Tension-reducing suture operation navigation method and system, computer equipment and storage medium

Abstract

The invention discloses a relaxation suture operation navigation method and system, computer equipment and a storage medium, and the method comprises the steps: collecting wound surface multi-dimensional image data through a multi-modal image sensing device, building a three-dimensional wound surface model after the analysis of an improved DeepLab semantic segmentation network, calculating the edge tension distribution of the wound surface, and calculating the edge tension distribution of the wound surface; an optimal suture path planning scheme is generated in combination with preset relaxation suture parameter constraint conditions, and finally a surgical navigation instruction set is output. The system comprises a plurality of cooperative working units such as a multi-modal image data acquisition unit, an improved semantic segmentation processing unit and a mechanical simulation unit. According to the method, through multi-modal image fusion and an improved algorithm, wound tissue characteristics are accurately analyzed, suture parameters are optimized, the accuracy and effectiveness of a relaxation suture operation are improved, and the method has good clinical application value.

Description

Technical Field

[0001] This invention relates to the field of tension-reducing suture surgical navigation, and more particularly to tension-reducing suture surgical navigation methods, systems, computer equipment, and storage media. Background Technology

[0002] In the field of surgery, tension-reducing suture techniques are crucial for promoting wound healing and reducing the risk of wound dehiscence. With the continuous development of medical technology, the requirements for the precision and safety of tension-reducing suture surgery are increasing. Traditional tension-reducing suture surgery relies on the surgeon's experience to determine parameters such as the needle insertion point, angle, suture depth, and suture tension. It lacks objective and accurate navigation support, making it difficult to fully consider the differences in tissue characteristics among different patients, resulting in significant individual variations in surgical outcomes and making it difficult to guarantee surgical quality. Meanwhile, although the development of medical imaging technology has provided some assistance to surgery, the existing single-modal image information cannot fully reflect the complexity of the wound surface, limiting the surgeon's precise understanding of the surgical site.

[0003] Currently, existing tension-reducing suture surgical navigation technologies have many shortcomings. In terms of image information processing, traditional single-modal image sensing devices acquire limited information. For example, ultrasound images alone cannot clearly present skin surface details, while RGB imaging cannot acquire subcutaneous tissue characteristics, failing to provide comprehensive and accurate tissue information for surgery, resulting in information blind spots when surgeons plan suture paths. Furthermore, conventional semantic segmentation algorithms lack the ability to identify wound edges and tissue layers, easily leading to misjudgments when dealing with complex wounds, affecting the accuracy of surgical planning.

[0004] In terms of surgical parameter determination and navigation, current technologies lack in-depth analysis of the relationship between tissue mechanical properties and suture parameters. They cannot accurately determine key parameters such as needle insertion point spacing, needle insertion angle, suture depth, and suture tension based on the actual tension distribution of the wound, leading to suboptimal suturing results and risks such as poor postoperative wound healing and scar hyperplasia. Furthermore, existing surgical navigation systems have low levels of intelligence, failing to effectively integrate multi-source information and perform real-time analysis, thus failing to meet the precise navigation requirements of tension-reducing suture surgeries. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a method, system, computer equipment and storage medium for tension-reducing suture surgery navigation.

[0006] The technical solution adopted in this invention is a tension-reducing suture surgical navigation method, which includes:

[0007] S1: Acquire multi-dimensional image data of the wound area to be sutured based on a multimodal image sensing device. The multimodal image sensing device includes an ultrasound imaging module, a near-infrared imaging module, and an RGB imaging module. Each module synchronously acquires image data of different characteristics of the same wound area.

[0008] S2: Input the multi-dimensional image data into the improved DeepLab semantic segmentation network for feature extraction and analysis. The improved DeepLab semantic segmentation network enhances the ability to identify wound edge features and subcutaneous tissue layer features by introducing spatial attention mechanism and channel attention mechanism.

[0009] S3: Based on the segmentation results output by the improved DeepLab semantic segmentation network, a three-dimensional wound model containing the skin layer, subcutaneous tissue layer and muscle layer is constructed. The three-dimensional wound model contains the geometric morphology information and physical property parameters of each tissue layer.

[0010] S4: Based on the three-dimensional wound model, calculate the tension distribution at each point on the edge of the wound. The tension distribution is obtained by analyzing the elastic modulus, thickness variation, and geometric parameters of the tissue surrounding the wound.

[0011] S5: Based on the tension distribution and the preset tension-reducing suture parameter constraints, generate an optimal suture path planning scheme. The tension-reducing suture parameter constraints include needle insertion point spacing, needle insertion angle, suture depth, and suture tension threshold.

[0012] S6: Map the optimal suture path planning scheme to the actual surgical space to generate a surgical navigation instruction set containing the needle entry point position, needle entry angle and suture tension value.

[0013] Furthermore, the improved DeepLab semantic segmentation network enhances its ability to recognize features of wounds of different sizes by introducing a multi-scale feature fusion module. This multi-scale feature fusion module employs an adaptive weight allocation strategy to fuse feature maps of different scales. The adaptive weight allocation strategy is determined based on the following model formula:

[0014]

[0015] Among them, W i Let A be the fusion weight of the feature map at the i-th scale. i Let α be the channel attention value of the i-th scale feature map, α be the weight adjustment parameter, and n be the number of feature maps; the channel attention value A i The feature map is obtained by analyzing the statistical characteristics of different channels, which is used to characterize the importance of each channel feature to the wound segmentation task.

[0016] Furthermore, the tension distribution is calculated by constructing a tissue biomechanical model, which is determined based on the following formula:

[0017]

[0018] Where T(x,y) is the tension tensor at coordinates (x,y) within the wound area, u(x,y) is the tissue displacement field, and λ and μ are the Lamé constants of the tissue, representing the bulk elastic modulus and shear elastic modulus of the tissue, respectively. The tissue displacement field u(x,y) is obtained by analyzing the tissue deformation features output by the improved DeepLab semantic segmentation network, and is used to describe the deformation of the tissue under stress.

[0019] Furthermore, the optimal suture path planning scheme is generated by constructing a multi-objective optimization model, which is determined based on the following formula:

[0020] minF(P)=w1·T avg (P)+w2·L(P)+w3·Δθ(P

[0021] F(P) = w1·T avg F(P) is the evaluation value calculated for "a certain path P". For a given path P, F(P) should be unique. My understanding is that minF(P) should be the path with the lowest evaluation value among all paths.

[0022] minF(p)=min P∈p (w1·T avg (P)+w2·L(P)+w3·Δθ(P)), where p is the set of all possible paths)

[0023] Where F(P) is the comprehensive evaluation function of path planning scheme P, and T avg (P) is the average tension value on path P, L(P) is the total length of path P, Δθ(P) is the rate of change of angle on path P, and w1, w2 and w3 are weighting coefficients, which respectively characterize the importance of tension uniformity, path length and angle smoothness in the optimization objective; the rate of change of angle Δθ(P) is obtained by calculating the angle difference between adjacent needle entry points and is used to evaluate the smoothness of the suture path.

[0024] Furthermore, the preset tension-reducing suture parameter constraints are determined by constructing a parameter correlation model, which is based on the following formula:

[0025]

[0026] Where d is the optimal spacing between adjacent needle insertion points, E is the elastic modulus of the tissue, h is the tissue thickness, and σ max β represents the maximum allowable stress of the tissue, and β is an empirical correction coefficient. The parameter correlation model determines the optimal parameter combination that meets the tension reduction requirements by analyzing the relationship between suture parameters under different tissue characteristics.

[0027] Furthermore, the construction process of the three-dimensional wound model includes: constructing three-dimensional point cloud data containing tissue hierarchical information based on the multimodal features output by the improved DeepLab semantic segmentation network; performing surface fitting on the three-dimensional point cloud data to generate a continuous three-dimensional wound surface model; and coupling the three-dimensional wound surface model with a pre-established tissue mechanics model to generate a three-dimensional wound model containing mechanical properties.

[0028] Furthermore, the process of generating the surgical navigation instruction set includes: converting the optimal suture path planning scheme into a motion trajectory that can be executed by the robot; determining the optimal suture tension value for each needle insertion point based on the tension distribution and the preset suture tension threshold; and encoding the motion trajectory and suture tension value to generate a surgical navigation instruction set containing spatial position information and mechanical parameter information.

[0029] Tension-reducing suture surgical navigation system, the system includes:

[0030] A multimodal image data acquisition unit is used to acquire multi-dimensional image data of the wound area to be sutured based on a multimodal image sensing device;

[0031] An improved semantic segmentation processing unit is used to input the multi-dimensional image data into an improved DeepLab semantic segmentation network for feature extraction and analysis;

[0032] The 3D model reconstruction unit is used to construct a 3D wound model containing the skin layer, subcutaneous tissue layer and muscle layer based on the segmentation results output by the improved DeepLab semantic segmentation network.

[0033] The tension analysis and calculation unit is used to calculate the tension distribution at each point on the edge of the wound based on the three-dimensional wound model.

[0034] The suture path planning unit is used to generate an optimal suture path planning scheme based on the tension distribution and preset tension reduction suture parameter constraints.

[0035] The navigation instruction generation unit is used to map the optimal suture path planning scheme to the actual surgical space and generate a set of surgical navigation instructions including the needle entry point position, needle entry angle and suture tension value.

[0036] The improved semantic segmentation processing unit includes: a multi-scale feature extraction module for extracting wound features at different scales; a spatial attention enhancement module for enhancing the ability to identify wound edge features; a channel attention optimization module for optimizing the expressive ability of features from different channels; and a feature fusion decision module for fusing features from different scales and channels based on an adaptive weight allocation strategy.

[0037] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a tension-reducing suture surgical navigation method.

[0038] A storage medium storing a computer program that, when executed by a processor, implements the steps of a tension-reducing suture surgical navigation method.

[0039] Beneficial Effects: This invention proposes a method, system, computer equipment, and storage medium for tension-reducing suture surgery navigation. In terms of the method, data is acquired using a multimodal image sensing device, and semantic segmentation is performed using an improved DeepLab algorithm. A three-dimensional wound model containing tissue layer information and physical property parameters is constructed, and then the tension distribution is calculated to generate an optimal suture path planning scheme and surgical navigation instruction set. This makes surgical navigation more accurate, fully considering tissue characteristics, rationally planning the suture path according to the wound condition, effectively reducing tissue tension, improving the tension-reducing suture effect, and reducing postoperative complications. In terms of the system, it consists of multiple units, including multimodal image data acquisition and improved semantic segmentation processing. These units work collaboratively, and by constructing multi-objective optimization models and parameter correlation models, the system's ability to identify wound features and optimize suture parameters is improved, making the system more intelligent and automated, and better assisting doctors in performing tension-reducing suture surgery. Regarding the computer equipment and storage medium, a computer program capable of executing the above method is stored, allowing the tension-reducing suture surgery navigation method to run on a computer device, facilitating operation and use by doctors, while ensuring the repeatability and stability of the method, which is conducive to the promotion and application of this technology. This invention, from the method to the system, and then to the computer equipment and storage medium, forms a complete and effective navigation system for tension-reducing suture surgery, which significantly improves the quality and efficiency of tension-reducing suture surgery and has important clinical application value and broad market prospects. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method steps of the present invention;

[0041] Figure 2 This is a diagram showing the system unit composition of the present invention. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, the tension-reducing suture surgical navigation method includes the following steps:

[0044] S1: Acquire multi-dimensional image data of the wound area to be sutured based on a multimodal image sensing device. The multimodal image sensing device includes an ultrasound imaging module, a near-infrared imaging module, and an RGB imaging module. Each module synchronously acquires image data of different characteristics of the same wound area.

[0045] Specifically, this step utilizes a multimodal image sensing device integrating an ultrasound imaging module, a near-infrared imaging module, and an RGB imaging module. The ultrasound imaging module employs a 10-15MHz high-frequency probe, capable of penetrating the skin 3-5mm to generate subcutaneous tissue tomographic images with a resolution of 0.1-0.3mm, clearly displaying deep structures such as the fascia and muscle layers. The near-infrared imaging module uses an 850-940nm wavelength light source, combined with a blood oxygen saturation sensitive algorithm, to identify the microcirculation status within a 2-3cm radius around the wound, detecting ischemic areas with tissue oxygen partial pressure below 40mmHg. The RGB imaging module uses a 4K resolution camera to acquire surface texture and color information of the wound, with a spatial resolution of 0.05mm, used to identify the boundaries of epidermal damage. The three modules achieve multi-dimensional data acquisition of the same wound area through time synchronization technology (error <10ms).

[0046] The tissue density information provided by ultrasound images can quantify the elastic modulus of different tissue layers (e.g., the elastic modulus of the skin layer is 10-50 kPa, and that of the muscle layer is 50-200 kPa), which can be used for subsequent tension analysis. The blood oxygen distribution map generated by near-infrared images can help determine the suture path to avoid ischemic areas and reduce the risk of postoperative necrosis. The surface features of RGB images are used to accurately extract the curvature of the wound edge (areas with a curvature radius <2 mm are defined as high stress concentration areas). After the three modal data are fused by a pixel-level registration algorithm (registration error <0.5 mm), a multidimensional dataset containing geometric morphology, physical properties, and physiological functions is formed.

[0047] In actual surgical scenarios, the multimodal device continuously acquires data at a frequency of 30 frames per second, covering the wound and a surrounding 5cm area. For wounds deeper than 0.5cm, the ultrasound module automatically switches to multi-plane scanning mode (scanning angle range ±30°) to generate three-dimensional volumetric data. The system monitors image quality in real time, and automatically triggers a gain compensation algorithm when the signal-to-noise ratio (SNR) of the ultrasound image is below 15dB. The acquired raw data is losslessly compressed (compression ratio 1:5) before being transmitted to the processing unit, with overall data latency controlled within 200ms to ensure the real-time nature of navigation information.

[0048] S2: Input the multi-dimensional image data into the improved DeepLab semantic segmentation network for feature extraction and analysis. The improved DeepLab semantic segmentation network enhances the ability to identify wound edge features and subcutaneous tissue layer features by introducing spatial attention mechanism and channel attention mechanism.

[0049] Specifically, the improved DeepLab network used in this step introduces a Spatial Attention Module (SAM) and a Channel Attention Module (CAM) on top of the original Xception backbone network. SAM generates a spatial attention weight matrix by calculating the correlation between each pixel and other pixels in the feature map, enhancing the response to slender features such as wound edges (width < 0.3 mm) and vascular structures (diameter > 0.2 mm). CAM adaptively adjusts the importance of features in each channel through global average pooling, global max pooling, and a shared multilayer perceptron, improving the ability to distinguish different tissue layers (such as epidermis, dermis, and subcutaneous fat). The network output layer employs a multi-task learning strategy, simultaneously generating a tissue category segmentation map (containing 7 tissue categories: normal skin, wound edges, necrotic tissue, blood vessels, nerves, fascia, and muscle) and a boundary probability map (with boundary localization accuracy up to 0.1 mm).

[0050] The input multimodal images are first subjected to feature dimensionality reduction using depthwise separable convolution, unifying the feature maps of ultrasound, near-infrared, and RGB modalities into a 256-dimensional feature space. In the encoder stage, the network captures contextual information at different scales through dilated convolutions (with dilation rates of 6, 12, and 18), targeting blood vessels with diameters of 1-5 mm and areas of 0.5-2 cm². 2 Necrotic tissue can be effectively identified. The decoder stage employs bilinear interpolation upsampling and fuses low-level features (such as edge gradients) to restore the feature map resolution to 1 / 4 of the original image. The final segmentation output is post-processed using a Conditional Random Field (CRF) to eliminate misclassification of small regions (area < 2 mm). 2 Isolated areas are automatically merged.

[0051] On a test set containing 200 cases of different types of wounds (cuts, lacerations, burns), the improved DeepLab network achieved a mean intersection-over-union (mIoU) of 89.7%, a 6.2 percentage point improvement over the traditional DeepLabv3+. The detection precision for wound edges was 93.4%, the recall was 91.2%, and the F1 score was 0.923. With GPU acceleration, the processing time for a single frame was 87ms, meeting the requirements for real-time navigation. Notably, for complex wounds deeper than 1cm, the network controlled the segmentation error of different tissue layers within 0.2mm, providing a high-precision foundation for subsequent 3D modeling.

[0052] S3: Based on the segmentation results output by the improved DeepLab semantic segmentation network, a three-dimensional wound model containing the skin layer, subcutaneous tissue layer and muscle layer is constructed. The three-dimensional wound model contains the geometric morphology information and physical property parameters of each tissue layer.

[0053] Specifically, based on the 2D segmentation results output by the improved DeepLab network, a disparity map of the multi-view images is first calculated using a stereo matching algorithm (using semi-global matching, SGM), generating a point cloud density of 500 points / mm. 2 The three-dimensional point cloud data was used. For missing data regions (such as the interior of deep cavities), radial basis function interpolation was employed to fill in the gaps, with an interpolation error <0.15mm. Then, moving least squares (MLS) was used for surface fitting to generate a smooth three-dimensional surface model, with a surface curvature calculation accuracy of 0.01 / mm. The two-dimensional segmented tissue category information was mapped to the three-dimensional model, assigning each vertex a tissue type label (e.g., skin, fat, muscle) and physical property parameters (elastic modulus, Poisson's ratio).

[0054] The elastic modulus parameter in the model was measured using ultrasound elastography. The elastic modulus of the skin layer ranged from 10 to 50 kPa, with higher values ​​indicating stiffer tissue. Poisson's ratio was set at 0.45-0.49 (close to incompressible materials), reflecting the lateral deformation characteristics of the tissue under stress. Thickness parameters were calculated based on interlaminar reflection signals from ultrasound images, with a measurement accuracy of ±0.05 mm. These parameters directly affect the accuracy of subsequent tension analysis; for example, adjacent tissue areas with an elastic modulus difference >30% are prone to forming stress concentration points.

[0055] The constructed 3D wound model can intuitively display the three-dimensional morphology of the wound, including burrowing wounds with a depth exceeding 1 cm and sinus tracts with a diameter >0.5 cm. Doctors can observe the internal structure of the wound and measure key parameters such as the wound floor area (>2 cm²) using the virtual dissection function. 2In cases of tension-reducing sutures or edge thickness differences (>1mm, requiring layered sutures), a three-dimensional model was used to identify a 5mm wound edge, guiding the surgeon to add deep anchoring sutures and avoiding the risk of hematoma that might occur with traditional surface sutures.

[0056] S4: Based on the three-dimensional wound model, calculate the tension distribution at each point on the edge of the wound. The tension distribution is obtained by analyzing the elastic modulus, thickness variation, and geometric parameters of the tissue surrounding the wound.

[0057] Specifically, this step calculates the wound tension distribution based on the finite element analysis (FEA) method, discretizing the three-dimensional wound model into tetrahedral elements (average side length 0.5 mm), with each element assigned the elastic modulus and Poisson's ratio parameters corresponding to the tissue. The Neo-Hookean hyperelastic model is used to describe the soft tissue mechanical behavior; this model has an error of <5% compared to experimental data when the strain is less than 30%. Boundary conditions are set as follows: the wound edges are fixed, and physiological loads (such as the natural skin tension of 0.05-0.1 N / mm) are applied to the surrounding tissue.

[0058] The calculated tension distribution is represented in tensor form, with the principal stress direction indicating the direction of maximum tissue tensile stress, and the stress amplitude reflecting the magnitude of the tension. Parameters of clinical interest include: maximum principal stress (>0.2 N / mm²). 2 (Tension reduction treatment required), stress gradient (>0.1 N / mm) 3 Indicates stress concentration), shear stress (>0.05 N / mm²). 2 (This may lead to tissue tearing). In the abdominal longitudinal incision model, tension analysis showed that the stress concentration factor at both ends reached 1.8 times, guiding the doctor to design Z-shaped tension-reducing incisions at both ends.

[0059] Tension distribution maps visually display high-stress areas around the wound, helping surgeons optimize suturing strategies. For example, in a V-shaped facial laceration, the high-stress area is usually located at the laceration tip; using a figure-of-eight tension-reducing suture can reduce the tip stress by more than 50%. For cross-joint wounds, tension analysis can indicate the effect of joint flexion and extension on wound tension (e.g., when the knee is flexed at 30°, wound tension is reduced by 25% compared to when it is extended), guiding the selection of the optimal suturing position.

[0060] S5: Based on the tension distribution and the preset tension-reducing suture parameter constraints, generate an optimal suture path planning scheme. The tension-reducing suture parameter constraints include needle insertion point spacing, needle insertion angle, suture depth, and suture tension threshold.

[0061] Specifically, the constraints for tension-reducing suture parameters in this step include: needle insertion point spacing (2-5 mm in the skin layer, 5-8 mm in the subcutaneous tissue), needle insertion angle (15°-30° angle with the normal line of the wound edge), suture depth (0.5-1.5 mm in the skin layer, 1-3 mm in the subcutaneous tissue), and suture tension threshold (0.1-0.3 N for absorbable sutures, 0.2-0.5 N for non-absorbable sutures). These parameters are dynamically adjusted based on tissue type, wound location, and patient age. For example, the needle insertion point spacing is reduced by 20% for pediatric patients, and the suture tension threshold is reduced by 30% for elderly patients.

[0062] The generation of the suture path uses tension distribution as an important parameter of the heuristic function, prioritizing areas with small tension gradients as needle entry points. Simultaneously, considering suture efficiency, the angle variation between adjacent needle entry points is limited to <45° to avoid suture crossing. In complex wounds, stress concentration areas (such as wound corners and undermining edges) are automatically identified, and the number of needle entry points is increased (spacing reduced to 1-2 mm) and deep fixation sutures are added at these locations.

[0063] In a case of a 6cm longitudinal laceration on the dorsum of the hand, the planned suture path avoided the high-stress tendon surface area, employing a "stepped" needle insertion strategy (3mm needle spacing in the epidermis, 4mm in the dermis, and 5mm in the subcutaneous tissue) to reduce the maximum tension from 0.25 N / mm². 2 Reduced to 0.12 N / mm 2 In cases of facial stellate lacerations, a radial tension-reducing suture path was designed to address the stress concentration at multiple wound angles, combined with subcutaneous anchoring sutures. This resulted in a 40% reduction in postoperative scar width compared to traditional sutures.

[0064] S6: Map the optimal suture path planning scheme to the actual surgical space to generate a surgical navigation instruction set containing the needle entry point position, needle entry angle and suture tension value.

[0065] Specifically, an optical tracking system (such as PolarisSpectra based on infrared markers) is used to map the virtual planning to the actual surgical space, achieving a positioning accuracy of ±0.3mm. First, 3-5 passive reflective markers are attached to the patient's skin, and a calibration algorithm is used to establish the transformation relationship between the world coordinate system and the patient coordinate system. Then, the planned suture path point cloud data is converted into Cartesian coordinates executable by the robot, with the transformation error controlled within 0.5mm.

[0066] The surgical navigation command set contains three types of information: spatial positioning information (three-dimensional coordinates of the needle insertion point), angle information (needle insertion direction vector), and mechanical parameters (target suture tension value). Each command point is accompanied by a confidence score (based on factors such as tissue identification accuracy and tension calculation stability). Points with a score below 85% automatically trigger replanning. For deep tissue suturing, the command also includes tissue layer information of the puncture path, prompting the surgeon when to switch suturing techniques (e.g., from continuous suturing to interrupted suturing).

[0067] In pig model experiments, tension-reducing suturing performed according to navigation instructions had an average needle entry point position error of 0.42 mm and an average needle entry angle error of 2.1°, both meeting clinical requirements (position error <1 mm, angle error <5°). In human trials, 20 cases of limb wounds were treated with navigation-assisted suturing. The average scar score (VSS scale) at 3 months postoperatively was 3.2 points, significantly lower than the 5.7 points in the traditional suturing group, indicating that the navigation system effectively reduced tension-related complications.

[0068] Preferably, the improved DeepLab semantic segmentation network enhances its ability to recognize features of wounds of different sizes by introducing a multi-scale feature fusion module. The multi-scale feature fusion module employs an adaptive weight allocation strategy to fuse feature maps of different scales. The adaptive weight allocation strategy is determined based on the following model formula:

[0069]

[0070] Among them, W i Let A be the fusion weight of the feature map at the i-th scale. i Let α be the channel attention value of the i-th scale feature map, α be the weight adjustment parameter, and n be the number of feature maps; the channel attention value A i The feature map is obtained by analyzing the statistical characteristics of different channels, which is used to characterize the importance of each channel feature to the wound segmentation task.

[0071] Specifically, the performance of the improved DeepLab semantic segmentation network is enhanced by introducing a multi-scale feature fusion module. This module employs an adaptive weight allocation strategy to dynamically fuse feature maps at different scales. In practical applications, small-scale feature maps (e.g., 1 / 8 resolution) are sensitive to detailed features such as wound edges, while large-scale feature maps (e.g., 1 / 32 resolution) contain richer contextual information. By calculating the channel attention values ​​of feature maps at each scale (reflecting the importance of the features to the segmentation task), the system can automatically adjust the fusion weights. For example, when processing vascular structures with a diameter <2mm, the weight of small-scale feature maps is automatically increased to 0.7; when the recognition area is >5cm², the weight of small-scale feature maps is increased to 0.7. 2When dealing with necrotic tissue, the weight of the large-scale feature map is increased to 0.6. This adaptive mechanism enables the network to maintain high recognition accuracy for target structures of different sizes. In the test set, the recall rate for detecting microvessels and large areas of necrotic tissue was improved by 12% and 8%, respectively, effectively reducing missed detections and false positives.

[0072] Preferably, the tension distribution is calculated by constructing a tissue mechanics model, and the mechanical response of the skin material is described by the strain energy function U. The tissue mechanics simulation is determined based on the following constitutive equation:

[0073]

[0074] Where μ and α are material constants, λ1, λ2, and λ3 are the principal tensile forces in three directions, referring to the strain of the material in three mutually perpendicular principal stress directions. λ1 is along the longest axis of the material, corresponding to the maximum principal tensile force. If the material is a cuboid, this direction is usually along the length of the cuboid; λ2 is along the middle axis of the material, corresponding to the middle principal tensile force. In a cuboid material, this direction is usually along the width of the cuboid; λ3 is along the shortest axis of the material, corresponding to the minimum principal tensile force. The tissue displacement field is obtained by analyzing the tissue deformation features output by the improved DeepLab semantic segmentation network, used to describe the deformation of the tissue under stress. The nominal stress T is given by the following formula:

[0075]

[0076] Specifically, the wound tension distribution is calculated based on a tissue mechanics model, which considers the Lamé constants (λ and μ) and displacement field of the tissue. The Lamé constants reflect the elastic properties of the tissue; for the skin layer, λ is approximately 10-20 kPa and μ is approximately 5-10 kPa; for the muscle layer, λ is 50-100 kPa and μ is 20-40 kPa. The tissue displacement field is obtained by analyzing the deformation characteristics output by the improved DeepLab, which quantifies the amount of tissue deformation under stress. In practical applications, this model can accurately calculate the tension values ​​at various points along the wound edge. For example, at the edge of an abdominal surgical incision, the error between the model's predicted maximum tension value and the actual measured value is <15%. Through tension distribution analysis, doctors can identify high-risk areas (e.g., tension > 0.2 N / mm²) in advance. 2 The area is targeted, and the suturing strategy is adjusted accordingly, such as increasing tension-reducing sutures or adjusting the needle insertion angle, thereby effectively reducing the risk of postoperative wound dehiscence.

[0077] Preferably, the optimal suture path planning scheme is generated by constructing a multi-objective optimization model, which is determined based on the following formula:

[0078] minF(P)=w1·Tavg (P)+w2·L(P)+w3·Δθ(P

[0079] F(P) = w1·T avg F(P) is the evaluation value calculated for "a certain path P". For a given path P, F(P) should be unique. My understanding is that minF(P) should be the path with the lowest evaluation value among all paths.

[0080] minF(p)=min P∈p (w1·T avg (P)+w2·L(P)+w3·Δθ(P)), where p is the set of all possible paths)

[0081] Where F(P) is the comprehensive evaluation function of path planning scheme P, and T avg (P) is the average tension value on path P, L(P) is the total length of path P, Δθ(P) is the rate of change of angle on path P, and w1, w2 and w3 are weighting coefficients, which respectively characterize the importance of tension uniformity, path length and angle smoothness in the optimization objective; the rate of change of angle Δθ(P) is obtained by calculating the angle difference between adjacent needle entry points and is used to evaluate the smoothness of the suture path.

[0082] Specifically, an optimal suture path is generated through a multi-objective optimization model, comprehensively considering tension uniformity, path length, and angle smoothness. Regarding tension uniformity, the model minimizes the average tension value along the path (target <0.15 N / mm). 2 To avoid stress concentration, the path length parameter limits the total suture path length, reducing tissue damage. Angle smoothness controls the angle variation between adjacent needle entry points (target <45°), ensuring the continuity of the suture operation. Weighting coefficients w1, w2, and w3 are dynamically adjusted according to the wound type. For example, in wounds near joints, w1 is set to 0.6 to prioritize tension uniformity; in areas with rich blood supply, w2 is set to 0.5 to reduce tissue damage. Experiments show that the suture path generated using this model can reduce the maximum wound tension by 25%-35%, while shortening suture time by 10%-15%, significantly improving surgical efficiency and quality.

[0083] Preferably, the preset tension-reducing suture parameter constraints are determined by constructing a parameter correlation model, which is based on the following formula:

[0084]

[0085] Where d is the optimal spacing between adjacent needle insertion points, E is the elastic modulus of the tissue, h is the tissue thickness, and σ maxβ represents the maximum allowable stress of the tissue, and β is an empirical correction coefficient. The parameter correlation model determines the optimal parameter combination that meets the tension reduction requirements by analyzing the relationship between suture parameters under different tissue characteristics.

[0086] Specifically, a parameter correlation model is constructed to determine the constraints of tension-reducing suture parameters. Core parameters include needle insertion point spacing, tissue elastic modulus, thickness, and maximum allowable stress. In practical applications, when the tissue elastic modulus is 20 kPa and the thickness is 2 mm, the optimal needle insertion point spacing calculated by the model is 4 mm; if the elastic modulus increases to 50 kPa, the spacing automatically adjusts to 5 mm. The empirical correction coefficient β is fine-tuned according to the wound location, set to 0.8 for the face (due to more fragile tissue) and 1.0 for the limbs. This model has been validated through extensive clinical data. In 100 cases of wounds at different locations, the wound dehiscence rate when sutured according to the model parameters was 2%, significantly lower than the 8% rate of traditional empirical sutures. Furthermore, the model can dynamically adjust parameters based on factors such as patient age and nutritional status; for example, the needle insertion point spacing is automatically reduced by 1 mm in elderly patients, further improving suture safety.

[0087] Preferably, the construction process of the three-dimensional wound model includes: constructing three-dimensional point cloud data containing tissue hierarchical information based on the multimodal features output by the improved DeepLab semantic segmentation network; performing surface fitting on the three-dimensional point cloud data to generate a continuous three-dimensional wound surface model; and coupling the three-dimensional wound surface model with a pre-established tissue mechanics model to generate a three-dimensional wound model containing mechanical properties.

[0088] Specifically, the construction process of the 3D wound model first involves generating 3D point cloud data using the improved multimodal features output by DeepLab, achieving a point cloud density of 500 points / mm. 2 To ensure the model's detailed accuracy, the surface fitting employs the moving least squares method, resulting in a surface model curvature calculation error of <0.01 / mm, accurately reflecting the wound's morphology. During the tissue mechanics model coupling process, parameters such as elastic modulus and Poisson's ratio are assigned to each mesh node, forming a three-dimensional model with continuously changing physical properties. For example, in burn wounds, the model can distinguish between normal skin (elastic modulus 20 kPa), superficial second-degree burns (10 kPa), and deep second-degree burns (5 kPa), providing precise guidance for layered suturing. In clinical applications, this model helps surgeons plan the anchoring suture positions for deep tissues in advance, increasing the surgical success rate by 18% in complex wound repair.

[0089] Preferably, the process of generating the surgical navigation instruction set includes: converting the optimal suture path planning scheme into a motion trajectory that can be executed by the robot; determining the optimal suture tension value for each needle insertion point based on the tension distribution and a preset suture tension threshold; and encoding the motion trajectory and suture tension value to generate a surgical navigation instruction set containing spatial position information and mechanical parameter information.

[0090] Specifically, the motion trajectory transformation employs a quintic spline interpolation algorithm to ensure the smoothness of the robot's path during execution, with an acceleration change rate <0.5 m / s². 3 The suture tension value is dynamically adjusted based on the calculated tension distribution, for example, in high-tension areas (>0.2 N / mm). 2 A suture tension of 0.3N is used, with low tension areas (<0.1N / mm). 2 The system employs a tension of 0.15N. The instruction encoding uses JSON format and includes position coordinates (accuracy 0.1mm), angle information (accuracy 0.5°), and tension value (accuracy 0.01N). In a smart operating room environment, this instruction set can directly drive the surgical robot to perform suturing operations, with a needle insertion position error of <0.5mm and an angle error of <2°, significantly improving the accuracy and repeatability of the surgery.

[0091] like Figure 2 As shown, the tension-reducing suture surgical navigation system includes:

[0092] A multimodal image data acquisition unit is used to acquire multi-dimensional image data of the wound area to be sutured based on a multimodal image sensing device;

[0093] An improved semantic segmentation processing unit is used to input the multi-dimensional image data into an improved DeepLab semantic segmentation network for feature extraction and analysis;

[0094] The 3D model reconstruction unit is used to construct a 3D wound model containing the skin layer, subcutaneous tissue layer and muscle layer based on the segmentation results output by the improved DeepLab semantic segmentation network.

[0095] The tension analysis and calculation unit is used to calculate the tension distribution at each point on the edge of the wound based on the three-dimensional wound model.

[0096] The suture path planning unit is used to generate an optimal suture path planning scheme based on the tension distribution and preset tension reduction suture parameter constraints.

[0097] The navigation instruction generation unit is used to map the optimal suture path planning scheme to the actual surgical space and generate a set of surgical navigation instructions including the needle entry point position, needle entry angle and suture tension value.

[0098] The output of the multimodal image data acquisition unit is connected to the input of the improved semantic segmentation processing unit, the output of the improved semantic segmentation processing unit is connected to the input of the three-dimensional model reconstruction unit, the output of the three-dimensional model reconstruction unit is connected to the input of the tension analysis and calculation unit, the output of the tension analysis and calculation unit is connected to the input of the suture path planning unit, and the output of the suture path planning unit is connected to the input of the navigation command generation unit.

[0099] The improved semantic segmentation processing unit includes: a multi-scale feature extraction module for extracting wound features at different scales; a spatial attention enhancement module for enhancing the ability to identify wound edge features; a channel attention optimization module for optimizing the expressive ability of features from different channels; and a feature fusion decision module for fusing features from different scales and channels based on an adaptive weight allocation strategy.

[0100] Specifically, the internal structure of the improved semantic segmentation processing unit includes a multi-scale feature extraction module that captures contextual information at different scales through parallel dilated convolutional layers with varying dilation rates (6, 12, and 18), enabling simultaneous identification of microvessels at the 0.5mm level and large areas of necrotic tissue at the 5cm level. The spatial attention enhancement module generates an attention weight matrix by calculating the correlations between locations in the feature map, enhancing the response to slender structures such as wound edges by 30% and improving boundary localization accuracy to 0.1mm. The channel attention optimization module adjusts the feature weights of each channel through global average pooling, global max pooling, and a shared multilayer perceptron, improving the differentiation accuracy of different tissue layers such as skin, fat, and muscle to 94%. The feature fusion decision module dynamically fuses multi-scale features based on an adaptive weight allocation strategy, processing a single frame image in less than 100ms under GPU acceleration, meeting real-time navigation requirements. This overall architecture enables the improved DeepLab network to achieve an average intersection-union ratio of 89.7% in wound segmentation tasks, a 6.2 percentage point improvement over traditional architectures.

[0101] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a tension-reducing suture surgical navigation method.

[0102] Specifically, the computer equipment includes a processor and memory. The processor uses high-performance chips such as the Intel Xeon E5-2680v4 (2.4GHz, 14 cores) or AMD Ryzen Threadripper 3970X (3.7GHz, 32 cores) to ensure real-time performance of multimodal image processing and complex algorithm calculations. The memory is equipped with 16GB of DDR4 ECC RAM (2400MHz read / write speed) and a 512GB NVMe SSD (3500MB / s read / write speed), supporting fast data access. The operating system uses Ubuntu 20.04 LTS or Windows Server 2019 to ensure system stability. This computer equipment connects to multimodal image sensing devices via a PCIe interface, with a data transmission bandwidth of 16GB / s, supporting real-time processing of multimodal data at 30 frames per second. When running the tension-reducing suture surgical navigation program, the system can complete the entire process from image acquisition to navigation command generation within 200ms, meeting the real-time requirements of clinical surgery.

[0103] A storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a tension-reducing suture surgical navigation method.

[0104] Specifically, the computer-readable storage medium employs non-volatile storage technologies such as NAND flash memory or solid-state drives (SSDs), with a storage capacity of ≥512GB, a data read speed of ≥500MB / s, and a write speed of ≥300MB / s, ensuring rapid loading and execution of program code. The stored computer program includes modules for multimodal image preprocessing, improved DeepLab network inference, 3D model reconstruction, tension analysis, path planning, and navigation command generation. The program code is optimized and compiled, occupying <2GB of storage space. This storage medium supports at least 100,000 erase / write cycles, with a data retention period of ≥10 years, ensuring the long-term reliability of medical data. In clinical applications, the storage medium can be connected to computer devices via a USB 3.1 Gen2 interface or a SATA III interface, enabling rapid deployment and updates of tension-reducing suture surgical navigation programs, providing stable software support for surgery.

[0105] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent 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 tension-reducing suture surgical navigation method, characterized in that, Includes the following steps: S1: Acquire multi-dimensional image data of the wound area to be sutured based on a multimodal image sensing device. The multimodal image sensing device includes an ultrasound imaging module, a near-infrared imaging module, and an RGB imaging module. Each module synchronously acquires image data of different characteristics of the same wound area. S2: Input the multi-dimensional image data into the improved DeepLab semantic segmentation network for feature extraction and analysis. The improved DeepLab semantic segmentation network enhances the ability to identify wound edge features and subcutaneous tissue layer features by introducing spatial attention mechanism and channel attention mechanism. S3: Based on the segmentation results output by the improved DeepLab semantic segmentation network, a three-dimensional wound model containing the skin layer, subcutaneous tissue layer and muscle layer is constructed. The three-dimensional wound model contains the geometric morphology information and physical property parameters of each tissue layer. S4: Based on the three-dimensional wound model, calculate the tension distribution at each point on the wound edge. The tension distribution is obtained by analyzing the mechanical simulation results (stress and strain) of the wound model. S5: Based on the tension distribution and the preset tension-reducing suture parameter constraints, generate an optimal suture path planning scheme. The tension-reducing suture parameter constraints include needle insertion point spacing, needle insertion angle, suture depth, and suture tension threshold. S6: Map the optimal suture path planning scheme to the actual surgical space to generate a surgical navigation instruction set containing the needle entry point position, needle entry angle and suture tension value.

2. The tension-reducing suture surgical navigation method according to claim 1, characterized in that, The improved DeepLab semantic segmentation network enhances its ability to recognize features of wounds of different sizes by introducing a multi-scale feature fusion module. This multi-scale feature fusion module employs an adaptive weight allocation strategy to fuse feature maps of different scales. The adaptive weight allocation strategy is determined based on the following model formula: Among them, W i Let A be the fusion weight of the feature map at the i-th scale. i Let α be the channel attention value of the i-th scale feature map, α be the weight adjustment parameter, and n be the number of feature maps; the channel attention value A i The feature map is obtained by analyzing the statistical characteristics of different channels, which is used to characterize the importance of each channel feature to the wound segmentation task.

3. The tension-reducing suture surgical navigation method according to claim 1, characterized in that, The tension distribution is calculated by constructing a tissue mechanics model. The mechanical response of the skin material is described by the strain energy function U. The tissue mechanics simulation is based on the following constitutive equation: Where μ and α are material constants, λ1, λ2, and λ3 are the principal tensile forces in three directions, referring to the strain of the material in three mutually perpendicular principal stress directions. λ1 is along the longest axis of the material, corresponding to the maximum principal tensile force. If the material is a cuboid, this direction is usually along the length of the cuboid; λ2 is along the middle axis of the material, corresponding to the middle principal tensile force. In a cuboid material, this direction is usually along the width of the cuboid; λ3 is along the shortest axis of the material, corresponding to the minimum principal tensile force. The tissue displacement field is obtained by analyzing the tissue deformation features output by the improved DeepLab semantic segmentation network, used to describe the deformation of the tissue under stress. The nominal stress T is given by the following formula:

4. The tension-reducing suture surgical navigation method according to claim 1, characterized in that, The optimal suture path planning scheme is generated by constructing a multi-objective optimization model, which is determined based on the following formula: minF(P)=w1·T avg (P)+w2·L(P)+w3·Δθ(P) Where F(P) is the comprehensive evaluation function of path planning scheme P, and T avg (P) is the average tension value on path P, L(P) is the total length of path P, Δθ(P) is the rate of change of angle on path P, and w1, w2 and w3 are weighting coefficients, which respectively characterize the importance of tension uniformity, path length and angle smoothness in the optimization objective; the rate of change of angle Δθ(P) is obtained by calculating the angle difference between adjacent needle entry points and is used to evaluate the smoothness of the suture path.

5. The tension-reducing suture surgical navigation method according to claim 1, characterized in that, The preset tension-reducing suture parameter constraints are determined by constructing a parameter correlation model, which is based on the following formula: Where d is the optimal spacing between adjacent needle insertion points, E is the elastic modulus of the tissue, h is the tissue thickness, and σ max β represents the maximum allowable stress of the tissue, and β is an empirical correction coefficient. The parameter correlation model determines the optimal parameter combination that meets the tension reduction requirements by analyzing the relationship between suture parameters under different tissue characteristics.

6. The tension-reducing suture surgical navigation method according to claim 1, characterized in that, The construction process of the three-dimensional wound model includes: constructing three-dimensional point cloud data containing tissue hierarchical information based on the multimodal features output by the improved DeepLab semantic segmentation network; performing surface fitting on the three-dimensional point cloud data to generate a continuous three-dimensional wound surface model; and coupling the three-dimensional wound surface model with a pre-established tissue mechanics model to generate a three-dimensional wound model containing mechanical properties.

7. The tension-reducing suture surgical navigation method according to claim 1, characterized in that, The process of generating the surgical navigation instruction set includes: converting the optimal suture path planning scheme into a motion trajectory that can be executed by the robot; determining the optimal suture tension value for each needle insertion point based on the tension distribution and the preset suture tension threshold; and encoding the motion trajectory and suture tension value to generate a surgical navigation instruction set containing spatial position information and mechanical parameter information.

8. A tension-reducing suture surgical navigation system, characterized in that, include: A multimodal image data acquisition unit is used to acquire multi-dimensional image data of the wound area to be sutured based on a multimodal image sensing device; An improved semantic segmentation processing unit is used to input the multi-dimensional image data into an improved DeepLab semantic segmentation network for feature extraction and analysis; The 3D model reconstruction unit is used to construct a 3D wound model containing the skin layer, subcutaneous tissue layer and muscle layer based on the segmentation results output by the improved DeepLab semantic segmentation network. The tension analysis and calculation unit is used to calculate the tension distribution at each point on the edge of the wound based on the three-dimensional wound model. The suture path planning unit is used to generate an optimal suture path planning scheme based on the tension distribution and preset tension reduction suture parameter constraints. The navigation instruction generation unit is used to map the optimal suture path planning scheme to the actual surgical space and generate a set of surgical navigation instructions including the needle entry point position, needle entry angle and suture tension value. The improved semantic segmentation processing unit includes: a multi-scale feature extraction module for extracting wound features at different scales; a spatial attention enhancement module for enhancing the ability to identify wound edge features; a channel attention optimization module for optimizing the expressive ability of features from different channels; and a feature fusion decision module for fusing features from different scales and channels based on an adaptive weight allocation strategy.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images