Meridian point foundation construction method for precise positioning of acupuncture robot

Abstract

The invention relates to the field of medical robots, and discloses a meridian point foundation construction method for precise positioning of an acupuncture robot, which comprises the following steps: scanning skin texture, muscle contour and skeleton mark features of a human body surface, and establishing a human body surface feature digital model; the method comprises the following steps: acquiring temperature distribution data of a human body surface, human body surface feature digital model data and elastic modulus data, performing weight mapping feature level fusion to generate an enhanced acupuncture point feature map, and establishing an acupuncture point-meridian association relationship matrix based on the enhanced acupuncture point feature map; compared with the prior art, the method has the advantages that three-dimensional body surface scanning, infrared thermal imaging and elastic modulus data are integrated through the multi-modal data fusion technology to generate the enhanced acupoint feature map, the dynamic meridian-acupoint twinborn model is constructed in combination with double-ellipse section fitting and implicit curved surface reconstruction, and tissue deformation is simulated in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of medical robots, and specifically to a precise positioning method for an acupuncture robot based on multimodal data fusion and intelligent decision-making, which is suitable for the intelligent identification and dynamic adjustment of acupoints in traditional Chinese medicine meridians. Background Art

[0002] Traditional acupuncture therapy relies on the physician's experience and has core problems such as low positioning accuracy (according to statistics from "Clinical Research on Traditional Chinese Medicine Acupuncture", the manual error is >±2mm, and the accuracy rate of deep acupoints is only 62%), poor adaptability to individual differences (the acupoint depth difference between obese and thin patients is 3-5 times), and lack of compensation for dynamic physiological interference (respiratory movement causes acupoint displacement of ±3mm).

[0003] Existing technologies such as ultrasound positioning (deep recognition rate of 78%) rely on a single physical property (acoustic impedance difference), and the deep acupoint recognition rate is only 78% (Journal of Medical Imaging, 2020). Electrical impedance imaging (error >1.5mm) is greatly affected by tissue water content, with an error >1.5mm. Deep learning methods (insufficient generalization for special body types) have insufficient model generalization ability, and the positioning failure rate for special body types (such as patients with edema) exceeds 30%.

[0004] In response to the above-mentioned technological gaps, the present invention solves the three major problems of low precision, poor adaptability and dynamic interference of traditional technologies through multimodal data fusion (integration of temperature, biomechanics, and anatomical structure characteristics), dynamic twin modeling (compensation for respiratory and muscle movement interference) and layered intelligent decision-making (CNN coarse positioning + anatomical attention fine positioning + reinforcement learning dynamic compensation). Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a basic construction method for meridian acupoints for precise positioning of acupuncture robots. By integrating three-dimensional body surface scanning, infrared thermal imaging and elastic modulus detection data to generate enhanced acupoint feature maps, an acupoint-meridian association matrix is constructed; multi-scale spatial attention segmentation and geodesic distance field algorithms are used to extract meridian centerlines and generate a hierarchical tree model; a dynamic meridian-acupoint twin model is established by combining double elliptical section fitting and implicit surface reconstruction technology to simulate tissue deformation; a hierarchical positioning decision framework is constructed to achieve high-precision positioning through CNN coarse positioning, multimodal data fine positioning and sub-millimeter compensation; at the same time, a needle insertion resistance model and a vascular bifurcation point optimization algorithm are established to avoid risk areas, and a dynamic knowledge graph and an adaptive evolutionary algorithm are used to continuously optimize the positioning strategy, ultimately achieving an acupuncture positioning solution with sub-millimeter precision, enhanced individual adaptability and intelligent safety control.

[0006] To achieve the above objectives, one of the present inventions is to provide a method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot, comprising scanning the skin texture, muscle contours, and bone landmark features of the human body surface, establishing a digital model of the human body surface features, collecting temperature distribution data of the human body surface, digital model data of the human body surface features, and elastic modulus data, performing weight mapping feature-level fusion, generating an enhanced acupoint feature map, and establishing an acupoint-meridian association matrix based on the enhanced acupoint feature map;

[0007] The fused enhanced acupoint feature map is segmented using multi-scale and spatial attention techniques. Geodesic distance field calculation and fast marching tree structure construction are performed based on the mask map to obtain the centerline candidate point set of meridian acupoints and the initial meridian acupoint tree model.

[0008] The double ellipse cross-section fitting technique was used to perform three-dimensional biomechanical modeling of the tissues surrounding the acupoints, and a dynamic meridian-acupoint twin model was generated through implicit surface reconstruction.

[0009] A hierarchical decision framework is constructed through the meridian-acupoint twin model and the following positioning operations are performed:

[0010] a. Coarse positioning: Extract key features from human body surface feature data and meridian acupoint models to form a feature set, and preliminarily determine the locations of acupuncture points using a convolutional neural network;

[0011] b. Precise positioning: This involves multimodal fusion of the human body surface feature digital model, the initial meridian and acupoint tree model data, and the acupuncture robot's own sensor data, combined with an anatomical structure attention mechanism to achieve precise positioning.

[0012] c. Positioning parameter adjustment: Generate acupoint location reliability heat map and combine it with the dynamic meridian-acupoint twin model to perform sub-millimeter positioning compensation.

[0013] A needle insertion resistance model is established through a six-dimensional force sensor, and dense neurovascular areas are avoided based on a vascular bifurcation point optimization algorithm. The needle insertion torque of the acupuncture robot is dynamically adjusted in combination with the needle insertion resistance model. A multi-dimensional evaluation matrix is established, and historical operation data and real-time feedback data are fused through a hierarchical encoder to construct a dynamic knowledge graph. Based on the evaluation matrix and knowledge graph, an adaptive evolutionary algorithm is used to continuously optimize the acupoint positioning strategy.

[0014] The second aspect of the present invention is to provide an acupuncture robot and a method for constructing a meridian acupoint foundation based on the precise positioning of the acupuncture robot. The device includes: a multimodal data acquisition module, a dynamic biomechanical modeling module, a hierarchical positioning control module, and a safe path planning module;

[0015] The multimodal data acquisition module is used to achieve high-precision digitization of human surface biometrics and includes three submodules: a 3D structured light scanning unit, an infrared thermal imaging acquisition unit, and a biomechanical detection unit;

[0016] The dynamic biomechanical modeling module is used to construct a three-dimensional model of human tissue deformation prediction and acupoints, and includes three core components: a double elliptical section fitting engine, an implicit surface reconstruction unit, and a transversely isotropic material library;

[0017] The hierarchical positioning control module is used to achieve sub-millimeter acupoint positioning and dynamic compensation, and includes three submodules: a coarse positioning unit, a fine positioning unit, and a dynamic compensation unit;

[0018] The safe path planning module is used to avoid nerves and blood vessels and optimize the needle insertion trajectory. The core technologies include a vascular bifurcation point detector, an RRT*-Connect path planner, and a six-dimensional force feedback system.

[0019] The third aspect of the present invention is to provide a computer device, which includes: a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to realize a basic construction method for meridian acupoints precise positioning of an acupuncture robot.

[0020] A fourth aspect of the present invention is to provide a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by a processor to implement a method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot.

[0021] Compared with the existing technology, the present invention provides a method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot, which has the following beneficial effects:

[0022] This approach uses multimodal data fusion technology to integrate three-dimensional body surface scanning, infrared thermal imaging, and elastic modulus data to generate enhanced acupoint feature maps. It then combines double elliptical cross-section fitting with implicit surface reconstruction to construct a dynamic meridian-acupoint twin model, simulating tissue deformation in real time. It uses multi-scale spatial attention segmentation and geodesic distance field algorithms to extract meridian centerlines and generate a hierarchical tree model. It also constructs a hierarchical positioning framework, achieving high-precision positioning through CNN coarse positioning, multimodal data fine positioning, and submillimeter compensation. It also establishes a needle insertion resistance model and a vascular bifurcation point optimization algorithm, combining a dynamic knowledge graph with an adaptive evolutionary algorithm for continuous optimization. This approach compensates for tissue deformation using a dynamic model, enhances feature recognition using multimodal data, and optimizes path planning using an intelligent algorithm. Ultimately, it achieves submillimeter positioning accuracy (error ≤ 0.6mm), a 98.7% safe path planning success rate, and a 40% improvement in treatment plan optimization efficiency, effectively addressing the core issues of traditional acupuncture positioning, such as strong subjectivity, poor adaptability, and insufficient safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1 This is a schematic diagram of the steps of the meridian acupoint foundation construction method of the present invention;

[0025] Figure 2 Schematic diagram of the steps of human body surface feature enhancement modeling according to the present invention;

[0026] Figure 3 Schematic diagram of the meridian topology reconstruction steps of the present invention. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0028] This paper aims to address the problems of insufficient accuracy (error ±3mm) caused by traditional acupuncture positioning's reliance on experience, the inability of existing equipment static models to adapt to individual differences (fat thickness differences of 5-15mm), and high clinical safety risks (accident rate 0.07%-0.12%).

[0029] This proposal proposes a precise positioning method based on multimodal data fusion and dynamic model optimization.

[0030] By integrating 3D body surface scanning, infrared thermal imaging and elastic modulus data, an enhanced acupoint feature map is generated and an acupoint-meridian association matrix is constructed.

[0031] Multi-scale spatial attention segmentation and geodesic distance field algorithm are used to extract meridian centerlines and generate a hierarchical tree model;

[0032] Combining double elliptical cross-section fitting with implicit surface reconstruction technology to establish a dynamic meridian-acupoint twin model and simulate tissue deformation in real time;

[0033] Build a hierarchical positioning decision framework to achieve high-precision positioning through CNN coarse positioning, multimodal data fine positioning and sub-millimeter compensation;

[0034] At the same time, a needle insertion resistance model and a vascular bifurcation point optimization algorithm are established to avoid risk areas, and a dynamic knowledge graph and adaptive evolutionary algorithm are used to continuously optimize the positioning strategy.

[0035] This solution uses dynamic models to compensate for tissue deformation, multimodal data to enhance feature recognition, and intelligent algorithms to optimize path planning, ultimately achieving submillimeter positioning accuracy (error ≤ 0.6mm), a 98.7% success rate in safe path planning, and a 40% improvement in treatment plan optimization efficiency, effectively addressing the core issues of traditional acupuncture positioning, such as strong subjectivity, poor adaptability, and insufficient safety.

[0036] Example 1, as Figure 1 As shown, the meridian acupoint foundation construction method provided in the embodiment of the present application is exemplified.

[0037] S100: It uses high-precision optical measurement technology to scan the skin texture, muscle contours, and bone landmarks on the human surface. Specifically, it uses a stereoscopic vision system consisting of a 635nm blue laser emitter and a 12-megapixel industrial-grade CMOS camera. It builds a digital model of human surface features, collects temperature distribution data, digital model data of human surface features, and elastic modulus data for weight mapping and feature-level fusion. Using an improved ICP algorithm, it introduces a dual optimization strategy of normal vector constraints and feature descriptor matching to generate an enhanced acupoint feature map. Based on the enhanced acupoint feature map and a hybrid neural network architecture, it establishes an acupoint-meridian association matrix.

[0038] S200: Using the Laplacian pyramid decomposition algorithm, the fused enhanced acupoint feature map is segmented using multi-scale and spatial attention. Based on the mask map, geodesic distance field calculation and fast marching tree structure construction are performed to obtain the centerline candidate point set of meridian acupoints and the initial meridian acupoint tree model.

[0039] S300: 3D biomechanical modeling of tissues surrounding acupoints is performed using a double elliptical cross-section fitting technique. Double elliptical cross-section modeling is based on tissue slice microscopic imaging data, using a nonlinear least squares fitting algorithm. Dynamic meridian-acupoint twin models are generated through implicit surface reconstruction using the moving least squares (MLS) method.

[0040] S400: Build a hierarchical decision framework using the meridian-acupoint twin model and perform the following positioning operations:

[0041] a. Coarse positioning: Extract key features from human body surface feature data and meridian acupoint models to form a feature set. A convolutional neural network is used to preliminarily determine the locations of acupuncture points. The coarse positioning network uses an improved feature pyramid network (FPN).

[0042] b. Precise positioning: This involves multimodal fusion of the human body surface feature digital model, the initial meridian and acupoint tree model data, and the acupuncture robot's own sensor data, combined with an anatomical structure attention mechanism to achieve precise positioning.

[0043] c. Positioning parameter adjustment: Generate acupoint location reliability heat map and combine it with the dynamic meridian-acupoint twin model to perform sub-millimeter positioning compensation.

[0044] S500: A needle insertion resistance model is established through a six-dimensional force sensor. Based on the vascular bifurcation point optimization algorithm, dense neurovascular areas are avoided. The needle insertion torque of the acupuncture robot is dynamically adjusted in combination with the needle insertion resistance model. A multi-dimensional evaluation matrix is established. Historical operation data and real-time feedback data are integrated through a hierarchical encoder to construct a dynamic knowledge graph. Based on the evaluation matrix and knowledge graph, an adaptive evolutionary algorithm is used to continuously optimize the acupoint positioning strategy.

[0045] like Figure 2 The step S100 is to enhance the modeling of human body surface features, including:

[0046] S110: The 3D structured light scanning process used to scan the skin texture, muscle contours, and bone landmark features on the human body surface uses high-precision optical measurement technology. The specific configuration is a stereo vision system composed of a 635nm blue laser emitter and a 12-megapixel industrial-grade CMOS camera. During the scanning process, coded stripes (including a composite pattern of sinusoidal phase grating and Gray code) are projected to achieve submillimeter 3D reconstruction. The point cloud density reaches 500 sampling points per square centimeter. The skin surface curvature feature extraction uses the moving least squares (MLS) method for local surface fitting, calculating the principal curvature radius of each sampling point:

[0047]

[0048] Where L, M, and N are the second fundamental form coefficients of the surface. Infrared thermal imaging data acquisition requires baseline calibration in a constant temperature laboratory (22±0.5°C), and the non-uniformity error of the thermal imager is corrected using a blackbody radiation source. Thermodynamic characteristic analysis uses Fourier's law to establish a heat flux density model: The elastic modulus is measured using a piezoelectric tactile sensor array. Each tactile unit applies a progressive load of 0.1-5N, and the tissue stiffness is inferred using the Hertz contact theory: Data synchronization uses hardware trigger signals to achieve microsecond-level time alignment of multiple devices and establish a unified space-time coordinate system.

[0049] S120: The improved ICP algorithm introduces a dual optimization strategy of normal vector constraint and feature descriptor matching. In the initial registration stage, the FPFH (Fast Point Feature Histogram) descriptor is used to establish a 400-dimensional feature vector for coarse matching:

[0050]

[0051] In the fine registration stage, curvature consistency constraints are added to construct the objective function:

[0052] E=∑w i ||Rp i +tq i || 2 +λ∑||n pi -Rn qi || 2

[0053] In the entropy weight fusion process, the Z-score of each modal data (temperature, elastic modulus, geometric shape) is first normalized to calculate the information entropy of each dimension:

[0054]

[0055] where p(x i ) is obtained through kernel density estimation. Dynamic weight update uses a gated recurrent unit (GRU) network. The input layer receives real-time collected bioelectric signals (such as electromyographic signals) and pressure feedback data. The hidden layer state update formula is:

[0056]

[0057] Realize adaptive adjustment of feature weights.

[0058] S130: The hybrid neural network architecture consists of four convolutional layers (kernel size 3×3×32, stride 2) and a three-layer graph convolutional network (GCN). During the feature extraction phase, the CNN branch processes local texture patterns, while the GCN branch constructs a human surface mesh (vertices ≈ 50,000, edge connection radius 5mm). The graph attention mechanism (GAT) is used to calculate the strength of the association between nodes:

[0059] e ij =LeakyReLU(a T [Wh i ||Wh j ])

[0060] Acupoint area detection uses an improved U-Net architecture, adding a multi-scale feature pyramid in the decoder stage, and the loss function is designed to be a composite form of the Dice coefficient and cross entropy:

[0061]

[0062] The dynamic meridian-acupoint twin model simulates tissue deformation using the finite element method (FEM), and the displacement of each vertex satisfies: Ku = F

[0063] The stiffness matrix K contains the nonlinear hyperelastic material parameters (Mooney-Rivlin model coefficients C 10 =0.3MPa,C 01 =0.1MPa).

[0064] like Figure 3 As shown, this is step S200, which includes:

[0065] The S100's 3D structured light scanning captures skin surface geometric features (such as bony landmarks and skin folds). This data, combined with "high-temperature zone" data captured by an infrared thermal imager (reflecting the path of Qi and blood circulation), is fed into the S200's multi-scale segmentation module. Elastic modulus data (S100 output) serves as a biomechanical constraint, guiding the S200's geodesic path planning, ensuring that the reconstructed meridian centerlines avoid high-hardness tissues.

[0066] S210: The image pyramid is constructed using the Laplacian pyramid decomposition algorithm. The HOG features are calculated in each scale space. The feature response of the first layer is calculated as:

[0067]

[0068] The spatial attention module introduces a channel-spatial dual-path mechanism. The channel attention branch generates channel weights through global average pooling:

[0069] w c=σ(W2·ReLU(W1·GAP(F)))

[0070] The spatial attention branch uses dilated convolution (dilation rate = 3) to expand the receptive field, and the final attention map is activated by Sigmoid to achieve feature enhancement.

[0071] S220: The numerical solution of the Fast Marching Algorithm (FMM) uses the entropy condition satisfaction format to discretize the Eikonal equation:

[0072]

[0073] The velocity function F(x) integrates curvature constraint and tissue stiffness:

[0074]

[0075] Path optimization uses the A* algorithm combined with biomechanical constraints, and the cost function is defined as:

[0076]

[0077] Among them, α=0.7 and β=0.3 are the empirical weight coefficients.

[0078] S230: The L-system grammar rule set contains 20 productions, for example:

[0079] A→A[+B][-B]C; B→BB; C→C[-A]

[0080] The parametric growth model introduces fluid dynamics equations to simulate the energy flow in the meridians:

[0081]

[0082] The degree of freedom of each node in the finite element model includes multiple physical field quantities such as displacement, temperature, and electric potential, and the stiffness matrix update frequency reaches 1kHz.

[0083] It is worth mentioning that the initial meridian tree model generated by S200 (including the topological relationship of 12 main meridians and 8 extraordinary meridians) is cross-validated with the acupoint-meridian association matrix of S100. If the confidence level of the model-predicted "Quchi" acupoint location and the association matrix is lower than the threshold (e.g. <85%), the weight mapping recalculation of S100 is triggered, and the infrared thermal image data weight of the area is increased first (adjusted from 0.4 to 0.6), and the enhanced feature map is regenerated.

[0084] The step S300 includes the following steps:

[0085] S310: Double elliptical cross-section modeling is based on tissue section microscopic imaging data (resolution 0.5μm) using a nonlinear least squares fitting algorithm. The cross-sectional profile of each acupoint area is obtained by scanning 12 sets of orthogonal laser planes to form a discrete point set. The ellipse parameter fitting objective function is defined as:

[0086]

[0087] where x' i =x i -h,y' i =y i -k, parameter to be optimized They represent the coordinates of the ellipse center, the major and minor axes, and the rotation angle, respectively. The Levenberg-Marquardt algorithm is used to iteratively solve the problem. The iterative step length correction formula is:

[0088] Δθ=(J T J+λdiag(J T J)) -1 J T r

[0089] The Jacobian matrix Residual vector The damping coefficient λ is dynamically adjusted according to the trust region radius.

[0090] The constitutive model of anisotropic materials adopts the transverse isotropy assumption, and the stiffness tensor is expressed as:

[0091] C ijkl =λδ ij δ kl +μ(δ ik δ jl +δ il δ jk )+α(a j a l δ kl +a k a l δ ij )+β(a i a k δ jl +a j a l δ ik )

[0092] Where a is the fiber direction vector, λ and μ are Lame constants, and α and β are anisotropy parameters. The parameters were calibrated by nanoindentation experiments: λ = 1.2 GPa, μ = 0.8 GPa, α = 0.3 GPa, and β = 0.2 GPa for the dermis.

[0093] S320: Implicit surface reconstruction uses the moving least squares (MLS) method and defines the implicit function:

[0094]

[0095] Among them, p i (x) = [1, x, y, z] T The linear basis function is used, and the weight function is the compactly supported Wendland function:

[0096]

[0097] Support radius r = 3 mm. By solving the weighted least squares problem:

[0098]

[0099] The coefficient vector a is obtained, and then a continuous implicit field is constructed.

[0100] Dynamic topology maintenance uses the α-shape algorithm to define a three-dimensional α complex:

[0101]

[0102] Where Del(p) is the Delaunay triangulation of the point set P, and the radius α is dynamically adjusted according to the elasticity of the tissue:

[0103]

[0104] E(t) is the Young's modulus of the tissue collected in real time, and α0 = 2 mm is the initial value.

[0105] The step S400 includes:

[0106] S410: The coarse positioning network uses an improved feature pyramid network (FPN). The backbone network is ResNet-101. The feature pyramid contains 5 levels (P3-P7). The multi-scale feature fusion formula is:

[0107] P l =Conv 1×1 (C l )+Upsample 2× (P l+1 )

[0108] Among them C l is the feature map of the lth layer. Uncertainty modeling is implemented through the Bayesian convolution layer, and the weight parameters follow the Gaussian distribution:

[0109]

[0110] The output probability distribution is approximated by Monte Carlo sampling:

[0111]

[0112] where f t (x) is the result of the t-th forward propagation, T = 50 samples.

[0113] S420: The Kalman-Transformer model consists of the following components:

[0114] S421: Kalman filter layer: The state equation and observation equation are:

[0115] x k =Fx k-1 +ω k ,z k =Hx k +v k

[0116] The process noise Observation noise The covariance matrix is calibrated by maximum likelihood estimation.

[0117] S422: Transformer encoder: Contains 6 self-attention layers, and the attention calculation adopts a multi-head mechanism:

[0118] MultiHead(Q,K,V)=Concat(head1,…,head h )W O

[0119] Each attention head has dimension d k =d v =64, the position encoding uses a three-dimensional sine function:

[0120]

[0121] The anatomical attention module introduces prior knowledge constraints and defines the spatial attention mask:

[0122]

[0123] Among them, σ = 5mm controls the attention range, p i are the coordinates of anatomical landmarks.

[0124] S430: The confidence heat map is generated using a Gaussian mixture model (GMM), and the probability density function is:

[0125]

[0126] Where k = 3, corresponding to shallow, middle, and deep tissues, and the covariance matrix ∑k is estimated by the EM algorithm.

[0127] Micro-displacement compensation adopts PID-fuzzy composite controller, and the fuzzy rule base contains 49 items such as:

[0128] IF e is A i ANDΔ e isB j THENu is C ij

[0129] The membership function uses trigonometric function, the defuzzification uses the centroid method, and the PID parameter self-tuning formula is:

[0130]

[0131] Among them, K p0 =0.5,α=0.1.

[0132] The step S500 includes:

[0133] S510: Six-dimensional force sensor data is decomposed by the rotation transformation matrix:

[0134]

[0135] Jacobian matrix Determined by the sensor structure parameters.

[0136] The nonlinear resistance model is fitted using a third-order polynomial:

[0137] R(d)=a0+a1d+a2d 2 +a3d 3

[0138] The coefficient identification is achieved by recursive least squares method, and the objective function is:

[0139]

[0140] The regularization parameter λ = 0.01 prevents overfitting.

[0141] S520: The response function of the improved Harris corner detection algorithm is:

[0142] R=det(M)-k·trace 2 (M)

[0143] in, The determination condition of vascular bifurcation point is R>0.01R max

[0144] The RRT*-Connect algorithm combines biomechanical constraints to define the extended step length:

[0145]

[0146] Among them, d base =2mm,E min =50kPa, the path optimization objective function includes a safety distance penalty term:

[0147]

[0148] Among them, γ = 10, v = 3 mm, and V is the blood vessel point set.

[0149] Experimental Example 1:

[0150] Step 1: Build the experimental platform:

[0151] A silica gel (Ecoflex00-30)-agarose (2% concentration) composite material was used, with a layered structure simulating the epidermis (thickness 0.5 mm, E = 1 MPa), dermis (thickness 2 mm, E = 50 kPa), and muscle (E = 20 kPa). An OptiTrack Prime41 camera array (120 Hz sampling rate, 0.01 mm positioning accuracy) was configured to calibrate the needle tip trajectory error.

[0152] Step 2: Acupoint positioning error analysis:

[0153]

[0154] Measured results: static error 0.08±0.03mm, dynamic error 0.25±0.07mm.

[0155] Step 3: Verify using the coefficient of determination:

[0156]

[0157] Resistance model R 2 =0.97, elastic deformation model R 2 =0.93.

[0158] Step 4: Vascular avoidance test: In a bionic model containing an artificial vascular network (diameter 0.5-2 mm), the minimum distance between the planned path and the blood vessel is ≥ 0.3 mm;

[0159] Force control response test: When applying a sudden resistance (0.5N→3N), the system adjustment time is ≤50ms and the overshoot is ≤5%.

[0160] The experimental technical parameters are as follows:

[0161]

[0162]

[0163] Verified by the China Medical Device Testing Center, this solution has a positioning error of ≤0.6mm (Report No.: CMDA-2023-0582), with a safe path planning success rate of 98.7%. This solution integrates multi-physics coupling modeling, hybrid intelligent algorithms, and high-precision control technologies to create a clinically applicable acupuncture robot system. All mathematical models were validated through in vitro experiments, and key parameters met or exceeded the operational accuracy standards of traditional manual acupuncture (manual acupuncture positioning error is approximately 1-2mm).

[0164] Experimental Example 2:

[0165] Data acquisition (S100): 3D scanning was used to obtain the contour of the tibialis anterior muscle. Infrared thermal imaging showed that there was a high-temperature area with a diameter of 8 mm 3 inches below the knee. Elasticity testing showed that the Young's modulus of the subcutaneous tissue in this area was 45 kPa.

[0166] Model construction (S200-S300): Multi-scale segmentation identifies the overlapping area of the "high temperature zone" and the "bone landmark" and generates the initial acupuncture point coordinates (x0, y0, z0);

[0167] Double ellipse fitting showed that the inclination angle of the fascia layer was 12°, and the implicit surface model predicted that the peak value of the needle insertion resistance was 0.4N.

[0168] Layered positioning (S400): The coarse positioning CNN outputs the candidate area (x0±1.5mm), and the fine positioning module combines the robot's tilt sensor data to correct the final coordinates to (x0+0.2mm, y0-0.3mm, z0);

[0169] The heat map shows deep tissue with a confidence level of 92%, triggering submillimeter compensation (z-axis fine-tuning +0.1mm).

[0170] Safety Control (S500):

[0171] The path planning avoided the anterior tibial artery branch (distance 1.2 mm), and the resistance value (0.38-0.42N) was monitored in real time during needle insertion, with a 95% agreement with the prediction model;

[0172] When the target depth is reached, the six-dimensional force sensor detects the fascia breakthrough feature (resistance drops by 20%), immediately stops the needle insertion and feeds back a completion signal.

[0173] Example 2: An acupuncture robot, based on a method for constructing meridian acupoints with precise positioning of the acupuncture robot, the device includes: a multimodal data acquisition module, a dynamic biomechanical modeling module, a hierarchical positioning control module, and a safe path planning module;

[0174] The multimodal data acquisition module is used to achieve high-precision digitization of human surface biometrics and includes three submodules: a 3D structured light scanning unit, an infrared thermal imaging acquisition unit, and a biomechanical detection unit;

[0175] The dynamic biomechanical modeling module is used to construct a three-dimensional model of human tissue deformation prediction and acupoints, and includes three core components: a double elliptical section fitting engine, an implicit surface reconstruction unit, and a transversely isotropic material library;

[0176] The hierarchical positioning control module is used to achieve sub-millimeter acupoint positioning and dynamic compensation, and includes three submodules: a coarse positioning unit, a fine positioning unit, and a dynamic compensation unit;

[0177] The safe path planning module is used to avoid nerves and blood vessels and optimize the needle insertion trajectory. The core technologies include a vascular bifurcation point detector, an RRT*-Connect path planner, and a six-dimensional force feedback system.

[0178] Embodiment 3: A computer device, comprising: a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement a basic construction method for meridian acupoints precise positioning of an acupuncture robot.

[0179] Embodiment 4: A computer-readable storage medium, characterized in that the storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement a basic construction method for meridian acupoints for precise positioning of an acupuncture robot.

[0180] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot, characterized by: include Scan the skin texture, muscle contours, and bone landmark features of the human body surface to establish a digital model of the human body surface features. Collect temperature distribution data, digital model data of the human body surface features, and elastic modulus data for weight mapping feature-level fusion to generate an enhanced acupoint feature map. Based on the enhanced acupoint feature map, establish an acupoint-meridian association matrix; The fused enhanced acupoint feature map is segmented using multi-scale and spatial attention techniques. Geodesic distance field calculation and fast marching tree structure construction are performed based on the mask map to obtain the centerline candidate point set of meridian acupoints and the initial meridian acupoint tree model. The double ellipse cross-section fitting technique was used to perform three-dimensional biomechanical modeling of the tissues surrounding the acupoints, and a dynamic meridian-acupoint twin model was generated through implicit surface reconstruction. A hierarchical decision framework is constructed through the meridian-acupoint twin model and the following positioning operations are performed: a. Coarse positioning: Extract key features from human body surface feature data and meridian acupoint models to form a feature set, and preliminarily determine the locations of acupuncture points using a convolutional neural network; b. Precise positioning: This involves multimodal fusion of the human body surface feature digital model, the initial meridian and acupoint tree model data, and the acupuncture robot's own sensor data, combined with an anatomical structure attention mechanism to achieve precise positioning. c. Positioning parameter adjustment: Generate acupoint location reliability heat map and combine it with the dynamic meridian-acupoint twin model to perform sub-millimeter positioning compensation. A needle insertion resistance model is established through a six-dimensional force sensor, and dense neurovascular areas are avoided based on a vascular bifurcation point optimization algorithm. The needle insertion torque of the acupuncture robot is dynamically adjusted in combination with the needle insertion resistance model. A multi-dimensional evaluation matrix is established, and historical operation data and real-time feedback data are fused through a hierarchical encoder to construct a dynamic knowledge graph. Based on the evaluation matrix and knowledge graph, an adaptive evolutionary algorithm is used to continuously optimize the acupoint positioning strategy.

2. The method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot according to claim 1, characterized in that: The weight mapping feature-level fusion includes: obtaining human body surface geometric point cloud data through 3D structured light scanning, with the scanning resolution not less than 0.05mm; simultaneously using an infrared thermal imager to capture temperature field distribution data, with the accuracy of the thermal element being ±0.1°C; and combining a piezoelectric tactile sensor array to measure the skin elastic modulus, with the sensor array grid density being 100×100 and the detection range being 1kPa to 10MPa.

3. The method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot according to claim 2, characterized in that: The double ellipse section fitting is based on the Levenberg-Marquardt algorithm to iteratively optimize the ellipse parameters, including the center coordinates, major and minor axes, and rotation angles; the implicit surface reconstruction adopts the moving least squares method, and the weight function uses the compactly supported Wendland function with a support radius of 3 mm.

4. The method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot according to claim 1, characterized in that: The hierarchical decision-making framework includes: in the coarse positioning stage, a feature pyramid network is used to extract multi-scale anatomical features. The network includes a ResNet-101 backbone and a 5-level feature pyramid. In the fine positioning stage, the robot posture data is fused with the biomechanical model output through a Kalman-Transformer model. The Transformer encoder includes a 6-layer self-attention mechanism.

5. The method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot according to claim 1, characterized in that: The precise positioning stage also includes: dynamically adjusting the positioning weight based on the anatomical attention mechanism, assigning a priority coefficient of 0.7-0.9 to bony landmarks, and assigning a suppression coefficient of 0.3-0.5 to vascular dense areas.

6. The method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot according to claim 1, characterized in that: The vascular bifurcation point optimization algorithm adopts the improved RRT*-Connect path planning, defines the path cost function to include the tissue elastic modulus constraint term, and dynamically adjusts the step size to: Among them E min =50kPa is the minimum allowable stiffness.

7. The method for constructing a meridian acupoint foundation for precise positioning of an acupuncture robot according to claim 6, characterized in that: The adaptive evolutionary algorithm includes: generating a Gaussian mixture model confidence heat map and dividing the shallow, middle and deep tissue confidence regions; when the needle tip touches the fascia layer, activating the PID-fuzzy composite controller, and dynamically adjusting the integral term gain with the stress change rate, with an adjustment range of 0.1-0.

5.

8. An acupuncture robot, based on the meridian acupoint foundation construction method for precise positioning of the acupuncture robot according to any one of claims 1 to 7, characterized in that: The device includes: a multimodal data acquisition module, a dynamic biomechanical modeling module, a hierarchical positioning control module, and a safe path planning module; The multimodal data acquisition module is used to achieve high-precision digitization of human surface biometrics and includes three submodules: a 3D structured light scanning unit, an infrared thermal imaging acquisition unit, and a biomechanical detection unit; The dynamic biomechanical modeling module is used to construct a three-dimensional model of human tissue deformation prediction and acupoints, and includes three core components: a double elliptical section fitting engine, an implicit surface reconstruction unit, and a transversely isotropic material library; The hierarchical positioning control module is used to achieve sub-millimeter acupoint positioning and dynamic compensation, and includes three submodules: a coarse positioning unit, a fine positioning unit, and a dynamic compensation unit; The safe path planning module is used to avoid nerves and blood vessels and optimize the needle insertion trajectory. The core technologies include a vascular bifurcation point detector, an RRT*-Connect path planner, and a six-dimensional force feedback system.

9. A computer device, characterized in that: The computer device includes: a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the meridian and acupoint basic construction method for precise positioning of an acupuncture robot as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the meridian and acupoint basic construction method for precise positioning of the acupuncture robot as described in any one of claims 1 to 7.

Images