Spinal orthosis automatic design method and system, storage medium and program

Abstract

The invention discloses an automatic design method and system for a spinal orthosis, a storage medium and a program. The method comprises the steps that three-dimensional point cloud data of the trunk surface of a wearer are acquired; acquiring body bone detection data of the wearer, and further acquiring scoliosis data of the wearer; carrying out registration and alignment on the three-dimensional point cloud data and the scoliosis data; an improved artificial intelligence model is adopted, and orthosis design parameters are generated in combination with the three-dimensional point cloud data and the scoliosis data; and importing the generated orthosis design parameters into aided design software, and automatically generating a three-dimensional model of the orthosis. Through the improved artificial intelligence model, the orthosis design parameters are automatically generated in combination with the human body three-dimensional point cloud data and the scoliosis data, and therefore the efficient, accurate and personalized design process is achieved.

Description

Technical Field

[0001] The present application relates to the field of orthosis design, and in particular, to a method, system, storage medium and program for automatic design of a spinal orthosis. Background Art

[0002] Scoliosis refers to the curvature of the spine in the coronal plane and is a disease that affects a wide range of people. This lesion not only affects the appearance, but can also cause serious health problems such as breathing difficulties and chronic pain. Currently, the treatments for scoliosis include physical therapy, orthosis therapy and surgical intervention, among which orthosis is a non-surgical treatment method used to support and correct the curvature of the spine.

[0003] In the prior art, the orthosis design process mainly relies on the experience and manual operation of the designers, which has the limitations of difficulty in customizing orthosis and high production cost. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a method, system, storage medium and program for automatic design of spinal orthosis, which automatically generates orthosis design parameters through an improved artificial intelligence model combined with human three-dimensional point cloud data and scoliosis data, thereby realizing an efficient, accurate and personalized design process.

[0005] Specifically, the technical solution of this application is as follows:

[0006] In a first aspect, the present application discloses a method for automatically designing a spinal orthosis, comprising:

[0007] Acquire three-dimensional point cloud data of the wearer's torso surface;

[0008] Obtaining the wearer's trunk bone detection data, and then obtaining the wearer's scoliosis data;

[0009] Registering and aligning the three-dimensional point cloud data with the scoliosis data;

[0010] Using an improved artificial intelligence model, combining the three-dimensional point cloud data with the scoliosis data to generate orthosis design parameters;

[0011] The generated design parameters of the orthosis are imported into the auxiliary design software to automatically generate a three-dimensional model of the orthosis.

[0012] In some embodiments, the improved artificial intelligence model is used to generate orthosis design parameters by combining the three-dimensional point cloud data with the scoliosis data, comprising the following steps:

[0013] The artificial intelligence model adopts Point Transformer network as the basic structure;

[0014] By performing multi-level convolution and pooling with the added pyramid structure, multi-scale features of the three-dimensional point cloud data are extracted; the multi-scale features include global features and local features;

[0015] By using position encoding and self-attention mechanism, we build a representation of query, key, and value in the local neighborhood of the point cloud.

[0016] The extracted multi-scale features are combined with the scoliosis data through a multi-layer perceptron, mapped into the orthosis design parameters and output.

[0017] In some embodiments, the multi-level convolution and pooling performed by adding a pyramid structure to extract multi-scale features of the three-dimensional point cloud data includes the following steps:

[0018] Using a pyramid feature network architecture, the three-dimensional point cloud data is divided into multiple levels for processing;

[0019] Using the farthest point sampling method, representative sampling points are selected from the three-dimensional point cloud data to ensure that the sampling points are evenly distributed;

[0020] For each sampling point, the k-NN algorithm is used to find its neighboring points, and the features of the neighboring points are aggregated to capture the local structure;

[0021] Applying convolution operations at different levels to extract spatial patterns of features of different sizes; using different combinations of convolution kernels and depths to enable the AI ​​model to flexibly extract features at multiple levels;

[0022] Features of different scales are combined through skip connections and weighted summation, so that the artificial intelligence model can utilize multi-level feature information.

[0023] In some embodiments, the scoliosis data includes angle, position, and rotation information of spinal curvature;

[0024] The orthosis design parameters include the force application area position, size, depression degree, rotation angle, release area position, size, and cutting position of the orthosis.

[0025] In some embodiments, the method for automatically designing a spinal orthosis further comprises the following steps:

[0026] Collect several groups of scoliosis data with different degrees of scoliosis and corresponding three-dimensional point cloud data as training sets;

[0027] Label the ideal orthosis design parameters according to each set of data as the target data for training;

[0028] Use the Tensor Flow framework to build and train the artificial intelligence model, and set appropriate hyperparameters for model training to improve the accuracy and stability of the model;

[0029] The artificial intelligence model is trained using a weighted loss function to adjust the attention paid to different categories according to the weights;

[0030] The performance of the artificial intelligence model was evaluated using cross-validation.

[0031] In some embodiments, the registering and aligning the three-dimensional point cloud data with the scoliosis data comprises the following steps:

[0032] placing markers at target locations on the wearer's torso;

[0033] Acquire a first position of the marker in the three-dimensional point cloud data and a second position of the marker in the scoliosis data;

[0034] The three-dimensional point cloud data and the scoliosis data are registered to the same coordinate system through an iterative closest point algorithm.

[0035] In some embodiments, the method for automatically designing a spinal orthosis further comprises the following steps: converting the three-dimensional model into a wearable physical orthosis;

[0036] Specifically, it includes: using finite element analysis technology to test the performance and comfort of orthotics under different force conditions;

[0037] 3D printing of biocompatible materials to create solid orthoses.

[0038] In a second aspect, the present application further discloses a spinal orthosis automatic design system, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the spinal orthosis automatic design method described in any one of the above embodiments.

[0039] In a third aspect, the present application further discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for automatic design of a spinal orthosis as described in any of the above embodiments.

[0040] In a fourth aspect, the present application further discloses a computer program product, including a computer program, which, when executed by a processor, implements the steps of the method for automatic design of a spinal orthosis described in any of the above embodiments.

[0041] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0042] The automatic design method of spinal orthosis proposed in this application utilizes an improved artificial intelligence model to realize the automated design process of spinal orthosis. This innovation not only greatly saves human resources and improves the efficiency of spinal orthosis design and production, but also can tailor orthosis for each patient through precise data analysis and personalized design. This customized design method enables the orthosis to better fit the patient's body curve, thereby significantly improving the wearer's comfort and orthotic effect. In addition, automated design also reduces the possibility of human error and ensures the quality and safety of the orthosis. Through the method disclosed in this application, the wearer can get a more comfortable orthotic wearing experience, and can also reduce the customization cost of the orthosis, which is conducive to the promotion of the orthosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The preferred implementation scheme will be described below in a clear and understandable manner with reference to the accompanying drawings to further illustrate the above-mentioned characteristics, technical features, advantages and implementation methods of the present application.

[0044] Figure 1 This is a schematic flow chart of the steps of an embodiment of a method for automatically designing a spinal orthosis according to the present application.

[0045] Figure 2 Schematic diagram of step S200 in another embodiment of a method for automatically designing a spinal orthosis of the present application. DETAILED DESCRIPTION

[0046] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0047] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections.

[0048] In order to simplify the drawings, only the parts related to the invention are schematically shown in each figure, and they do not represent the actual structure of the product. In addition, in order to simplify the drawings and facilitate understanding, in some figures, only one of the parts with the same structure or function is schematically drawn or marked. In this article, "one" not only means "only one", but also means "more than one".

[0049] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0050] In this document, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0051] In addition, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the specific implementation methods of the present application will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0053] The human body model can be roughly divided into the head, limbs and torso. The human body other than the head and limbs can be called the torso. The torso: the front is the chest and abdomen, and the back is the back and waist. The inner cavity of the torso is divided into the thoracic cavity and the abdominal cavity, which contains other human organs except the brain.

[0054] The trunk bones of the human body model mainly play a supporting and protective role. The trunk bones are the main bone components of the human body, including the cervical vertebrae, thoracic vertebrae, lumbar vertebrae, sacral vertebrae, coccygeal vertebrae, sternum, and ribs, a total of 51 bones, which are connected by bones to form the spine and thorax. Among them, the cervical spine has 7 vertebrae, of which the first cervical vertebra has no vertebrae, but only a ring-shaped bone. The posterior process of the seventh cervical vertebra is relatively long, which is the surface positioning mark of the human vertebrae. The thoracic vertebrae have 12 cones, and each thoracic vertebra is connected to the ribs. There are 12 ribs on each side, which together with the thoracic vertebrae and sternum form the thorax. The sternum is divided into three parts: the sternal head, the sternal manubrium, and the sternal tail. There are 5 vertebrae in the lumbar vertebrae. The intervertebral discs between the lumbar vertebrae are larger, which play the role of cushioning, buffering, and decompression. The sacrum is divided into the 1st, 2nd, 3rd and 4th sacral vertebrae in childhood, and fused into a whole sacral vertebrae in adulthood. The coccyx is composed of the 1st, 2nd and 3rd coccygeal vertebrae.

[0055] The spine is the main supporting structure of the human body. There is a spinal canal behind the vertebral body, and spinal nerves run in the spinal canal. Nerves can send out nerve roots to control the sensory and motor functions of the whole body and limbs. The thorax contains important organs, such as the heart and lungs, which are protected from the possibility of damage when subjected to huge external forces, while protecting the integrity of their internal structures.

[0056] Due to long hours of study and desk work, more and more people are hunched over (including many teenagers). Deformation of the thoracic spine, disorder of the facet joints, ligament and muscle strain are the main causes of stubborn pain in the thoracic spine and back. It may also squeeze the internal organs and cause many diseases such as palpitations, chest tightness, stomachache, etc.

[0057] Scoliosis is an abnormal curvature of the spine. A normal spine has a backward curve at the shoulders and a forward curve at the waist.

[0058] Typical scoliosis involves a three-dimensional deformity of the spine and ribs. Depending on the degree, the spine bends sideways and sometimes the vertebrae rotate slightly, causing imbalances in the hips or shoulders. It can develop into:

[0059] With either a primarily sideways curve (similar to the letter C) or two curves (a primary curve and a secondary curve, resulting in an S-shape), scoliosis most often occurs between the upper back (thoracic region) and the lower back (lumbar region). It is also possible for scoliosis to occur only in the upper or lower back.

[0060] In order to avoid correcting scoliosis deformity and stop its further progression, spinal braces came into being. Spinal braces are used to restore the physiological curvature of the spine, obtain stability, maintain trunk balance, change the appearance deformity, minimize the range of fusion, reduce or relieve low back pain, and maximize the improvement and maintenance of the heart and lungs.

[0061] The spinal corrector uses a soft material reinforced with elastic strips to wrap the trunk, exert a certain amount of pressure on the waist and abdominal soft tissues, increase the intra-abdominal pressure, reduce the load on the lumbar spine caused by body weight, and limit spinal movement, thereby achieving the purpose of eliminating pain. It is mainly divided into two categories: soft and hard, which are used to increase spinal stability, limit spinal movement, relieve pain, reduce vertebral load, support muscles, prevent and correct deformities. Prices vary greatly due to differences in materials, processes, designs, etc., and the prices of customized products vary. For mild and moderate scoliosis, the corrector has obvious therapeutic effects, can prevent the development of lesions, partially or completely correct them, and avoid surgery. When choosing, you need to consider the condition, materials, comfort and cost. It is recommended to customize under the guidance of a professional doctor. The application of 3D printing technology makes the corrector more fitting, comfortable, and beautiful, and the treatment cost is relatively low.

[0062] Soft spinal braces play an important role in treating mild scoliosis, disc disease, lumbar spine injuries, etc. However, for severe scoliosis or complex cases, a more rigid brace or surgical treatment may be required.

[0063] The traditional orthosis design process mainly relies on the designer's experience and manual operation. This method has the following limitations:

[0064] 1. Complex design: It involves the adjustment and optimization of multiple variables, including the configuration of the force application area and the release area, which requires the designer to have rich experience.

[0065] 2. Lack of personalization: Since each wearer has a different body shape and degree of curvature, traditional designs are difficult to fully meet personalized needs, which may lead to unsatisfactory orthopedic effects.

[0066] 3. High time and cost: The manual design and manufacturing process is lengthy, unable to respond quickly to wearer needs, and has high production costs.

[0067] With the advancement of 3D scanning and printing technology, automated and personalized orthosis design has become possible. However, existing technologies are still insufficient in processing complex geometric data and automatically generating accurate design parameters.

[0068] This application proposes an automatic design method for spinal orthosis based on a generative model, which aims to automatically generate orthosis design parameters through an improved Point Transformer model combined with human body scanning data and scoliosis data, thereby achieving an efficient, accurate and personalized design process.

[0069] Reference Manual Attached Figure 1 An embodiment of a spinal orthosis automatic design method provided by the present application comprises the following steps:

[0070] S100, obtaining three-dimensional point cloud data of the wearer's torso surface.

[0071] Specifically, 3D point cloud data acquisition: Use a 3D scanning device to obtain high-precision 3D point cloud data of the wearer's entire torso. The scanning device must have high precision and fast acquisition capabilities to ensure the accuracy and completeness of the data. The 3D scanning device includes a laser scanning device or a structured light scanning device.

[0072] S200, obtaining the wearer's trunk bone detection data, and then obtaining the wearer's scoliosis data.

[0073] Specifically, in some embodiments, scoliosis data is acquired using non-radiation equipment, such as equipment based on infrared or ultrasonic technology, to acquire detailed scoliosis data, including bending angle, position, direction, etc.

[0074] In other embodiments, the scoliosis data is acquired by scanning the trunk bones of the wearer using medical imaging equipment to acquire trunk bone images. The scoliosis data is obtained based on the trunk bone image evaluation. The trunk bone image is acquired using medical imaging equipment such as X-ray imaging equipment, computer tomography (CT) equipment, magnetic resonance imaging (MRI) equipment, etc. The Cobb angle on the trunk bone image is then measured by machine measurement or manual measurement to determine the spinal curvature angle, position, direction, etc. The specific equipment used is determined based on the actual patient situation and is not limited in this application.

[0075] S300, aligning the three-dimensional point cloud data with the scoliosis data.

[0076] Specifically, a marker-based feature matching algorithm is used to accurately align the data by identifying the marker positions in the scanned point cloud and the marker positions in the scoliosis data. Aligning the three-dimensional point cloud data of the wearer's torso surface with the wearer's scoliosis data to the same coordinate system helps the artificial intelligence model obtain more accurate internal bone structure and external morphological information. When designing orthotics, accurate alignment of human and bone data can ensure that the device perfectly matches the patient's body.

[0077] S400, using an improved artificial intelligence model to combine the three-dimensional point cloud data with the scoliosis data to generate orthosis design parameters.

[0078] Specifically, an improved artificial intelligence model (Point Transformer model) is used to extract 3D point cloud features and generate orthosis design parameters in combination with scoliosis data. The improved artificial intelligence model includes at least the following modules:

[0079] 1) Multi-scale feature extraction: A multi-scale feature extraction mechanism is introduced to capture geometric features at different scales and improve adaptability to complex shapes.

[0080] 2) Enhanced self-attention mechanism: Enhance the self-attention mechanism of the model by increasing position encoding and channel attention to improve the ability to capture the relationship between different points.

[0081] 3) Parameter generation module: A dedicated parameter generation module is designed to generate specific orthosis design parameters from the extracted features, including the force application area position, size, degree of concavity, rotation angle, release area position, size, cutting position, etc.

[0082] S500, importing the generated design parameters of the orthosis into auxiliary design software to automatically generate a three-dimensional model of the orthosis.

[0083] Specifically, S510, design parameter application: importing the generated design parameters into auxiliary design software to automatically generate a three-dimensional model of the orthosis. Optionally, the auxiliary design software includes SolidWorks or AutoCAD, etc., which can automatically generate a three-dimensional model of the orthosis. The position and shape of the force application area and the release area are configured according to the parameters to ensure the correction effect of the orthosis.

[0084] S520, Model Optimization: Optimize the structure and performance of the orthosis through simulation and emulation to ensure its comfort and correction effect. Optionally, use simulation technology such as finite element analysis (FEA) to test the performance and comfort of the orthosis under different force conditions. Based on the simulation results, iteratively optimize the design to ensure the durability and user experience of the orthosis.

[0085] Another embodiment of the automatic design method of a spinal orthosis of the present application is based on the above method embodiment, and the step S300 includes the following sub-steps:

[0086] S310, setting a marker at a target position on the wearer's torso.

[0087] Specifically, marker setting: Before scanning the patient, identifiable markers are placed at the wearer's cervical vertebrae 7 (C7) and lumbar vertebrae 5 (L5). These markers are used to align the 3D point cloud data and scoliosis data to the same coordinate system.

[0088] S320, obtaining a first position of the marker in the three-dimensional point cloud data and a second position of the marker in the scoliosis data.

[0089] S330, aligning the three-dimensional point cloud data and the scoliosis data to the same coordinate system through an iterative closest point algorithm.

[0090] Specifically, feature matching: Use a marker-based feature matching algorithm to align the point cloud data and scoliosis data to a unified coordinate system. By identifying the position of the marker, accurate alignment of the data is achieved. Preferably, the marker is used as a reference point and the data is aligned using the ICP (Iterative Closest Point) algorithm. The ICP algorithm is a point cloud matching algorithm for geometric alignment of three-dimensional models and is widely used in computer vision tasks, including three-dimensional scanning, robot navigation, and augmented reality. The basic idea of ​​the ICP algorithm is to achieve alignment between point clouds by finding the closest point correspondence between two sets of point clouds and iteratively optimizing the rigid body transformation between them.

[0091] Optionally, aligning the three-dimensional point cloud data and the scoliosis data to the same coordinate system means aligning the two to the three-dimensional point cloud coordinate system; or aligning the two to the scoliosis coordinate system; or aligning the two to another new coordinate system. In other embodiments, the technical effect achieved by step S300 of the present application can also be achieved by using markers placed at other locations.

[0092] Another embodiment of the method for automatic design of a spinal orthosis of the present application, based on any one of the embodiments of the above method, further includes the following steps: S600, converting the three-dimensional model into a wearable physical orthosis.

[0093] Specifically, it includes: using finite element analysis technology to test the performance and comfort of orthoses under different force conditions. Using biocompatible materials for 3D printing to manufacture physical orthoses.

[0094] Optionally, 3D printing utilizes biocompatible materials such as PLA (polylactic acid), TPU (thermoplastic polyurethane) or other types of materials to transform the 3D model into a wearable physical orthosis.

[0095] In some other embodiments, the following steps are also included: clinical verification, trying on and evaluating the orthosis in a clinical environment, adjusting the model based on wearer feedback, trying on and verifying the effect in real wearers, and collecting feedback for further optimization.

[0096] Another embodiment of the automatic design method of a spinal orthosis of the present application is based on an embodiment of the above method, and the artificial intelligence model adopts a Point Transformer network as a basic structure.

[0097] Reference Manual Attached Figure 2 , step S200 includes the following steps:

[0098] S210, performing multi-level convolution and pooling by adding a pyramid structure to extract multi-scale features of the three-dimensional point cloud data. The multi-scale features include global features and local features.

[0099] Specifically, in 3D point cloud data, features of different scales are crucial to accurately describe the shape and structure. The architecture of the pyramid feature network (PFN) processes point cloud data through multiple levels. High-level features: represent large-scale features to provide global shape information, such as overall bending trends. Low-level features: describe small-scale features to reflect local details, such as tiny bumps on the skin surface.

[0100] S220, by using position encoding and self-attention mechanism, establish the representation of query, key and value in the local area of ​​point cloud.

[0101] Specifically, the self-attention mechanism allows the model to focus on the important parts of the input data and ignore the unimportant parts when processing data. This is particularly important for analyzing the geometry of specific areas in 3D point clouds. In PointTransformer, the self-attention mechanism allows the network to calculate the attention weights directly in the local neighborhood of the point cloud instead of calculating on the entire point cloud, which effectively reduces the computational complexity and video memory consumption. This self-attention mechanism enables PointTransformer to effectively capture the local structure and global contextual information in point cloud data. Through the self-attention mechanism, Point Transformer is able to process unordered point cloud data and does not require a fixed receptive field like a convolutional neural network. This makes Point Transformer more flexible and adaptable when processing point cloud data.

[0102] S230, combining the extracted multi-scale features with the scoliosis data through a multi-layer perceptron, mapping them into the orthosis design parameters and outputting them.

[0103] Specifically, a multilayer perceptron (MLP) is a basic form of neural network, which consists of multiple layers, including an input layer, one or more hidden layers, and an output layer. MLP is trained using the supervised learning technique of back propagation to map features to specific design parameter outputs. The ultimate goal of feature extraction is to generate specific orthosis design parameters. These parameters include the location, size, shape, etc. of the force application area and the release area.

[0104] Another embodiment of the automatic design method of a spinal orthosis of the present application is based on the above embodiment of the method, and the step S210 specifically includes the following details:

[0105] S211, using a pyramid feature network architecture, dividing the three-dimensional point cloud data into multiple levels for processing.

[0106] S212, using the farthest point sampling method, select representative sampling points from the three-dimensional point cloud data to ensure that the sampling points are evenly distributed.

[0107] S213, for each sampling point, use the k-NN algorithm to find its neighborhood points, and aggregate the features of the neighborhood points to capture the local structure.

[0108] S214, applying convolution operations at different levels to extract spatial patterns of features of different sizes. Using different convolution kernels and depth combinations allows the artificial intelligence model to flexibly extract features at multiple levels.

[0109] S215, combining features of different scales through skip connections and weighted summation, so that the artificial intelligence model can utilize multi-level feature information.

[0110] Specifically, the core of the pyramid structure is the ability to process and extract multi-scale features. In Point Transformer, this can be achieved by introducing local neighborhoods of different sizes at different stages. For example, local neighborhoods of different sizes (defined by KNN or other sampling strategies) can be used at different levels of the network to simulate multi-scale feature extraction in the pyramid structure. In the pyramid structure, high-resolution features at lower levels need to be effectively fused with low-resolution but more abstract features at higher levels. In Point Transformer, features of different scales can be combined by designing specific fusion modules, such as using feature fusion techniques or attention mechanisms to strengthen the interaction between features of different scales. In Point Transformer, the process of downsampling and upsampling can be simulated by adjusting the size of the local neighborhood and the feature dimension.

[0111] Another embodiment of the automatic design method of a spinal orthosis of the present application is based on the above embodiment of the method, and the step S220 specifically includes the following details:

[0112] S221, generating point cloud position codes. For each point, a position code is generated in combination with its spatial position.

[0113] Specifically, the Point Transformer network is a network based on the self-attention mechanism. In Point Transformer, position encoding is crucial to capturing structural information in point clouds. Position encoding can be adjusted with the introduction of the pyramid structure to adapt to features of different scales. You can consider using relative position encoding or position encoding strategies that can adapt to different resolutions. Generating point cloud position encoding helps the model distinguish points at different positions and understand their relative relationships.

[0114] In practical applications, position encoding can be generated in many ways. For example: Feature-based methods: First, convolutional neural networks are used to extract the features of the point cloud, and then these features are enhanced through position encoding to contain spatial location information.

[0115] S222, the self-attention module inputs the feature map, performs a linear transformation on the input features, and generates three different representations: Query, Key, and Value.

[0116] Specifically, Point Transformer designs a self-attention layer for point clouds and combines it with position encoding to build a Transformer block. The self-attention mechanism enables the model to focus on the important parts of the input data while ignoring the unimportant parts when processing data. This is especially important for analyzing the geometry of specific areas in 3D point clouds. Through the self-attention mechanism of the local neighborhood, the representation of Query, Key, and Value in the local neighborhood of the point cloud is established, and a feature learning model with point pairs as attention units is established.

[0117] S223, attention calculation: Calculate the dot product of each pair of Query and Key, and get the attention weight through the Softmax function. These weights are used to perform weighted summation on the Value and output a new feature representation.

[0118] Specifically, the self-attention mechanism calculates the contextual information of each point through the interaction of query (Query, Q), key (Key, K) and value (Value, V). In Point Transformer, each point generates representations of Q, K and V, and then obtains attention scores by calculating the similarity between Q and K. These scores are then used to weight V to generate new point representations.

[0119] S224, uses a multi-head attention mechanism to calculate multiple attention outputs in parallel and then merge them.

[0120] Specifically, multi-head attention increases the representation capability of the model, allowing it to focus on multiple different feature patterns at the same time. The multi-head attention mechanism means that there are multiple attention heads working in parallel, and each head learns a different representation subspace.

[0121] S225 calculates the attention weight on each channel and dynamically adjusts the feature importance of each channel, thereby improving the model's ability to capture complex features.

[0122] Another embodiment of the automatic design method of a spinal orthosis of the present application is based on the above embodiment of the method, and the step S230 specifically includes the following details:

[0123] S231, using a multi-layer perceptron (MLP) to map the extracted features into specific orthosis design parameters.

[0124] S232, each MLP contains several fully connected layers, which enhance the expression ability of nonlinear features through nonlinear activation functions (such as ReLU).

[0125] S233, target parameter generation, outputs multiple design parameters, including the following:

[0126] Force application zone: The area of ​​the spine where the orthosis applies corrective force.

[0127] Release Zone: An area of ​​the body where no pressure is applied.

[0128] Concavity: The degree of concavity or convexity of the inner shape of the orthosis.

[0129] Angle of rotation: The degree to which the orthosis rotates around the body.

[0130] Cutting position: The border of the orthosis is designed to ensure comfort and functionality.

[0131] Specifically, the features output by the self-attention mechanism are processed by the MLP layer. The multilayer perceptron (MLP) generally consists of multiple layers, including an input layer, one or more hidden layers, and an output layer. The input layer is used to receive raw data input. The hidden layer consists of multiple neurons, each of which is connected to the neurons in the previous layer and is parameterized by weights and biases. Activation function: The neurons in the hidden layer usually contain nonlinear activation functions such as ReLU, sigmoid, or tanh, which enable the MLP to learn and perform nonlinear tasks. Output layer: The last layer provides the prediction results of the model, and the number of its neurons depends on the task.

[0132] In order to ensure that the model can effectively generate high-quality design parameters, the training process must optimize the adaptability of the model to complex forms. This application discloses another embodiment, and the model training process for the improved artificial intelligence model (Point Transformer network) is as follows:

[0133] S010, collecting several groups of scoliosis data with different degrees of scoliosis and corresponding three-dimensional point cloud data as training sets.

[0134] Optionally, when organizing the training set, randomly rotate, translate, scale, and add noise to expand the diversity of the training data to enhance the generalization ability of the model.

[0135] S020, labeling the ideal orthosis design parameters according to each set of data as the target data for training.

[0136] S030, use the Tensor Flow framework to build and train the artificial intelligence model, and set appropriate hyperparameters for model training to improve the accuracy and stability of the model.

[0137] Hyperparameters are parameters that are manually set before the algorithm runs to control the behavior and performance of the model. For example, learning rate, number of iterations, regularization parameters, number of neurons in the hidden layer, etc. are all common hyperparameters. The choice of these hyperparameters will affect the training speed, convergence, capacity, and generalization ability of the model. The selection of hyperparameters is usually a trial-and-error process that needs to be adjusted based on experience and domain knowledge.

[0138] S040, using a weighted loss function to train the artificial intelligence model, and adjusting the attention paid to different categories according to the weights.

[0139] Optionally, a weighted loss function is used to combine feature matching loss and structural similarity loss to improve the accurate extraction of different parameters.

[0140] Adaptive learning rate scheduling: Dynamically adjust the learning rate to accelerate model convergence. Regularization strategy: Use techniques such as Dropout and Batch Normalization to prevent model overfitting.

[0141] S050, use cross-validation method to evaluate the performance of the artificial intelligence model.

[0142] Optionally, in some other implementations of this embodiment, the model training process also includes model experiments and evaluations. Specifically, experiments are conducted through actual data sets to verify the performance of the improved Point Transformer in extracting multi-scale features and generating orthosis design parameters.

[0143] The model training effect is evaluated by the following items:

[0144] Accuracy: Compare the generated orthosis design parameters to the design criteria of the professional designer.

[0145] Efficiency: The computational time and resource consumption required to evaluate the generated model parameters.

[0146] Comfort and functionality: Tests are conducted on real wearers to evaluate the wearing comfort and corrective effect of the orthosis.

[0147] Based on the same technical concept, the present application also discloses a spinal orthosis automatic design system, which can be used to implement any of the above-mentioned spinal orthosis automatic design methods. Specifically, an embodiment of the spinal orthosis automatic design system of the present application includes:

[0148] The point cloud acquisition module is used to obtain three-dimensional point cloud data of the wearer's torso surface.

[0149] The scoliosis acquisition module is used to acquire the wearer's trunk bone detection data, and then acquire the wearer's scoliosis data.

[0150] An alignment module is used to align the three-dimensional point cloud data with the scoliosis data.

[0151] An intelligent design module is used to generate orthosis design parameters by using an improved artificial intelligence model to combine the three-dimensional point cloud data with the scoliosis data.

[0152] The orthosis generation module is used to import the generated orthosis design parameters into the auxiliary design software to automatically generate a three-dimensional model of the orthosis.

[0153] Another embodiment of an automatic spinal orthosis design system provided by the present application is based on the above system embodiment, wherein the alignment module is used to set a marker at a target position on the wearer's torso.

[0154] The method is also used to obtain a first position of the marker in the three-dimensional point cloud data and a second position of the marker in the scoliosis data, and to register the three-dimensional point cloud data and the scoliosis data to the same coordinate system by an iterative closest point algorithm.

[0155] Another embodiment of the automatic spinal orthosis design system provided by the present application is based on the above system embodiment, and the artificial intelligence model adopts the Point Transformer network as the basic structure.

[0156] The intelligent design module is specifically used to implement the steps of the following method:

[0157] By adding a pyramid structure to perform multi-level convolution and pooling, multi-scale features of the three-dimensional point cloud data are extracted. The multi-scale features include global features and local features.

[0158] By using position encoding and self-attention mechanism, we build a representation of query, key, and value in the local neighborhood of the point cloud.

[0159] The extracted multi-scale features are combined with the scoliosis data through a multi-layer perceptron, mapped into the orthosis design parameters and output.

[0160] Specifically, the scoliosis data includes information on the angle, position and rotation of the spinal curvature.

[0161] The orthosis design parameters include the force application area position, size, depression degree, rotation angle, release area position, size, and cutting position of the orthosis.

[0162] Another embodiment of the automatic design system for spinal orthosis provided by the present application, based on the above system embodiment, the automatic design system for spinal orthosis further includes: a model training module for training the artificial intelligence model in the above embodiment. Specifically, the model training module is used to implement the following steps:

[0163] Several groups of scoliosis data with different degrees of scoliosis and corresponding three-dimensional point cloud data are collected as training sets.

[0164] The ideal orthosis design parameters are labeled according to each set of data as the target data for training.

[0165] The artificial intelligence model is built and trained using the Tensor Flow framework, and appropriate hyperparameters are set for model training to improve the accuracy and stability of the model.

[0166] The artificial intelligence model is trained using a weighted loss function, and the attention paid to different categories is adjusted according to the weights.

[0167] The performance of the artificial intelligence model was evaluated using cross-validation.

[0168] Another embodiment of an automatic spinal orthosis design system provided by the present application, based on the above system embodiment, the automatic spinal orthosis design system also includes: a 3D printing module for converting a three-dimensional model into a wearable physical orthosis.

[0169] Specifically used for: Using finite element analysis technology to test the performance and comfort of orthoses under different force conditions. Using biocompatible materials for 3D printing to manufacture physical orthoses.

[0170] Based on the same concept, the present application also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for automatically designing a spinal orthosis described in any of the above embodiments are implemented.

[0171] Based on the same concept, the present application also discloses a computer program product, including a computer program, which, when executed by a processor, implements the steps of the method for automatically designing a spinal orthosis described in any of the above embodiments.

[0172] The automatic spinal orthosis design method, system, storage medium and program of the present application have the same technical concept, and the technical details of the embodiments of the two are applicable to each other. In order to reduce repetition, they will not be repeated here.

[0173] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned program modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program units or modules to complete all or part of the functions described above. The program modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into a processing unit, and the above-mentioned integrated unit can be implemented in the form of hardware or in the form of software program units. In addition, the specific names of the program modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application.

[0174] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0175] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for automatic design of a spinal orthosis, characterized in that: The steps include: Acquire three-dimensional point cloud data of the wearer's torso surface; Obtaining the wearer's trunk bone detection data, and then obtaining the wearer's scoliosis data; Registering and aligning the three-dimensional point cloud data with the scoliosis data; Using an improved artificial intelligence model, combining the three-dimensional point cloud data with the scoliosis data to generate orthosis design parameters; The generated design parameters of the orthosis are imported into the auxiliary design software to automatically generate a three-dimensional model of the orthosis.

2. The method for automatically designing a spinal orthosis according to claim 1, characterized in that: The improved artificial intelligence model is used to generate orthosis design parameters by combining the three-dimensional point cloud data with the scoliosis data, comprising the following steps: The artificial intelligence model adopts Point Transformer network as the basic structure; By performing multi-level convolution and pooling with the added pyramid structure, multi-scale features of the three-dimensional point cloud data are extracted; the multi-scale features include global features and local features; By using position encoding and self-attention mechanism, we build a representation of query, key, and value in the local neighborhood of the point cloud. The extracted multi-scale features are combined with the scoliosis data through a multi-layer perceptron, mapped into the orthosis design parameters and output.

3. The automatic design method of a spinal orthosis according to claim 2, characterized in that: The multi-level convolution and pooling are performed by adding a pyramid structure to extract the multi-scale features of the three-dimensional point cloud data, including the following steps: Using a pyramid feature network architecture, the three-dimensional point cloud data is divided into multiple levels for processing; Using the farthest point sampling method, representative sampling points are selected from the three-dimensional point cloud data to ensure that the sampling points are evenly distributed; For each sampling point, the k-NN algorithm is used to find its neighboring points, and the features of the neighboring points are aggregated to capture the local structure; Applying convolution operations at different levels to extract spatial patterns of features of different sizes; using different combinations of convolution kernels and depths to enable the AI ​​model to flexibly extract features at multiple levels; Features of different scales are combined through skip connections and weighted summation, so that the artificial intelligence model can utilize multi-level feature information.

4. The method for automatically designing a spinal orthosis according to any one of claims 1 to 3, characterized in that: The scoliosis data includes information on the angle, position and rotation of the spinal curvature; The orthosis design parameters include the force application area position, size, depression degree, rotation angle, release area position, size, and cutting position of the orthosis.

5. The method for automatically designing a spinal orthosis according to claim 1, characterized in that: The following steps are also included: Collect several groups of scoliosis data with different degrees of scoliosis and corresponding three-dimensional point cloud data as training sets; Label the ideal orthosis design parameters according to each set of data as the target data for training; Use the Tensor Flow framework to build and train the artificial intelligence model, and set appropriate hyperparameters for model training to improve the accuracy and stability of the model; The artificial intelligence model is trained using a weighted loss function to adjust the attention paid to different categories according to the weights; The performance of the artificial intelligence model was evaluated using cross-validation.

6. The method for automatically designing a spinal orthosis according to claim 1, characterized in that: The registering and aligning the three-dimensional point cloud data with the scoliosis data comprises the following steps: placing markers at target locations on the wearer's torso; Acquire a first position of the marker in the three-dimensional point cloud data and a second position of the marker in the scoliosis data; The three-dimensional point cloud data and the scoliosis data are registered to the same coordinate system through an iterative closest point algorithm.

7. The method for automatically designing a spinal orthosis according to claim 1, characterized in that: The steps are also included: converting the three-dimensional model into a wearable physical orthosis; Specifically, it includes: using finite element analysis technology to test the performance and comfort of orthotics under different force conditions; 3D printing of biocompatible materials to create solid orthoses.

8. A spinal orthosis automatic design system, comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method for automatic design of a spinal orthosis according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for automatically designing a spinal orthosis according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for automatically designing a spinal orthosis according to any one of claims 1 to 7 are implemented.

Images