Intelligent scoliosis screening method and system

Abstract

Disclosed in the present invention are an intelligent scoliosis screening method and system. The method comprises: performing three-dimensional scanning on the back of a subject by a depth sensing device, and acquiring 3D point cloud data comprising back geometry and quantified surface appearance; collecting static standing posture and dynamic forward flexion posture data of the subject; extracting a back key point feature vector, a spinal profile feature vector, and a skin tension feature vector; generating, according to the static standing posture and dynamic forward flexion posture data, a spinal biomechanical feature vector; and inputting the back key point feature vector, the spinal profile feature vector, the skin tension feature vector, and the spinal biomechanical feature vector into a trained scoliosis screening model, outputting a degree of scoliosis of the subject and a corresponding risk score, and generating a scoliosis screening report. According to the present invention, scoliosis can be screened only by means of the 3D point cloud data collected by the depth sensing device, thereby significantly improving the screening efficiency and accuracy.

Description

An intelligent method and system for scoliosis screening Technical Field

[0001] This invention relates to the field of medical devices and medical assistive technologies, specifically to an intelligent scoliosis screening method and system, and a computing device. Background Technology

[0002] Scoliosis is a common skeletal deformity characterized by lateral curvature of the spine. Early detection and intervention are crucial to preventing the condition from worsening. Currently, scoliosis screening methods mainly include visual inspection, X-ray examination, and CT scan. Visual inspection is the most common initial screening method, where doctors or professionals observe the subject's standing and forward bending postures to determine the presence of scoliosis. However, it is easily influenced by subjective medical experience, resulting in low screening efficiency. X-ray examination uses X-rays as the gold standard for diagnosing scoliosis (including parameters such as the Cobb angle). However, X-ray examination carries radiation exposure risks, making it particularly unsuitable for populations requiring frequent screening, such as children and adolescents. Furthermore, X-ray images require professional interpretation, leading to low efficiency and hindering large-scale screening needs. CT scans assess the severity and type of scoliosis through three-dimensional skeletal structure images. However, CT scans involve higher radiation doses than X-rays, making them unsuitable for routine screening. They require specialized equipment and professional personnel, limiting their widespread application.

[0003] With the development of artificial intelligence and image technology, some methods based on neural network models have emerged. However, since they often only focus on the geometric features of the back surface and ignore important information such as skin tension and spinal biomechanics, it is difficult to distinguish between structural and functional scoliosis, resulting in the need to improve the accuracy and reliability of screening.

[0004] To address the aforementioned problems, there is an urgent need for a non-invasive, efficient, and accurate method for scoliosis screening to overcome the shortcomings of existing technologies. Based on this, this invention proposes an intelligent scoliosis screening method that can screen for scoliosis using only 3D point cloud data collected by a depth sensing device, thereby improving screening efficiency and accuracy. Summary of the Invention

[0005] In view of the above problems, the present invention provides an intelligent scoliosis screening method and system, and a computing device.

[0006] According to one aspect of the present invention, an intelligent scoliosis screening method is provided, comprising:

[0007] The subject's back is scanned in three dimensions using a depth sensing device to obtain 3D point cloud data including the geometry of the back and the quantified surface appearance; and the subject's static standing posture and dynamic forward flexion posture data are also collected.

[0008] Based on the 3D point cloud data, feature vectors of key back points, spinal curves, and skin tension are extracted; and spinal biomechanical feature vectors are generated based on the static standing posture and dynamic forward flexion posture data.

[0009] The back key point feature vector, spinal curve feature vector, skin tension feature vector and spinal biomechanical feature vector are input into the trained scoliosis screening model, and the scoliosis degree and corresponding risk score of the subject are output. The scoliosis degree includes normal, abnormal posture and suspected scoliosis.

[0010] A scoliosis screening report is generated based on the subject's back geometry, spinal curvature, degree of scoliosis, and corresponding risk score.

[0011] In an alternative approach, the method for extracting the skin tension feature vector further includes:

[0012] Based on the 3D point cloud data, the curvature and normal vector change rate of each point's neighborhood are calculated to construct a strain tensor field characterizing minute deformations on the skin surface; wherein, the strain tensor field reflects the degree of stretching and compression on the skin surface in different directions to reveal the tension distribution of potential soft tissue.

[0013] The strain tensor field is decomposed into three layers according to the Daubechies wavelet basis method to extract the approximation coefficients and detail coefficients of skin texture features;

[0014] The entropy features of the approximation coefficient and the detail coefficient are calculated separately and fused to obtain the skin tension feature vector; wherein, the entropy features include information entropy, energy entropy and singular value entropy.

[0015] In an alternative approach, the method for extracting the spinal biomechanical feature vector further includes:

[0016] A multi-rigid-body spinal dynamics model is constructed based on the static and dynamic posture data, wherein the multi-rigid-body spinal dynamics model divides the spine into multiple rigid segments;

[0017] The torque and angle acting on each spinal segment during dynamic flexion were calculated using inverse dynamics.

[0018] The muscle strength of the erector spinae and rectus abdominis muscles in the back is estimated based on simulated electromyography signals, and the stability parameters of the spine are calculated based on the torque, angle and muscle strength; wherein, the stability parameters include spinal flexural stiffness, flexural strength and flexion mode;

[0019] The torque, angle, muscle strength, and spinal stability parameters are combined to form the spinal biomechanical feature vector.

[0020] In an alternative approach, the method further includes:

[0021] The influence of soft tissue on spinal morphology is assessed based on the skin tension feature vector to differentiate between structural scoliosis and functional scoliosis.

[0022] Specifically, a soft tissue model including back skin, muscles, and adipose tissue is constructed. The skin tension feature vector is input into the soft tissue model, and the pressure distribution of the soft tissue on the spine and the degree of curvature of the spine under different pressure distributions are output. If the soft tissue tension distribution exceeds a preset abnormal threshold and causes spinal curvature, it is judged as functional scoliosis; otherwise, it is judged as structural scoliosis.

[0023] In an alternative approach, constructing the strain tensor field characterizing minute deformations on the skin surface further includes:

[0024] For each point p in the 3D point cloud data i Find its k nearest neighbors;

[0025] Calculate using point p i Using a local coordinate system centered at point p, we transform the k nearest neighbor points to this local coordinate system and perform regression fitting to obtain the local tangent plane; where point p... i normal vector n i The z-axis direction;

[0026] According to point p i The gradient on the local tangent plane constructs the strain tensor field E. i .

[0027] In an alternative approach, calculating the spinal stability parameters based on the torque, angle, and muscle strength further includes:

[0028] The bending stiffness of the spine is calculated based on the moments and angles of the spinal segments. The formula for calculating the bending stiffness of the spine is as follows:

[0029] Among them, M i Let θ be the torque of the i-th spinal segment; i θi represents the angle of the i-th spinal segment; θ0 represents the reference angle threshold; n represents the total number of spinal segments; and α represents the dynamic adjustment coefficient.

[0030] Based on the spinal flexural stiffness, the flexural strength of the spine is calculated using the following formula:

[0031] Among them, K d(x) represents the dynamic bending stiffness of the spine at position x; β is the bending strength adjustment coefficient; L is the total length of the spine; C(x) is the curvature of the spine at position x;

[0032] Based on the spinal bending strength, the flexion mode of the spine is calculated. The formula for calculating the flexion mode of the spine is as follows:

[0033] Where ρ is the density of the spine; A is the cross-sectional area of ​​the spine; γ is the pressure distribution adjustment coefficient; and P(x) is the pressure on the spine at position x.

[0034] In one alternative approach, the scoliosis screening model includes:

[0035] The angular offset tracking module is used to extract and track the angular offset changes of the spinal vertebral body center in three-dimensional space to capture subtle bending and rotation information of the spine.

[0036] The ROI module is used to locate and crop point cloud data containing the spine region;

[0037] The downsampling module, which includes convolutional and pooling layers, is used to extract deep features from the input features.

[0038] The U-Net module includes an encoder-decoder section, where the encoder section consists of multiple downsampling modules and the decoder section consists of multiple upsampling modules. Features from corresponding levels of the encoder and decoder are fused through a skip connection structure.

[0039] The upsampling module uses a deconvolutional network layer to locate the position and degree of scoliosis;

[0040] Curvature sensing module, used to extract curvature features of the spinal surface;

[0041] The posture fusion module includes a multilayer perceptron, which is used to fuse biomechanical feature vectors extracted from static standing posture and dynamic forward flexion posture, learn the correlation between different posture features and output a comprehensive representation of posture features.

[0042] The prediction output module is used to output the degree of scoliosis of the subject and the corresponding risk score.

[0043] In one alternative approach, the loss function of the scoliosis screening model is:

[0044] Where N1 is the total number of pixels in the heatmap; h is the height of the heatmap; w is the width of the heatmap; y ij p represents the ideal depth value of the spine surface at coordinates (i,j); ijLet (i,j) be the actual depth value of the spine surface in the state of pixel (i,j); N2 is the number of keypoints; O xi O is the offset of the i-th keypoint in the x-direction; yi Let be the offset of the i-th keypoint in the y-direction.

[0045] According to another aspect of the present invention, an intelligent scoliosis screening system is provided, comprising:

[0046] The posture acquisition module is used to perform a three-dimensional scan of the subject's back using a depth sensing device to acquire 3D point cloud data including the geometry of the back and the quantified surface appearance; and to acquire the subject's static standing posture and dynamic forward flexion posture data.

[0047] The feature extraction module is used to extract back key point feature vectors, spinal curve feature vectors, and skin tension feature vectors from the 3D point cloud data; and to generate spinal biomechanical feature vectors from the static standing posture and dynamic forward flexion posture data.

[0048] The screening and assessment module is used to input the back key point feature vector, spinal curve feature vector, skin tension feature vector and spinal biomechanical feature vector into the trained scoliosis screening model, and output the degree of scoliosis and the corresponding risk score of the subject. The degree of scoliosis includes normal, abnormal posture and suspected scoliosis.

[0049] The report generation module is used to generate a scoliosis screening report based on the subject's back geometry, spinal curve, degree of scoliosis, and corresponding risk score.

[0050] According to another aspect of the present invention, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0051] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described intelligent scoliosis screening method.

[0052] According to the solution provided by the present invention, a three-dimensional scan of the subject's back is performed using a depth sensing device to acquire 3D point cloud data including the back's geometry and quantified surface appearance; and static standing posture and dynamic forward flexion posture data of the subject are collected; key back feature vectors, spinal curve feature vectors, and skin tension feature vectors are extracted from the 3D point cloud data; spinal biomechanical feature vectors are generated based on the static standing posture and dynamic forward flexion posture data; the key back feature vectors, spinal curve feature vectors, skin tension feature vectors, and spinal biomechanical feature vectors are input into a pre-trained scoliosis screening model, and the model outputs the subject's degree of scoliosis and corresponding risk score, wherein the degree of scoliosis includes normal, abnormal posture, and suspected scoliosis; a scoliosis screening report is generated based on the subject's back geometry, spinal curve, degree of scoliosis, and corresponding risk score. The present invention can screen for scoliosis using only 3D point cloud data collected by a depth sensing device, significantly improving screening efficiency and accuracy.

[0053] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0054] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0055] Figure 1 shows a flowchart of the intelligent scoliosis screening method according to an embodiment of the present invention;

[0056] Figure 2 shows a schematic diagram of dataset annotation according to an embodiment of the present invention;

[0057] Figure 3 shows a schematic diagram of angular offset prediction according to an embodiment of the present invention;

[0058] Figure 4 shows a schematic diagram of scoliosis diagnosis according to an embodiment of the present invention;

[0059] Figure 5 shows a schematic diagram of the framework of the intelligent scoliosis screening system according to an embodiment of the present invention;

[0060] Figure 6 shows a schematic diagram of the structure of a computing device according to an embodiment of the present invention. Detailed Implementation

[0061] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0062] Figure 1 shows a flowchart of the intelligent scoliosis screening method according to an embodiment of the present invention. Specifically, as shown in Figure 1, it includes the following steps:

[0063] Step S101: A three-dimensional scan of the subject's back is performed using a depth sensing device to obtain 3D point cloud data including the geometry of the back and the quantified surface appearance; and static standing posture and dynamic forward flexion posture data of the subject are collected.

[0064] In this embodiment, depth sensing is a non-invasive, radiation-free examination method, particularly suitable for screening adolescents and children, avoiding the potential risks of X-ray radiation. Three-dimensional point cloud data provides objective spinal geometric parameters (such as Cobb angle, vertebral rotation angle, vertebral offset distance, etc.), avoiding the subjectivity of manual measurement, and can detect minor spinal deformities and muscle imbalances, improving screening sensitivity. The entire scanning process typically takes only a few seconds or minutes, shortening screening time and making it suitable for large-scale screening. Furthermore, the three-dimensional point cloud data can be visualized in three dimensions, allowing doctors to intuitively observe the geometry of the spine.

[0065] Step S102: Extract back key point feature vectors, spinal curve feature vectors, and skin tension feature vectors from the 3D point cloud data; generate spinal biomechanical feature vectors from the static standing posture and dynamic forward flexion posture data.

[0066] In this embodiment, the 3D point cloud data contains rich surface geometry information, enabling the extraction of key back points, spinal curves, skin tension, and other features. Combining static standing and dynamic forward flexion postures to assess spinal performance in different states provides a more realistic assessment and helps detect early or mild scoliosis.

[0067] Specifically, a 3D scanner was used to acquire back point cloud data of the subjects during static standing and dynamic flexion. The dynamic flexion process employed continuous or interval scanning. As shown in Table 1, point cloud processing algorithms (such as curvature analysis, normal vector analysis, or PointNet++) were used to detect key back points such as the inferior angle of the scapula, the iliac crest, and the spinous processes of the spine. The distances, angles, and relative positions between these key points were calculated to construct feature vectors. Based on the spinous processes, spline curves and B-spline curves were used to fit the spinal curves. The curvature, torsion, length, and radius of curvature of the spinal curves were calculated to construct feature vectors. Skin tension distribution was estimated based on point cloud density and normal vector changes. Statistical features (such as mean, variance, maximum, and minimum) or spatial distribution features of skin tension were extracted to construct feature vectors. Based on point cloud data or synchronized video, the maximum standing and flexion amplitudes during static standing and dynamic flexion were identified. A biomechanical model of the spine was established using finite element or multibody dynamics. The torque, stress, strain, and joint range of motion biomechanical parameters of the spine were calculated under different postures. The above biomechanical parameters are combined into a spinal biomechanical feature vector.

[0068] Table 1

[0069] In an alternative approach, the method for extracting the skin tension feature vector further includes:

[0070] Based on the 3D point cloud data, the curvature and normal vector change rate of each point's neighborhood are calculated to construct a strain tensor field characterizing minute deformations on the skin surface; wherein, the strain tensor field reflects the degree of stretching and compression on the skin surface in different directions to reveal the tension distribution of potential soft tissue.

[0071] The strain tensor field is decomposed into three layers according to the Daubechies wavelet basis method to extract the approximation coefficients and detail coefficients of skin texture features;

[0072] The entropy features of the approximation coefficient and the detail coefficient are calculated separately and fused to obtain the skin tension feature vector; wherein, the entropy features include information entropy, energy entropy and singular value entropy.

[0073] In this embodiment, early detection and assessment of the condition are crucial in scoliosis screening. Utilizing skin tension information allows for more sensitive detection of potential soft tissue and skeletal abnormalities, enabling quantitative assessment of the severity of scoliosis and thus improving screening accuracy. Since changes in skin tension may precede structural changes, this embodiment can provide early warning of scoliosis.

[0074] Specifically, for each point in the 3D point cloud data, its neighboring points are determined. The neighborhood can be determined using the k-nearest neighbor algorithm or a fixed-radius neighborhood search. The curvature (e.g., Gaussian curvature, mean curvature) and rate of change of the normal vector are calculated for each point and its neighborhood to reflect the degree and trend of curvature of the skin surface. A strain tensor is constructed based on the curvature and rate of change of the normal vector; for example, the strain tensor is a 2x2 matrix. This indicates the degree of stretching and compression of the skin surface in different directions, where εxx and εy represent stretching / compression in the x and y directions, respectively, and εxy and εyx represent shear deformation.

[0075] A suitable Daubechies wavelet basis (such as db4) is selected, and each component of the strain tensor field is decomposed using a three-level Daubechies wavelet decomposition. This decomposition yields an approximation coefficient matrix and a detail coefficient matrix. The information entropy, energy entropy, and singular value entropy of the approximation coefficient matrix are calculated. Higher entropy values ​​indicate a more uniform information distribution, energy entropy reflects the concentration of the signal, and singular value entropy is obtained by decomposing the approximation coefficient matrix using Singular Value Decomposition (SVD). Similarly, the information entropy, energy entropy, and singular value entropy of the detail coefficient matrix are calculated.

[0076] In an alternative approach, constructing the strain tensor field characterizing minute deformations on the skin surface further includes:

[0077] For each point p in the 3D point cloud data i Find its k nearest neighbors;

[0078] Calculate using point p i Using a local coordinate system centered at point p, we transform the k nearest neighbor points to this local coordinate system and perform regression fitting to obtain the local tangent plane; where point p... i normal vector n i The z-axis direction;

[0079] According to point p i The gradient on the local tangent plane constructs the strain tensor field E. i .

[0080] In this embodiment, fitting the local tangent plane using k-nearest neighbor points can better adapt to the complex curvature changes of the skin surface and avoid errors caused by global fitting. Directly using the gradient on the local tangent plane to construct the strain tensor more clearly expresses the direction and intensity of skin surface deformation, improving the sensitivity to minute deformations. In particular, the local tangent plane and gradient more accurately capture subtle skin deformations caused by early or slight changes in scoliosis, thereby improving the early detection rate of screening.

[0081] Specifically, for each point p in the 3D point cloud data iUse algorithms such as kd-tree or ball tree to find its k nearest neighbors N i ={p1,p2,...,p k}, calculate the centroid of the neighborhood point set: centroid = (1 / k) × Σp j For j = 1 to k, move the neighborhood points and the center to the origin: p′ j =p j -centroid,j=1 to k.

[0082] Principal component analysis (PCA) is used to estimate point p. i normal vector n i , set N i ′={p1′,p2′,...,p k The covariance matrix is ​​formed by ′} Calculate the eigenvectors and eigenvalues ​​of the covariance matrix C, where the eigenvector corresponding to the smallest eigenvalue is the normal vector n. i .

[0083] Define a local coordinate system, where the z-axis is the normal vector n. i The x-axis can be any vector perpendicular to the z-axis; for example, a vector not parallel to n can be chosen. i Given a vector v, then calculate x. i =normalize(v-dot(v,n) i )×n i (where normalize represents normalization). Calculate the y-axis y i =cross(z i ,x i The local coordinate system is composed of vector (x) i ,y i ,z i )constitute.

[0084] Find the k nearest neighbors N i Transform from the global coordinate system to a coordinate system based on point p i Using a local coordinate system centered at the center, fit a plane in the local coordinate system with the goal of minimizing the squared error to obtain the plane equation z = ax + by + c, where a and b represent the gradients in the x and y directions, respectively.

[0085] Construct the strain tensor E based on the gradient. i ;

[0086] In an alternative approach, the method further includes:

[0087] The influence of soft tissue on spinal morphology is assessed based on the skin tension feature vector to differentiate between structural scoliosis and functional scoliosis.

[0088] Specifically, a soft tissue model including back skin, muscles, and adipose tissue is constructed. The skin tension feature vector is input into the soft tissue model, and the pressure distribution of the soft tissue on the spine and the degree of curvature of the spine under different pressure distributions are output. If the soft tissue tension distribution exceeds a preset abnormal threshold and causes spinal curvature, it is judged as functional scoliosis; otherwise, it is judged as structural scoliosis.

[0089] In this embodiment, the pressure distribution and curvature of the spine by the soft tissue are quantified by the soft tissue model. The parameters of the soft tissue model can be adjusted according to the individual characteristics of the patient (age, height, weight) to achieve personalized assessment.

[0090] Specifically, a geometric model containing the skin, muscles, and adipose tissue of the back is constructed using finite element analysis (FEA) software, and corresponding material properties (such as elastic modulus and Poisson's ratio) are assigned to different soft tissues. Boundary conditions of the model are defined; for example, the vertebrae at the base of the spine can be fixed to restrict their displacement and rotation.

[0091] The skin tension feature vector is transformed into a pressure distribution on the surface of a soft tissue model. For example, each component of the feature vector is associated with the pressure magnitude of the corresponding area on the model surface, and an interpolation method is used to map the information of the feature vector onto the entire model surface. The pressure distribution is adjusted according to different individual circumstances (considering the pressure influence of adipose tissue for obese individuals).

[0092] Finite element analysis software was used to calculate the stress, strain, and displacement of a soft tissue model under a given pressure distribution. The degree of spinal curvature was extracted from the analysis results using the Cobb angle or other spinal curvature metrics. The pressure distribution of the soft tissue on the spine was analyzed to assess which areas experienced excessive or insufficient pressure. Statistical analysis of data from a large number of healthy individuals and scoliosis patients established pre-defined abnormal thresholds for soft tissue tension distribution and spinal curvature. If the soft tissue tension distribution exceeded the pre-defined abnormal thresholds and caused spinal curvature, it was classified as functional scoliosis. If the soft tissue tension distribution was within the normal range, or even if it exceeded the thresholds but could not explain the degree of spinal curvature, it was classified as structural scoliosis.

[0093] In an alternative approach, the method for extracting the spinal biomechanical feature vector further includes:

[0094] A multi-rigid-body spinal dynamics model is constructed based on the static and dynamic posture data, wherein the multi-rigid-body spinal dynamics model divides the spine into multiple rigid segments;

[0095] The torque and angle acting on each spinal segment during dynamic flexion were calculated using inverse dynamics.

[0096] The muscle strength of the erector spinae and rectus abdominis muscles in the back is estimated based on simulated electromyography signals, and the stability parameters of the spine are calculated based on the torque, angle and muscle strength; wherein, the stability parameters include spinal flexural stiffness, flexural strength and flexion mode;

[0097] The torque, angle, muscle strength, and spinal stability parameters are combined to form the spinal biomechanical feature vector.

[0098] In this embodiment, instead of relying solely on surface attitude information, a multi-rigid-body dynamics model is constructed to calculate the torques and angles acting on the spinal segments, thereby providing a deeper understanding of the forces acting on the spine during motion. Spinal flexural stiffness, flexural strength, and bending modal stability parameters directly reflect the spine's ability to resist deformation and damage.

[0099] Specifically, motion capture systems (such as Vicon) or inertial measurement units (IMUs) are used to record the subject's posture data (including the trajectory and angle changes of spinal segments) during dynamic forward flexion. Surface electromyography (sEMG) sensors are used to record the electromyographic signals of the erector spinae and rectus abdominis muscles during dynamic forward flexion. The spine is divided into multiple rigid segments, such as cervical, thoracic, lumbar, and sacral vertebrae, connected by joints. The mass and moment of inertia of each rigid segment are determined based on the subject's height, weight, and skeletal dimensions. Incorporating physiological constraints such as the range of rotation of intervertebral discs and the range of motion of joints, inverse dynamic methods (Newton-Euler equations or Lagrange equations) are used to calculate the torque and angle acting on each spinal segment during dynamic forward flexion based on the dynamic posture data. Based on the recorded electromyographic signals and the established relational model, the muscle strength of the erector spinae and rectus abdominis muscles is estimated (e.g., linear envelope method, physiological model-based optimization methods, etc.). By analyzing the torque-angle relationship, the bending stiffness of the spine is calculated, the maximum bending moment that the spine can withstand is estimated, and the bending mode of the spine is determined by finite element analysis, reflecting the deformation mode of the spine when it is unstable.

[0100] In this embodiment, calculating the stability parameters of the spine based on the torque, angle, and muscle strength further includes:

[0101] The bending stiffness of the spine is calculated based on the moments and angles of the spinal segments. The formula for calculating the bending stiffness of the spine is as follows:

[0102] Among them, M i Let θ be the torque of the i-th spinal segment; iθi represents the angle of the i-th spinal segment; θ0 represents the reference angle threshold; n represents the total number of spinal segments; and α represents the dynamic adjustment coefficient.

[0103] Based on the spinal flexural stiffness, the flexural strength of the spine is calculated using the following formula:

[0104] Among them, K d (x) represents the dynamic bending stiffness of the spine at position x; β is the bending strength adjustment coefficient; L is the total length of the spine; C(x) is the curvature of the spine at position x;

[0105] Based on the spinal bending strength, the flexion mode of the spine is calculated. The formula for calculating the flexion mode of the spine is as follows:

[0106] Where ρ is the density of the spine; A is the cross-sectional area of ​​the spine; γ is the pressure distribution adjustment coefficient; and P(x) is the pressure on the spine at position x.

[0107] Step S103: Input the back key point feature vector, spinal curve feature vector, skin tension feature vector and spinal biomechanical feature vector into the trained scoliosis screening model, and output the degree of scoliosis and corresponding risk score of the subject. The degree of scoliosis includes normal, abnormal posture and suspected scoliosis.

[0108] In this embodiment, the back key point feature vector, spinal curve feature vector, skin tension feature vector, and spinal biomechanical feature vector reflect the state of the spine from different perspectives. Specifically, the back key point and spinal curve features focus on morphology, skin tension reflects muscle imbalances, and biomechanical features reflect spinal function (for example, even with small errors in posture detection, biomechanical features still provide crucial information). Early scoliosis may only manifest as slight postural abnormalities or asymmetrical skin tension, while biomechanical indicators may have already changed, facilitating early detection of scoliosis and enabling early diagnosis and treatment.

[0109] In one alternative approach, the scoliosis screening model includes:

[0110] The angular offset tracking module is used to extract and track the angular offset changes of the spinal vertebral body center in three-dimensional space to capture subtle bending and rotation information of the spine.

[0111] The ROI module is used to locate and crop point cloud data containing the spine region;

[0112] The downsampling module, which includes convolutional and pooling layers, is used to extract deep features from the input features.

[0113] The U-Net module includes an encoder-decoder section, where the encoder section consists of multiple downsampling modules and the decoder section consists of multiple upsampling modules. Features from corresponding levels of the encoder and decoder are fused through a skip connection structure.

[0114] The upsampling module uses a deconvolutional network layer to locate the position and degree of scoliosis;

[0115] Curvature sensing module, used to extract curvature features of the spinal surface;

[0116] The posture fusion module includes a multilayer perceptron, which is used to fuse biomechanical feature vectors extracted from static standing posture and dynamic forward flexion posture, learn the correlation between different posture features and output a comprehensive representation of posture features.

[0117] The prediction output module is used to output the degree of scoliosis of the subject and the corresponding risk score.

[0118] In this embodiment, biomechanical features of static standing postures and dynamic forward flexion postures are fused to learn subtle changes in the spine under different postures. An angular offset tracking module and a curvature sensing module capture subtle bending, rotation, and surface curvature changes of the spine in three-dimensional space, which is crucial for the early detection and diagnosis of scoliosis, especially in the early stages of the disease when it is difficult to detect with the naked eye. Specifically, as shown in Figure 3, the ROI module locates and crops point cloud data containing the spinal region from the cervical to the lumbar spine, restoring the vertebral body center points to the spinal ROI. The angular offset tracking module extracts and tracks the angular offset changes of the vertebral body center in three-dimensional space. The curvature sensing module extracts the curvature features of the spinal surface. The downsampling module uses convolutional and pooling layers to extract deep features from the input features. The U-Net module captures subtle changes in the spine through an encoder-decoder structure and fuses features from corresponding layers of the encoder and decoder through a skip connection structure. The upsampling module uses deconvolutional network layers to locate the position and degree of scoliosis. The posture fusion module fuses the angular offset information, curvature features, and the positioning information output by the upsampling module. The prediction output module outputs the degree of scoliosis and the corresponding risk score. As shown in Figure 2, to avoid the model labeling the cervical spine region as the true value of the spine during training, non-vertebral regions need to be masked before Cobb angle measurement, and datasets of vertebral body center points and spinal position anchor points need to be created. The image processing process is shown in Figure 4. The image is input into the model to obtain vertebral body center point images and spinal position anchor point images (two images provide the location information of key points). The cervical spine portion in the vertebral body center points is filtered using the spinal position anchor points to obtain the vertebral body center point images.

[0119] In one alternative approach, the loss function of the scoliosis screening model is:

[0120] Where N1 is the total number of pixels in the heatmap; h is the height of the heatmap; w is the width of the heatmap; y ij p represents the ideal depth value of the spine surface at coordinates (i,j); ij Let (i,j) be the actual depth value of the spine surface in the state of pixel (i,j); N2 is the number of keypoints; O xi O is the offset of the i-th keypoint in the x-direction; yi Let be the offset of the i-th keypoint in the y-direction.

[0121] In this embodiment, a heatmap cross-entropy loss is used to learn and predict whether each pixel belongs to the spinal surface, and a keypoint offset loss is used to learn and predict the offset of spinal keypoints (such as the center of the vertebral body), thereby more accurately measuring the Cobb angle index of the spine. Specifically, a corresponding heatmap is generated for each training sample (spine point cloud data), and each pixel in the heatmap represents the probability that the pixel belongs to the spinal surface. Ideally, the depth values ​​of the spinal surface are generated using a Gaussian kernel function to form a smooth probability distribution around the keypoints. For example, the pixel value at the center of the vertebral body is set to 1, and the values ​​of surrounding pixels decay according to distance. Here, the keypoint offset represents the difference between the predicted keypoint position and the actual position, the heatmap cross-entropy loss measures the difference between the predicted heatmap and the actual heatmap, and the keypoint offset loss measures the difference between the predicted keypoint position and the actual position.

[0122] Step S104: Generate a scoliosis screening report based on the subject's back geometry, spinal curve, degree of scoliosis, and corresponding risk score.

[0123] For example, one subject's back showed significant asymmetry, with the right side higher than the left. The spinal curvature was S-shaped, with a more pronounced right-sided curvature of 15 degrees. The risk score was 75 out of 100, classifying the risk level as moderate.

[0124] According to the solution provided by the present invention, a three-dimensional scan of the subject's back is performed using a depth sensing device to acquire 3D point cloud data including the back's geometry and quantified surface appearance; and static standing posture and dynamic forward flexion posture data of the subject are collected; key back feature vectors, spinal curve feature vectors, and skin tension feature vectors are extracted from the 3D point cloud data; spinal biomechanical feature vectors are generated based on the static standing posture and dynamic forward flexion posture data; the key back feature vectors, spinal curve feature vectors, skin tension feature vectors, and spinal biomechanical feature vectors are input into a pre-trained scoliosis screening model, and the model outputs the subject's degree of scoliosis and corresponding risk score, wherein the degree of scoliosis includes normal, abnormal posture, and suspected scoliosis; a scoliosis screening report is generated based on the subject's back geometry, spinal curve, degree of scoliosis, and corresponding risk score. The present invention can screen for scoliosis using only 3D point cloud data collected by a depth sensing device, significantly improving screening efficiency and accuracy.

[0125] Figure 5 shows a schematic diagram of the framework of the intelligent scoliosis screening system according to an embodiment of the present invention. The intelligent scoliosis screening system includes:

[0126] The posture acquisition module 510 is used to perform a three-dimensional scan of the subject's back using a depth sensing device to acquire 3D point cloud data including the geometry of the back and the quantified surface appearance; and to acquire the subject's static standing posture and dynamic forward flexion posture data.

[0127] Feature extraction module 520 is used to extract back key point feature vectors, spinal curve feature vectors and skin tension feature vectors based on the 3D point cloud data; and to generate spinal biomechanical feature vectors based on the static standing posture and dynamic forward flexion posture data.

[0128] The screening and assessment module 530 is used to input the back key point feature vector, spinal curve feature vector, skin tension feature vector and spinal biomechanical feature vector into the trained scoliosis screening model, and output the degree of scoliosis and corresponding risk score of the subject, wherein the degree of scoliosis includes normal, abnormal posture and suspected scoliosis.

[0129] The report generation module 540 is used to generate a scoliosis screening report based on the subject's back geometry, spinal curve, degree of scoliosis and corresponding risk score.

[0130] Figure 6 shows a schematic diagram of the structure of an embodiment of the computing device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computing device.

[0131] As shown in Figure 6, the computing device may include: a processor 602, a communications interface 604, a memory 606, and a communications bus 608.

[0132] The processor 602, communication interface 604, and memory 606 communicate with each other via communication bus 608. Communication interface 604 is used to communicate with other network elements such as clients or other servers. The processor 602 executes program 610, specifically performing the relevant steps in the above-described intelligent scoliosis screening method embodiment.

[0133] Specifically, program 610 may include program code that includes computer operation instructions.

[0134] Processor 602 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0135] Memory 606 is used to store program 610. Memory 606 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0136] According to the solution provided by the present invention, a three-dimensional scan of the subject's back is performed using a depth sensing device to acquire 3D point cloud data including the back's geometry and quantified surface appearance; and static standing posture and dynamic forward flexion posture data of the subject are collected; key back feature vectors, spinal curve feature vectors, and skin tension feature vectors are extracted from the 3D point cloud data; spinal biomechanical feature vectors are generated based on the static standing posture and dynamic forward flexion posture data; the key back feature vectors, spinal curve feature vectors, skin tension feature vectors, and spinal biomechanical feature vectors are input into a pre-trained scoliosis screening model, and the model outputs the subject's degree of scoliosis and corresponding risk score, wherein the degree of scoliosis includes normal, abnormal posture, and suspected scoliosis; a scoliosis screening report is generated based on the subject's back geometry, spinal curve, degree of scoliosis, and corresponding risk score. The present invention can screen for scoliosis using only 3D point cloud data collected by a depth sensing device, significantly improving screening efficiency and accuracy.

[0137] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination of all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed can be employed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be construed as limiting the order of execution.

Claims

1. An intelligent scoliosis screening method, characterized in that, include: The subject's back is scanned in three dimensions using a depth sensing device to obtain 3D point cloud data, including the geometry of the back and the quantified surface appearance. In addition, data on the static standing posture and dynamic forward flexion posture of the subjects were collected; Based on the 3D point cloud data, feature vectors of key back points, spinal curves, and skin tension are extracted; and spinal biomechanical feature vectors are generated based on the static standing posture and dynamic forward flexion posture data. The back key point feature vector, spinal curve feature vector, skin tension feature vector and spinal biomechanical feature vector are input into the trained scoliosis screening model, and the scoliosis degree and corresponding risk score of the subject are output. The scoliosis degree includes normal, abnormal posture and suspected scoliosis. A scoliosis screening report is generated based on the subject's back geometry, spinal curvature, degree of scoliosis, and corresponding risk score.

2. The intelligent scoliosis screening method according to claim 1, characterized in that, The method for extracting the skin tension feature vector further includes: Based on the 3D point cloud data, the curvature and normal vector change rate of each point's neighborhood are calculated to construct a strain tensor field characterizing minute deformations on the skin surface; wherein, the strain tensor field reflects the degree of stretching and compression on the skin surface in different directions to reveal the tension distribution of potential soft tissue. The strain tensor field is decomposed into three layers according to the Daubechies wavelet basis method to extract the approximation coefficients and detail coefficients of skin texture features; The entropy features of the approximation coefficient and the detail coefficient are calculated separately and fused to obtain the skin tension feature vector; wherein, the entropy features include information entropy, energy entropy and singular value entropy.

3. The intelligent scoliosis screening method according to claim 1, characterized in that, The method for extracting the spinal biomechanical feature vector further includes: A multi-rigid-body spinal dynamics model is constructed based on the static and dynamic posture data, wherein the multi-rigid-body spinal dynamics model divides the spine into multiple rigid segments; The torque and angle acting on each spinal segment during dynamic flexion were calculated using inverse dynamics. The muscle strength of the erector spinae and rectus abdominis muscles in the back is estimated based on simulated electromyography signals, and the stability parameters of the spine are calculated based on the torque, angle and muscle strength; wherein, the stability parameters include spinal flexural stiffness, flexural strength and flexion mode; The torque, angle, muscle strength, and spinal stability parameters are combined to form the spinal biomechanical feature vector.

4. The intelligent scoliosis screening method according to claim 1, characterized in that, The method further includes: The influence of soft tissue on spinal morphology is assessed based on the skin tension feature vector to differentiate between structural scoliosis and functional scoliosis. Specifically, a soft tissue model including back skin, muscles, and adipose tissue is constructed. The skin tension feature vector is input into the soft tissue model, and the pressure distribution of the soft tissue on the spine and the degree of curvature of the spine under different pressure distributions are output. If the soft tissue tension distribution exceeds a preset abnormal threshold and causes spinal curvature, it is judged as functional scoliosis; otherwise, it is judged as structural scoliosis.

5. The intelligent scoliosis screening method according to claim 2, characterized in that, The construction of the strain tensor field characterizing minute deformations on the skin surface further includes: For each point p in the 3D point cloud data i find its k nearest neighbors; Calculate using point p i Using a local coordinate system centered at point p, we transform the k nearest neighbor points to this local coordinate system and perform regression fitting to obtain the local tangent plane; where point p... i normal vector n i The z-axis direction; According to point p i The gradient on the local tangent plane constructs the strain tensor field E. i .

6. The intelligent scoliosis screening method according to claim 3, characterized in that, The calculation of spinal stability parameters based on the torque, angle, and muscle strength further includes: The bending stiffness of the spine is calculated based on the moments and angles of the spinal segments. The formula for calculating the bending stiffness of the spine is as follows: Among them, M i Let θ be the torque of the i-th spinal segment; i θi represents the angle of the i-th spinal segment; θ0 represents the reference angle threshold; n represents the total number of spinal segments; and α represents the dynamic adjustment coefficient. Based on the spinal flexural stiffness, the flexural strength of the spine is calculated using the following formula: Among them, K d (x) represents the dynamic bending stiffness of the spine at position x; β is the bending strength adjustment coefficient; L is the total length of the spine; C(x) is the curvature of the spine at position x; Based on the spinal bending strength, the flexion mode of the spine is calculated. The formula for calculating the flexion mode of the spine is as follows: Where ρ is the density of the spine; A is the cross-sectional area of ​​the spine; γ is the pressure distribution adjustment coefficient; and P(x) is the pressure on the spine at position x.

7. The intelligent scoliosis screening method according to claim 1, characterized in that, The scoliosis screening model includes: The angular offset tracking module is used to extract and track the angular offset changes of the spinal vertebral body center in three-dimensional space to capture subtle bending and rotation information of the spine. The ROI module is used to locate and crop point cloud data containing the spine region; The downsampling module, which includes convolutional and pooling layers, is used to extract deep features from the input features. The U-Net module includes an encoder-decoder section, where the encoder section consists of multiple downsampling modules and the decoder section consists of multiple upsampling modules. Features from corresponding levels of the encoder and decoder are fused through a skip connection structure. The upsampling module uses a deconvolutional network layer to locate the position and degree of scoliosis; Curvature sensing module, used to extract curvature features of the spinal surface; The posture fusion module includes a multilayer perceptron, which is used to fuse biomechanical feature vectors extracted from static standing posture and dynamic forward flexion posture, learn the correlation between different posture features and output a comprehensive representation of posture features. The prediction output module is used to output the degree of scoliosis of the subject and the corresponding risk score.

8. The intelligent scoliosis screening method according to claim 1 or 7, characterized in that, The loss function of the scoliosis screening model is: Where N1 is the total number of pixels in the heatmap; h is the height of the heatmap; w is the width of the heatmap; y ij p represents the ideal depth value of the spine surface at coordinates (i,j); ij Let (i,j) be the actual depth value of the spine surface in the state of pixel (i,j); N2 is the number of keypoints; O xi O is the offset of the i-th keypoint in the x-direction; yi Let be the offset of the i-th keypoint in the y-direction.

9. An intelligent scoliosis screening system, characterized in that, include: The posture acquisition module is used to perform a three-dimensional scan of the subject's back using a depth sensing device to acquire 3D point cloud data, including the geometry of the back and the quantified surface appearance. In addition, data on the static standing posture and dynamic forward flexion posture of the subjects were collected; The feature extraction module is used to extract back key point feature vectors, spinal curve feature vectors, and skin tension feature vectors from the 3D point cloud data; and to generate spinal biomechanical feature vectors from the static standing posture and dynamic forward flexion posture data. The screening and assessment module is used to input the back key point feature vector, spinal curve feature vector, skin tension feature vector and spinal biomechanical feature vector into the trained scoliosis screening model, and output the degree of scoliosis and the corresponding risk score of the subject. The degree of scoliosis includes normal, abnormal posture and suspected scoliosis. The report generation module is used to generate a scoliosis screening report based on the subject's back geometry, spinal curve, degree of scoliosis, and corresponding risk score.

10. A computing device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described intelligent scoliosis screening method.

Images