Method for designing orthosis based on AI artificial intelligence technology

Abstract

The invention belongs to the technical field of application of an artificial intelligence technology to medical rehabilitation instruments, and relates to a method for designing an orthosis based on an AI artificial intelligence technology, and the method comprises the steps: S1, data collection and preprocessing: converting patient data into structured data; s2, feature extraction: extracting geometric and biomechanical features related to orthosis design, and constructing a feature set; s3, processing the features by using a deep learning model, and predicting personalized parameters; s4, designing and optimizing an orthopedic module, and optimizing pressure distribution; s5, a three-dimensional orthopedic CAD system is developed and deployed, and patient information and orthopedic device design are managed; and S6, evaluation and optimization: fusing clinical data, and optimizing the system according to an evaluation result. According to the method, the AI technology is utilized to improve the design efficiency and the adaptation accuracy, the personalized requirements are met, the rehabilitation effect is optimized, digital management of data and iteration of an AI model are achieved, a treatment basis is provided for rehabilitation medical employees such as a brace teacher and a rehabilitation therapist, and the clinical effectiveness of a rehabilitation treatment scheme is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of applying artificial intelligence technology to medical rehabilitation devices, and particularly relates to a method for designing an orthosis based on AI artificial intelligence technology. Background Art

[0002] As an important part of medical rehabilitation devices, orthoses play a key role in improving the physical functions of patients and promoting rehabilitation. By applying force to specific parts of the human body, they correct body deformities, relieve pain, assist limb movement, etc., and are widely used in the rehabilitation treatment of multiple parts such as the spine and limbs. However, there are many drawbacks in the traditional orthosis manufacturing process, which severely restrict its application effect in the medical field. In the traditional manufacturing process, technicians mostly rely on manual operations to obtain patients' body data. For example, they obtain a negative mold by soaking a plaster bandage to wrap the patient's body part, then pour plaster slurry to make a positive mold. Subsequently, a cumbersome mold trimming process is required. Finally, materials such as resin are used to make the orthosis. This not only makes the working environment dirty and messy, but the generated dust also threatens the physical health of technicians. Moreover, this manufacturing method highly depends on manual experience and lacks precise quantitative standards, making it difficult to conduct inspections and verifications through digital means. The manufactured orthoses have poor adaptability and cannot meet the personalized needs of patients. At the same time, the traditional process has a weak awareness of preserving patients' data and cannot effectively trace and evaluate the treatment process later, which is not conducive to optimizing the treatment plan according to the actual situation of patients. In addition, the traditional manufacturing process is inefficient, with a long production cycle and difficult to achieve concurrent operations. This not only prolongs the time for patients to obtain orthoses but may also delay the treatment opportunity.

[0003] In recent years, as a complex and severely harmful three-dimensional spinal deformity disease, scoliosis has a high incidence rate. More than 1,000 international literature studies have been published to research and practically verify the clinical effectiveness of orthoses in the recovery of scoliosis. It is one of the important application fields of orthoses. Scoliosis patients not only bear huge psychological pressure due to obvious body asymmetries such as uneven shoulders, pelvic tilt, and hunchback, but also face many physical health problems. As the condition progresses, the curvature of the spine will continuously worsen, leading to chest deformity, which in turn compresses important organs such as the heart and lungs, restricting the patient's breathing function and significantly decreasing exercise endurance. In severe cases, it even endangers life. As an important non-surgical treatment method, scoliosis orthoses are of particular significance for adolescent patients who are in a critical period of growth and development. It can provide effective external support force, play a corrective or control role in the development of scoliosis, significantly slow down or even prevent the further deterioration of the spinal curvature, greatly reduce the patient's dependence on surgical treatment, and bring hope for the patient's rehabilitation. It is an important guarantee for improving the patient's quality of life. However, the traditional manufacturing process of scoliosis orthoses also has many of the above problems, such as an environmentally unfriendly manufacturing process, poor adaptability, lack of data management, and low efficiency.

[0004] With the rapid development of artificial intelligence technology, its application in the medical field has shown great potential. AI technology can be used to quickly analyze and process a large amount of patient data, extract key features, and thus realize the intelligent and precise design of orthotics, which can significantly improve the design efficiency and adaptability, and optimize the patient's rehabilitation effect. In terms of orthotic design, AI technology can accurately extract the geometric and biomechanical features related to orthotic design by analyzing multi-source information such as patient imaging data and three-dimensional scanning data, construct a parametric model, and design an orthosis that is highly consistent with the patient's normal biomechanical morphology and physical condition. However, the research and practice of applying AI technology to the design of various orthotics is still in the exploratory stage, and a mature and complete technical system has not yet been formed. Therefore, it is of great practical significance to develop a method for designing orthotics based on AI artificial intelligence technology, which can effectively solve the problems existing in traditional design methods, improve the design quality and efficiency of orthotics, better meet the personalized needs of patients, and promote the development of the field of medical rehabilitation. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for designing orthotics based on AI artificial intelligence technology, aiming to solve the design problems of traditional orthotics with the help of AI artificial intelligence technology, improve design efficiency by using AI's ability to quickly process data, accurately extract features through in-depth analysis of multi-source data, build parametric models to improve adaptation accuracy, tap into patients' unique data features to meet personalized needs, optimize rehabilitation effects based on biomechanical principles combined with AI prediction parameters, and realize digital management of patient data and continuous iteration of AI models, so as to provide medical personnel with a scientific basis for treatment and promote the continuous improvement of rehabilitation and orthopedic treatment plans. Specifically:

[0006] 1. With the help of AI's ability to quickly process large amounts of data, the process from patient data collection to orthosis design generation is optimized, the design cycle is greatly shortened, the time-consuming status of traditional manual design is changed, the clinical demand for rapid delivery of orthotics is met, and delays in patient treatment are avoided.

[0007] 2. Through in-depth analysis of multi-source information such as patient imaging data and 3D scanning data, we can accurately extract geometric and biomechanical features related to orthosis design and build a parametric model. Based on this, we can design orthosis that is highly consistent with the patient's morphology and physical condition, solve the problem of poor adaptability caused by traditional design relying on experience and lack of quantitative standards, and enhance the correction effect.

[0008] 3. Utilize AI technology to mine the unique data characteristics of each patient, and provide customized orthosis design solutions according to the differences in the types, degrees, and physical development stages of different patients' needs, so that the orthosis can better adapt to individual needs during the correction process, improving the comfort and compliance of patients wearing it.

[0009] 4. Based on the principles of biomechanics, combined with the personalized design parameters predicted by the AI model, reasonably plan the pressurization, release, and anti-rotation modules of the orthosis to ensure that the correction of the orthosis is more scientific and effective, improving the overall rehabilitation effect.

[0010] To achieve the above object, the technical solution adopted by the present invention is: a method for designing an orthosis based on AI artificial intelligence technology, including the following steps: Step S1: Data collection and preprocessing, collect the relevant data of the patient, and preprocess the collected data to convert it into structured data suitable for subsequent processing;

[0011] Step S2: Feature extraction, extract various features related to orthosis design from the preprocessed data, including geometric features and biomechanical features, and construct a feature set for subsequent design and modeling;

[0012] Step S3: Construct a deep learning model, process the extracted features, and train the model to predict personalized parameters related to orthosis design;

[0013] Step S4: Design and optimization of the orthosis module, design and optimize different functional modules of the orthosis according to the model prediction results, including designs that meet biomechanical standards and correction effects, as well as optimization of pressure distribution;

[0014] Step S5: System development and deployment, develop and deploy a 3D orthosis CAD system containing corresponding functional modules for storing and managing patient information, performing orthosis design operations, and displaying correction plans;

[0015] Step S6: Evaluation and optimization, through the evaluation of the actual use effect, fuse the clinical data with the collected data, and optimize the entire system according to the evaluation results.

[0016] Further, in the step S1, data collection includes obtaining the imaging data of the patient and the three-dimensional scan data of the contour of the body trunk part, that is, obtaining point cloud data. The imaging data includes one or more of X-ray films, CT scans, and MRI images. Preprocessing includes data cleaning, denoising, missing data completion, cropping, and smoothing processing;

[0017] Use the Delaunay triangular mesh algorithm to convert the point cloud data into a polygon mesh, where

[0018] For point cloud data: P = {(x i , y i , z i ) | i = 1, 2, …, N};

[0019] The transformed mesh model is: M = {V, E, F}, where: V = {v1, v2, …, v m} is the vertex set;

[0020] E = {e1, e 2, …, e k} is the edge set; F = {f1, f 2, …, f l} is the face set;

[0021] The adopted mesh generation algorithm is Delaunay triangular patch tessellation;

[0022] It is used for triangulation of point clouds to generate triangular meshes.

[0023] Given the point set P, the triangle set T satisfies:

[0024]

[0025] Furthermore, after data cleaning, geometric features related to orthosis design are extracted to construct parametric features for subsequent design and modeling; the step S2 includes the following sub-steps:

[0026] Step S21: Geometric feature extraction, extracting the external surface contour of the point cloud to generate a smooth skin surface model. By screening and fitting the original point cloud data, a smooth surface representing the external contour of the patient's body part is generated, and the discrete point cloud data is transformed into a continuous surface model;

[0027] Step S22: Biomechanical feature extraction, calculating the curvature distribution of the scanned contour surface, including Gaussian curvature and principal curvature, for evaluating the support of the contact area;

[0028] Assume the surface is parametrically defined as r(u, v):

[0029] r(u, v) = (x(u, v), y(u, v), z(u, v))

[0030] The first fundamental form: used to describe the length on the surface:

[0031] E = r u · r u , F = r u · r v , G = r v · r v , where r u and rv , is the partial derivative of the surface;

[0032] Second fundamental form: Describes the normal variation of the surface:

[0033] e = r uu ·n, f = r uv ·n, g = r vv ·n where n is the normal vector of the surface;

[0034] Curvature calculation:

[0035] Gaussian curvature

[0036] Mean curvature

[0037] Principal curvatures k1, k2: The principal curvatures are obtained by solving a quadratic eigenvalue equation:

[0038] λ 2 -2Hλ + K = 0

[0039] The solution is:

[0040] Step S23: Parametric feature construction, parametric modeling of the bone shape using principal component shape analysis, extracting relevant feature parameters including bending angle, cross-sectional shape, protrusion / depression depth, providing high-dimensional features for the input of the AI model; collecting sample shape point cloud data and alignment:

[0041] Standardization of training data:

[0042] Collect a set of sample shape point cloud data Si, align each sample point cloud;

[0043] Shape mean calculation:

[0044] Calculate the shape mean:

[0045] Covariance matrix construction and eigen-decomposition:

[0046] Calculate the covariance matrix of the shape samples and perform PCA decomposition to obtain eigenvalues and eigenvectors;

[0047] Parametric representation:

[0048] Use the mean shape and the principal components to reconstruct the shape:

[0049]

[0050] where b j, is the shape parameter, which describes the degree of deformation on a specific principal component; the number of principal components k is selected according to the required shape change range and accuracy, and different shape changes can be generated by adjusting the value of b j to achieve a parametric representation of the bone shape;

[0051] Step S24: According to the above three-dimensional model feature extraction, including bony prominence markings and other detection and evaluation data, all training data are labeled, and the following labels can be further updated according to clinical research findings. The current implemented application functions include: patient race, gender, age, disease type, Risser grade, pain grade, flexibility grade, thoracic / lumbar apical vertebra, thoracic / lumbar end vertebra, thoracic / lumbar Cobb angle, thoracic / lumbar ATR angle, trunk deviation direction and distance, pelvic rotation direction and angle, pelvic lordosis / kyphosis angle, thoracic lordosis / kyphosis angle, lumbar lordosis / kyphosis angle.

[0052] Furthermore, the step S3 includes the following sub-steps:

[0053] Step S31: Build a deep learning model based on a 3D convolutional neural network, where

[0054] The mathematical formula applied to the convolutional layer:

[0055]

[0056] where, X j is the input image, W j is the convolutional kernel, b is the bias, and Y i is the convolutional output.

[0057] For 3D convolution, the dimensions of the input and output are 3D, and the convolutional kernel is also 3D. The calculation formula is

[0058]

[0059] where, i, j, k are the positions of the output dimensions, and p, q are the sizes of the convolutional kernel.

[0060] Step S32: According to the requirements of the orthosis design task, use the regression task to output the orthosis design parameters. Adjust the output layer from classification to regression, and predict the design parameters related to the normal force line morphology of the body, including at least the position and intensity of the pressure release module; change the output layer to the form of multiple design parameters;

[0061] Step S33: Use the labeled data to train the model, and perform regression training using the MSE loss function.

[0062] Loss function: Mean Squared Error (MSE) loss, which represents the difference between the model output and the actual design parameters:

[0063]

[0064] Among them, y i is the true value, is the model prediction value, and N is the number of samples.

[0065] Model optimization: Training is performed using the Adam optimizer:

[0066] m t = β1m t-1 + (1 - β1)g t

[0067]

[0068] Among them, m t and v t are the momenta representing the first moment and the second moment respectively, g t is the current gradient, α is the learning rate, β1 and β2 are the momentum coefficients, and ∈ is a constant to avoid division-by-zero errors.

[0069] The data will be divided into a training set, a validation set, and a test set; in each training step, the model will update the parameters through the backpropagation algorithm to minimize the loss function; batch normalization and dropout techniques are used during training to avoid overfitting;

[0070] Step S34: Evaluate the model using the validation dataset, calculate the loss value, and check whether the model is overfitting.

[0071] Furthermore, the step S4 includes the following sub-steps:

[0072] Step S41: Design of the pressurization / release module

[0073] Take the segmented area as the orthosis pressurization module, anti-rotation module, and release module, and design the positions and shapes of the corresponding modules according to the curvature distribution and biomechanical requirements of the body part;

[0074] Curvature analysis: Determine the application methods of various modules by calculating the local curvature of the body. The calculation formula for curvature is

[0075] K = k1 * k2

[0076] where: k1 and k2 are the principal curvatures of the body surface in the local area respectively

[0077] And determine the shape and strength of the pressurization module according to the curvature magnitude.

[0078] Step S42: Use a vector database to store and process vector data, and recommend corresponding design templates for relevant diseases and malformation features. The vector data stored in the vector database includes body geometry and biomechanical feature vectors. Find the most similar vectors through similarity measurement, and generate corresponding orthopedic design templates based on the stored vector data;

[0079] Step S43: Design a transition area between the pressurizing module and the releasing module to optimize the distribution of pressure and release. The pressure distribution formula: P(x) = ∫ Ω σ(x)·ndA, where

[0080] P(x) is the pressure distribution, σ(x) is the stress of the material, and n is the surface normal vector, representing the direction of the pressure.

[0081] Furthermore, in the said Step S5, the 3D orthopedic CAD system includes a data system module for establishing case files, uploading imaging data and 3D files, and graphical prescriptions, and the data is traceable. The specific data storage structure is a hierarchical structure, including levels such as patient information, imaging data, design data, treatment process records, etc. Each level stores different types of data. During the storage process, a database system is used to ensure the security and integrity of the data; it also has the functions of identifying, cropping, and marking 3D files, comparing and calibrating X-ray films, adjusting biomechanical alignment, and loading the correction design module.

[0082] Furthermore, in the said Step S6, by comparing the X-ray films before and after the patient wears a spinal orthosis and the reexamination results at each stage, verify the effect of the model in the actual patient sample, fuse the clinical data with the collected data, and optimize the entire system according to the evaluation results. Specifically, calculate the change in the scoliosis angle Δθ = θ after -θ before , where θ after is the scoliosis angle after wearing the orthosis, and θ before is the scoliosis angle before wearing the orthosis;

[0083] The change in the vertebral rotation angle Δα = α after -α before ; where α after is the vertebral rotation angle after wearing the orthosis, and α before is the vertebral rotation angle before wearing the orthosis;

[0084] The change in the spinal length ΔL = L after -L before , where L after is the spinal length after wearing the orthosis, and L before is the spinal length before wearing the orthosis;

[0085] Evaluate the correction effect of the orthosis and the performance of the model according to these indicators. When fusing clinical data and 3D scan data, match according to the timestamps and patient information of different data, and update the parameters of the model according to the evaluation results.

[0086] Further, in step S6, by comparing the gait data of the patient before and after wearing the lower limb orthosis, including step length, step width, walking speed, gait cycle, changes in the angles of the ankle joint, knee joint, and hip joint, and changes in the center of pressure trajectory COP of the sole; by comparing the changes in range of motion ROM of the joints, surface electromyogram sEMG analysis, and strength test of the patient before and after wearing the upper limb orthosis; evaluate the correction effect of the orthosis and the performance of the model according to these indicators. When fusing clinical data and 3D scan data, match according to the timestamps and patient information of different data, and update the parameters of the model according to the evaluation results.

[0087] The present invention adopts the above technical solutions and has at least the following beneficial effects:

[0088] 1. Precise and efficient data processing and analysis: By comprehensively collecting the patient's imaging data and 3D scan point cloud data, and using advanced data preprocessing technologies, it can effectively remove data noise, complete missing information, and convert it into high-quality meshed data. This not only ensures the accuracy of subsequent analysis, but also greatly improves the data processing efficiency and saves time costs. At the same time, innovatively combining 3D data processing with artificial intelligence deep learning technology realizes the efficient and intelligent conversion from data to design, laying a solid foundation for the precise design of orthoses.

[0089] 2. Comprehensive and in-depth feature extraction: Extract features from multiple dimensions such as geometry and biomechanics, and construct parameterized features using principal component shape analysis, which can more comprehensively and in-depthly reflect the actual situation of scoliosis, providing rich and accurate basis for subsequent orthosis design.

[0090] 3. Significant advantages of the deep learning model: Build a deep learning model based on 3D convolutional neural network, combined with training and optimization strategies, which can accurately predict personalized orthosis design parameters. This model has strong feature learning ability and can adapt to the complex situations of different patients. Compared with traditional methods, it significantly improves the accuracy and pertinence of design. In addition, by adopting multi-model fusion and distributed training, the model performance and design efficiency are further improved, enabling the model to process more complex data in a shorter time and quickly generate high-quality design solutions.

[0091] 4. The orthopedic module is scientifically and reasonably designed: According to the curvature distribution and biomechanical requirements of body parts, the pressurizing, anti-rotation and release modules are designed, and the pressure distribution is optimized, which can better meet the correction needs of patients. This scientific and reasonable design not only improves the correction effect, but also effectively enhances the comfort of patients wearing, and reduces the discomfort caused by wearing orthotics.

[0092] 5. The system functions are complete and practical: The developed 3D orthopedic CAD system integrates various functions, such as case file management, image file processing, biomechanical alignment adjustment, etc., providing a convenient and efficient operation platform for doctors and patients. At the same time, the data traceability of the system also helps the management of medical records and subsequent research and analysis. Moreover, by constructing a design process that combines clinical data and patients' personalized needs, the design is more in line with the actual situation of patients, significantly improving patient satisfaction and rehabilitation effects.

[0093] 6. Broad application prospects: The method of the present invention can be widely applied to the intelligent design fields of other types of orthotics, prosthetic sockets and related rehabilitation equipment, significantly reducing the design cost, improving the rehabilitation effects of patients, and providing a new solution for personalized medicine. By migrating successful technical experiences to the design of other rehabilitation equipment, it is expected to promote the intelligent development of the entire rehabilitation medical industry and benefit more patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0095] Figure 1 It is a flowchart of the method for designing orthotics based on AI artificial intelligence technology of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0096] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are only examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0097] Embodiment 1:

[0098] As Figure 1 shown, this embodiment provides a method for designing orthotics based on AI artificial intelligence technology, including the following steps:

[0099] Step S1: Data collection and preprocessing, collecting relevant data of the patient and preprocessing the collected data to convert it into structured data suitable for subsequent processing;

[0100] Step S2: Feature extraction, extracting various features related to orthosis design from the preprocessed data, including geometric features and biomechanical features, and constructing a feature set for subsequent design and modeling;

[0101] Step S3: construct a deep learning model, process the extracted features, and train the model to predict personalized parameters related to orthosis design;

[0102] Step S4: Design and optimization of orthosis modules, design and optimize different functional modules of the orthosis according to the model prediction results, including the design that meets the biomechanical standards and the correction effect, and the optimization of the pressure distribution;

[0103] Step S5: System development and deployment, developing and deploying a three-dimensional orthopedic CAD system containing corresponding functional modules for storing and managing patient information, performing orthotic design operations, and displaying correction plans;

[0104] Step S6: Evaluation and optimization: by evaluating the actual use effect, the clinical data is integrated with the collected data, and the entire system is optimized according to the evaluation results.

[0105] As a preferred implementation, in step S1 of this embodiment, data acquisition includes obtaining imaging data of the patient and three-dimensional scanning data of the contours of body parts, i.e., obtaining point cloud data, wherein the imaging data includes one or more of X-rays, CT scans, and MRI images, and preprocessing includes data cleaning, denoising, missing data completion, cropping, and smoothing processing;

[0106] The Delaunay triangulation algorithm is used to convert point cloud data into polygonal meshes.

[0107] For point cloud data: P = {(x i ,y i ,z i )|i=1,2,…,N};

[0108] The transformed grid model is: M = {V, E, F}, where: V = {v1, v2, ..., v m} is a vertex set;

[0109] E={e1,e2,…,e k} is an edge set; F = {f1,f2,…,f l} is a face set;

[0110] The adopted grid generation algorithm is Delaunay triangular patch meshing;

[0111] It is used for triangulation of point clouds to generate triangular meshes.

[0112] Given a set of points P, the set of triangles T satisfies:

[0113]

[0114] As a preferred implementation, in this embodiment, geometric features related to orthosis design are extracted after data cleaning to construct parametric features for subsequent design and modeling; the step S2 includes the following sub-steps:

[0115] Step S21: Geometric feature extraction. Extract the external surface contour of the point cloud to generate a smooth skin surface model. By screening and fitting the original point cloud data, a smooth surface representing the external contour of the patient's body part is generated, and the discrete point cloud data is converted into a continuous surface model;

[0116] Step S22: Biomechanical feature extraction. Label relevant parameters of the patient's three-dimensional data, use a deep learning model to assist in identifying key points, input the labeled data into a pre-trained deep learning network, and the network outputs the coordinate information of the key points. According to the anatomical structure and skeletal key points, it provides an accurate position reference for subsequent curvature calculation and feature extraction; calculate the curvature distribution of the scanned contour surface, including Gaussian curvature and principal curvature, to evaluate the support of the contact area; generate a triangular mesh representation of the scanned data; for each vertex, use the neighboring triangular patches to fit a local quadratic surface; use the formula to calculate the Gaussian curvature and principal curvature of each vertex;

[0117] Assume that the surface is parametrically defined as r(u,v):

[0118] r(u,v) = (x(u,v), y(u,v), z(u,v))

[0119] The first fundamental form: used to describe the length on the surface:

[0120] E = r u ·r u , F = r u ·r v , G = r v ·r v , where r u and r v , are the partial derivatives of the surface;

[0121] The second fundamental form: describes the normal variation of the surface:

[0122] e = r uu ·n, f = r uv ·n, g = r vv ·n where n is the normal vector of the surface;

[0123] Curvature calculation:

[0124] Gaussian curvature

[0125] Mean curvature

[0126] Principal curvatures k1, k2: The principal curvatures are obtained by solving the quadratic eigenvalue equation:

[0127] λ 2 -2Hλ + K = 0

[0128] The solution is:

[0129] Step S23: Parametric feature construction. Use principal component shape analysis to parametrically model the bone shape, extract relevant feature parameters including bending angle, cross-sectional shape, protrusion / depression depth, and provide high-dimensional features for the input of the AI model; collect sample shape point cloud data and align them:

[0130] Standardization of training data:

[0131] Collect a set of sample shape point cloud data Si and align each sample point cloud;

[0132] Shape mean calculation:

[0133] Calculate the shape mean:

[0134] Covariance matrix construction and eigenvalue decomposition:

[0135] Calculate the covariance matrix of the shape samples and perform PCA decomposition to obtain eigenvalues and eigenvectors;

[0136] Parametric representation:

[0137] Use the mean shape and principal components to reconstruct the shape:

[0138]

[0139] where b j , is the shape parameter, describing the degree of deformation on a specific principal component; select the number k of principal components according to the required shape change range and accuracy, and different shape changes can be generated by adjusting the value of b j to achieve the parametric representation of the bone shape.

[0140] Step S24: Based on the above extraction of 3D model features, including bony prominence markers and other detection and evaluation data, all training data are labeled. Subsequently, the following labels can be further updated according to clinical research findings. The currently implemented application functions include: patient ethnicity, gender, age, disease type, Risser grade, pain grade, flexibility grade, thoracic / lumbar apical vertebra, thoracic / lumbar end vertebra, thoracic / lumbar Cobb angle, thoracic / lumbar ATR angle, trunk deviation direction and distance, pelvic rotation direction and angle, pelvic lordosis / kyphosis angle, thoracic lordosis / kyphosis angle, lumbar lordosis / kyphosis angle.

[0141] Furthermore, the said Step S3 includes the following sub-steps:

[0142] Step S31: Construct a deep learning model based on a 3D convolutional neural network, where

[0143] The mathematical formula applied to the convolutional layer:

[0144]

[0145] where, X j is the input image, W j is the convolutional kernel, b is the bias, and Y i is the convolutional output.

[0146] For 3D convolution, the dimensions of the input and output are 3D, and the convolutional kernel is also 3D. The calculation formula is:

[0147]

[0148] where, i, j, k are the positions of the output dimensions, and p, q, r are the sizes of the convolutional kernel.

[0149] Step S32: According to the requirements of the orthosis design task, use the regression task to output the orthosis design parameters. Adjust the output layer from classification to regression, and predict the design parameters related to the normal force line morphology of the body, including at least the position and intensity of the pressure release module; change the output layer to the form of multiple design parameters;

[0150] Step S33: Use the labeled data to train the model, and perform regression training using the MSE loss function.

[0151] Loss function: Mean Squared Error (MSE) loss, which represents the difference between the model output and the actual design parameters:

[0152]

[0153] where, y i is the true value, is the model prediction value, and N is the number of samples.

[0154] Model optimization: Training is performed using the Adam optimizer:

[0155] m t = β1m t-1 + (1 - β1)g t

[0156]

[0157] where m t and v t are the momenta representing the first - order moment and the second - order moment respectively, g t is the current gradient, α is the learning rate, β1, β2 are the momentum coefficients, and ∈ is a constant to avoid division - by - zero errors.

[0158] The data will be divided into a training set, a validation set, and a test set; in each training step, the model updates the parameters through the backpropagation algorithm to minimize the loss function; batch normalization and dropout techniques are used during training to avoid overfitting;

[0159] Step S34: Evaluate the model using the validation dataset, calculate the loss value, and check whether the model is overfitting.

[0160] As a preferred implementation manner, step S4 in this embodiment includes the following sub - steps:

[0161] Step S41: Design of the pressurization / release module

[0162] Take the segmented regions as the orthosis pressurization module, anti - rotation module, and release module, and design the positions and shapes of the corresponding modules according to the curvature distribution and biomechanical requirements of the body part;

[0163] Curvature analysis: Determine the application methods of various modules by calculating the local curvature of the body. The formula for curvature is:

[0164] K = k1 * k2

[0165] where: k1 and k2 are the principal curvatures of the body surface in the local region respectively, and the shape and intensity of the pressurization module are determined according to the curvature magnitude.

[0166] Step S42: Use the vector database to store and process vector data, recommend corresponding design templates for related diseases and deformity features. The vector data stored in the vector database includes body geometry and biomechanical feature vectors, find the most similar vectors through similarity measurement, and generate corresponding orthosis design templates according to the stored vector data;

[0167] Step S43: Design a transition area between the pressurizing module and the releasing module to optimize the distribution of pressure and release. The pressure distribution formula is: P(x) = ∫ Ω σ(x)·ndA, where

[0168] P(x) is the pressure distribution, σ(x) is the stress of the material, and n is the surface normal vector representing the direction of the pressure.

[0169] As a preferred implementation, in step S5 of this embodiment, the three-dimensional orthopedic CAD system includes a data system module for establishing case files, uploading image data and three-dimensional files, and graphical prescriptions, and the data is traceable. The specific data storage structure is a hierarchical structure, including levels such as patient information, image data, design data, and treatment process records. Each level stores different types of data. During the storage process, a database system is used to ensure the security and integrity of the data; it also has the functions of identifying, cropping, and marking three-dimensional files, comparing and calibrating X-ray films, adjusting biomechanical alignment, and loading correction design modules.

[0170] As a preferred implementation, in step S6 of this embodiment, by comparing the X-ray films before and after the patient wears the spinal orthosis and the results of stage-by-stage reexaminations, the effect of the model is verified in actual patient samples, the clinical data is fused with the collected data, and the entire system is optimized according to the evaluation results. Specifically, the change in the scoliosis angle is calculated as Δθ = θ after -θ before , where θ after is the scoliosis angle after wearing the orthosis, and θ before is the scoliosis angle before wearing the orthosis;

[0171] The change in the vertebral rotation angle is Δα = α after -α before ; where α after is the vertebral rotation angle after wearing the orthosis, and α before is the vertebral rotation angle before wearing the orthosis;

[0172] The change in the spinal length is ΔL = L after -L before , where L after is the spinal length after wearing the orthosis, and L before is the spinal length before wearing the orthosis;

[0173] Evaluate the correction effect of the orthosis and the performance of the model based on these indicators. When fusing the clinical data and the three-dimensional scan data, match them according to the timestamps and patient information of different data, and update the parameters of the model according to the evaluation results.

[0174] As a preferred embodiment, in step S6 of this embodiment, by comparing the gait data of the patient before and after wearing the lower limb orthosis, including step length, step width, walking speed, gait cycle, changes in ankle, knee, and hip joint angles, and changes in the center of pressure (COP) trajectory of the plantar pressure; by comparing the changes in range of motion (ROM) of joints, surface electromyography (sEMG) analysis, and strength testing of the patient before and after wearing the upper limb orthosis; according to these indicators, the correction effect of the orthosis and the performance of the model are evaluated. When fusing clinical data and three-dimensional scan data, they are matched according to the timestamps and patient information of different data, and the parameters of the model are updated according to the evaluation results.

[0175] Embodiment 2

[0176] Data acquisition and preprocessing: In actual operation, high-precision medical imaging devices such as X-ray, CT, and MRI are used to obtain the patient's imaging data. At the same time, a three-dimensional laser scanner is used to collect the three-dimensional scan point cloud data of the patient's body contour. The collected data is imported into CAD software for data cleaning and denoising. For missing data, an interpolation algorithm based on neighboring points is used to estimate and complete the missing data according to the information of known points. According to the pre-set region of interest, the data is cropped to only retain the part of the data related to the orthosis design. Finally, the data is smoothed to make the data surface smoother, and the processed point cloud data is converted into a polygon mesh to obtain structured data suitable for subsequent processing.

[0177] Feature extraction: The external surface contour of the preprocessed point cloud data is extracted, and through screening and fitting operations, a smooth skin surface model is generated for subsequent geometric feature analysis. Using a pre-trained deep learning model, a key point recognition model based on a convolutional neural network, key points of the patient's three-dimensional data are labeled. According to the definition of surface parameterization, the curvature distribution of the scanned contour surface is calculated, including the first fundamental form, the second fundamental form, Gaussian curvature, mean curvature, and principal curvature. Through these biomechanical features, the force-bearing situation and the deformation characteristics of the force position and force line of the patient's body are deeply understood. A certain number of sample shape point cloud data are collected, and the iterative closest point algorithm (ICP) is used to align them to a common reference coordinate system. The shape mean and covariance matrix are calculated, and the covariance matrix is eigen-decomposed to obtain eigenvalues and eigenvectors. According to the required shape change range and accuracy, an appropriate number of principal components are selected, and the shape is reconstructed using the mean shape and principal components to construct parameterized features for subsequent design and modeling.

[0178] Deep learning model processing: A deep learning model is constructed based on a 3D convolutional neural network. The extracted parametric features are used as input data and fed into the model. According to experience and experimental results, the size and number of convolutional kernels are adjusted, and the number of network layers is determined. The output layer of the model is adjusted from classification to regression, and the activation function of the output layer is modified to a linear function. The number of neurons is determined according to the number of orthosis design parameters to be predicted. A large amount of labeled patient data is used to perform regression training on the model with the mean squared error loss function and the Adam optimizer. The training set, validation set, and test set are divided in a ratio of 7:2:1. During the training process, batch normalization is used to normalize the data after each convolutional layer to prevent gradient vanishing or explosion. At the same time, the dropout technique is adopted, and neurons are randomly discarded with a probability of 0.3 to avoid overfitting. During the training process, the loss value is calculated using the validation dataset, and the loss curves of the validation set and the training set are observed. When the loss value of the validation set starts to rise while the loss value of the training set continues to decline, it is determined as overfitting, and the model parameters and training strategy are adjusted in a timely manner.

[0179] Orthosis module design and optimization: According to the curvature distribution and biomechanical requirements of the corresponding body parts predicted by the model, a pressurization module, a release module, and an anti-rotation module are designed, and the positions and shapes of each module are determined. A vector database is used to store and process vector data containing body geometry and biomechanical feature vectors. By similarity measurement methods such as Euclidean distance or cosine similarity, the most similar vectors are searched to recommend corresponding design templates for each orthosis design. A transition region is designed between the pressurization module and the release module. According to the pressure distribution formula, an optimization function is used to make the actual pressure distribution close to the target pressure distribution. An interpolation function is used to achieve a smooth transition of pressure in the transition region. According to clinical feedback and experimental results, the weights and target pressure distribution are adjusted to achieve the best pressure distribution effect.

[0180] System Development and Deployment: Develop a 3D orthopedic CAD system. In the expert data system module, a hierarchical data storage structure is adopted, and the MySQL database system is used to store patient information, imaging data, design data, treatment process records, etc. When establishing a case file, basic information, diagnosis information, etc. of the patient are entered; during the upload of imaging materials and 3D files, files in different formats are converted into a unified format that can be processed within the system. Using a file format recognition algorithm, 3D files are recognized based on file header information and file characteristics; data in the files are cropped according to the areas specified by the user; marker information is added at the specified positions of the data. When comparing X-ray films, the position differences are calculated based on pre-marked feature points such as vertebral apex points, pedicle points, etc., and calibration evaluation is carried out. According to the mechanical model and patient data, biomechanical alignment adjustment parameters are calculated to adjust the angle and position of the orthosis. The design module is integrated into the orthosis model, and the geometric shape and internal structure of the orthosis are adjusted according to the design parameters of the orthosis. The developed system is deployed on the corresponding server to provide a convenient operation interface for doctors and patients.

[0181] Model Evaluation and Optimization: At specific time nodes, 3 months and 6 months, after the patient wears the orthosis, X-ray films are taken and rechecked. Calculate the change in the scoliosis angle, the change in the vertebral rotation angle, and the change in the spinal length. Calculate the gait data of the patient before and after wearing the lower limb orthosis, including step length, step width, walking speed, gait cycle. Calculate the angle changes of the patient's ankle joint, knee joint, and hip joint, and the change in the center of pressure trajectory COP of the sole. Calculate the range of motion ROM of the joints, surface electromyogram sEMG analysis, strength test, etc. of the patient before and after wearing the upper limb orthosis, and other evaluation indicators. Evaluate the correction effect of the orthosis and the performance of the model based on these indicators. Integrate clinical data with 3D scan data, and match data from different sources according to the unique identifier and timestamp of the patient. According to the evaluation results, optimization algorithms such as gradient descent are used to update the parameters of the model. Continuously iterate and optimize the model based on new patient data and clinical feedback to improve the accuracy and adaptability of the model. Regularly maintain and upgrade the system to ensure the stability and performance of the system.

[0182] Example 3

[0183] This example provides an example for scoliosis and elaborates on the design process of a scoliosis orthosis.

[0184] Data Collection and Preprocessing:

[0185] Use high-precision medical imaging equipment, including X-ray, CT, and MRI, to obtain image data of the patient's spine and trunk. Use a 3D laser scanner to collect 3D scanned point cloud data of the patient's trunk contour. Transmit the collected data to professional CAD software, first perform data cleaning to remove abnormal points caused by equipment noise or external interference. For data missing problems, adopt an interpolation algorithm based on deep learning, such as the method based on generative adversarial network (GAN), to generate the missing data using the characteristics of surrounding data. According to the standard attention areas of scoliosis orthosis design, accurately crop the data to remove data of irrelevant body parts. Use the Gaussian filtering algorithm to smooth the data to make the data surface smoother. Use the Delaunay triangular patch triangulation algorithm to convert the point cloud data into a polygon mesh. During the conversion process, strictly follow the algorithm principle to ensure that the generated triangular mesh can accurately reflect the geometric shape of the patient's trunk.

[0186] Feature extraction:

[0187] Extract the external surface contour from the preprocessed point cloud data through advanced edge detection algorithms and surface fitting techniques to generate a smooth skin surface model. With the help of a key point recognition model based on convolutional neural network (CNN), label the key points of the patient's 3D data, accurately label key anatomical positions such as vertebral body vertices, pedicle points, and spinous process points. According to the definition of surface parameterization, calculate the curvature distribution of the scanned contour surface, including the first fundamental form, the second fundamental form, Gaussian curvature, mean curvature, and principal curvature, so as to deeply analyze the force and deformation characteristics of the spine. Collect a large amount of sample shape point cloud data of scoliosis patients, and use the iterative closest point algorithm (ICP) to align them to a unified reference coordinate system. Calculate the shape mean, covariance matrix, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors. According to clinical experience and research on scoliosis correction, select an appropriate number of principal components, use the mean shape and principal components to reconstruct the shape, and construct parametric features for subsequent design and modeling. At the same time, combined with clinical detection and evaluation data, label all training data, and the labeling content covers patient race, gender, age, scoliosis type (such as idiopathic scoliosis, congenital scoliosis, etc.), Risser grade, pain grade, spinal flexibility grade, thoracic / lumbar apical vertebra, thoracic / lumbar end vertebra, thoracic / lumbar Cobb angle, thoracic / lumbar ATR angle, trunk offset direction and distance, pelvic rotation direction and angle, pelvic lordosis / kyposis angle, thoracic lordosis / kyposis angle, lumbar lordosis / kyposis angle and other information.

[0188] Deep learning model processing:

[0189] Based on a 3D convolutional neural network, a deep learning model is built. According to the characteristics of the scoliosis orthosis design task, the size, number of convolutional kernels, and number of network layers are adjusted. The output layer is changed from classification to regression, the activation function is modified to a linear function, and the number of neurons is determined according to the number of design parameters such as the position and intensity of the pressurization and release modules. Using a large amount of labeled data of scoliosis patients, the model is trained for regression with the mean squared error loss function (MSE) and the Adam optimizer. The training set, validation set, and test set are divided in the ratio of 7:2:1. During the training process, batch normalization is used to normalize the data after each convolutional layer, effectively preventing gradient vanishing or explosion. At the same time, the dropout technique is adopted, and neurons are randomly discarded with a probability of 0.3 to avoid overfitting of the model. During the training process, the loss value is calculated using the validation dataset, and the loss curves of the validation set and the training set are closely observed. Once it is found that the loss value of the validation set starts to rise while the loss value of the training set continues to decline, it is determined as overfitting, and the model parameters and training strategy are adjusted in a timely manner.

[0190] Orthosis Module Design and Optimization:

[0191] According to the predicted spinal curvature distribution and compression requirements of the model, a pressurization module is designed at the convex part of the scoliosis (i.e., the area with a larger curvature), and a release module is designed at the corresponding concave part (the area with a smaller curvature or the opposite direction). According to the torsion angle and the tangent direction of the curve of the spine, the anti-rotation module is accurately designed to ensure that the position and shape of the module can effectively counteract the torsion trend of the spine. A vector database is used to store and process vector data containing the geometric and biomechanical feature vectors of scoliosis. Through similarity measurement methods such as cosine similarity, the most similar vectors are searched to recommend corresponding design templates for different types of scoliosis (such as different degrees of curvature and different types of curvature). A transition area is designed between the pressurization module and the release module. According to the pressure distribution formula P(x) = ∫α(z)·ndA, the actual pressure distribution is made close to the target pressure distribution through an optimization function. The spline interpolation function is used to achieve a smooth transition of the pressure in the transition area. According to the clinical feedback and the results of simulation experiments, the weights and the target pressure distribution are continuously adjusted to achieve the best pressure distribution effect and ensure the comfort and correction effect of the patient's wearing.

[0192] System Development and Deployment:

[0193] Develop a 3D orthopedic CAD system. In the data system module, a hierarchical data storage structure is adopted, and the MySQL database system is used to store patient information, imaging data, design data, treatment process records, etc. When establishing a case file, basic patient information, diagnosis information, past treatment records, etc. are entered; during the upload process of imaging materials and 3D files, files in different formats are converted into a unified format that can be processed within the system. Using a file format recognition algorithm, 3D files are recognized based on file header information and file characteristics; data in the files are cropped according to the areas specified by the user; marking information is added at the specified positions of the data. When comparing X-ray films, the position differences are calculated based on pre-annotated feature points such as vertebral apex points, pedicle points, etc., and calibration evaluation is carried out. According to the mechanical model and patient data, biomechanical alignment adjustment parameters are calculated to adjust the angle and position of the orthosis. The design module is integrated into the orthosis model, and the geometric shape and internal structure of the orthosis are adjusted according to the design parameters of the orthosis. The developed system is deployed on a cloud server, and through a secure network connection, a convenient operation interface is provided for doctors and patients, facilitating doctors to perform orthosis design operations and display correction plans, and patients can also view their treatment-related information through authorized access.

[0194] Model evaluation and optimization:

[0195] At key time points such as 3 months and 6 months after the patient wears a scoliosis orthosis, X-ray films are taken and a comprehensive review is carried out. Calculate the change amount of the scoliosis angle Δθ = θ after - θ before 、the change amount of the vertebral rotation angle Δα = α after - α before and the change amount of the spinal length ΔL = L after - L before and other evaluation indicators. Integrate clinical data with 3D scan data, and match data from different sources according to the unique identifier and timestamp of the patient. According to the evaluation results, optimization algorithms such as stochastic gradient descent are used to update the parameters of the model. According to new patient data and clinical feedback, continuously iterate and optimize the model to improve the accuracy and adaptability of the model for scoliosis orthosis design. At the same time, regularly maintain and upgrade the system to ensure the stability and performance of the system. According to the problems found in actual use and new requirements, expand and improve the functions of the system, such as adding more data analysis and visualization functions to facilitate doctors to more intuitively understand the patient's correction progress and model effects.

[0196] Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for designing an orthosis based on AI artificial intelligence technology, characterized in that: The following steps are involved: Step S1: Data collection and preprocessing, collecting relevant data of the patient and preprocessing the collected data to convert it into structured data suitable for subsequent processing; Step S2: Feature extraction, extracting various features related to orthosis design from the preprocessed data, including geometric features and biomechanical features, and constructing a feature set for subsequent design and modeling; Step S3: construct a deep learning model, process the extracted features, and train the model to predict personalized parameters related to orthosis design; Step S4: Design and optimization of orthosis modules: design and optimize different functional modules of the orthosis according to the model prediction results, including the design of pressurization, release and anti-rotation modules that meet biomechanical standards and correction effects, and optimization of pressure distribution; Step S5: System development and deployment, developing and deploying a three-dimensional orthopedic CAD system containing corresponding functional modules for storing and managing patient information, performing orthotic design operations, and displaying correction plans; Step S6: Evaluation and optimization: by evaluating the actual use effect, the clinical data is integrated with the collected data, and the entire system is optimized according to the evaluation results.

2. The method according to claim 1, characterized in that: In step S1, data acquisition includes obtaining imaging data of the patient and three-dimensional scanning data of the contours of body parts, i.e., obtaining point cloud data, wherein the imaging data includes one or more of X-rays, CT scans, and MRI images, and preprocessing includes data cleaning, denoising, missing data completion, cropping, and smoothing; The Delaunay triangulation algorithm is used to convert point cloud data into polygonal meshes. For point cloud data: P = {(x i ,y i ,z i )|i=1,2,…,N}; The transformed grid model is: M = {V, E, F}, where: V = {v1, v2, ..., v m } is a vertex set; E={e1,e 2, …, e k } is an edge set; F = {f1,f 2, …, f l } is a face set; The mesh generation algorithm used is Delaunay triangulation; Used for triangulation of point clouds to generate triangular meshes; Given a set of points P, a set of triangles T satisfies:

3. The method according to claim 1, characterized in that: After data cleaning, geometric features related to orthosis design are extracted to construct parametric features for subsequent design and modeling; step S2 includes the following sub-steps: Step S21: geometric feature extraction, extracting the external surface contour of the point cloud, generating a smooth skin surface model, generating a smooth surface representing the external contour of the patient's body part by screening and fitting the original point cloud data, and converting the discrete point cloud data into a continuous surface model; Step S22: extracting biomechanical features, calculating the curvature distribution of the scanned contour surface including Gaussian curvature and principal curvature, which is used to evaluate the support of the contact area; Assume that the surface is defined by parameterization as r(u,v): r(u,v)=(x(u,v),y(u,v),z(u,v)) First basic form: used to describe the length on the surface: E=r u ·r u ,F=r u ·r v ,G=r v ·r v , where r u and r v , is the partial derivative of the surface; The second basic form: describes the change of surface normal: e=r uu n,f=r uv n,g=r vv n where n is the normal vector of the surface; Curvature calculation: Gaussian curvature Mean curvature Principal curvatures k1, k2: The principal curvatures are obtained by solving the quadratic characteristic equation: l 2 -2Hλ+K=0 The solution is: Step S23: parametric feature construction, using principal component shape analysis to perform parametric modeling on the bone shape, extracting relevant feature parameters including bending angle, cross-sectional shape, and convex / concave depth, to provide high-dimensional features for AI model input; Collect sample shape point cloud data and align: Normalization of training data: Collect a set of sample shape point cloud data Si and align each sample point cloud; Shape mean calculation: Compute the shape mean: Covariance matrix construction and eigendecomposition: Calculate the covariance matrix of the shape samples and perform PCA decomposition to obtain eigenvalues ​​and eigenvectors; Parameterized representation: Use Mean Shape And principal components to reconstruct the shape: where b j , is a shape parameter that describes the degree of deformation on a specific principal component; the number of principal components k is selected according to the required shape variation range and accuracy, and b is adjusted by j The value of can generate different shape changes to achieve parameterized representation of bone shape; Step S24: All training data are labeled based on the above three-dimensional model feature extraction, including bone markers and other detection and evaluation data. The following annotations can be further updated based on clinical research findings. The current application functions include: patient race, gender, age, disease type, Risser grade, pain grade, flexibility grade, thoracic / lumbar apical vertebra, thoracic / lumbar distal vertebra, thoracic / lumbar Cobb angle, thoracic / lumbar ATR angle, trunk offset direction and distance, pelvic rotation direction and angle, pelvic lordosis / kyphosis angle, thoracic lordosis / kyphosis angle, lumbar lordosis / kyphosis angle.

4. The method according to claim 1, characterized in that: The step S3 comprises the following sub-steps: Step S31: Building a deep learning model based on a 3D convolutional neural network, where Mathematical formula applied to the convolution layer: Among them, X j is the input image, W j is the convolution kernel, b is the bias, Y i is the convolution output. For 3D convolution, the dimensions of input and output are 3D, the convolution kernel is also 3D, and the calculation formula is: Among them, i, j, k are the positions of the output dimension, and p, q, r are the sizes of the convolution kernel. Step S32: according to the requirements of the orthosis design task, the orthosis design parameters are output using the regression task; the output layer is adjusted from classification to regression, and the design parameters related to the normal force line shape of the body are predicted to include at least the position and strength of the pressurization release module; the output layer is changed into the form of multiple design parameters; Step S33: Use the labeled data to train the model and use the MSE loss function for regression training; Loss function: Mean squared error (MSE) loss, which represents the difference between the model output and the actual design parameters: Among them, y i is the true value, is the model prediction value, and N is the number of samples. Model optimization: Use Adam optimizer for training: Among them, m t and v t are the momentum representing the first and second order moments, g t is the current gradient, α is the learning rate, β1, β2 are momentum coefficients, and ∈ is a constant to avoid division by zero errors. The data will be divided into training set, validation set, and test set. In each training step, the model will update the parameters through the back-propagation algorithm to minimize the loss function. Batch normalization and dropout techniques are used during training to avoid overfitting. Step S34: Evaluate the model using the validation dataset, calculate the loss value and check whether the model is overfitting.

5. The method according to claim 1, characterized in that: The step S4 comprises the following sub-steps: Step S41: Pressurization / Release Module Design The segmented areas are used as orthosis compression modules, anti-rotation modules, and release modules. The positions and shapes of the corresponding modules are designed according to the curvature distribution of body parts and biomechanical requirements. Curvature analysis: The application of various modules is determined by calculating the local curvature of the body; the curvature calculation formula is: K=k1*k2 Where: k1 and k2 are the principal curvatures of the body surface in the local area, and the shape and strength of the pressurization module are determined according to the curvature; Step S42: using a vector database to store and process vector data, recommending corresponding design templates for related diseases and deformity characteristics, wherein the vector data stored in the vector database includes body geometry and biomechanical feature vectors, searching for the most similar vectors through similarity measurement, and generating corresponding orthopedic design templates according to the stored vector data; Step S43: Design a transition area between the pressurizing module and the releasing module to optimize the distribution of pressure and release. The pressure distribution formula is: in P(x) is the pressure distribution, σ(x) is the stress in the material, and n is the surface normal vector, indicating the direction of the pressure.

6. The method according to claim 1, characterized in that: In step S5, the three-dimensional orthopedic CAD system includes a data system module for establishing case files, uploading imaging data and three-dimensional files, and graphical prescriptions, and the data is traceable. The specific data storage structure is a hierarchical structure, including patient information, imaging data, design data, treatment process records and other levels. Each level stores different types of data. The database system is used in the storage process to ensure the security and integrity of the data; it also has the functions of three-dimensional file recognition, cutting, marking, X-ray film comparison, calibration evaluation, biomechanical alignment adjustment and loading correction design modules.

7. The method according to any one of claims 1 to 6, characterized in that: In step S6, the effect of the model is verified in actual patient samples by comparing the X-rays of the patient before and after wearing the spinal orthosis and the staged review results, and the clinical data is integrated with the collected data. The entire system is optimized according to the evaluation results, specifically, the change in the scoliosis angle Δθ=θ is calculated. after -θ before , where θ after is the scoliosis angle after wearing the orthosis, θ before is the scoliosis angle before wearing the orthosis; Change in vertebral rotation angle △α=α after -α before , where α after is the vertebral rotation angle after wearing the orthosis, α before is the vertebral rotation angle before wearing the orthosis; Change in spine length △L=L after -L before , where L after is the length of the spine after wearing the orthosis, L before is the length of the spine before wearing the orthosis; The corrective effect of the orthosis and the performance of the model are evaluated based on these indicators. When the clinical data and 3D scanning data are fused, they are matched according to the timestamps and patient information of the different data, and the parameters of the model are updated according to the evaluation results.

8. The method according to any one of claims 1 to 6, characterized in that: In step S6, the gait data of the patient before and after wearing the lower limb orthosis are compared, including step length, step width, step speed, gait cycle, ankle joint, knee joint, hip joint angle changes, and plantar pressure center trajectory COP changes; By comparing the changes in joint range of motion (ROM), sEMG analysis, and strength testing of patients before and after wearing upper limb orthoses; The corrective effect of the orthosis and the performance of the model are evaluated based on these indicators. When the clinical data and 3D scanning data are fused, they are matched according to the timestamps and patient information of the different data, and the parameters of the model are updated according to the evaluation results.

Images