the candidate connected region is marked as a suspected neural region; 3) If the intensity of the bone edge feature is not greater than the second feature intensity threshold (i.e., less than or equal to 0.3), then the candidate connected region is directly eliminated.

[0045] The threshold setting and grading logic mentioned above are determined based on the significant differences in acoustic characteristics of different tissues in ultrasound imaging. Specifically, bone tissue, due to its high density and high acoustic impedance, presents as a strong echo with a bright boundary during ultrasound imaging. Its bone edge feature intensity value approaches 1. Setting a first feature intensity threshold can accurately lock the bone reference benchmark and avoid misidentifying ordinary high-echo tissue as bone. The sacral nerve and perineum present as medium-intensity echoes, with contour regularity between bone and soft tissue. Its bone edge feature intensity value is usually in the middle range. However, some structural artifacts or local stray soft tissue echoes generated by ultrasound probe disturbances may occasionally show similar image features and fall into this range. Therefore, they are initially marked as suspected nerve regions and retained as candidate targets. Further identification is then performed through spatial connectivity to balance the integrity of target retention with the risk of misjudgment. When the bone edge feature intensity value is lower than the second feature intensity threshold, the corresponding region has weak echoes, blurred edges, and random shapes. These are mostly muscle soft tissue, fat layer, or ultrasound speckle noise artifacts. These regions are neither bone anchor points nor nerve target areas, and direct removal can effectively eliminate artifact interference.

[0046] S104. For each suspected neural region in the cross-sectional image, perform region matching in consecutive adjacent cross-sections and calculate the mean spatial connectivity between the matched regions; perform artifact removal based on the mean spatial connectivity to determine each target neural region in the cross-sectional image.

[0047] like Figure 6As shown, artifact removal and target neural extraction are achieved through cross-frame spatial connectivity analysis. The left figure shows the overlap area ratio matrix and connectivity distribution of suspected neural regions between adjacent frames, and the right figure shows the target neural regions after screening in the cross-section of frame 42. Summary statistics show a total of 119 suspected neural regions, 119 cross-frame matching sequences, a connectivity threshold of 0.45, and a minimum sequence length of 3 frames. True neural structures must maintain stable positional overlap and morphological continuity in consecutive frames. Based on this, isolated artifacts that do not meet the connectivity requirements are removed, retaining the target neural regions with anatomical continuity.

[0048] The real sacral nerve is an anatomically continuous cord-like entity extending through space. Therefore, in consecutive adjacent cross-sectional images acquired by a uniformly scanning ultrasound probe, the real nerve will inevitably exhibit a high degree of spatial connectivity across slices. Specifically, the cross-sections of the same nerve in adjacent slices generally have significant positional overlap, and due to the continuity of the nerve's course, the edge morphology (such as contour curvature) of its cross-sections shows a high degree of inheritance and similarity over short distances. Conversely, ultrasound artifacts or occasional local stray echoes are essentially random interferences in the acoustic imaging process. These artifacts are isolated and randomly generated, lacking the support of physical continuity, and therefore cannot form stable positional overlap and morphological continuity in consecutive adjacent slices.

[0049] As an exemplary implementation, determining each target neural region in a cross-sectional image includes: The first step is to perform region matching in consecutive adjacent cross sections for each suspected neural region in the cross section image and calculate the mean spatial connectivity between the matched regions.

[0050] This step quantifies spatial connectivity by extracting the overlap ratio and contour curvature difference of suspected regions between adjacent slices. If a suspected neural region can be stably matched in multiple consecutive frames and has a high mean connectivity, it indicates that it is a real extended physical entity (target nerve); conversely, if it disappears rapidly or undergoes drastic morphological changes in adjacent slices, it can be identified as a random artifact and removed, thereby achieving high-fidelity extraction of the target neural structure.

[0051] Each suspected neural region has a corresponding mean spatial connectivity. The steps for calculating the mean spatial connectivity include: The first sub-step involves temporally associating suspected neural regions that span multiple consecutive cross-sectional images and have overlapping locations, thereby constructing a suspected neural region matching sequence.

[0052] In this embodiment, the cross-sectional image sequence is traversed frame by frame according to the time sequence of ultrasound scans. For any suspected neural region in the cross-sectional image at time t, its contour coordinates are directly projected onto the cross-sectional image at time t+1, and the overlap area between each suspected neural region and the projected area in the cross-sectional image at time t+1 is calculated, i.e., the number of overlapping pixels. If there is a suspected neural region in the cross-sectional image at time t+1 such that the overlap area is greater than a preset overlap area threshold, such as half the area of ​​any suspected neural region in the cross-sectional image at time t, it is determined to be a continuation of the same physical structure on adjacent slices and is determined to be a pair of adjacent matching regions. Subsequently, the matching region determined at time t+1 is used as a new tracking starting point, and the same projection and overlap determination rules are used to continue tracking and matching in the cross-sectional image at time t+2. In this way, all suspected neural regions that are successfully matched in consecutive frames along the time axis are sequentially associated to form a complete suspected neural region matching sequence.

[0053] It should be noted that if the number of regions in a suspected neural region matching sequence is less than or equal to the minimum preset number, it indicates that the physical structure represented by the sequence lacks sufficient spatial extension. This clearly violates the anatomical morphology of the real sacral nerve as a continuous cord-like entity. It is highly likely due to transient artifacts caused by instantaneous ultrasound probe jitter, or stray echoes from isolated small soft tissues (such as microvascular cross-sections or fat lobules). In this case, the matching sequence is directly identified as an isolated artifact or invalid tissue and discarded, terminating the tracking and analysis. The minimum preset number can be set by combining the ultrasound sampling rate and the shortest anatomical length of the nerve; its empirical value can be 3 to 5.

[0054] Only when the number of regions matching the suspected neural region sequence is greater than the minimum preset number is the sequence considered to have the basic spatial span as a neural entity and retained for the next step of the specific analysis and calculation process of spatial connectivity between adjacent matching regions.

[0055] The second sub-step involves obtaining the ratio of the overlapping area of ​​each pair of adjacent matching regions and the curvature difference between the matching contour points for each pair of adjacent matching regions in the suspected neural region matching sequence.

[0056] Taking the first region at time t and the immediately adjacent second region at time t+1 in a suspected neural region matching sequence as an example, the first region and the second region are a pair of adjacent matching regions: First, count the total number of pixels contained in the first region (i.e., the area of ​​the first region) and the total number of pixels in the overlapping part of the first region and the second region in the same spatial coordinate system (i.e., the overlapping area). Divide the overlapping area by the area of ​​the first region to calculate the ratio of the overlapping areas of the pair of adjacent matching regions.

[0057] Next, for any reference contour point on the outer edge contour of the first region, calculate its Euclidean distance to all candidate points on the outer edge contour of the second region; select the candidate point with the shortest Euclidean distance as the matching contour point of the reference contour point in the second region, thus binding the two to form a matching contour point pair. Traverse all reference contour points in the first region to construct a set of matching contour point pairs for adjacent matching regions.

[0058] Finally, for each constructed matching contour point pair, the curvature of the reference contour point in the first region contour and the curvature of the matching contour point in the second region contour are extracted respectively, and the curvature difference between the reference contour point and the matching contour point in a matching contour point pair is calculated. The matching contour point pair set is traversed and the arithmetic mean of all curvature differences is calculated. This mean is used as the final curvature difference between the first region and the second region to represent the degree of distortion of the local morphology between the two slices.

[0059] The third sub-step involves calculating the spatial connectivity between each pair of adjacent matching regions within the suspected neural region matching sequence, based on the extracted overlap area ratio and curvature difference.

[0060] Here, the essence of spatial connectivity lies in evaluating the degree of smooth spatial continuity of physical entities between adjacent slices in ultrasound images. The actual sacral nerve, as a regular-shaped, gently curving continuous cord-like structure, exhibits dual stability in both position and morphology during slice analysis. Specifically: On the one hand, the stability of the position is positively correlated with the overlap area ratio. Within a continuous short time interval of uniform probe scanning, the projection position of the real neural entity on adjacent cross-sectional slices does not change abruptly, and there is a high proportion of physical overlap between the two slices. Therefore, the larger the overlap area ratio, the more stable the cross-section of the structure is in the longitudinal scanning direction, the higher the probability that it belongs to a real coherent physical entity, that is, the higher the spatial connectivity.

[0061] On the other hand, the stability of morphological inheritance is negatively correlated with the curvature difference. Real nerves, due to their tough outer sheath, exhibit highly consistent cross-sectional contours over short distances, with corresponding contour points in adjacent slices exhibiting extremely similar curvature. The curvature difference characterizes the degree of distortion in the local morphology; a smaller difference indicates a smoother contour transition. Conversely, a large curvature difference usually signifies drastic contour distortion, often caused by speckle noise or artifacts.

[0062] As an example, the formula for calculating the spatial connectivity between the nth pair of adjacent matching regions can be: In the formula, The norm represents the spatial connectivity between the nth pair of adjacent matching regions, and norm represents the linear normalization function. Let represent the ratio of the overlapping areas of the nth pair of adjacent matching regions, and exp represent an exponential function with the natural constant as the base. This represents the curvature difference between the nth pair of adjacent matching regions.

[0063] In the formula for calculating spatial connectivity, It utilizes the negative power property of the natural exponent to transform the curvature difference, which represents morphological distortion, into a range within... The attenuation weights between them. A pair of adjacent matching regions not only have high overlap ( It approaches 1), and the contour distortion is minimal ( Approaching 0, making When the product of the two terms approaches 1), it reaches its maximum, indicating that the structure has extremely strong physical coherence in space. The final norm can be a Min-Max normalization function. This normalization function uses the spatial connectivity calculation results of all matching region pairs within the current cross-sectional image sequence as a global sample set for maximum and minimum value normalization. It obtains the maximum and minimum values ​​used for normalization processing from the global sample set, which can smoothly stretch and map all calculated product results within the same sequence to... Standard intervals are then used to obtain the final standard spatial connectivity. Spatial connectivity is used in subsequent calculations of the mean spatial connectivity of the sequence and for comparison with the artifact threshold.

[0064] The fourth sub-step involves obtaining the spatial connectivity of all adjacent matching region pairs within the suspected neural region matching sequence, and calculating the average of all spatial connectivity as the overall average spatial connectivity of the suspected neural region matching sequence.

[0065] In this embodiment, for any suspected neural region matching sequence, the average value of all spatial connectivity corresponding to the suspected neural region matching sequence is calculated and used for subsequent artifact determination.

[0066] The second step involves artifact removal based on the mean of spatial connectivity to identify the target neural regions in the cross-sectional image.

[0067] In this embodiment, a connectivity threshold, such as 0.5, is set. Suspected neural region matching sequences with a spatial connectivity mean less than or equal to the connectivity threshold are identified as randomly generated speckle artifact sequences and removed from the candidate targets. Suspected neural region matching sequences with a spatial connectivity mean greater than the connectivity threshold are identified as real continuous physical entities, and all regions contained in the suspected neural region matching sequence are identified as successfully extracted target neural regions.

[0068] It should be noted that setting the connectivity threshold to the midpoint of the interval, i.e., 0.5, essentially establishes a physical entity confidence level. If the threshold is greater than the connectivity threshold, it indicates that the spatiotemporal continuity features of the structure corresponding to the suspected neural region matching sequence break the statistical distribution law of random noise and have a very high confidence level of being a real anatomical entity. If the threshold is less than or equal to the threshold, it indicates that the continuity features of the suspected neural region matching sequence have not significantly deviated from the random noise base, and therefore should be completely removed as an artifact. This can effectively ensure a high signal-to-noise ratio and high fidelity in the extraction of the target neural region.

[0069] S105, based on the bones and target nerve regions in the cross-sectional image sequence, a three-dimensional model of the sacrum is reconstructed.

[0070] As an exemplary implementation, reconstructing a three-dimensional model of the sacrum includes: The first step is to establish a three-dimensional spatial coordinate mapping: combining the ultrasound probe pose data recorded synchronously during the acquisition of the cross-sectional image sequence, and the reference coordinates of the multimodal markers placed on the patient's body surface, each frame of the cross-sectional image is transformed and mapped from a two-dimensional pixel coordinate system to a unified three-dimensional world coordinate system. Thus, the skeletal regions and target neural regions distributed on each frame of the image obtain their absolute three-dimensional spatial point sets in real physical space.

[0071] The second step is to perform categorized surface reconstruction: For the mapped 3D spatial point set, an isosurface extraction algorithm (such as the Marching Cubes algorithm) is used to reconstruct the 3D surface. Specifically, for all point sets marked as skeletal regions, a 3D skeletal sub-model representing the rigid cortex of the sacrum is reconstructed through surface smoothing and meshing; for all point sets marked as target neural regions, a 3D neural pipeline sub-model representing the neural pathways is reconstructed through curve fitting and pipelined surface generation.

[0072] The third step is to perform multimodal model fusion: the independently generated 3D skeletal sub-model and the 3D neural pipeline sub-model are spatially fused and meshed in the same coordinate system, and finally a personalized 3D sacral model containing skeletal topology and neural anatomical features is output.

[0073] like Figure 7As shown, the reconstructed 3D model of the sacrum has a smooth, continuous surface after Laplacian smoothing. The figure displays the detailed rendering effects from six perspectives: front, back, left, right, top, and bottom. The bottom right corner shows a comparison between the original mesh (with visible edges) and the smoothed model (with a smooth, continuous surface). The model contains a skeletal mesh (14,476 vertices, 29,026 faces) generated by the Marching Cubes algorithm, with dimensions ranging from 117×158×249mm, enabling dual-track visualization of skeletal topography and neural pathways, providing an intuitive anatomical reference for subsequent target mapping and surgical navigation.

[0074] Step S1 addresses the problem in existing technologies where significant individual differences in sacral morphology and nerve pathways lead to misalignment when blindly puncturing based solely on surface landmarks. This is addressed by constructing a personalized 3D model of the sacrum that integrates the real spatial topology of the skeleton and nerves, achieving transparent target localization. This model visualizes deep, previously invisible anatomical structures, ensuring that the acupuncture target area is determined based on the patient's actual internal physiological structure rather than general anatomical estimations. This effectively eliminates blind spots caused by individual physiological differences, achieving precise targeting. Regarding the risk of tissue injury or accidental entry into the sacral canal due to the dense blood vessels and nerves surrounding the sacral nerve modulation target area, this embodiment accurately marks the 3D spatial orientation of nerve pathways while constructing a high-fidelity sacral topography, forming a visible dual-track structure of skeleton and nerves. Based on this, the system can calculate the 3D spatial distance between the needle tip and dangerous structures in real time during minimally invasive blind puncture, assisting in planning the optimal safe needle insertion path, significantly reducing the incidence of intraoperative complications and improving surgical safety.

[0075] S2, the sacral 3D model is partitioned based on the bone morphology mutation features, the partitioned sacral 3D model is registered with the preset standard reference model, and the standard target points in the standard reference model are mapped to the sacral 3D model based on the spatial mapping relationship to obtain a sacral navigation model containing the target nerve target points.

[0076] The above steps begin when the acupuncture needle touches the skin (the moment of needle insertion). At this time, the binocular camera is activated, and the real-time captured surface deformation model is initially registered with the preoperative static model. By real-time monitoring of the needle handle protrusion length, needle body tilt angle, and needle insertion point coordinates, the real-time position of the needle tip in the body is calculated, thus obtaining a sacral navigation model containing the target nerve target.

[0077] At the anatomical level, although there are significant individual differences in the overall size and local curvature of the sacrum among different patients, as the base skeleton of the spine, the biomechanical connection formed by the fusion of multiple sacral vertebrae will inevitably retain specific morphological mutation features (such as obvious bony ridges, foramen edges, or concave-convex structures with dramatic curvature changes). These morphological mutation features correspond to the posterior sacral foramen margin structure in anatomy. These morphological mutation areas are geometrically characterized by extremely high local curvature and rich details, making them excellent anchor points for constructing spatial registration. If a rigid registration is directly applied to the entire 3D model of the sacrum, it is easy for macroscopic differences in the overall spinal curvature or pelvic tilt angle between individual patients to lead to serious spatial displacement errors in the registration of local key target areas.

[0078] Based on this, this embodiment uses local morphological mutation features as the segmentation plane to divide the sacral 3D model into multiple anatomical partitions with relatively consistent local features. Through this dimensionality reduction process, each independent local partition with morphological mutations is precisely registered with the corresponding partition of the standard reference model. This not only completely eliminates the influence of macroscopic body shape differences on matching accuracy but also greatly improves the matching overlap of key anatomical sites. Finally, using the partition space mapping matrix established after successful registration, the standard acupuncture target points marked by experts in the standard reference model are accurately projected back into the patient's individual sacral 3D model. This allows for the precise location of the actual neural target area coordinates suitable for the patient's own anatomical structure, thereby completing the construction of a high-precision sacral navigation model.

[0079] As an exemplary implementation, the process of step S2 described above is as follows: Figure 8 As shown, it includes: S201, Obtain the skeletal morphological mutation features in each cross-sectional image.

[0080] Within a single cross-sectional image, the skeletal contour lines in abrupt change regions are no longer gentle curves, but rather exhibit sharp angles or abrupt bends, indicating extremely high local curvature at these contour points. Greater curvature signifies a sharper and more unique local structure, resulting in higher distinguishability as a local feature point. In the smooth pelvic region, the skeletal cross-sectional shape remains almost unchanged between adjacent cross-sectional images (i.e., high inter-layer connectivity and smoothness). However, once the scanning probe passes through abrupt change locations such as vertebral interfaces or foramina, the skeletal morphology of adjacent cross-sections undergoes drastically different tomographic changes, indicating a sharp decrease in spatial connectivity. Based on these physical and geometric principles, accurately capturing the abrupt changes in sacral morphology requires a comprehensive consideration of the curvature of the contour points in the skeletal region and the spatial connectivity corresponding to the matching sequence of the skeletal region to which it belongs.

[0081] The first step is to extract the local curvature of the contour points of the sacral region in each cross-sectional image of the sacral 3D model, as well as the mean spatial connectivity of the matching sequence of the sacral region to which the sacral region belongs.

[0082] The first sub-step involves traversing the constructed 3D model of the sacrum. For any cross-sectional image of the model at any given moment, the edge contour point set of each bone region within it is extracted. For any contour point of a bone region, a calculation window is formed by selecting several neighboring pixels on the current cross-sectional contour line. A curvature estimation algorithm (e.g., based on polynomial fitting or discrete curvature calculation methods) is used to calculate the local curvature at the location of the contour point. The larger the local curvature value of a contour point of a bone region, the sharper or more severely curved the bone surface at that point.

[0083] Since the skeletal structure also exhibits physical continuity across slices during continuous ultrasound scanning, this embodiment uses the same temporal tracking logic for the skeletal region as for the target neural region, corresponding to the second sub-step below.

[0084] The second sub-step involves combining overlapping skeletal regions in the cross-sectional image along a temporal sequence to construct a skeletal region matching sequence. Subsequently, the same algorithm for calculating spatial connectivity between matching regions is invoked to calculate the spatial connectivity between each pair of adjacent matching skeletal regions within the skeletal region matching sequence. Finally, the arithmetic mean of the spatial connectivity values ​​of all adjacent skeletal region pairs within the skeletal matching sequence is calculated to obtain the mean skeletal spatial connectivity value, which characterizes the smoothness of the macroscopic interlayer morphology of the bone segment.

[0085] The third sub-step involves mapping and binding the local curvature of the obtained individual skeletal region contour point to the mean value of the interlayer spatial connectivity of the entire skeletal matching sequence to which the skeletal region belongs, in order to jointly evaluate whether there is an anatomical three-dimensional morphological mutation at the location of the skeletal region contour point.

[0086] The second step is to determine the deformation feature contribution of the bone region contour points based on the local curvature of the bone region contour points and the mean interlayer spatial connectivity of the entire bone matching sequence to which the bone region containing the bone region contour points belongs.

[0087] As an example, the formula for calculating the deformation feature contribution of the i-th skeletal region contour point can be: In the formula, This represents the contribution of the deformation features of the i-th skeletal region contour point. This represents the local curvature of the i-th skeletal region contour point. This represents the maximum curvature of all bone region contour points in the cross-sectional image where the i-th bone region contour point is located. It represents the mean interlayer spatial connectivity of the entire bone matching sequence to which the bone region containing the contour point of the i-th bone region belongs.

[0088] In the formula for calculating the contribution of deformation characteristics, This represents the relative curvature term, used to convert dimensional absolute curvature into a quantity. The dimensionless ratio between them reflects the curvature significance of the i-th skeletal region contour point within the current cross-sectional image. Generally, There is no case where the value is zero. If there is an extreme case, the maximum curvature value is selected from all cross-sectional images as the denominator of the relative curvature term. The value of is also between, Indicates the reverse connectivity. The smaller the value, the worse the connectivity between the front and back skeletal regions, and the greater its contribution to deformation feature analysis.

[0089] The third step is to obtain the deformation feature contribution of each skeletal region contour point, compare it with the preset contribution threshold to obtain local feature points, and use the local feature points as skeletal morphological mutation features.

[0090] In this embodiment, the contribution threshold can be set to 0.5 through empirical determination, and the implementer can set it according to the specific situation. For each skeletal region contour point, if the contribution of the deformation feature of a certain skeletal region contour point is greater than the contribution threshold, then the skeletal region contour point is identified as a local feature point that can characterize the anatomical mutation features, and it is highlighted in the three-dimensional coordinate system; conversely, if the contribution of the deformation feature of a skeletal region contour point is not greater than the contribution threshold, then it is determined as a contour point of a normal smooth skeletal region. Finally, all cross-sectional images in the entire sacral three-dimensional model are traversed, and all successfully marked local feature points are summarized to form a three-dimensional feature point cloud that characterizes the topological relationship of sacral morphological mutations in this patient, which serves as the key anatomical basis (i.e., skeletal morphological mutation features) for subsequent model partitioning.

[0091] S202, the sacral 3D model is divided into partitions based on the skeletal morphological mutation characteristics to obtain the partitioned sacral 3D model.

[0092] like Figure 9 As shown, model partitioning is completed based on skeletal morphological mutation features. Figure 9 middle, Figure 9 (a) Shows the spatial distribution of the three-dimensional feature point cloud and 12 partition sections (the sections with a morphological mutation density score >0.330 are highlighted in red). Figure 9 (b) The morphological mutation density scoring curves for each cross section are shown; Figure 9(c) The final partitioning scheme (13 partitions) is presented in Gantt chart format. The partitioning report shows: a total of 60 frames, 1737 feature points, 12 mutation surfaces, 13 partitions, and a quantile threshold of 80%. Using these high-density mutation surfaces as the segmentation benchmark, the patient's personalized sacral 3D model is divided into multiple anatomical partitions with relatively consistent local features, providing a spatial benchmark for subsequent fine-grained registration with the standard reference model.

[0093] The first step is to statistically analyze the density distribution of local feature points in each cross-sectional image.

[0094] In this embodiment, the entire 3D model of the sacrum containing a cluster of 3D feature points is traversed. Taking the k-th cross-sectional image as an example, the total number of local feature points marked in the k-th cross-sectional image is counted and recorded as the number of feature points. The total number of contour points of all skeletal regions in the k-th cross-sectional image is also counted and recorded as the base number of skeletal contour points. The ratio of the number of feature points in the k-th cross-sectional image to the base number of skeletal contour points is used as the basic ratio characterizing the density of local feature points. To assign appropriate weight to the basic ratio, it is multiplied by the average contribution of all local feature points in the cross-sectional image to calculate the morphological mutation density score of the k-th cross-sectional image, that is, the density distribution of local feature points in the k-th cross-sectional image.

[0095] The second step is to extract cross-sectional images of feature point clusters based on density distribution and mark them as morphological abrupt change surfaces.

[0096] In this embodiment, after obtaining the morphological mutation density scores corresponding to all cross-sectional images, all scores are arranged in ascending (or descending) order according to their numerical values. Through statistical quantile analysis, cross-sectional images whose scores rank in the top quantile (e.g., the top 20% of those with quantiles greater than 0.8) are extracted. These images represent the anatomical layers in the entire sacral 3D model where local feature points are highly concentrated and curvature distortion is most significant. These extracted high-quantile cross-sections are explicitly marked as morphological mutation surfaces.

[0097] The third step is to segment the sacral 3D model along the morphological change plane to obtain the partitioned sacral 3D model.

[0098] In this embodiment, the Z-axis coordinates of each morphological mutation surface in the three-dimensional coordinate system are used as the reference plane. The three-dimensional mesh Boolean operation or model cutting algorithm is called to cut and segment the complete patient-personalized sacral three-dimensional model solid mesh along these parallel reference planes to obtain a set of independent sub-blocks containing complete local anatomical features. These sub-blocks together constitute the partitioned sacral three-dimensional model for subsequent fine local registration.

[0099] S203, the partitioned 3D model of the sacrum is registered with the preset standard reference model, and the standard target points in the standard reference model are mapped to the 3D model of the sacrum based on the spatial mapping relationship to obtain a sacral navigation model containing the target nerve target points.

[0100] As an exemplary implementation, obtaining a sacral navigation model containing target neural targets includes: The first step involves pre-constructing a standard reference model of the sacrum. This model can be obtained by averaging 3D MRI or CT data of the sacrum from multiple healthy individuals of the same age and sex. For this standard reference model, clinical experts pre-determine the ideal needle insertion point coordinates in the model's 3D space as the standard target point. Alternatively, the standard target point can be determined based on the geometric center of the posterior sacral foramen margin. Simultaneously, using the same morphological change surface determination rules as the patient model or a pre-set anatomical segmentation ratio, the standard reference model is divided into multiple standard partitions corresponding to the patient model.

[0101] The second step involves extracting all local feature point clouds from the surface of any independent sub-block in the 3D model of the sacrum after patient partitioning, using this as the set of points to be registered. Subsequently, the feature point cloud of the corresponding standard sub-block is used as the target point set. These two sets of point sets are then input into the 3D point cloud registration algorithm model. A coarse registration and noise reduction algorithm, such as RANSAC (Random Sample Consensus), is preferably used, followed by fine registration and fine-tuning using the ICP (Iterative Closest Point) algorithm. After multiple iterations, the 3D point cloud registration algorithm model calculates an optimal 3D spatial transformation matrix that maximizes the spatial overlap between the feature point groups of the patient sub-block and the standard sub-block, including translation, rotation, and scaling parameters.

[0102] The third step involves establishing a local spatial mapping relationship between the patient's personalized model and the standard reference model after completing the registration of all corresponding sub-blocks. This is achieved using the optimal 3D spatial transformation matrix of each independent sub-block output by the aforementioned algorithm. Based on the obtained optimal 3D spatial transformation matrix and its inverse matrix, the spatial coordinates of the preset standard target points within each sub-block of the standard reference model are inversely mapped and transformed to the original coordinate system of the sub-block of the patient's corresponding sacral 3D model. After this mapping and transformation, the patient's personalized sacral 3D model generates target nerve targets that conform to the patient's actual anatomical characteristics, i.e., the optimal puncture and control area specific to the patient. The final output is a sacral navigation model that can be used to guide clinical surgical pathways.

[0103] Step S2 above, through local feature registration and inverse mapping of standard target points, translates and anchors the vast amount of general anatomical experience (standard target points) determined by clinical experts into the patient's unique physiological structural coordinate system with high precision. This ensures that the final determined neural target point is no longer an estimated, vague area, but a millimeter-precise puncture target point with absolute spatial coordinates, customized according to the patient's individual skeletal morphology, fundamentally eliminating the blind spots caused by individual differences. Based on the sacral navigation model, doctors can plan a perfect needle insertion trajectory before surgery, avoiding the nerve trunk and directly striking the target point, greatly improving the error tolerance and operational safety of minimally invasive blind puncture, and effectively avoiding the risk of nerve mechanical damage.

[0104] S3: Acquire the real-time three-dimensional model of the patient's body surface, the exposed length of the needle handles of several acupuncture needles, and the needle insertion posture parameters during acupuncture. Then, perform real-time positioning and registration between the real-time three-dimensional model of the body surface and the sacral navigation model to obtain the registration residual of the real-time positioning and registration.

[0105] In minimally invasive sacral nerve modulation surgery, the challenge of precise positioning lies in the disconnect and dynamic drift between the absolute space inside the body and the relative operating space outside. Although step S2 above has constructed a high-precision sacral navigation model inside the body, the surgeon's needle insertion operation depends on the real-time physical space outside the patient's body. Therefore, it is necessary to use a binocular camera to capture a three-dimensional model of the body surface containing multimodal markers in real time. Through the coordinate transformation matrix of the markers, the external operating space and the internal navigation space are initially registered in real time. Combined with the needle handle protrusion length extracted by binocular vision and the needle insertion posture parameters (such as tilt angle) obtained by IMU, the real-time coordinates of the needle tip hidden under the skin are calculated using spatial geometry.

[0106] However, the non-rigid deformation of the human body is the biggest source of interference disrupting this mapping between reality and illusion. During surgery, when the doctor applies needling techniques such as lifting, thrusting, and twisting the acupuncture needles, or when the patient's chest and abdomen rise and fall due to breathing, the skin and subcutaneous soft tissue around the needle insertion point undergo significant elastic traction and compression. This causes the body surface depth captured by the binocular camera to include false displacements caused by skin depressions or bulges. At the registration algorithm level, this geometric mismatch caused by the non-rigid traction of soft tissue is precisely quantified as a sudden increase in the fluctuation of the registration residual.

[0107] To eliminate spurious displacements caused by skin traction and restore the true depth of the needle tip within the body, a dynamic correction mechanism based on historical steady-state conditions is introduced. The core principle is that when the registration residual and its temporal fluctuations exhibit significant abnormal interference, the system no longer trusts the surface depth reading contaminated by skin traction at the current moment. Instead, it backtracks to historical timelines, extracting the physical depth measured at the most recent steady-state skin condition, and uses this as a reliable absolute benchmark. Based on this absolute benchmark, relying on core physical quantities that do not change with skin deformation—namely, the change in needle handle length measured by the binocular camera and the needle tilt angle monitored in real-time by the IMU—geometric algebraic compensation is performed using trigonometric functions. This effectively shields against the interference of elastic deformation in soft tissue, ensuring the absolute puncture depth of the needle tip in the output sacral navigation coordinate system.

[0108] As an exemplary implementation, step S3 above can be achieved through the following steps: The first step is to obtain a real-time three-dimensional model of the patient's body surface during acupuncture, the exposed length of the needle handles of several acupuncture needles, and the needle insertion posture parameters.

[0109] In this embodiment, the pressure of the ultrasound probe is first removed, and the soft tissue is allowed to rebound to a natural stress-free state after a preset time. The coordinates of the multimodal markers at this time are used as the initial positioning and registration reference for binocular vision. Then, a binocular camera is used to acquire real-time images of the patient's sacrum and coccyx during acupuncture by attaching multiple multimodal markers. Depth information is calculated using a stereo matching algorithm to generate a real-time point cloud sequence, which is then reconstructed into a real-time three-dimensional model of the body surface through triangulation. Computer vision algorithms are used to identify the needle handle end and the skin insertion point of each acupuncture needle in the image. The pixel distance from the end to the insertion point is calculated using a preset physical scale of the needle body as a reference scale, and the physical length is obtained. The quaternion or Euler angle output by the needle handle integrated IMU is read in real time to obtain the tilt angle of the needle body relative to the gravity vector, or the spatial direction vector of the needle body axis in the binocular image is extracted, and the angle between this vector and the local normal vector of the skin insertion point is calculated as the needle insertion posture parameter.

[0110] Among these, the stereo matching algorithm can employ the SGBM (Semi-Global Block Matching) algorithm, which aggregates matching costs through dynamic programming in multiple directions. This effectively reduces noise and incorrect matching in the disparity map while maintaining high computational efficiency, making it particularly suitable for handling targets with continuous curved surfaces but weak textures, such as human skin. Point cloud processing and mesh reconstruction can also utilize the Poisson Surface Reconstruction algorithm. The Poisson reconstruction algorithm can generate a smooth triangular mesh that better approximates the real surface by solving the Poisson equation based on the normal vector information of the point cloud. For needle handle and insertion point identification, template matching or feature point detection (such as the ORB algorithm) can be used. During the identification process, since the needle handle usually has high contrast or a specific shape, image templates of the needle handle end and the vicinity of the insertion point can be pre-stored. Pixel distance conversion can utilize the camera pinhole imaging model and the principle of similar triangles.

[0111] The second step is to perform real-time positioning and registration between the real-time 3D model of the body surface and the sacral navigation model to obtain the registration residuals of the real-time positioning and registration.

[0112] Binocular cameras can only capture the external body surface and needle body; the internal target area (bones, nerves) exists only in the preoperative ultrasound model. Registration maps the real-time external acupuncture coordinates to the internal ultrasound coordinate system, calculating the distance from the needle tip to the target area. Furthermore, needle insertion, lifting, twisting, or patient breathing can cause deformation of the skin and subcutaneous tissue. When deformation occurs, the real-time body surface model no longer perfectly matches the preoperative static model; the magnitude of the registration residual directly reflects the severity of the current soft tissue deformation. A sudden increase in the registration residual indicates significant skin traction deformation, leading to optical positioning distortion. This serves as a trigger condition, and subsequent insertion depth is corrected using a compensation algorithm based on the needle handle protrusion length and IMU tilt angle. Obtaining the registration residual for real-time positioning registration specifically includes: The first sub-step involves extracting the first spatial coordinates of the multimodal markers in the sacral navigation model and the second spatial coordinates in the real-time three-dimensional body surface model.

[0113] In this embodiment, intensity threshold segmentation is performed on the sacral navigation model to extract the echo regions of markers with strong reflectivity. A three-dimensional connected component analysis algorithm is used to separate the independent voxel clusters of each marker, and the spatial geometric centroid of each voxel cluster is calculated. The coordinates of this centroid are used as the first spatial coordinates of the corresponding marker in the sacral navigation model (ultrasound coordinate system). In the real-time three-dimensional body surface model, the two-dimensional center pixel coordinates of each marker are identified based on color space threshold or contour feature extraction algorithms. Combined with the depth map generated by binocular stereo matching, the two-dimensional center pixel coordinates are inversely projected onto the camera's three-dimensional space based on camera intrinsic parameters. Local point clouds on the marker surface are extracted, and the three-dimensional geometric centroid of these local point clouds is calculated. This local point cloud is used as the second spatial coordinates of the corresponding marker in the real-time three-dimensional body surface model (optical coordinate system).

[0114] The second sub-step involves real-time positioning and registration of the real-time body surface 3D model and the sacral navigation model based on feature alignment of the first and second spatial coordinates.

[0115] In this embodiment, the Singular Value Decomposition (SVD) algorithm is used to solve for the optimal rigid transformation matrix from the real-time 3D body surface model to the sacral navigation model based on the first and second spatial coordinates of all multimodal markers. This matrix includes rotation and translation vectors. Subsequently, the optimal rigid transformation matrix is ​​applied to all point cloud data of the real-time 3D body surface model, transforming the real-time 3D body surface model to the coordinate system of the sacral navigation model, thus completing real-time positioning and registration. Furthermore, to accommodate possible slight patient movements during surgery, the above feature alignment and transformation updates are repeated at a frequency of at least 30Hz to continuously maintain spatial consistency between the real-time 3D body surface model and the sacral navigation model, providing accurate reference for subsequent surgical navigation.

[0116] The third sub-step is to obtain the registration residual of real-time positioning and registration.

[0117] In this embodiment, all the second spatial coordinates are mapped to the ultrasonic coordinate system through the above rigid transformation matrix. The Euclidean distance between the mapped spatial coordinates and the corresponding first spatial coordinates is calculated. The mean or root mean square error (RMSE) of the Euclidean distances corresponding to all marker points is obtained and used as the registration residual for real-time positioning and registration.

[0118] It is worth noting that if the calculated root mean square error exceeds the preset residual threshold (e.g., 2mm), a prompt message can be issued to remind the operator to check whether the surface markers have shifted or whether the sensor has been interfered with, and the feature alignment step can be re-executed to ensure the accuracy of real-time positioning and registration.

[0119] Step S3 above establishes a mapping relationship between the external visual space and the internal anatomical space by acquiring a real-time surface model and registering it with the sacral navigation model, enabling the externally visible acupuncture needles to be oriented to the internal target area. Obtaining the registration residual essentially constructs a soft tissue deformation monitoring mechanism, as the magnitude of the residual directly quantifies the degree of skin and subcutaneous tissue compression and displacement caused by acupuncture techniques (insertion, lifting, etc.). Acquiring the needle handle protrusion length and needle insertion posture parameters extracts an absolute physical rigidity reference quantity unaffected by soft tissue deformation, providing a calculation basis for subsequently eliminating visual errors and correcting needle tip depth in cases of severe deformation. Step S3 overcomes the shortcomings of traditional optical positioning, which is easily interfered with by skin traction deformation. By accurately identifying the moment of deformation that causes positioning distortion through the registration residual and switching to depth compensation using rigid parameters, positioning errors caused by false displacement can be fundamentally eliminated, significantly improving the accuracy and safety of acupuncture targeting the sacral nerve and avoiding the risk of accidental damage to blood vessels or nerves.

[0120] S4. Calculate the first distance from the acupuncture point of each acupuncture needle to the corresponding target nerve target point. Determine the current deviation index based on the temporal fluctuation of the real-time registration residual and the degree of abrupt change in the first distance. If the current deviation index meets the preset historical fluctuation approximation condition, then combine the change in the needle handle leakage length and the needle insertion posture parameters to correct and obtain the current acupuncture depth of each acupuncture needle.

[0121] The registration residual characterizes the degree of soft tissue deformation, and the first distance characterizes the needle insertion progress from the optical perspective. If the registration residual suddenly increases and the first distance also changes abruptly, it indicates that the increase in depth observed optically is actually a pseudo-displacement caused by skin pressure depression. The deviation index is used to quantify the probability of this pseudo-displacement. Matching historical fluctuation approximation conditions is used to confirm that the current deformation belongs to normal interference caused by periodic acupuncture techniques (such as regular lifting and thrusting), excluding accidental camera slippage or calculation errors. After confirming the deformation interference, since the needle body is a rigid body, the needle handle protrusion length and needle insertion angle are not affected by the sinking of the skin origin. Using this as an absolute geometric benchmark, a compensation model is constructed, which can isolate the epidermal deformation error and restore the true needle tip position.

[0122] As an exemplary implementation, step S4 above can be achieved through the following steps: The first step is to calculate the first distance from the acupuncture point of each acupuncture needle to the corresponding target nerve point.

[0123] In this embodiment, the puncture point of the acupuncture needle on the skin surface is identified in real time by a binocular camera, and the three-dimensional coordinates of the puncture point in the real-time three-dimensional model coordinate system of the body surface are obtained. The rigid transformation matrix obtained by real-time positioning and registration is used to map the coordinates of the puncture point to the ultrasound coordinate system of the sacral navigation model. The three-dimensional coordinates of the centroid of the corresponding target nerve target point are obtained in the sacral navigation model, and the Euclidean distance between the mapped three-dimensional coordinates of the puncture point and the three-dimensional coordinates of the centroid is calculated and used as the first distance of the acupuncture needle.

[0124] The second step is to determine the current deviation index based on the temporal fluctuations of the real-time registration residuals and the degree of abrupt change in the first distance.

[0125] The first sub-step involves obtaining the current registration residual for any acupuncture needle at the current moment, and extracting the historical registration residual from the previous moment.

[0126] The second sub-step is to calculate the distance difference between the first distance at the current moment and the historical distance sequence, and then filter the target moment based on the distance difference.

[0127] Here, the distance difference value of the historical distance sequence is essentially achieved by subtracting the first distance of the previous time from the first distance of the current time. Both of these time points belong to the historical distance sequence that is continuously cached in the time dimension.

[0128] In this embodiment, the first distance of the acupuncture needle at the current moment and the first distance of the adjacent previous moment are extracted; the difference between the first distance at the current moment and the first distance at the previous moment is calculated as the distance difference value; in order to filter out sub-pixel level measurement noise and system interference, it is determined whether the absolute value of the distance difference value is greater than the preset static noise filtering threshold. If it is greater, the current moment is marked as the target moment.

[0129] The static noise filtering threshold ranges from 0.1 mm to 1.0 mm. Preferably, when using a high-precision depth camera, the threshold is set to 0.2 mm. When the absolute value of the distance difference is less than or equal to the static noise filtering threshold, the current distance difference can be corrected to 0, thereby creating a dead zone in the time domain. This effectively filters out false displacement signals caused by image sensor noise or sub-pixel positioning jitter, ensuring the accuracy of needle depth calculation.

[0130] The third sub-step involves obtaining the current deviation index for the target time based on the current registration residual, historical registration residual, and the absolute value of the distance difference.

[0131] In acupuncture navigation, whether it's the doctor's hand tremor or the patient's body movement, the greater the actual physical displacement of the needle, the greater the possibility that the needle tip will deviate from the preset safe path. The absolute value of the distance difference directly reflects the displacement amplitude of the acupuncture needle per unit time. The current registration residual reflects the alignment error between the real-time body surface model and the sacral navigation model at the current moment. The larger the residual, the less reliable the current coordinate positioning. The historical registration residual represents the positioning accuracy of the system at the previous moment and is usually used as a benchmark or stability reference. The smaller the historical residual, the lower the original noise floor and the more stable the benchmark. When the benchmark is very stable, any increase in the current residual or needle displacement will appear more significant and abnormal.

[0132] As an example, the formula for calculating the current deviation index of the acupuncture needle can be: In the formula, This indicates the current deviation index of the acupuncture needle at the target time. This represents the current registration residual of the acupuncture needle at the target time. This represents the historical registration residual of the acupuncture needle at the previous time step before the target time. This represents the distance difference of the acupuncture needle at the target time. This represents a non-zero constant, used to avoid the possibility of the denominator of a fraction being zero. An empirical value for this constant can be 0.01. This represents the function for finding the absolute value.

[0133] In the current formula for calculating the deviation index, It is the physical basis of the deviation index, which can directly reflect the magnitude of the displacement of the acupuncture needle between adjacent moments. The larger the value, the more vigorous the needle movement, and the higher the potential risk of deviation. This represents the relative rate of change of the registration residual, which is an indicator of the alignment accuracy between the real-time body surface model and the navigation model. A larger residual indicates less reliable spatial positioning. When this occurs, it indicates a decrease in registration quality and an increase in positioning uncertainty. In this case, the sensitivity to motion should be increased; that is, for the same displacement, the deviation index should be higher. When the system is stable, the registration quality is improved and more reliable, and the response to small displacements can be appropriately suppressed. When the system registration is unstable, even if the needle displacement is small, false deviations may occur due to positioning drift. When the system is stable, the deviation index is mainly driven by the real displacement.

[0134] The third step is to correct the current acupuncture depth of each acupuncture needle by combining the change in the needle handle protrusion length and the needle insertion posture parameters if the current deviation index meets the preset historical fluctuation approximation conditions.

[0135] During acupuncture, the techniques used by practitioners, such as lifting, thrusting, and twisting, typically exhibit a regular cycle. This cyclical nature can lead to similar skin deformation and positioning errors. Triangular projection compensation based on the needle handle protrusion length and insertion posture parameters can serve as an engineering approximation model for real-time surgical navigation, providing a baseline depth correction during periods of severe soft tissue deformation. Here, the preset historical fluctuation approximation condition refers to the positioning fluctuation characteristics exhibited during the current continuous displacement time period, which shares multi-dimensional similarity with the fluctuation characteristics of historical displacement time periods already occurring in this acupuncture procedure. Specifically, the current acupuncture depth of each acupuncture needle is obtained, including: The first sub-step is to merge consecutive target moments into a target time period, and obtain the target time period containing the current moment as the current target time period.

[0136] In this embodiment, detection begins at the start of acupuncture. When the first target time is detected, it is marked as the start point of the time period. When N consecutive non-target times are detected, the previous target time is marked as the end point of the time period. All consecutive times within the start and end intervals are merged into a complete target time period. The target time period containing the latest sampling time is extracted as the current target time period for subsequent similarity comparison. Here, N is a positive integer, which can be taken as an empirical value of 3.

[0137] It should be noted that if the current time is not within any target time period, the estimated needle penetration depth observed by the binocular camera at the current time will be directly output without historical similarity judgment or formula correction.

[0138] The second sub-step involves calculating the differences between the current target time period and the reference historical time period in terms of deviation index, duration, and distance difference sequence, to obtain the approximation of historical fluctuations.

[0139] In this embodiment, all target time period data completed in the acupuncture task are cached in real time. Whenever a target time period ends (i.e., the needle insertion action transitions to a static state), the target time period and its corresponding first-order distance difference sequence, deviation exponent sequence, and duration data are stored in the historical behavior database. When performing similarity analysis on the current target time period, all stored time period data are retrieved from this database as reference historical time periods. If the current stage is the initial stage of the task and there are no stored historical time periods, the system skips the similarity judgment logic by default.

[0140] As an example, the formula for calculating the approximation of historical fluctuations can be: In the formula, This indicates the similarity between the current target time period and the historical fluctuations of the reference historical time period, where exp represents an exponential function with the natural constant as the base. This represents the difference in deviation index between the current target time period and the x-th reference historical time period. This represents the difference in duration between the current target time period and the x-th reference historical time period. This represents the difference between the current target time period and the x-th reference historical time period in the distance difference sequence. This represents the function for finding the absolute value. and All of these are preset weighting coefficients, which can be empirically set to 0.3, 0.3, and 0.4. Implementers can set these coefficients according to the importance of the differences in each dimension, and ensure that all weighting coefficients add up to 1. This represents the largest historical deviation from the index. This represents the historical maximum difference in duration. This represents the historical maximum difference value of the distance difference sequence. It is used to eliminate the influence of dimensions. Generally, it is not zero. If there is an extreme case, the maximum difference value of the deviation index, which is preset by the user, is added to the denominator of the fraction.

[0141] In the formula for calculating the approximation of historical fluctuations, It can represent the absolute value of the difference between the current mean deviation index of the current target time period and the mean deviation index of the xth reference target time period, showing the difference in the magnitude of deviation; It can represent the absolute value of the difference between the duration of the current target time period and the duration of the xth reference historical time period, showing the difference in time scale; The distance values ​​between the distance difference sequence within the current target time period calculated using the Dynamic Time Warping (DTW) algorithm and the distance difference sequence of the xth reference historical time period exhibit differences in morphological trends.

[0142] The third sub-step involves extracting the baseline acupuncture depth of each acupuncture needle when the historical fluctuation approximation is greater than the preset fluctuation threshold, for each acupuncture needle.

[0143] In this embodiment, the fluctuation threshold is set to 0.5. When the historical fluctuation approximation is greater than the preset fluctuation threshold, it indicates that the acupuncture needle depth at the current moment is affected by morphological fluctuation interference. The system backtracks forward in the cached time-series coordinate database and extracts the moment with the most recent distance difference value of zero (i.e., in a static and stable state). The first distance corresponding to this non-target moment is used as the reference acupuncture depth.

[0144] It should be noted that the non-target moment refers to the stable state of the skin without compression or displacement. Since acupuncture may cause elastic deformation of the skin, the surface depth directly identified by the binocular camera at this time may include skin displacement and is a false depth. Therefore, it is necessary to extract a stable depth before deformation as the initial benchmark, and then perform physical compensation by the change in the length of the needle handle and the depth projection coefficient to restore the true needle tip depth and eliminate the measurement error caused by skin deformation.

[0145] It should also be noted that 0.5 represents a moderate level of fluctuation. When the historical fluctuation approximation is greater than 0.5, it means that the current fluctuation characteristics deviate from the historical stable characteristics by more than 50%, which is highly likely due to a substantial change in the morphology of the target area (such as skin stretching or tissue displacement). In this case, the currently disturbed data must be discarded, and a historical stable benchmark must be sought instead. Setting the preset fluctuation threshold to 0.5 aims to ensure that when significant morphological fluctuation interference is detected, the backtracking mechanism can be reliably triggered to extract the first distance in the static stable state as the benchmark acupuncture depth, thereby ensuring the accuracy of subsequent depth calculations and the safety of navigation.

[0146] The fourth sub-step involves determining the depth projection coefficient based on the angle of the acupuncture needle, and then performing data fusion processing on the change in the length of the needle handle protruding from the acupuncture needle and the depth projection coefficient to obtain the depth compensation amount of the acupuncture needle.

[0147] The fifth sub-step involves using depth compensation to correct the baseline acupuncture depth corresponding to the acupuncture needle, thereby obtaining the current acupuncture depth of the acupuncture needle.

[0148] As an example, the formula for calculating the current acupuncture depth of an acupuncture needle can be: In the formula, This indicates the current acupuncture depth of the acupuncture needle. This indicates the reference acupuncture depth corresponding to the acupuncture needle. This indicates the length of the needle handle protruding from the body at the current moment. This represents the length of the needle handle protruding from the acupuncture needle corresponding to the nearest non-target time. This indicates the angle between the acupuncture needles at the current moment. This represents the depth projection coefficient of the acupuncture needle at the current moment. This indicates the decrease in the length of the needle handle protruding from the current time relative to the reference time (i.e., the non-target time closest to the current time). This indicates the depth compensation amount of the acupuncture needle at the current moment.

[0149] like Figure 10As shown, when soft tissue deformation causes a sudden increase in registration residuals, the system automatically activates the depth compensation mechanism. The figure simultaneously displays a comparison between the measured depth and the true depth (including optically measured depth and compensated depth), the registration residual fluctuation curve, the deviation index (the red area represents compensation activation), the needle handle protrusion length, the needle insertion angle, and a real-time illustration of the needle tip position (distance from the target point). When the deviation index exceeds a threshold, the system backtracks to the most recent non-target moment (skin stable state), using the baseline puncture depth at that moment as a basis, and combines the change in needle handle protrusion length with the depth projection coefficient to perform triangular projection compensation, effectively eliminating the false displacement introduced by skin deformation and restoring the true puncture depth of the needle tip within the body.

[0150] In current formulas for calculating acupuncture depth, because the needle is not always perpendicular to the skin surface during acupuncture, the change in the length of the needle handle protruding when the needle is tilted is not exactly equal to the change in the depth of the needle tip within the tissue. Therefore, according to trigonometric relationships, only by projecting the axial displacement of the needle onto the normal direction perpendicular to the skin surface (or the direction of needle depth) can an accurate depth change be obtained. The physical displacement of the needle body along the axial direction is converted into an effective depth component, eliminating measurement errors caused by needle body tilt.

[0151] Calculating the difference in needle handle length between the current moment and the most recent stable moment (not the target moment) is actually calculating the movement of the needle body since the last stable state. This relative calculation can effectively offset the effects of individual needle differences and initial installation errors, focusing only on the amount of change, thereby improving the robustness of the measurement.

[0152] The revised calculation formula described above, by introducing a depth projection coefficient, addresses to some extent the technical challenge of inaccurate depth measurement under oblique insertion conditions; and by introducing a baseline depth and incremental calculation, it addresses to some extent the problem of absolute measurement being susceptible to environmental interference. The combination of these two approaches facilitates real-time, accurate, and dynamic tracking of the current depth of the acupuncture needle.

[0153] S5, based on the current acupuncture depth of each acupuncture needle, determines the second distance from the tip of each acupuncture needle to the corresponding target nerve point, and outputs the positioning information.

[0154] In this embodiment, after determining the current acupuncture depth of each acupuncture needle, the preset anatomical depth of the target nerve target point corresponding to the acupuncture needle in the human coordinate system is first obtained based on the preset acupoint scheme and the human three-dimensional anatomical model database. This preset anatomical depth is the target depth obtained based on standard human anatomical data or individualized preoperative image reconstruction of the patient. Subsequently, the preset anatomical depth of the target nerve target point is obtained and the difference is calculated with the real-time monitored current acupuncture depth of the acupuncture needle to accurately calculate the remaining distance from the needle tip to the target nerve target point, i.e., the second distance. Based on the sign and magnitude of the second distance, the relative spatial relationship between the current position of the needle tip and the target target point is determined.

[0155] After calculating the second distance, it is further converted into intuitive positioning information and output in a multimodal manner. Specifically, on the display screen of the human-computer interaction interface, the text prompt "Distance to target nerve: X mm" is presented in real time in numerical form. At the same time, the graphical interface of the progress bar or dashboard dynamically reflects the process of the needle tip approaching the target through color gradient. In addition, an audio-visual feedback mechanism can be activated to emit different frequency prompts or different color light signals according to the range of the second distance. When the needle tip reaches the target depth, a specific "in place" prompt is issued. When the needle tip exceeds the target depth, an "over-depth" alarm is triggered. This continuously and intuitively assists the doctor in accurately controlling the needle insertion depth and ensuring that the needle tip accurately reaches the area around the target nerve.

[0156] Another embodiment of the present invention provides a minimally invasive sacral nerve modulation acupuncture-assisted positioning system based on three-dimensional reconstruction, including a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement a minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction.

[0157] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A minimally invasive sacral nerve modulation acupuncture-assisted localization method based on three-dimensional reconstruction, characterized in that, Includes the following steps: The patient's three-dimensional ultrasound transverse image sequence was acquired. Based on the image features, the bone and suspected nerve regions were distinguished. The spatial connectivity between adjacent transverse sections was combined to remove scanning artifacts and extract the distribution of target nerves. A three-dimensional model of the sacrum was reconstructed. The sacral 3D model is partitioned based on the bone morphology mutation features. The partitioned sacral 3D model is then registered with a preset standard reference model. Based on the spatial mapping relationship, the standard target points in the standard reference model are mapped to the sacral 3D model to obtain a sacral navigation model containing target nerve targets. The real-time three-dimensional model of the patient's body surface, the exposed length of the needle handles of several acupuncture needles, and the needle insertion posture parameters are obtained during acupuncture. The real-time three-dimensional model of the body surface is then registered with the sacral navigation model in real time to obtain the registration residual of the real-time registration. Calculate the first distance from the acupuncture point of each acupuncture needle to the corresponding target nerve point, and determine the current deviation index based on the temporal fluctuation of the real-time registration residual and the degree of abrupt change of the first distance; If the current deviation index meets the preset historical fluctuation approximation conditions, then the current acupuncture depth of each acupuncture needle is obtained by combining the change in the needle handle leakage length and the needle insertion posture parameters. Based on the current acupuncture depth of each acupuncture needle, the second distance from the tip of each acupuncture needle to the corresponding target nerve point is determined, and the positioning information is output.

2. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 1, characterized in that, The reconstruction yields a three-dimensional model of the sacrum, including: Edge detection is performed on the cross-sectional images in the cross-sectional image sequence to obtain each candidate connected component; For any candidate connected component, the number of effective edge pixels, the average gray level, and the average gray level gradient magnitude of the candidate connected component are obtained, and data fusion processing is performed to obtain the skeletal edge feature intensity of the candidate connected component; wherein, the effective edge pixels are the contour direction change points of the candidate connected component; Set a first and a second feature intensity threshold, with the first feature intensity threshold being greater than the second feature intensity threshold; based on the comparison results of the bone edge feature intensity with the first and second feature intensity thresholds, determine the bone region and suspected nerve region in the cross-sectional image; For each suspected neural region in the cross-sectional image, region matching is performed in consecutive adjacent cross-sections, and the mean spatial connectivity between the matched regions is calculated; artifact removal is performed based on the mean spatial connectivity to determine each target neural region in the cross-sectional image. Based on the bones and target nerve regions in the cross-sectional image sequence, a three-dimensional model of the sacrum is reconstructed.

3. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 2, characterized in that, Obtaining the effective edge pixels includes: Extract the geometric feature values ​​of each region contour point in the candidate connected domain, and perform first-order difference processing on all geometric feature values ​​to obtain the difference values; Obtain the region contour points with non-zero difference values ​​as the effective edge pixels of the candidate connected domain.

4. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 2, characterized in that, For each suspected neural region in the cross-sectional image, region matching is performed in consecutive adjacent cross-sections, and the mean spatial connectivity between the matched regions is calculated, including: The suspected neural regions that span multiple consecutive frames of cross-sectional images and have overlapping positions are temporally correlated to construct a suspected neural region matching sequence. For each pair of adjacent matching regions in the suspected neural region matching sequence, obtain the ratio of the overlapping area of ​​each pair of adjacent matching regions and the curvature difference between the matching contour points; Based on the extracted overlap area ratio and curvature difference, the spatial connectivity between each pair of adjacent matching regions in the suspected neural region matching sequence is calculated sequentially; wherein, the spatial connectivity is positively correlated with the overlap area ratio and negatively correlated with the curvature difference. Obtain the spatial connectivity of all adjacent matching region pairs within the suspected neural region matching sequence, and calculate the average of all spatial connectivity as the overall spatial connectivity mean of the suspected neural region matching sequence.

5. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 1, characterized in that, Before partitioning the sacral 3D model based on bone morphological mutation features, the method further includes obtaining the bone morphological mutation features: Extract the local curvature of the contour points of the sacral region in each cross-sectional image of the sacral 3D model, and the mean spatial connectivity of the matching sequence of the sacral region to which the sacral region belongs; The deformation feature contribution of the bone region contour points is determined based on the local curvature of the bone region contour points and the mean interlayer spatial connectivity of the entire bone matching sequence to which the bone region containing the bone region contour points belongs. The deformation feature contribution of each skeletal region contour point is obtained, and compared with the preset contribution threshold to obtain local feature points. These local feature points are then used as the skeletal morphological mutation features.

6. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 5, characterized in that, The partitioning of the sacral 3D model based on bone morphological mutation features includes: Statistically analyze the density distribution of the local feature points in each cross-sectional image; Based on the density distribution, extract the cross-sectional image of the feature point cluster and mark it as a morphological abrupt change surface; The sacral three-dimensional model is segmented along the morphological mutation plane to obtain the partitioned sacral three-dimensional model.

7. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 1, characterized in that, When acquiring the transverse image sequence of a patient's three-dimensional ultrasound, multiple multimodal markers are attached to the patient's body surface. These multimodal markers are visible in both ultrasound and binocular visual images. The real-time positioning and registration of the real-time body surface three-dimensional model with the sacral navigation model includes: The first spatial coordinates of the multimodal markers in the sacral navigation model and the second spatial coordinates in the real-time three-dimensional body surface model are extracted respectively. Based on the feature alignment of the first spatial coordinates and the second spatial coordinates, real-time positioning and registration of the real-time body surface 3D model and the sacral navigation model are performed.

8. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 1, characterized in that, The determination of the current deviation index based on the temporal fluctuation of the real-time registration residual and the degree of abrupt change in the first distance includes: For any acupuncture needle, obtain the current registration residual of the real-time positioning and registration at the current moment, and extract the historical registration residual of the previous moment at the current moment; Calculate the distance difference between the first distance at the current moment and the historical distance sequence, and filter the target time based on the distance difference; For the target time, the current deviation index is obtained based on the current registration residual, the historical registration residual, and the absolute value of the distance difference; wherein, the current deviation index is positively correlated with both the current registration residual and the absolute value of the distance difference, and negatively correlated with the historical registration residual.

9. The minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction according to claim 8, characterized in that, The needle insertion posture parameter characterizes the angle between the insertion direction of each acupuncture needle and the normal to the skin surface at the insertion point; if the current deviation index meets the preset historical fluctuation approximation condition, then the current acupuncture depth of each acupuncture needle is corrected by combining the change in the needle handle leakage length and the needle insertion posture parameter, including: The consecutive target moments are merged into a target time period, and the target time period including the current moment is obtained as the current target time period; The differences between the current target time period and the reference historical time period in terms of deviation index, duration, and distance difference sequence are calculated to obtain the approximation of historical fluctuations. For each acupuncture needle, when the historical fluctuation approximation is greater than the preset fluctuation threshold, the reference acupuncture depth of the acupuncture needle corresponding to the non-target time closest to the current time is extracted. The depth projection coefficient is determined based on the included angle of the acupuncture needle. The change in the length of the needle handle protruding from the acupuncture needle is fused with the depth projection coefficient to obtain the depth compensation amount of the acupuncture needle. The current acupuncture depth of the acupuncture needle is obtained by correcting the reference acupuncture depth corresponding to the acupuncture needle using the depth compensation amount.

10. A minimally invasive sacral nerve modulation acupuncture-assisted positioning system based on three-dimensional reconstruction, characterized in that, It includes a processor and a memory, the processor being used to process instructions stored in the memory to implement a minimally invasive sacral nerve modulation acupuncture-assisted positioning method based on three-dimensional reconstruction as described in any one of claims 1-9.

Images