Artificial skin flap customization design method fusing mirror image optimization and biomechanical analysis

Abstract

The invention provides an artificial skin flap customization design method fusing mirror image optimization and biomechanical analysis. The method comprises the following steps: S1, carrying out dual-region multi-modal data acquisition; s2, fusion generation of a defect area three-dimensional model; s3, performing healthy area mirroring and aesthetic optimization design; s4, designing a bionic blood vessel channel network; and S5, outputting and manufacturing. According to the method, the epidermal form and the subcutaneous fat thickness are integrally and accurately reconstructed, so that the designed skin flap is close to a natural tissue in the aspects of surface profile and internal structure, and the tissue for transplantation is perfectly matched with a recipient defect in the aspects of three-dimensional form and volume.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomedical engineering and digital medical cross, and particularly relates to an artificial flap customized design method fusing mirror image optimization and biomechanical analysis. BACKGROUND

[0002] At present, for complex skin and subcutaneous tissue defects caused by trauma, tumor resection or burns, the gold standard for clinical repair is autologous tissue flap transplantation. This treatment method will cause secondary trauma and new scars to the patient's body, and the autologous tissue available for transplantation is difficult to perfectly match the defect in three-dimensional shape, volume, especially in complex contour parts such as face and joints, the aesthetic effect is often not ideal, and the existing three-dimensional model of the flap generated by the computer has problems such as contour deformity and suture wrinkle, and the tension suture problem between the flap and the recipient area in the natural state increases the risk of complications. SUMMARY

[0003] The present application relates to the technical field of biomedical engineering and digital medical cross, and particularly relates to an artificial flap customized design method fusing mirror image optimization and biomechanical analysis.

[0004] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: an artificial flap customized design method fusing mirror image optimization and biomechanical analysis, comprising the following steps: S1: dual-region multi-modal data acquisition; Synchronously acquire the surface geometric information of the defect area and the healthy symmetrical area of the patient by an optical three-dimensional scanner, and synchronously acquire the subcutaneous structure information of the defect area and the healthy symmetrical area of the patient by a high-frequency ultrasonic imaging device, to generate an epidermis point cloud data file and a DICOM image file of the subcutaneous structure; S2: three-dimensional model fusion generation; Align the epidermis point cloud and the subcutaneous DICOM image in the epidermis point cloud data file and the DICOM image file of the subcutaneous structure in space by recognizing anatomical landmark points, automatically segment the skin, fat and muscle in the DICOM image by using a U-Net network based on deep learning, map the segmented structure data to the three-dimensional grid constructed by the epidermis point cloud to form a healthy area three-dimensional model and a defect area three-dimensional model; S3: healthy area mirror image and aesthetic optimization design; Mirror image the healthy area model with the body median line as the symmetry plane, calculate the boundary contour of the mirror image model and the defect area model by a boundary fusion algorithm, perform smooth transition by using a minimum energy surface algorithm, perform finite element analysis on the model grid according to the input skin tension data collected by the mechanical sensor, adjust the model edge shape, and form a repair body model; S4: bionic vascular channel network design; Through CT scanning, CTA image data of the blood vessels of the patient's affected area is obtained, and the center line and diameter information of the blood vessels of the affected area are extracted from the CTA image by a skeletonization algorithm. On the side of the prosthesis model close to the blood vessels of the affected area, a main entrance seed point is set, and a channel is generated in the model to avoid the edge of the model in a fractal manner from the main entrance to the inside of the model to form a final flap three-dimensional model. S5: output manufacturing; The final flap three-dimensional model is confirmed, the final flap three-dimensional model is layered and converted into a three-dimensional printing file format using slicing software, and output to an external manufacturing device for printing.

[0005] Preferably, the step S2 includes the following steps: S21: multi-modal data space alignment, using SIFT-3D algorithm to extract robust key points and their feature descriptors in the epidermis point cloud data file, extracting skin surface mesh from DICOM sequence in DICOM image file of subcutaneous structure by Marching Cubes algorithm, and extracting key points on the mesh using key point detection algorithm, using nearest neighbor search, matching the descriptors of the two groups of key points to find the initial corresponding point pair, using RANSAC algorithm to randomly select a small number of matching point pairs, calculating a rigid transformation matrix, counting the number of inliers that meet the transformation, finding the transformation matrix with the most inliers through iteration, and filtering out the outliers, using all the inliers filtered out by RANSAC, solving the optimal rigid transformation matrix by SVD method, setting the rotation matrix as R, the translation vector as T, the point in the original point cloud as P_cloud and the point aligned to the DICOM coordinate system as P_dicom_aligned, and the transformation formula as P_dicom_aligned= R * P_cloud + T; S22: automatic segmentation of subcutaneous tissue based on U-Net, adjusting the contrast of the image according to the CT value, normalizing the pixel value to the interval [0, 1], adopting an online enhancement strategy to randomly rotate, translate, scale, elastically deform and add Gaussian noise to increase the robustness of the model for preprocessing DICOM slices, using 2D / 3D U-Net with DiceLoss + Focal Loss on the preprocessed DICOM slices to generate a segmentation probability map with the same size as the input, each pixel containing the probability of belonging to background, skin, fat and muscle categories, performing argmax operation on the output probability map to get the final class label of each pixel, using connected component analysis to remove small, isolated noise regions, using morphological operations to smooth the tissue boundary, filling small holes, and using AdamW optimizer to train the U-Net network; S23: Multi-layer structure data mapping and three-dimensional reconstruction, using the Poisson surface reconstruction algorithm to generate a closed and smooth triangular mesh model from the epidermis point cloud that has been aligned with the DICOM space, voxelizing the space inside the epidermis mesh to create a three-dimensional voxel grid, for each voxel, finding its corresponding position in the original DICOM coordinate system through coordinate transformation, according to the position, querying the result of the U-Net segmentation in step S22, and assigning a tissue label to the voxel, taking the epidermis mesh as the starting point, generating a series of nested meshes inward along the reverse direction of the vertex normal vector through Laplace smoothing and mesh shrinking algorithm, and assigning properties to the nested meshes according to the tissue labels stored in the voxelized data; S24: Health zone and defect zone model separation, the doctor directly outlines the boundary of the defect area on the surface of the model through the interactive device on the reconstructed health zone three-dimensional model, adds a defect category in the U-Net segmentation stage, and after reconstruction, the defect area naturally appears as a hole on the model, establishes a standard filler template mesh similar to the patient's anatomical structure, deforms the template mesh using the non-rigid iterative closest point algorithm to make its boundary accurately match the boundary of the defect area on the health zone model, and the deformed template is the three-dimensional model of the defect area. Based on geometric operations, the defect boundary marked on the health zone model is triangulated to generate an initial and flat patch mesh, and the patch mesh is deformed according to the curvature of the surrounding healthy tissue to form a naturally transitioned and curved surface.

[0006] Preferably, the step S3 comprises the following steps: S31: Based on the geometric mirror of the median line, a series of anatomical landmark points located on the median sagittal plane of the human body are automatically selected as feature points on the three-dimensional model, including the glabella, the tip of the nose, the lip pearl, the chin midpoint, the sternal angle, and the navel. A three-dimensional plane equation Ax + By + Cz + D = 0 is fitted using the least squares method, and the plane is the mirror symmetry plane. Principal component analysis is adopted, and the data points are all the vertices of the entire model. The first principal component direction represents the length, the second represents the width, and the third represents the thickness. The symmetric point V_mirror of each vertex V_health = (x, y, z) on the healthy area model to the mirror plane is calculated, the directed distance of the vertex to the plane is calculated: d = (A*x + B*y +C*z + D) / sqrt(A^2 + B^2 + C^2), the projection of the vertex in the direction of the plane normal vector N = (A, B, C) is calculated: V_proj = V_health - d * N_normalized, N_normalized is the unit normal vector, and the symmetric point coordinates are: V_mirror = V_health - 2 * d * N_normalized; S32: Boundary fusion and smooth transition, on the defect area model, a closed or non-closed polygon boundary ring composed of a series of ordered vertices is extracted, on the mirror model, the area corresponding to the defect area boundary is found through ray collision detection, and the boundary ring matching the area is cut out, the existing ring defect area boundary C_defect and the mirror model boundary C_mirror, using the triangulation algorithm to generate new triangular facets between the two boundary rings, suture, using a smoothing algorithm based on minimum energy surface to process the just sutured area, improvement of Laplace smoothing: solving Δ² (x) = 0, Δ is the Laplace operator, using the conjugate gradient method to solve the linear system: A * X = B, A is a matrix dependent on the connection relationship of the grid, and B is composed of boundary conditions; S33: biomechanical morphology optimization, convert the fused model obtained in step S32 into a finite element mesh using tetrahedral elements, assign corresponding hyperelastic material constitutive models and mechanical parameters to the cells in different regions according to the different tissues segmented in step S2, apply fixed constraints to the healthy tissue regions of the model away from the defect area, map the skin tension data collected by the mechanical sensor to the force vector acting on the corresponding nodes of the defect area boundary, solve the static mechanics equilibrium equation by Newton-Raphson iteration method: [K] {u} = {F}, [K] is the global stiffness matrix related to material properties and geometric shape, {u} is the node displacement vector to be solved, {F} is the node load vector, and the displacement field {u} obtained by finite element analysis is applied to the initial repair body geometric model for model morphology adjustment. For each vertex on the repair body, the new position is: V_optimized = V_initial - u, u is the displacement corresponding to the vertex.

[0007] Preferably, the step S4 comprises the following steps: S41: extraction of the blood vessel topology of the affected area, using the U-Net network trained in step S2, the blood vessel structure in the CTA image is semantically segmented to obtain a binary three-dimensional mask of the blood vessel region, the segmented blood vessel mask is subjected to three-dimensional distance transformation, the voxel value inside each blood vessel is assigned to its distance to the nearest blood vessel boundary, a topology thinning algorithm is used to iteratively erode the blood vessel surface voxel until it only remains a single voxel wide skeleton while ensuring that its topology is not destroyed, the center line of the blood vessel is obtained, the skeleton points are connected to construct a graph structure G = (V, E), the nodes V are the bifurcation points and end points of the blood vessel, the edges E are the blood vessel segments connecting these nodes, the diameter information of each blood vessel segment in the graph is obtained by twice the distance transformation value at all points on the center line of the segment, and the median value is taken as the representative diameter of the blood vessel segment, and the blood vessel graph of the affected area containing node coordinates, edge connection relationship and corresponding diameter of each blood vessel segment is output; S42: main blood vessel channel generation, define a three-dimensional space inside the flap mesh model as the domain of blood vessel growth, set the main entrance seed point P0 as the starting point of growth, and virtually connect it with the nearest trunk blood vessel node in the recipient blood vessel graph, marking the future anastomosis position, define the growth target by uniformly generating a large number of target points inside the model, provide uniform perfusion for the entire flap volume, generate target points: randomly and uniformly sprinkle a large number of target points {T_i} inside the flap model, initialize growth points: the initial growth point set S only contains the main entrance seed point P0, iterative growth: for each target point T_i, find the nearest growth point in the growth point set S, for each growth point S_j, count all target points that take it as the nearest growth point, grow a small distance in the average direction from S_j to these target points to produce a new growth point S_new, diameter calculation: calculate the diameter of the new blood vessel segment according to Murray's law, at the bifurcation, the cube of the parent blood vessel diameter D_p is equal to the sum of the cubes of the child blood vessel diameters D_c1 and D_c2: D_p^3 = D_c1^3 + D_c2^3, termination condition: when the distance between all target points and their nearest growth points is less than a certain threshold, or when the maximum number of iterations is reached, the algorithm stops, in the growth process, introduce distance field constraints to avoid the model edge, calculate the distance from the growth point to the model surface; S43: channel network integration and model generation, convert the blood vessel centerline graph generated by the algorithm into a three-dimensional solid model, each edge in the graph is represented as a capsule or a circular truncated cone, the capsule is defined by the centerline and a constant diameter, the circular truncated cone is defined by the diameters at both ends, use scanning technology to convert these capsules or circular truncated cones into smoothly connected triangular meshes, form the blood vessel channel solid, perform three-dimensional Boolean difference operation between the flap solid model and the blood vessel channel solid model: Final_Flap_Model = Solid_Flap_Model - Vascular_Channel_Solid, excavate the blood vessel channel from the solid flap model to form a hollow pipe network, perform local re-triangulation at the interface generated by the Boolean operation, perform Laplace smoothing on the internal surface of the blood vessel channel and tissue interface to eliminate the printing step effect, finally check whether the final model is a closed manifold.

[0008] Preferably, the step S5 comprises the following steps: S51: Model preprocessing and orientation, use ray casting algorithm to check whether the model is a closed manifold, respectively emit a plurality of rays from the outer part of the model, calculate the number of intersection points with the model, if all the rays are even intersection points, it is a closed manifold, rotate the model around the X and Y axes, find the optimal rotation angle through the simulated annealing algorithm, so that the objective function is minimized: F(orientation) = alpha * (contact area) + beta * (total support volume) + gamma * (Z axis height), alpha, beta, gamma are weight coefficients, the goal is to maximize the contact area of the model and the build platform, while minimizing the volume and total height of the support structure; S52: Slicing, cut a series of planes parallel to the printing platform from the bottom to the top of the model at a fixed layer height, for each cutting plane Z = h, find its intersection line with all triangular facets, use bounding box spatial index to quickly exclude facets that do not intersect with the current plane, for each triangular facet that intersects the plane, calculate the intersection line segment, connect the line segments to form a series of closed polygon rings, which are the cross-sectional profiles of the model at that height, according to the diameter of the printing nozzle, offset the original profile inward to generate the offset outer wall profile, determine the nesting relationship between multiple profiles and plan the printing order, starting from the innermost profile; S53: Path planning, generate continuous outer wall paths along the offset profiles generated in step S52, generate high-density fill patterns inside the solid tissue region, generate several closed paths around the inner profile for the blood vessel channel region to form a solid blood vessel wall, keep the inside of the blood vessel hollow without generating any fill paths, compare the area changes of adjacent slices to automatically detect overhanging structures that need support, generate easy-to-remove support structures below the detected overhanging areas; S54: G-code generation and post-processing, discretize all motion paths into tiny straight line segments, generate G1 commands for each line segment, including: coordinates (x, y, z), speed F, extrusion amount E = (line segment length * nozzle cross-sectional area * extrusion multiplier) / filament_cross_section_area, filament_cross_section_area refers to the cross-sectional area of the printing material inside the nozzle that is about to be extruded, but has not been extruded yet, transfer the final G-code file to the printing device through wired or wireless means.

[0009] To achieve the above purpose, the application also adopts the following technical scheme: a system for customizing and designing artificial flaps by applying fusion image optimization and biomechanical analysis, comprising a multi-modal data acquisition module, a central processing and calculation module, a three-dimensional visualization and human-computer interaction module, and a manufacturing data output module. The multi-modal data acquisition module is an input end, used for acquiring surface geometric information and subcutaneous structure information of the defect area and the healthy symmetrical area of a patient and transmitting to the central processing and calculation module; The central processing and calculation module is used for receiving data from the multi-modal data acquisition module, performing serialization processing by internal sub-modules, outputting a three-dimensional model to the three-dimensional visualization and human-computer interaction module, and receiving modification instructions from the three-dimensional visualization and human-computer interaction module, and the central processing and calculation module comprises a data registration and fusion sub-module, a mirror and aesthetic optimization sub-module, and a bionic blood vessel network design sub-module. The data registration and fusion sub-module is used for receiving data from the multi-modal data acquisition module, accurately registering surface geometric information and subcutaneous structure information, and generating a three-dimensional grid model with thickness attributes. The mirror and aesthetic optimization sub-module is used for geometric mirroring of the healthy area model, calculating the boundary profile of the mirror model and the defect area model, performing smooth transition, performing finite element analysis on the model grid according to skin tension data, adjusting the model edge shape, and forming a repair body model. The bionic blood vessel network design sub-module is used for generating a set of tree-like hierarchical, one end thick and one end small final flap three-dimensional model. The three-dimensional visualization and human-computer interaction module is bidirectionally connected with the central processing and calculation module, used for real-time display of the three-dimensional model generated by the central processing and calculation module, and doctors view, rotate, scale, edit and confirm the automatically generated design. The manufacturing data output module is connected with the central processing and calculation module, after completion of confirmation in the three-dimensional visualization and human-computer interaction module, the manufacturing data output module converts the final flap three-dimensional model into a three-dimensional printing file format and outputs to an external manufacturing equipment for printing.

[0010] Preferably, the multi-modal data acquisition module comprises an optical three-dimensional scanner and a high-frequency ultrasonic imaging device, the optical three-dimensional scanner is used for synchronously acquiring surface geometric information of the defect area and the healthy symmetrical area of a patient, generating epidermis point cloud data files, the surface geometric information comprises surface point cloud data, and the high-frequency ultrasonic imaging device is used for synchronously acquiring subcutaneous structure information of the defect area and the healthy symmetrical area of a patient, the subcutaneous structure information comprises B-ultrasound image sequences of fat layers and fascia layers, and generates DICOM image files of subcutaneous structures.

[0011] Preferably, the three-dimensional visualization and human-computer interaction module comprises a display and an interactive device, and the interactive device is a mouse, a keyboard, a touch screen, a brush or a force feedback device.

[0012] Compared with the prior art, the beneficial effects of the present application are: the present application integrates the epidermis morphology and subcutaneous fat thickness for precise reconstruction by steps S2: defect area three-dimensional model fusion generation, S3: healthy area mirror and aesthetic optimization design and S4: bionic vascular channel network design, so that the designed flap is close to the natural tissue not only in surface contour but also in internal structure, laying a foundation for restoring natural appearance and touch; the deformation caused by tissue tension is automatically compensated by Newton-Raphson iteration method, so that the designed flap can seamlessly connect with the recipient area, effectively avoiding the problems such as contour deformity and suture wrinkles commonly seen in traditional repair, and significantly improving the naturalness of the postoperative appearance; through biomechanical finite element analysis, the influence of skin tension on the morphology of the repair body is simulated in advance, and the edge morphology of the model is adjusted accordingly, so that the flap and the recipient area can be sutured under low tension in a natural state, reducing the risk of complications caused by suture tension; step S4: bionic vascular channel network design is based on the real topological structure of the blood vessels of the patient's recipient area to generate tree-like hierarchical vascular channels, and strictly follows the Murray law to design the diameter of the blood vessels, simulates the perfusion rules of natural blood vessels, provides precise guidance for the rapid ingrowth of host blood vessels, greatly shortens the blood vessel reconstruction time, and improves the survival rate of the flap; the reasonable branch distribution of the bionic vascular network can ensure the uniform and stable blood supply of each region of the transplanted tissue, and reduce long-term complications such as tissue atrophy and pigmentation caused by local insufficient blood supply; through the sequential processing of the central processing and computing module, the present application automates the whole process from data acquisition, model reconstruction, optimization design to manufacturing data output, directly converts clinical needs into executable manufacturing instructions, realizes the automation and standardization of personalized customization, greatly improves the diagnosis and treatment efficiency, and at the same time realizes the perfect match of the transplanted tissue in three-dimensional morphology and volume with the recipient defect; through the precise manufacturing of personalized artificial flaps, the use of autologous tissue is completely replaced, not only avoiding the trauma and scar formation of the donor area, but also solving the problem of insufficient supply of autologous tissue, and being suitable for patients with large area defects or poor donor area conditions. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 The flowchart of the artificial flap customization design method of the present application is shown in the figure.

[0014] Figure 2 The module schematic diagram of the system for realizing the artificial flap customization design method of the present application is shown in the figure. DETAILED DESCRIPTION

[0015] In order to further understand the purpose, structure, features and functions of the present application, the following detailed description is given in conjunction with the embodiments.

[0016] Please refer to Figure 1The application provides a kind of artificial flap customized design method applying fusion mirror image optimization and biomechanics analysis, comprising the following steps: S1: dual-zone multi-modal data acquisition; The surface geometric information of the defect area and the healthy symmetrical area of the patient is synchronously collected by an optical three-dimensional scanner, and the subcutaneous structure information of the defect area and the healthy symmetrical area of the patient is synchronously collected by a high-frequency ultrasonic imaging device, to generate epidermis point cloud data files and subcutaneous structure DICOM image files; S2: three-dimensional model fusion generation; By recognizing anatomical landmarks, the epidermis point cloud in the epidermis point cloud data file and the subcutaneous DICOM image file is spatially aligned with the subcutaneous DICOM image, a U-Net network based on deep learning is used to automatically segment the skin, fat and muscle in the DICOM image, and the segmented structure data is mapped to the three-dimensional grid constructed by the epidermis point cloud to form a healthy area three-dimensional model and a defect area three-dimensional model; S21: multi-modal data spatial alignment, SIFT-3D algorithm is used in the epidermis point cloud data file to extract robust key points and their feature descriptors, skin surface grid is extracted from the DICOM sequence in the subcutaneous structure DICOM image file by Marching Cubes algorithm, and key points are extracted on the grid using key point detection algorithm, nearest neighbor search is used to match the descriptors of the two groups of key points, find the preliminary corresponding point pairs, RANSAC algorithm is used to randomly select a small number of matching point pairs, calculate a rigid transformation matrix, count the number of inliers that meet the transformation, find the transformation matrix with the most inliers through iteration, and filter out the outliers, use all the inliers filtered by RANSAC, solve the optimal rigid transformation matrix by SVD method, set the rotation matrix as R, the translation vector as T, the point in the original point cloud as P_cloud and the point aligned to the DICOM coordinate system as P_dicom_aligned, and the transformation formula is P_dicom_aligned=R*P_cloud+T; S22: Based on the U-Net subcutaneous tissue automatic segmentation, according to the CT value adjustment image contrast, pixel value is normalized to [0, 1] interval, using online enhancement strategy, random rotation ± 10°, translation ± 10%, scaling 0.9-1.1 times, elastic deformation and add Gaussian noise, increase the robustness of the model for pretreatment DICOM slice, using 2D / 3D U-Net on the DICOM slice after pretreatment using Dice Loss + Focal Loss, generate the same size segmentation probability map, each pixel contains the probability of belonging to background, skin, fat, muscle category, the output probability map is argmax operation, get the final class label of each pixel, using connected component analysis to remove small, isolated noise area, using morphological operation smooth tissue boundary, fill small holes, using AdamW optimizer training U-Net network; S23: Multi-layer structure data mapping and three-dimensional reconstruction, the epidermis point cloud has been aligned with the DICOM space, using Poisson surface reconstruction algorithm to generate closed, smooth triangular mesh model, the space inside the above epidermis grid is voxelized to create a three-dimensional voxel grid, for each voxel, through coordinate transformation, find its corresponding position in the original DICOM coordinate system, according to the position, query the result of U-Net segmentation in step S22, give a tissue label to the voxel, take the epidermis grid as the starting point, generate a series of nested grids inward along the inverse direction of the vertex normal vector through Laplace smoothing and grid contraction algorithm, according to the tissue label stored in the voxelization data, assign properties to the nested grid; S24: Health area and defect area model separation, the doctor directly outlines the boundary of the defect area on the model surface through the interactive device on the reconstructed health area three-dimensional model, adds a defect category in the U-Net segmentation stage, after reconstruction, the defect area naturally appears as a hole on the model, establishes a standard, similar to the patient's anatomic structure, fills the template grid, uses non-rigid iterative closest point algorithm to deform the template grid, so that its boundary is accurately matched with the boundary of the defect area on the health area model, the deformed template is the three-dimensional model of the defect area, based on geometric operation, triangulates the defect boundary marked on the health area model to generate the initial, flat patch grid, deforms the patch grid according to the curvature of the surrounding healthy tissue, so that it forms a naturally transitioned, curved surface with the surrounding tissue; S3: Health area mirroring and aesthetic optimization design; Taking the body median line as the symmetry plane, the healthy area model is geometrically mirrored, the boundary contour of the mirrored model and the defect area model is calculated by the boundary fusion algorithm, the smooth transition is performed using the minimum energy surface algorithm, the defect area peripheral skin tension data is collected by the mechanical sensor, the model grid is adjusted according to the input skin tension data, and the edge shape of the model is adjusted to form the restoration model; S31: Based on the geometric mirror of the median line, a series of anatomical landmark points located on the median sagittal plane of the human body are automatically selected as feature points on the three-dimensional model, the anatomical landmark points include the eyebrow center, the nose tip, the lip pearl, the chin midpoint, the sternal angle, and the navel, a three-dimensional plane equation Ax + By + Cz + D = 0 is fitted to the feature points using the least square method, the plane is the mirror symmetry plane, principal component analysis is adopted, the data points are all the vertices of the entire model, the first principal component direction represents the length, the second principal component direction represents the width, and the third principal component direction represents the thickness, the symmetry point V_mirror of each vertex V_health = (x, y, z) on the healthy area model to the mirror plane is calculated, the directed distance of the vertex to the plane is calculated: d = (A*x + B*y + C*z + D) / sqrt(A^2 + B^2 + C^2), the projection of the vertex in the direction of the plane normal vector N = (A, B, C) is calculated: V_proj = V_health - d * N_normalized, N_normalized is the unit normal vector, and the symmetry point coordinates are: V_mirror = V_health - 2 * d * N_normalized; S32: Boundary fusion and smooth transition, on the defect area model, a closed or non-closed polygon boundary ring composed of a series of ordered vertices is extracted, on the mirrored model, the area corresponding to the defect area boundary is found through ray collision detection, and the boundary ring matching the area is cut out, the existing ring defect area boundary C_defect and the mirrored model boundary C_mirror are used to generate new triangular facets between the two boundary rings using a triangulation algorithm, the area is sutured, a smooth algorithm based on the minimum energy surface is used to process the just sutured area, and the improved Laplace smoothing is solved: Δ² (x) = 0, Δ is the Laplace operator, a conjugate gradient method is used to solve the linear system: A * X = B, A is a matrix dependent on the connection relationship of the grid, and B is composed of boundary conditions; S33: Based on biomechanical morphology optimization, the fused model obtained in step S32 is converted into a finite element mesh using tetrahedral elements, different tissues segmented in step S2 are assigned corresponding hyperelastic material constitutive models and mechanical parameters for different regions of the unit, fixed constraints are applied to the healthy tissue area of the model away from the defect area, the skin tension data collected by the mechanical sensor is mapped to the force vector acting on the corresponding nodes of the defect area boundary, and the static mechanical equilibrium equation is solved by Newton-Raphson iteration method: [K] {u} = {F}, [K] is the global stiffness matrix related to material properties and geometric shape, {u} is the node displacement vector to be solved, {F} is the node load vector, and the displacement field {u} obtained by finite element analysis is applied to the initial repair body geometric model for model morphology adjustment. For each vertex on the repair body, the new position is: V_optimized = V_initial - u, u is the displacement corresponding to the vertex; S4: Bionic vascular channel network design; Through CT scanning, the CTA image data of the blood vessels in the patient's affected area is obtained, the center line and diameter information of the blood vessels in the affected area are extracted from the CTA image through a skeletonization algorithm, a main entrance seed point is set on the side of the repair body model close to the blood vessels in the affected area, and a channel is generated in a fractal manner from the main entrance to the inside of the model while avoiding the edge of the model to form a final flap three-dimensional model; S41: Affected area vascular topology extraction, using the U-Net network trained in step S2, the blood vessel structure in the CTA image is semantically segmented to obtain a binary three-dimensional mask of the blood vessel region, the segmented blood vessel mask is subjected to three-dimensional distance transformation, and each blood vessel inside voxel value is assigned its distance to the nearest blood vessel boundary. Using a topology thinning algorithm, iteratively erode the blood vessel surface voxels until only a single voxel wide skeleton is left while ensuring that the topology is not destroyed, the center line of the blood vessel is obtained, the skeleton points are connected to form a graph structure G = (V, E), the nodes V are the bifurcation points and end points of the blood vessels, and the edges E are the blood vessel segments connecting these nodes. The diameter information of each blood vessel segment in the graph is obtained by performing distance transformation on all points on the center line of the segment, and the median value is taken as the representative diameter of the blood vessel segment. The affected area blood vessel graph containing node coordinates, edge connection relationships and corresponding diameters of each blood vessel segment is output; S42: main blood vessel channel generation, define a three-dimensional space inside the flap grid model as the domain of blood vessel growth, set the main entrance seed point P0 as the starting point of growth, and virtually connect it with the nearest trunk blood vessel node in the recipient vascular graph, marking the future anastomosis position, define the growth target by uniformly generating a large number of target points inside the model, provide uniform perfusion for the entire flap volume, generate target points: randomly and uniformly sprinkle a large number of target points {T_i} inside the flap model, initialize growth points: the initial growth point set S only contains the main entrance seed point P0, iterative growth: for each target point T_i, find the nearest growth point in the growth point set S, for each growth point S_j, count all target points that use it as the nearest growth point, grow a small distance in the average direction from S_j to these target points to produce a new growth point S_new, diameter calculation: calculate the diameter of the new blood vessel segment according to Murray's law, at the bifurcation, the cube of the parent blood vessel diameter D_p is equal to the sum of the cubes of the child blood vessel diameters D_c1 and D_c2: D_p^3 = D_c1^3 + D_c2^3, termination condition: when the distance between all target points and their nearest growth points is less than a certain threshold, or when the maximum number of iterations is reached, the algorithm stops, during the growth process, introduce distance field constraints to avoid the model edge, calculate the distance from the growth point to the model surface; S43: channel network integration and model generation, convert the blood vessel centerline graph generated by the algorithm into a three-dimensional solid model, each edge in the graph is represented as a capsule or a circular truncated cone, the capsule is defined by the centerline and a constant diameter, easier to generate, the circular truncated cone is defined by the diameters at both ends, more accurate, use scanning technology to convert these capsules or circular truncated cones into smoothly connected triangular meshes, form the blood vessel channel solid, perform three-dimensional boolean difference operation between the flap solid model and the blood vessel channel solid model: Final_Flap_Model = Solid_Flap_Model - Vascular_Channel_Solid, excavate the blood vessel channel from the solid flap model to form a hollow pipe network, perform local re-triangulation at the interface generated by the boolean operation, perform Laplace smoothing on the internal surface of the blood vessel channel and tissue interface to eliminate the printing step effect, finally check whether the final model is a closed manifold; S5: output manufacturing; Confirm the final flap three-dimensional model, use slicing software to perform layering on the final flap three-dimensional model and convert it into a three-dimensional printing file format, and output to an external manufacturing device for printing.

[0017] Preferably, the step S5 comprises the following steps: S51: Model preprocessing and orientation, use ray casting algorithm to check if the model is a closed manifold, respectively emit multiple rays from the outside of the model, calculate the number of intersection points with the model, if all rays are even intersection points, it is a closed manifold, rotate the model around the X and Y axes, find the optimal rotation angle through the simulated annealing algorithm, so that the objective function is minimized: F(orientation) = α * (contact area) + β * (total support volume) + γ * (Z-axis height), α, β, γ are weight coefficients, the goal is to maximize the contact area of the model with the build platform to increase stability, while minimizing the volume of support structure to save material, reduce damage, and total height to shorten printing time; S52: Slicing, cut the model from bottom to top with a series of planes parallel to the printing platform at a fixed layer height, for each cutting plane Z = h, find its intersection line with all triangular facets, use bounding box spatial index to quickly exclude facets that do not intersect with the current plane, for each triangular facet that intersects the plane, calculate the intersection line segment, connect the line segments to form a series of closed polygon rings, which are the cross-sectional profiles of the model at that height, according to the diameter of the printing nozzle, offset the original profile inward to generate the offset outer wall profile, ensuring the dimensional accuracy of the printed part, determine the nesting relationship between multiple profiles and plan the printing order, starting from the innermost profile; S53: Path planning, generate continuous outer wall paths along the offset profiles generated in step S52, solid tissue regions inside the profile generate high-density fill patterns, including grid, straight line, honeycomb, blood vessel channel regions around the inner profile generate several closed paths to form a solid blood vessel wall, the inside of the blood vessel remains hollow and does not generate any fill paths, compare the area changes of adjacent slices to automatically detect overhanging structures that need support, generate easy-to-remove support structures below the detected overhanging areas, the support material uses hydrogel for easy post-dissolution or physical removal, the support structure uses tree-like support or straight-line support and is filled with a low-density grid pattern to save material and reduce contact points with the model; S54: G-code generation and post-processing, discretize all motion paths, including outer walls, fills, supports, into tiny straight line segments, generate G1 commands for each line segment, including: coordinates (x, y, z), speed F, extrusion amount E = (line segment length * nozzle cross-sectional area * extrusion multiplier) / filament_cross_section_area, filament_cross_section_area refers to the cross-sectional area of the printing material inside the nozzle before it is extruded, transfer the final G-code file to the printing device through wired or wireless means.

[0018] Please refer to Figure 2 The application provides an artificial flap customized design method combining mirror image optimization and biomechanical analysis, comprising a multi-modal data acquisition module, a central processing and calculation module, a three-dimensional visualization and human-computer interaction module and a manufacturing data output module. The multi-modal data acquisition module is an input end, used to collect surface geometric information and subcutaneous structure information of the defect area and the healthy symmetrical area of the patient, and transmit to the central processing and calculation module. The central processing and calculation module is used to receive data from the multi-modal data acquisition module, and after the serialization processing of the internal sub-modules, output the three-dimensional model to the three-dimensional visualization and human-computer interaction module, and receive the modification instructions from the three-dimensional visualization and human-computer interaction module, the central processing and calculation module comprises a data registration and fusion sub-module, a mirror image and aesthetic optimization sub-module, and a bionic blood vessel network design sub-module. The data registration and fusion sub-module is used to receive data from the multi-modal data acquisition module, accurately register the surface geometric information and the subcutaneous structure information, generate a three-dimensional grid model with thickness attribute, and use the multi-modal data acquisition module and the data registration and fusion sub-module to accurately reconstruct the skin morphology and subcutaneous fat thickness, so that the designed flap is not only close to the natural tissue in the surface contour, but also in the internal structure, laying a foundation for restoring the natural appearance and touch; The mirror image and aesthetic optimization sub-module is used to perform geometric mirroring on the healthy area model, calculate the boundary contour of the mirror image model and the defect area model, perform smooth transition, perform finite element analysis on the model grid according to the skin tension data, adjust the model edge shape, form a repair body model, automatically compensate the deformation caused by tissue tension through algorithm, so that the designed flap can seamlessly connect with the recipient area, significantly improve the naturalness of the postoperative appearance, and reduce the risk of complications caused by suture tension; The bionic blood vessel network design sub-module is used to generate a set of tree-like hierarchical, one end thick and one end small final flap three-dimensional model, and pre-design an optimized blood vessel channel in the flap, which can greatly promote the rapid ingrowth of host blood vessels after implantation, solve the vascularization problem of tissue engineering products, and improve the survival rate of the flap; The three-dimensional visualization and human-computer interaction module is bidirectionally connected with the central processing and calculation module, used to display the three-dimensional model generated by the central processing and calculation module in real time, and the doctor checks, rotates, scales, edits and confirms the automatically generated design. The manufacturing data output module is connected with the central processing and calculation module, after confirmation in the three-dimensional visualization and human-computer interaction module, the manufacturing data output module converts the final skin flap three-dimensional model into a three-dimensional printing file format, and outputs to an external manufacturing equipment for printing, directly converts clinical requirements into executable manufacturing instructions, realizes automation and standardization of personalized customization, and greatly improves diagnosis and treatment efficiency.

[0019] In an embodiment, the multi-modal data acquisition module comprises an optical three-dimensional scanner and a high-frequency ultrasonic imaging device, the optical three-dimensional scanner is used for synchronously acquiring surface geometric information of a defect area and a healthy symmetrical area of a patient, generating epidermal point cloud data files, the surface geometric information comprises surface point cloud data, the high-frequency ultrasonic imaging device is used for synchronously acquiring subcutaneous structure information of the defect area and the healthy symmetrical area of the patient, the subcutaneous structure information comprises B-ultrasonic image sequences of a fat layer and a fascia layer, and generating DICOM image files of the subcutaneous structure.

[0020] In an embodiment, the three-dimensional visualization and human-computer interaction module comprises a display and an interactive device, the interactive device is a mouse, a keyboard, a touch screen, a brush or a force feedback device, the display and the interactive device are not innovated in the application, and existing products are adopted.

[0021] The application has been described by the above-mentioned related embodiments, however, the above-mentioned embodiments are only examples for implementing the application. It must be pointed out that the disclosed embodiments do not limit the scope of the application. On the contrary, changes and modifications made without departing from the spirit and scope of the application are all within the scope of the patent protection of the application.

Claims

1. A method of custom design of artificial skin flap by fusing mirror image optimization and biomechanical analysis, characterized in that: Comprising the following steps: S1: Dual-zone multi-modal data acquisition; Synchronously acquiring the surface geometry information of the defect area and the healthy symmetrical area of the patient through an optical three-dimensional scanner, and synchronously acquiring the subcutaneous structure information of the defect area and the healthy symmetrical area of the patient through a high-frequency ultrasonic imaging device, to generate an epidermis point cloud data file and a DICOM image file of the subcutaneous structure; S2: Defect area three-dimensional model fusion generation; By recognizing anatomical landmark points, the epidermis point cloud in the epidermis point cloud data file and the subcutaneous DICOM image file is spatially aligned with the subcutaneous DICOM image, a U-Net network based on deep learning is used to automatically segment the skin, fat and muscle in the DICOM image, the segmented structure data of each layer is mapped to the three-dimensional grid constructed by the epidermis point cloud to form a healthy area three-dimensional model and a defect area three-dimensional model; S3: Healthy area mirroring and aesthetic optimization design; Taking the body median line as the symmetry plane, the healthy area model is geometrically mirrored, the boundary profile of the mirrored model and the defect area model is calculated through a boundary fusion algorithm, a minimum energy surface algorithm is used for smooth transition, the skin tension data of the defect area periphery collected by a mechanical sensor is input, the model grid is subjected to finite element analysis according to the input skin tension data, the model edge morphology is adjusted, and a restoration model is formed; S4: Bionic vascular channel network design; Through CT scanning, CTA image data of the blood vessels of the patient's affected area is obtained, the center line and diameter information of the blood vessels of the affected area are extracted from the CTA image through a skeletonization algorithm, a main entrance seed point is set on the side of the restoration model close to the blood vessels of the affected area, the blood vessels in nature are simulated to grow from the main entrance, the channels are generated in a fractal manner to the inside of the model while avoiding the edge of the model, and a final flap three-dimensional model is formed; S5: Output manufacturing; Confirming the final flap three-dimensional model, using slicing software to perform layering processing on the final flap three-dimensional model, converting it into a three-dimensional printing file format, and outputting it to an external manufacturing device for printing.

2. The method of claim 1, wherein the method is a method of custom design of a free-form skin flap by fusion of mirror image optimization and biomechanical analysis. The step S2 comprises the following steps: S21: Multimodal data space alignment, using SIFT-3D algorithm to extract robust key points and their feature descriptors in the epidermal point cloud data file, extracting the skin surface mesh from the DICOM sequence in the DICOM image file of the subcutaneous structure by the Marching Cubes algorithm, and using the key point detection algorithm on the mesh to extract the key points, using the nearest neighbor search, matching the descriptors of the two sets of key points to find the preliminary corresponding point pairs, using the RANSAC algorithm to randomly select a small number of matching point pairs, calculating a rigid transformation matrix, counting the number of inliers that meet the transformation, finding the transformation matrix with the most inliers through iteration, and filtering out the outliers, using all the inliers filtered out by RANSAC to solve the optimal rigid transformation matrix by SVD method, setting the rotation matrix as R, the translation vector as T, the point in the original point cloud as P_cloud and the point aligned to the DICOM coordinate system as P_dicom_aligned, the transformation formula is P_dicom_aligned = R * P_cloud + T; S22: Automatic segmentation of subcutaneous tissue based on U-Net, adjusting the image contrast according to the CT value, normalizing the pixel value to the [0, 1] interval, using an online enhancement strategy to randomly rotate, translate, scale, elastically deform and add Gaussian noise to increase the robustness of the model for preprocessing DICOM slices, using 2D / 3D U-Net with Dice Loss + Focal Loss on the preprocessed DICOM slices to generate a segmentation probability map with the same size as the input, each pixel contains the probability of belonging to background, skin, fat and muscle categories, performing argmax operation on the output probability map to get the final class label of each pixel, using connected component analysis to remove small, isolated noise regions, using morphological operations to smooth the tissue boundary and fill small holes, using AdamW optimizer to train the U-Net network; S23: Multilayer structure data mapping and three-dimensional reconstruction, using the Poisson surface reconstruction algorithm to generate a closed and smooth triangular mesh model from the epidermal point cloud that has been aligned with the DICOM space, voxelizing the space inside the above epidermal mesh to create a three-dimensional voxel grid, for each voxel, finding its corresponding position in the original DICOM coordinate system through coordinate transformation, according to the position, querying the U-Net segmentation result in step S22, assigning a tissue label to the voxel, taking the epidermal mesh as the starting point, generating a series of nested meshes inward along the inverse direction of the vertex normal vector through Laplace smoothing and mesh contraction algorithm, and assigning properties to the nested meshes according to the tissue labels stored in the voxelization data. S24: The healthy zone and the defect zone model are separated. On the reconstructed healthy zone three-dimensional model, the doctor directly sketches the boundary of the defect area on the model surface through an interactive device. A defect category is added in the U-Net segmentation stage. After reconstruction, the defect area naturally appears as a hole on the model. A standard filling template grid similar to the patient's anatomic structure is established. The non-rigid iterative closest point algorithm is used to deform the template grid so that its boundary is accurately matched with the boundary of the defect area on the healthy zone model. The deformed template is the three-dimensional model of the defect zone. Based on geometric operations, the defect boundary marked on the healthy zone model is triangulated to generate an initial flat patch grid. According to the curvature of the surrounding healthy tissue, the patch grid is deformed to form a naturally transitioned curved surface with the surrounding tissue.

3. The method of claim 1, wherein the method further comprises: determining a thickness of the flap based on the flap thickness map; and determining a donor site based on the flap thickness map. The step S3 comprises the following steps: S31: Based on the geometric mirror of the median line, a series of anatomic landmark points located on the median sagittal plane of the human body are automatically selected as feature points on the three-dimensional model. The anatomic landmark points include the glabella, the tip of the nose, the lip pearl, the chin midpoint, the sternal angle, and the navel. A least squares method is used to fit the feature points into a three-dimensional plane equation Ax + By + Cz + D = 0. The plane is the mirror symmetry plane. Principal component analysis is adopted. The data points are all the vertices of the entire model. The first principal component direction represents the length, the second represents the width, and the third represents the thickness. The symmetric point V_mirror of each vertex V_health = (x, y, z) on the healthy zone model to the mirror plane is calculated. The directed distance of the vertex to the plane is calculated: d = (A*x + B*y + C*z + D) / sqrt(A^2 + B^2 + C^2). The projection of the vertex in the direction of the plane normal vector N = (A, B, C) is calculated: V_proj = V_health - d * N_normalized, N_normalized is the unit normal vector. The symmetric point coordinates are: V_mirror = V_health - 2 * d * N_normalized; S32: Boundary fusion and smooth transition. On the defect zone model, a closed or non-closed polygon boundary ring composed of a series of ordered vertices is extracted. On the mirror model, the area corresponding to the defect zone boundary is found through ray collision detection, and the boundary ring matching the area is cut out. The existing ring defect zone boundary C_defect and the mirror model boundary C_mirror are used to generate new triangular facets between the two boundary rings using a triangulation algorithm. The area is sutured. A smoothing algorithm based on the minimum energy surface is used to process the just sutured area. The improved Laplace smoothing: solving Δ² (x) = 0, Δ is the Laplace operator. The conjugate gradient method is used to solve the linear system: A * X = B. A is a matrix dependent on the connection relationship of the grid, and B is composed of boundary conditions. S33: biomechanical morphology optimization, using tetrahedral elements to convert the fused model obtained in step S32 into a finite element mesh, assigning different tissue regions with corresponding hyperelastic material constitutive models and mechanical parameters, applying fixed constraints to the healthy tissue regions far from the defect area, mapping the skin tension data collected by the mechanical sensor to the force vector acting on the corresponding nodes on the boundary of the defect area, solving the static equilibrium equation by Newton-Raphson iteration method: [K] {u} = {F}, [K] is the global stiffness matrix related to material properties and geometric shape, {u} is the node displacement vector to be solved, {F} is the node load vector, and the displacement field {u} obtained by finite element analysis is applied to the initial repair body geometric model for model morphology adjustment. For each vertex on the repair body, the new position is: V_optimized = V_initial - u, u is the displacement corresponding to the vertex.

4. The method of claim 1, wherein the method further comprises: determining a thickness of the flap based on the flap thickness map; and determining a donor site based on the flap thickness map. The step S4 includes the following steps: S41: extraction of blood vessel topology in the affected area, using the U-Net network trained in step S2 to perform semantic segmentation on the blood vessel structure in the CTA image to obtain a binary 3D mask of the blood vessel region, performing 3D distance transformation on the segmented blood vessel mask, assigning each blood vessel internal voxel value with its distance to the nearest blood vessel boundary, using a topology thinning algorithm to iteratively erode the blood vessel surface voxel until it only remains a single voxel wide skeleton while ensuring that its topology is not destroyed, obtaining the centerline of the blood vessel, connecting the skeleton points to construct a graph structure G = (V, E), the nodes V are the bifurcation points and end points of the blood vessel, and the edges E are the blood vessel segments connecting these nodes, obtaining the diameter information of each blood vessel segment in the graph by doubling the distance transformation value at all points on the centerline of the segment, taking the median value as the representative diameter of the blood vessel segment, and outputting the blood vessel graph in the affected area containing node coordinates, edge connection relationships, and corresponding diameters of each blood vessel segment; S42: main blood vessel channel generation, define a three-dimensional space inside the flap mesh model as the domain of blood vessel growth, set the main entrance seed point P0 as the starting point of growth, and virtually connect it with the nearest trunk blood vessel node in the recipient vascular graph, marking the future anastomosis position, define the growth target by uniformly generating a large number of target points inside the model, provide uniform perfusion for the entire flap volume, generate target points: randomly and uniformly sprinkle a large number of target points {T_i} inside the flap model, initialize growth points: the initial growth point set S only contains the main entrance seed point P0, iterative growth: for each target point T_i, find the nearest growth point in the growth point set S, for each growth point S_j, count all target points that use it as the nearest growth point, grow a small distance in the average direction from S_j to these target points to produce a new growth point S_new, diameter calculation: calculate the diameter of the new blood vessel segment according to Murray's law, at the bifurcation, the cube of the parent blood vessel diameter D_p is equal to the sum of the cubes of the child blood vessel diameters D_c1 and D_c2: D_p^3 = D_c1^3 + D_c2^3, termination condition: when the distance between all target points and their nearest growth points is less than a certain threshold, or when the maximum number of iterations is reached, the algorithm stops, during the growth process, introduce distance field constraints to avoid the model edge, calculate the distance from the growth point to the model surface; S43: channel network integration and model generation, convert the blood vessel centerline graph generated by the algorithm into a three-dimensional solid model, each edge in the graph is represented as a capsule or a circular truncated cone, the capsule is defined by the centerline and a constant diameter, the circular truncated cone is defined by the diameters at both ends, use scanning technology to convert these capsules or circular truncated cones into smoothly connected triangular meshes, forming the blood vessel channel solid, perform a three-dimensional boolean difference operation between the flap solid model and the blood vessel channel solid model: Final_Flap_Model = Solid_Flap_Model - Vascular_Channel_Solid, excavate the blood vessel channel from the solid flap model to form a hollow pipe network, perform local re-triangulation at the interface generated by the boolean operation, and perform Laplace smoothing on the internal surface of the blood vessel channel and the tissue interface to eliminate the printing step effect.

5. The method of claim 1, wherein the method further comprises: determining a thickness of the flap based on the flap thickness map; and determining a donor site based on the flap thickness map. The step S5 includes the following steps: S51: model preprocessing and orientation, use the ray casting algorithm to check whether the model is a closed manifold, respectively emit a plurality of rays from the outside of the model, calculate the number of intersection points with the model, if all rays are even intersection points, it is a closed manifold, rotate the model around the X and Y axes, find the optimal rotation angle through the simulated annealing algorithm to minimize the objective function: F(orientation) = α * (contact area) + β * (total support volume) + γ * (Z-axis height), α, β, γ are weight coefficients, the objective is to maximize the contact area of the model with the build platform, while minimizing the volume and total height of the support structure; S52: slicing, slicing with a series of planes parallel to the printing platform from the bottom to the top of the model at a fixed layer height, for each cutting plane Z = h, find its intersection with all triangular facets, quickly exclude the facets that do not intersect the current plane using the bounding box spatial index, for each triangular facet that intersects the plane, calculate the intersection line segment, connect the line segments to form a series of closed polygon rings, that is, the cross-sectional profile of the model at this height, according to the diameter of the printing nozzle, offset the original profile inward to generate the offset outer wall profile, determine the nesting relationship between multiple profiles and plan the printing order, starting from the innermost profile; S53: path planning, generate continuous outer wall paths along the offset profiles generated in step S52, generate high-density filling patterns inside the solid tissue area, generate several closed paths around the inner profile to form a solid blood vessel wall, and keep the inside of the blood vessel hollow without generating any filling paths, compare the area changes of adjacent slices to automatically detect overhanging structures that need support, and generate easy-to-remove support structures below the detected overhanging areas; S54: G-code generation and post-processing, discretize all motion paths into tiny straight line segments, generate G1 commands for each line segment, including: coordinates (x, y, z), speed F, extrusion amount E = (line segment length * nozzle cross-sectional area * extrusion multiplier) / filament_cross_section_area, filament_cross_section_area refers to the cross-sectional area of the printing material inside the nozzle before being extruded, and finally transmit the G-code file to the printing device through wired or wireless means.

6. A system for customizing and designing an artificial flap using the fusion mirror image optimization and biomechanical analysis method according to claims 1-5, characterized in that: The multi-modal data acquisition module, the central processing and calculation module, the three-dimensional visualization and human-computer interaction module, and the manufacturing data output module are included. The multi-modal data acquisition module is an input end for acquiring surface geometric information and subcutaneous structure information of a patient's defect area and healthy symmetric area and transmitting the information to the central processing and calculation module. The central processing and calculation module is configured to receive data from the multi-modal data acquisition module, perform sequential processing by internal sub-modules, output a three-dimensional model to the three-dimensional visualization and human-computer interaction module, and receive modification instructions from the three-dimensional visualization and human-computer interaction module. The data registration and fusion sub-module is configured to receive data from the multi-modal data acquisition module, accurately register surface geometric information and subcutaneous structure information, and generate a three-dimensional grid model with thickness attributes. The mirror and aesthetic optimization sub-module is configured to geometrically mirror the healthy area model, calculate the boundary profile of the mirrored model and the defect area model, perform smooth transition, perform finite element analysis on the model grid according to skin tension data, adjust the model edge shape, and form a restoration model. The mirror and aesthetic optimization sub-module is configured to geometrically mirror the healthy area model, calculate the boundary profile of the mirrored model and the defect area model, perform smooth transition, perform finite element analysis on the model grid according to skin tension data, adjust the model edge shape, and form a restoration model. The bionic vascular network design submodule is used for generating a set of final flap three-dimensional models with tree-like hierarchical structure, one end thick and the other end thin; The three-dimensional visualization and human-computer interaction module is bidirectionally connected with the central processing and calculation module, and is used for displaying the three-dimensional model generated by the central processing and calculation module in real time, and a doctor views, rotates, scales, edits and confirms the three-dimensional model; The manufacturing data output module is connected with the central processing and calculation module, and after the confirmation in the three-dimensional visualization and human-computer interaction module, the manufacturing data output module converts the final flap three-dimensional model into a three-dimensional printing file format and outputs to an external manufacturing equipment for printing.

7. The method of claim 6, wherein the method further comprises: determining a thickness of the flap based on the flap thickness map; and determining a donor site based on the flap thickness map. The multi-modal data acquisition module comprises an optical three-dimensional scanner and a high-frequency ultrasonic imaging device, the optical three-dimensional scanner is used for synchronously acquiring surface geometric information of a defect area and a healthy symmetrical area of a patient, generating an epidermis point cloud data file, the surface geometric information comprises surface point cloud data, and the high-frequency ultrasonic imaging device is used for synchronously acquiring subcutaneous structure information of the defect area and the healthy symmetrical area of the patient, the subcutaneous structure information comprises a B-ultrasound image sequence of a fat layer and a fascia layer, and generating a DICOM image file of the subcutaneous structure.

8. The method of claim 6, wherein the method further comprises: determining a thickness of the flap based on the flap thickness map; and determining a donor site based on the flap thickness map. The three-dimensional visualization and human-computer interaction module comprises a display and an interactive device, and the interactive device is a mouse, a keyboard, a touch screen, a brush or a force feedback device.

Images