A minimally invasive posterior wall of the pelvic floor layer suturing method based on three-dimensional positioning

Abstract

The present application relates to a kind of based on three-dimensional positioning minimally invasive pelvic floor posterior wall layered suture method, belong to pelvic floor surgical navigation technical field.The method is constructed pelvic floor posterior wall fine model by preoperative multi-modal image three-dimensional reconstruction, intraoperative endoscopy image and three-dimensional ultrasound data are fused and registered, and real-time generation dynamic three-dimensional model and automatic planning layered suture path;Distance from simultaneously tracking suture instrument tip position and rectal anterior wall is calculated and graded early warning is sent.On this basis, the actual needle point and depth deviation are collected, on the one hand for adaptive path correction, on the other hand by extending kalman filter online update local deformation field, form "suture-deviation collection-deformation field update-path re-planning" closed loop feedback.The present application realizes the dynamic compensation of intraoperative soft tissue deformation and adaptive correction of suture path, significantly improves the positioning accuracy and operation safety of layered suture, reduces the risk of rectal injury and nerve injury.

Description

Technical Field

[0001] This invention relates to the field of pelvic floor surgical navigation technology, specifically a minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning. Background Technology

[0002] Posterior pelvic floor defects (including posterior vaginal wall prolapse, enterohernia, perineal body defects, etc.) are common types of female pelvic floor dysfunction, and are often treated clinically with layered suture repair. This procedure requires sequential suturing of the mucosa, muscle, fascia, and levator ani muscle layers within the confined pelvic or vaginal operating space to restore the normal anatomical layers and supporting structures of the posterior pelvic floor. However, traditional surgery relies heavily on the surgeon's tactile sense and the two-dimensional endoscopic view to determine the extent of the defect, tissue layer boundaries, and needle depth, resulting in significant subjective experience dependence. Because the posterior pelvic floor is adjacent to important structures such as the anterior rectal wall, pudendal nerve, and pelvic vessels, excessively deep sutures or directional deviations can easily lead to rectal injury, nerve damage, or hematoma formation; while superficial sutures or excessively wide spacing make effective tissue alignment difficult, resulting in a high recurrence rate. Furthermore, intraoperative pneumoperitoneum pressure, instrument traction, and changes in patient position can all cause deformation and displacement of the pelvic floor soft tissues, further increasing the uncontrollability of the surgical procedure.

[0003] To overcome the aforementioned shortcomings, some surgical navigation systems have been attempted for application in pelvic floor surgery in recent years. Existing navigation technologies typically rely on preoperative MRI or CT images for three-dimensional reconstruction and employ optical or electromagnetic tracking systems to locate surgical instruments, displaying the relationship between instrument position and anatomical structures on a monitor. However, most of these systems use rigid registration methods, assuming that intraoperative tissue morphology is completely consistent with preoperative images. This fails to compensate for nonlinear soft tissue deformation caused by pneumoperitoneum, traction, and tissue resection, leading to a gradual misalignment between the preoperative three-dimensional model and the actual intraoperative anatomy, with positioning errors accumulating over time. Furthermore, existing systems lack quantitative path planning functions for specific procedures like layered sutures; the distance between needle entry and exit points, and the suture depth are still manually determined by the surgeon, failing to achieve individualized automatic planning. Regarding intraoperative feedback, existing systems only provide simple distance threshold alarms, lacking a dynamic correction mechanism based on actual suture deviations. Once operational deviations occur, the remaining suture path cannot be adjusted in real time. Therefore, a minimally invasive suture navigation method that can compensate for soft tissue deformation in real time, automatically plan layered suture paths, and possess closed-loop feedback correction capabilities is urgently needed. Summary of the Invention

[0004] In order to solve the problems of the prior art, the present invention provides a minimally invasive pelvic floor posterior wall layered suturing method based on three-dimensional positioning.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: Firstly, a minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning, comprising the following steps: Step S1: Acquire preoperative pelvic floor three-dimensional imaging data of the patient, the pelvic floor three-dimensional imaging data including at least T2-weighted MRI sequence and three-dimensional ultrasound volume data; and acquire intraoperative real-time endoscopic image sequence and intraoperative real-time three-dimensional ultrasound data; Step S2: Segment and reconstruct the preoperative pelvic floor three-dimensional image data to generate a preoperative three-dimensional model. The preoperative three-dimensional model includes the spatial location information of the mucosa, muscle, fascia, and anterior rectal wall boundaries of the posterior pelvic floor wall. Step S3: Perform non-rigid registration of the intraoperative real-time endoscopic image sequence and intraoperative real-time three-dimensional ultrasound data with the preoperative three-dimensional model to generate an intraoperative dynamic three-dimensional model; the non-rigid registration includes: using a global registration network based on deep learning, taking intraoperative ultrasound volume data as input, and outputting a global deformation field; using a local registration network based on dense optical flow method, taking continuous endoscopic image frames as input, and outputting a local deformation field; the global deformation field and the local deformation field are weighted and fused to obtain a total deformation field, which is then applied to the preoperative three-dimensional model to generate the intraoperative dynamic three-dimensional model; Step S4: Based on the intraoperative dynamic three-dimensional model, automatically calculate the suture path for layered suture, which includes a first-layer suture path and a second-layer suture path; wherein, calculate the minimum safe distance threshold between the suture needle tip and the anterior rectal wall. , The value range is 3mm to 8mm; a first set of needle insertion points is generated on the mucosal surface of the intraoperative dynamic three-dimensional model. and needle exit point set The horizontal distance between the needle entry point and the needle exit point is 3mm to 15mm; a second layer of needle entry point clusters is generated on the surface of the muscle layer. The horizontal distance between the needle entry point and the needle exit point is 3mm to 20mm; set the depth of the first layer of suture. The second layer of suture is 3mm to 5mm deep. The thickness is 3mm to 10mm, and ; Step S5: Real-time acquisition of the position and orientation of the suture instrument tip in three-dimensional space; Step S6: Calculate the spatial Euclidean distance between the tip of the suture instrument and the anterior wall of the rectum in real time. ,when At that time, visual warning signals and / or auditory warning signals are generated; Step S7: Overlay the intraoperative dynamic 3D model with the suture path onto the intraoperative endoscopic image, and display the position of the suture instrument tip and the spatial Euclidean distance in real time. ; Step S8: During the actual suturing process, collect the actual needle entry point position for each stitch. and actual suture depth ; Step S9: Calculate the actual needle entry point and the corresponding planned needle entry point in step S4. Positional deviation vector between and the difference between actual depth and planned depth Depth deviation between ; Step S10: Transfer the position deviation vector and depth deviation Simultaneously used for: (i) Generate the offset compensation amount for subsequent unstitched needle entry points and depth correction amount ,in The compensation function represents the number of stitches already sutured. Using an exponentially weighted moving average, It is the attenuation factor and its value ranges from 0.3 to 0.7; (ii) Using an extended Kalman filter to process the local deformation field in step S3 The local deformation field is updated online. ,in ,in The Kalman gain matrix; Step S11: Utilize the updated local deformation field The intraoperative dynamic 3D model is regenerated, and step S4 is re-executed based on the updated model to calculate the suture path for the remaining unsutured area, while simultaneously adjusting the offset compensation amount. and depth correction amount This is superimposed onto the recalculated path to generate the final corrected stitching path; Step S12: Repeat steps S8 to S11 after each 1 to 3 stitches are completed, until all stitches are completed.

[0006] In one specific implementation of the first aspect, the global registration network adopts a neural network consisting of a U-Net architecture and a cascaded spatial transformation network, and its loss function during training is:

[0007] in, Intraoperative ultrasound volume data, Simulated ultrasound images generated from the projection of the preoperative 3D model. The transformation function predicted by the global registration network. For the global deformation field, To normalize the cross-correlation similarity measure, For the spatial smoothing regularization term of the deformation field, This is the balance coefficient, with a value ranging from 0.01 to 0.1.

[0008] In one specific implementation of the first aspect, the weighted fusion employs adaptive weight coefficients. and ,in:

[0009] In the formula, The quality evaluation index for intraoperative ultrasound images is 0 to 1. This is the scaling factor, with a value ranging from 5 to 20; The quality threshold is set, with a value ranging from 0.5 to 0.7. When the ultrasonic quality is higher than the threshold, the weight of the global deformation field is increased.

[0010] In one specific implementation of the first aspect, the minimum safe distance threshold Adaptive adjustment based on individual patient anatomical characteristics:

[0011] in, The baseline safety distance is set at 5mm. This refers to the thickness of the anterior rectal wall measured in real time during the operation. The average thickness of the anterior rectal wall in a normal population is 2.5 mm. This represents the minimum distance between the pelvic floor nerve plexus and the target suture area. For reference distance, a value of 10mm is used; and These are weighting coefficients, with values ​​ranging from 0.5 to 1.5 and from 0.2 to 0.8, respectively.

[0012] In one specific implementation of the first aspect, the compensation function Using an exponentially weighted moving average:

[0013] in, For the first Needle position deviation vector This represents the number of stitches already placed. The forgetting factor ranges from 0.1 to 0.5; and when When calculating the error, only the deviation from the third and subsequent stitches is weighted and averaged; the first two stitches are not included in the compensation.

[0014] In one specific embodiment of the first aspect, the process noise covariance matrix of the extended Kalman filter Values Observation noise covariance matrix Values Initial state covariance matrix Values ,in The identity matrix; the Kalman gain matrix Automatic adjustment based on residual covariance ensures that the local deformation field converges to a steady-state error of less than 0.5 mm within 5 updates.

[0015] In one specific implementation of the first aspect, the triggering frequency of repeating steps S8 to S11 in step S12 is adaptively adjusted according to the actual deviation: when the absolute value of any axial deviation is greater than 2 mm or the absolute value of the depth deviation is greater than 1.5 mm, a feedback update is triggered immediately; when all deviations are less than the above threshold, a feedback update is triggered every 3 stitches; and a feedback update is forcibly triggered when suturing crosses tissue layers.

[0016] In one specific implementation of the first aspect, the warning signal includes three levels: Level 1 warning: When When this occurs, a yellow visual warning is generated, and an intermittent sound alert with a frequency of 1Hz is emitted; Level 2 warning: When At that time, an orange visual warning is generated, and an intermittent sound alert with a frequency of 5Hz is emitted; Level 3 Warning: When At that time, a red visual warning is generated, and a continuous audible prompt is issued. At the same time, the text warning "Danger! Please stop needle insertion" is displayed on the navigation interface.

[0017] In one specific implementation of the first aspect, the overlay display in step S7 adopts an augmented reality approach, overlaying the intraoperative dynamic three-dimensional model and suture path onto the real endoscopic image in the form of a semi-transparent color layer, wherein: the anterior rectal wall is rendered with a red semi-transparent surface; the first layer of suture path is represented by a green curve with a line width of 2 pixels; the second layer of suture path is represented by a blue curve with a line width of 2 pixels; the tip of the suture instrument is marked with a yellow crosshair and displays the value of the current distance d.

[0018] In a second aspect, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning.

[0019] The beneficial effects of this invention are as follows: 1. This invention utilizes preoperative multimodal image 3D reconstruction and non-rigid registration technology to dynamically fuse intraoperative real-time endoscopic images with 3D ultrasound data. Employing a weighted compensation method combining global and local deformation fields, it effectively counteracts model drift caused by intraoperative soft tissue deformation, ensuring real-time consistency between the preoperative 3D model and the actual anatomical structure. This provides a precise spatial positioning basis for layered suturing. Based on this, individualized safety distance thresholds are automatically calculated using the intraoperative dynamic 3D model, and first and second layer suturing paths are generated according to tissue layers. The distance between the needle entry and exit points and the suturing depth are quantitatively controlled, avoiding the subjective biases of traditional methods that rely on surgeon experience. Simultaneously, by tracking the spatial position and posture of the suturing instrument tip in real time, the spatial distance between it and the anterior rectal wall is dynamically calculated. Based on multi-level safety thresholds, graded warning signals are triggered, promptly alerting the surgeon when the instrument approaches a danger zone, significantly reducing the risk of rectal wall injury and pelvic floor nerve injury. Augmented reality navigation displays overlay the virtual model, suturing path, and instrument position onto the real endoscopic image, achieving intraoperative visual guidance and improving the controllability and accuracy of the operation.

[0020] 2. This invention further introduces a dual dynamic closed-loop feedback mechanism based on actual suture deviation. By real-time acquisition of the actual needle entry point position and suture depth for each stitch, the positional deviation vector and depth deviation between these and the planned path are calculated. On one hand, the deviation is input into the adaptive path correction module, which uses an exponentially weighted moving average method to generate offset compensation for subsequent unsutured needle entry points, dynamically adjusting the remaining suture trajectory. On the other hand, the deviation is fed as an observation into an extended Kalman filter to update the local deformation field in the non-rigid registration module online, enabling the intraoperative dynamic three-dimensional model to be corrected in real time according to tissue deformation and instrument traction effects during the actual suture process. The updated deformation field re-drives the path planning module to generate a remaining suture path that matches the current actual anatomical state, which, when superimposed with the path compensation, forms the final guiding trajectory. This closed-loop feedback mechanism achieves iterative optimization of "suture—deviation acquisition—deformation field update—path replanning—re-suture," enabling the system to adaptively eliminate accumulated errors and compensate for the adverse effects of tissue creep, instrument traction, and operational deviations. This maintains consistency between three-dimensional positioning accuracy and path guidance throughout the entire surgical process, significantly improving the overall quality of layered suture and surgical safety. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall system structure of the present invention.

[0022] Figure 2 This is a schematic diagram of the overall process of the method of the present invention.

[0023] Figure 3 This is a schematic diagram of the path planning and dynamic feedback closed loop of the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] like Figures 1 to 3 This paper presents a minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning.

[0026] I. System Overall Architecture This embodiment provides a minimally invasive suturing navigation system for the posterior pelvic floor wall based on non-rigid registration, including: Data acquisition module: communicates with MRI equipment, 3D ultrasound machine, endoscopic camera and instrument tracking equipment; 3D Reconstruction Module: Runs on a workstation equipped with a GPU; Non-rigid registration module: includes global deformation estimation submodule, local deformation estimation submodule, and deformation field fusion submodule; Path planning module: includes a safety area calculation submodule, a needle entry / exit point generation submodule, and a depth control submodule; The instrument tracking module includes a coordinate transformation unit and an attitude calculation unit. Distance calculation and early warning module: includes distance calculation unit and hierarchical early warning unit; Navigation display module: using an augmented reality head-mounted display or an external display; Deviation acquisition module, deviation calculation module, adaptive path correction module, deformation field online update module, and closed-loop controller.

[0027] The above modules work together to implement the steps of the method of the present invention.

[0028] II. Detailed Implementation Steps Step S1: Data Acquisition Input: Patient basic information, preoperative pelvic floor MRI and 3D ultrasound volume data.

[0029] Output: Preoperative image sequences in DICOM format, intraoperative real-time endoscopic images (1920×1080, ≥30fps) and intraoperative real-time three-dimensional ultrasound data (≥10Hz).

[0030] Specific procedures: 1-3 days prior to surgery, the patient is placed in the supine or lithotomy position. A 3.0T MRI scanner is used for T2-weighted scanning of the pelvic floor, with the following parameters: TR 4000ms, TE 120ms, matrix 320×320, FOV 240mm×240mm, slice thickness 1mm, and no-interval scanning. Simultaneously, a transperineal 3D ultrasound examination is performed (frequency 7.5MHz, automatic scanning to acquire volumetric data). During the procedure, images are acquired in real-time using a laparoscopy or transvaginal endoscopy, while a transperineal 3D ultrasound probe is used to continuously acquire volumetric data streams. All data are timestamped.

[0031] Step S2: 3D Reconstruction Input: Preoperative MRI and 3D ultrasound data.

[0032] Output: Preoperative 3D model (.stl or .obj format), including triangular patch meshes of the mucosa, muscle layer, fascia layer, and anterior rectal wall boundaries, as well as the relative position matrix between each layer.

[0033] Specific procedures: A deep learning-based multi-organ segmentation network (nnU-Net) was used to automatically segment MRI data. The network input was the original MRI image, and the output was a 5-channel probability map: background, mucosa, muscle layer, fascia, and anterior rectal wall. After segmentation, the Marching Cubes algorithm was used to extract isosurfaces to generate triangular meshes, which were then smoothed 10 times using Laplacian smoothing to reduce the number of triangles to 100,000-200,000. The same network architecture was used for the 3D ultrasound data, but anisotropic diffusion filtering and contrast-limited adaptive histogram equalization preprocessing were performed beforehand. Finally, the MRI and ultrasound segmentation results were fused: using the MRI model as a reference, the ultrasound model was registered to the MRI space using the mutual information maximization method, and the minimum bounding box of the anatomical boundaries of the two models was taken as the final model. The boundary of the anterior rectal wall was defined as the potential gap between the rectal muscle layer and the posterior vaginal fascia.

[0034] Step S3: Non-rigid registration (1) Global Deformation Estimation Input: Intraoperative 3D ultrasound volume data (Voxel size 0.5mm, volume 64mm×64mm×48mm); Simulated ultrasound images generated by preoperative 3D model projection (Generated using a ray casting algorithm).

[0035] Processing: The global registration network uses U-Net (5 downsampling layers for the encoder, 5 upsampling layers for the decoder, with skip connections) followed by a spatial transformation network (STN). The network predicts the dense displacement field. : The loss function is:

[0036] in

[0037] For normalized cross-correlation (window 11×11×11). Let the second derivative norm of the displacement field be . The network was first pre-trained on 2000 pairs of simulation data, and then fine-tuned on 50 pairs of clinical data.

[0038] Output: Global deformation field (Discrete grid, 1mm interval).

[0039] (2) Local deformation estimation Input: Three consecutive endoscopic images (Distortion corrected, grayscale converted).

[0040] Processing: The local registration network is based on a recursive fully convolutional network (RFCN). The encoder contains 7 convolutional layers (stride 2), the recurrent module is a ConvLSTM (hidden layer dimension 64), and the decoder contains 5 deconvolutional layers to output the optical flow field. The loss function includes photometric loss. . Maintain regularity at the edges The local deformation field relative to the preoperative reference frame is obtained by accumulating the optical flow of adjacent frames. .

[0041] Output: Local deformation field .

[0042] (3) Deformation field fusion Adaptive weighting coefficients are used:

[0043] In the formula The intraoperative ultrasound image quality evaluation index (combining SNR and CNR, output in real time by the support vector regression model) is taken as follows: Total deformation field The deformation field was applied to each vertex of the preoperative 3D model to obtain the intraoperative dynamic 3D model. Update frequency ≥ 5Hz.

[0044] Step S4: Path Planning Input: Intraoperative dynamic 3D model .

[0045] Output: Suturing path of the first layer (mucosal layer) and the second layer (muscle layer), including the coordinates of the needle entry point, the coordinates of the needle exit point, the suture depth, and the suture sequence.

[0046] Specific operations: Safe distance threshold: Adaptive formula used:

[0047] in The thickness of the anterior rectal wall was measured in real time during the operation (using three-dimensional ultrasound). This represents the minimum distance between the pelvic floor nerve plexus and the target suture area. Example measurement Calculated .

[0048] Needle entry and exit point generation: Samples are taken at equal intervals along the long axis of the defect on the mucosal surface, with an interval of 10 mm (range 5~15 mm), to form the needle entry point set. Needle exit point The first layer is located 8mm off to the opposite side. The second layer is on the surface of the muscle layer, with a spacing of 12mm (range 8~20mm), and the needle entry point is about 4mm off from the first layer.

[0049] Depth control: Depth of the first layer of sutures (Range 3~5mm), second layer (Range 6~10mm), direction along the tissue surface normal vector.

[0050] The output JSON file stores the 3D coordinates, depth, and order of all suture points.

[0051] Step S5: Real-time instrument tracking Input: Raw data from an electromagnetic positioning system (such as NDI Aurora) or an optical positioning system (such as NDI Polaris).

[0052] Output: The position of the tip of the suture instrument in the surgical space coordinate system. (Unit: mm) and orientation (Euler angles or quaternions).

[0053] Specific procedures: Install a 6-DOF electromagnetic sensor (1.8mm diameter, 0.5mm accuracy) or three non-collinear optical marker spheres near the handle and tip of the needle holder. Preoperatively, use cusp calibration to determine the spatial offset between the sensor and the instrument tip; calibration error <0.3mm. Coordinate transformation: Obtain the transformation matrix from the sensor coordinate system to the world coordinate system through preoperative point-to-point ICP registration (using a reference array fixed to the operating table); transformation error <0.5mm. Attitude calculation: For optical tracking, solve the pitch, yaw, and roll angles using the least squares method based on the coordinates of the three marker points; angular accuracy better than 0.5°.

[0054] Step S6: Real-time distance calculation and hierarchical early warning Input: Instrument tip coordinates Triangular mesh on the anterior wall of the rectum (from extract).

[0055] Output: Spatial Euclidean distance Warning level (0-3).

[0056] Specific operation: The bounding box hierarchy (BVH) is used to accelerate the nearest point search, and the calculation... Frequency ≥ 100Hz. The warning logic is as follows: Level 1 warning: When When this occurs, a yellow visual warning is generated, and an intermittent sound alert with a frequency of 1Hz is emitted; Level 2 warning: When At that time, an orange visual warning is generated, and an intermittent sound alert with a frequency of 5Hz is emitted; Level 3 Warning: When At that time, a red visual warning is generated, and a continuous audible prompt is issued. At the same time, the text warning "Danger! Please stop needle insertion" is displayed on the navigation interface.

[0057] Step S7: Augmented Reality Navigation Display Input: Intraoperative endoscopic images Dynamic model Suture path curve, instrument tip position ,distance .

[0058] Output: Augmented reality image with overlaid virtual information.

[0059] Specific operation: HoloLens 2 is used as the display device. First, hand-eye calibration (chessboard method) is performed to determine the transformation matrix between the endoscope camera and HoloLens. Rendering settings: The anterior rectal wall is rendered as a red semi-transparent (0.4 opacity); the first layer of suture path is a green curve (2mm line width, with a halo); the second layer is a blue curve; the instrument tip is a yellow crosshair (5mm diameter), with the current distance value displayed next to it (white text, semi-transparent black background). All virtual objects are transformed to the HoloLens world coordinate system through the calibration matrix, with a refresh rate of 60Hz and a latency of <50ms.

[0060] Step S8: Collection of actual suture deviation Input: The position and attitude continuously output by the instrument tracking module.

[0061] Output: The actual needle insertion point position for each stitch. and actual suture depth .

[0062] Specific operation: The system automatically determines the needle insertion event by detecting the speed of the instrument tip and its distance from the model surface: when the tip speed is less than a threshold and the distance from the tissue surface is <1mm for 0.5s, it is determined as the start of needle insertion; when the tip exits the tissue, it is determined as the end of needle withdrawal. The tip coordinates at the start of needle insertion are recorded as... The deepest distance the needle tip travels within the tissue during insertion (by integrating the path length and considering changes in instrument posture) is used as... .

[0063] Step S9: Deviation Calculation Input: Plan the needle entry point Planning depth Actual needle insertion point Actual depth .

[0064] Output: Position deviation vector Depth deviation .

[0065] Step S10: Dual Utilization of Bias (i) Adaptive path correction Using the exponentially weighted moving average compensation function:

[0066] in (Range 0.1~0.5), the first two stitches are not included in the compensation (i.e., calculation only occurs after n≥3). Depth correction amount. (Range 0.3~0.7).

[0067] (ii) Online updating of deformation field An extended Kalman filter (EKF) is used. State vector. (Discretized to a 20mm control point grid, approximately 500 control points), observation vector (3D). Observation matrix Interpolation maps the state to the deviation location. The process noise covariance is set. Initial state covariance After each injection, perform an EKF update:

[0068]

[0069]

[0070] After five updates, the local deformation field estimation error decreased from 2.1 mm to 0.4 mm (<0.5 mm).

[0071] Step S11: Regeneration of Model and Path The updated local deformation field Reintegrate with the current global deformation field (with the same weights as in step S3) to obtain a new total deformation field. It acts on the preoperative model to generate the updated dynamic model. Then re-execute step S4, based on... Calculate the suture path for the remaining unsutured area and apply the offset compensation amount given in step S10(i). and depth correction amount The corrected suture path is then superimposed onto the recalculated path.

[0072] Step S12: Loop Feedback The system monitors the absolute value of the deviation of the most recent stitch in real time. If the absolute value of any axial deviation is >2mm or the absolute value of the depth deviation is >1.5mm, steps S10-S11 are executed immediately; otherwise, they are executed once every 3 stitches. When the suture crosses tissue layers (from the mucosa to the muscle layer), a feedback update is forcibly triggered. Steps S8-S11 are repeated until all planned stitches are completed.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A minimally invasive layered suturing method for the posterior pelvic floor wall based on three-dimensional positioning, characterized in that, Includes the following steps: Step S1: Acquire preoperative pelvic floor three-dimensional imaging data of the patient, the pelvic floor three-dimensional imaging data including at least T2-weighted MRI sequence and three-dimensional ultrasound volume data; and acquire intraoperative real-time endoscopic image sequence and intraoperative real-time three-dimensional ultrasound data; Step S2: Segment and reconstruct the preoperative pelvic floor three-dimensional image data to generate a preoperative three-dimensional model. The preoperative three-dimensional model includes the spatial location information of the mucosa, muscle, fascia, and anterior rectal wall boundaries of the posterior pelvic floor wall. Step S3: Perform non-rigid registration of the intraoperative real-time endoscopic image sequence and intraoperative real-time three-dimensional ultrasound data with the preoperative three-dimensional model to generate an intraoperative dynamic three-dimensional model; the non-rigid registration includes: using a global registration network based on deep learning, taking intraoperative ultrasound volume data as input, and outputting a global deformation field; using a local registration network based on dense optical flow method, taking continuous endoscopic image frames as input, and outputting a local deformation field; the global deformation field and the local deformation field are weighted and fused to obtain a total deformation field, which is then applied to the preoperative three-dimensional model to generate the intraoperative dynamic three-dimensional model; Step S4: Based on the intraoperative dynamic three-dimensional model, automatically calculate the suture path for layered suture, which includes a first-layer suture path and a second-layer suture path; wherein, calculate the minimum safe distance threshold between the suture needle tip and the anterior rectal wall. , The value range is 3mm to 8mm; a first set of needle insertion points is generated on the mucosal surface of the intraoperative dynamic three-dimensional model. and needle exit point set The horizontal distance between the needle entry point and the needle exit point is 3mm to 15mm; a second layer of needle entry point clusters is generated on the surface of the muscle layer. The horizontal distance between the needle entry point and the needle exit point is 3mm to 20mm; set the depth of the first layer of suture. The second layer of suture is 3mm to 5mm deep. The thickness is 3mm to 10mm, and ; Step S5: Real-time acquisition of the position and orientation of the suture instrument tip in three-dimensional space; Step S6: Calculate the spatial Euclidean distance between the tip of the suture instrument and the anterior wall of the rectum in real time. ,when At that time, visual warning signals and / or auditory warning signals are generated; Step S7: Overlay the intraoperative dynamic 3D model with the suture path onto the intraoperative endoscopic image, and display the position of the suture instrument tip and the spatial Euclidean distance in real time. ; Step S8: During the actual suturing process, collect the actual needle entry point position for each stitch. and actual suture depth ; Step S9: Calculate the actual needle entry point and the corresponding planned needle entry point in step S4. Positional deviation vector between and the difference between actual depth and planned depth Depth deviation between ; Step S10: Convert the position deviation vector and depth deviation Simultaneously used for: (i) Generate the offset compensation amount for subsequent unstitched needle entry points and depth correction amount ,in The compensation function represents the number of stitches already sutured. Using an exponentially weighted moving average, It is the attenuation factor and its value ranges from 0.3 to 0.7; (ii) Using an extended Kalman filter to process the local deformation field in step S3 The local deformation field is updated online. ,in ,in The Kalman gain matrix; Step S11: Utilize the updated local deformation field The intraoperative dynamic 3D model is regenerated, and step S4 is re-executed based on the updated model to calculate the suture path for the remaining unsutured area, while simultaneously adjusting the offset compensation amount. and depth correction amount This is superimposed onto the recalculated path to generate the final corrected stitching path; Step S12: Repeat steps S8 to S11 after each 1 to 3 stitches are completed, until all stitches are completed.

2. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The global registration network is a neural network consisting of a U-Net architecture and a cascaded spatial transformation network. Its loss function during training is: in, Intraoperative ultrasound volume data, Simulated ultrasound images generated from the projection of the preoperative 3D model. The transformation function predicted by the global registration network. For the global deformation field, To normalize the cross-correlation similarity measure, For the spatial smoothing regularization term of the deformation field, This is the balance coefficient, with a value ranging from 0.01 to 0.

1.

3. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The weighted fusion uses adaptive weight coefficients. and ,in: In the formula, The quality evaluation index for intraoperative ultrasound images is 0 to 1. This is the scaling factor, with a value ranging from 5 to 20; The quality threshold is set, with a value ranging from 0.5 to 0.

7. When the ultrasonic quality is higher than the threshold, the weight of the global deformation field is increased.

4. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The minimum safe distance threshold Adaptive adjustment based on individual patient anatomical characteristics: in, The baseline safety distance is set at 5mm. This refers to the thickness of the anterior rectal wall measured in real time during the operation. The average thickness of the anterior rectal wall in a normal population is 2.5 mm. This represents the minimum distance between the pelvic floor nerve plexus and the target suture area. For reference distance, a value of 10mm is used; and These are weighting coefficients, with values ​​ranging from 0.5 to 1.5 and from 0.2 to 0.8, respectively.

5. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The compensation function Using an exponentially weighted moving average: in, For the first Needle position deviation vector This represents the number of stitches already placed. The forgetting factor ranges from 0.1 to 0.5; and when When calculating the error, only the deviation from the third and subsequent stitches is weighted and averaged; the first two stitches are not included in the compensation.

6. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The process noise covariance matrix of the extended Kalman filter Values Observation noise covariance matrix Values Initial state covariance matrix Values ,in The identity matrix; the Kalman gain matrix Automatic adjustment based on residual covariance ensures that the local deformation field converges to a steady-state error of less than 0.5 mm within 5 updates.

7. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The triggering frequency of repeating steps S8 to S11 in step S12 is adaptively adjusted according to the actual deviation: when the absolute value of any axial deviation is greater than 2mm or the absolute value of the depth deviation is greater than 1.5mm, a feedback update is triggered immediately; when all deviations are less than the above threshold, a feedback update is triggered every 3 stitches; a feedback update is forcibly triggered when suturing crosses tissue layers.

8. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The warning signal includes three levels: Level 1 warning: When When this occurs, a yellow visual warning is generated, and an intermittent sound alert with a frequency of 1Hz is emitted; Level 2 warning: When At that time, an orange visual warning is generated, and an intermittent sound alert with a frequency of 5Hz is emitted; Level 3 Warning: When At that time, a red visual warning is generated, and a continuous audible prompt is issued. At the same time, the text warning "Danger! Please stop needle insertion" is displayed on the navigation interface.

9. The minimally invasive posterior pelvic floor layered suturing method based on three-dimensional positioning according to claim 1, characterized in that: The overlay display in step S7 adopts an augmented reality approach, overlaying the intraoperative dynamic 3D model and suture path onto the real endoscopic image in the form of a semi-transparent color layer. Specifically, the anterior rectal wall is rendered with a red semi-transparent surface; the first layer of suture path is represented by a green curve with a line width of 2 pixels; the second layer of suture path is represented by a blue curve with a line width of 2 pixels; and the tip of the suture instrument is marked with a yellow crosshair and displays the current distance d value.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images