Scoliosis detection method based on muscle movement and body posture information

Abstract

The invention discloses a scoliosis detection method based on muscle movement and body posture information. The method comprises the following steps: acquiring a corresponding electromyographic signal, a near-infrared signal and an inertial signal for a target spine; inputting the electromyographic signals, the near-infrared signals and the inertial signals into a trained graph neural network model to obtain an end cone of each section of spine; determining the central point of the spine according to the end cone of each section of spine, and calculating the Cobb angle of the scoliosis bending section; and based on the Cobb angle, determining the lateral bending section information of the target spine. According to the method, the accuracy of predicting the scoliosis angle by the neural network model is improved, and early low-cost screening of scoliosis is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of electrophysiological detection technology, and more specifically, to a scoliosis detection method based on muscle movement and body posture information. Background Art

[0002] Scoliosis is a spinal disease that occurs in adolescents during the rapid growth and development period. The spine deviates from the midline and bends sideways, accompanied by rotation of the vertebral bodies. The disease generally develops progressively, the scoliosis angle continues to increase, and the deformity gradually worsens, seriously affecting the patient's physical and mental health. Early detection and early treatment are the key to preventing scoliosis from worsening.

[0003] The existing technology mainly uses magnetic resonance imaging (MRI), X-rays, optical images, spinal measuring rulers, electromyography and near-infrared blood oxygen to carry out the detection and early diagnosis of scoliosis. For example, patent application CN117765062A (Image processing method and system for adolescent scoliosis detection) uses MRI and X-rays to detect images, obtains a training set of a classification model to train the classification model, and obtains a trained classification model; inputs the scoliosis image to be classified into the trained classification model to classify the scoliosis image; inputs the scoliosis image to be predicted into the trained deep learning model to predict the center point of the spine; the scoliosis image includes a single-bend image of the spine, a double-bend image of the spine, and a triple-bend image of the spine; different models are selected according to different image structures to predict the center point of the spine. This solution involves radioactive radiation, is complex to operate, and has high costs.

[0004] Patent application CN118629068A (a method, device, terminal and medium for detecting abnormal human spine posture) performs image segmentation and edge extraction processing on the human posture detection image based on the human posture detection image taken by the camera to obtain a human body feature image of the human body part. Then, a number of human body posture detection points for detecting the human spine are determined based on the human body feature image, and human body posture feature parameters are calculated based on the human body posture detection points. The calculated human body posture feature parameters are then input into a human body posture detection model trained based on a preset human body posture feature parameter sample, and the human body posture feature parameters of the detection object are detected based on the prior knowledge formed by the training of the human body posture detection model to obtain the human body posture detection result of the detection object. This image-based detection method has a low detection accuracy rate for scoliosis, and cannot accurately calculate the angle of scoliosis, and can only serve as a preliminary basis for auxiliary diagnosis.

[0005] In addition, the electromyography and near-infrared blood oxygen detection methods require limb movement to generate relatively obvious electromyography and blood oxygen information. The classification and recognition of different movements are relatively accurate, but the estimation errors of angles such as bending angle and spinal Cobb angle are large. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a scoliosis detection method based on muscle movement and body posture information. The method comprises the following steps:

[0007] For the target spine, collect the corresponding electromyographic signals, near infrared signals and inertial signals;

[0008] Inputting the electromyographic signal, the near infrared signal and the inertial signal into a trained graph neural network model to obtain the end cone position of each vertebra;

[0009] According to the position of the end cone of each vertebra, the center point of the spine is determined, and the Cobb angle of the scoliotic curvature segment is calculated;

[0010] Based on the Cobb angle, scoliosis curve segment information of the target spine is determined.

[0011] Compared with the prior art, the advantage of the present invention is that, based on the existing electromyographic acquisition electrodes, a new integrated electrode acquisition module is designed, which can synchronously acquire electromyographic signals, near-infrared signals and inertial signals. Compared with the prior art, the measurement of inertial signals in different postures of the human body is added, and the real-time and accurate acquisition of limb movements and their spatial posture information is realized, which can further improve the accuracy of the neural network model in predicting the scoliosis angle. At the same time, it is easy to carry and has no radiation to the human body, which is helpful for the early and low-cost screening of scoliosis. In short, the present invention combines multimodal measurement signals and neural network models to achieve accurate positioning of the end cone of each vertebra on the back of the human body, thereby achieving high-precision detection of scoliosis.

[0012] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0014] Figure 1 is a flow chart of a scoliosis detection method based on muscle movement and body posture information according to an embodiment of the present invention;

[0015] Figure 2 is a top view of a comprehensive electrode acquisition module according to an embodiment of the present invention;

[0016] Figure 3 is a bottom view of a comprehensive electrode acquisition module according to one embodiment of the present invention;

[0017] Figure 4 is a plan view of a comprehensive electrode acquisition module according to an embodiment of the present invention;

[0018] Figure 5 is a side view of a comprehensive electrode acquisition module according to an embodiment of the present invention;

[0019] Figure 6 It is a schematic diagram of a comprehensive electrode acquisition module arranged on both sides of a bisecting line of the human back spine according to an embodiment of the present invention;

[0020] Figure 7 It is a schematic diagram of a comprehensive electrode acquisition module arranged on both sides of a bisecting line of the human back spine according to an embodiment of the present invention;

[0021] Figure 8 is a schematic diagram of a human back spine bisection line according to an embodiment of the present invention;

[0022] Fig. 9 is a schematic diagram of a modeling method of a neural network model according to an embodiment of the present invention;

[0023] Fig.10 4 is a schematic diagram of Cobb angle calculation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0025] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0026] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0027] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0028] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0029] In general, the present invention designs a new collection electrode based on the existing electromyographic collection electrode, which can collect electromyographic signals, near-infrared signals and IMU inertial signals at the same time. Then, a neural network model is established and trained, and the electromyographic signals, near-infrared signals and inertial signals are used as inputs to realize the positioning of the end cone of each vertebra in the back of the human body, thereby improving the accuracy of scoliosis detection.

[0030] Specifically, see Figure 1 As shown, the provided scoliosis detection method based on muscle movement and body posture information comprises the following steps:

[0031] Step S110, designing a comprehensive electrode acquisition module to integrate the functions of measuring electromyographic signals, near-infrared signals and inertial signals.

[0032] See also Figures 2 to 5 As shown, the present invention designs a comprehensive electrode acquisition module (or simply comprehensive electrode), wherein Figure 2 This is a top view of the integrated electrode acquisition module. Figure 3 This is a bottom view of the integrated electrode acquisition module. Figure 4 This is a plan view of the integrated electrode acquisition module. Figure 5 This is a side view of the integrated electrode acquisition module. The integrated electrode acquisition module integrates near infrared and inertial measurement functions in the existing electromyography acquisition module. The upper part is the existing distributed electromyography acquisition module, and the lower part is provided with an arc-shaped cavity. The arc-shaped cavity can be integrated with infrared acquisition modules and inertial measurement modules.

[0033] In one embodiment, the near-infrared acquisition module integrated in the arc-shaped cavity of the lower part includes a semiconductor light source, three photodetectors and a light shield, etc. After the semiconductor light source emits light of a preset wavelength, it penetrates the skin and is reflected, and then received by the photodetector to generate a near-infrared signal.

[0034] For example, the semiconductor light source is an integrated light emitting diode, which integrates 12 light emitting diodes (LEDs) inside, of which 4 LEDs have peak wavelengths of 850nm, 805nm, and 735nm. The choice of the wavelength of the irradiation light can be determined based on the absorption spectrum of the main absorption chromophore in human tissue. Because oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (HHb) in human blood have an isosbestic point near 805nm, the use of light with a wavelength of 805nm can reflect the change information of blood volume in tissues. The wavelengths of the other two light sources (735nm, 850nm) are selected on both sides of 805nm to ensure the differential effect of HbO2 and HHb on light absorption when blood oxygen changes. At the same time, an independent constant current drive power supply is set for the integrated light emitting diode.

[0035] In practical applications, the semiconductor light source is controlled by generating a control signal to make the semiconductor light source amplitude modulated to radio frequency. The detection section needs to measure not only the light intensity information, but also the phase deviation of the outgoing light relative to the incident light and the change in modulation depth. When the frequency is lower than 200MHz, the change in phase angle is linearly related to the average optical path length. Therefore, if the phase difference angle between the incident light and the outgoing light is known, the attenuation change of the light in the tissue can be directly converted into the concentration change of the absorbing chromophore.

[0036] In one embodiment, the photodetector uses a monolithic photodiode and integrates a transconductance amplifier, and the output voltage of the photodetector is linearly related to the intensity of the detected light. For example, there are three photodetectors arranged in a triangle, and the three photodetectors receive photoelectric signals and superimpose and couple to form a near-infrared signal.

[0037] In order to prevent the photodetector from directly receiving the light emitted by the semiconductor light source, a light shielding plate is set between the semiconductor light source and the photodetector. The light shielding plate is made of a flexible material and its height is consistent with the arc-shaped cavity of the near-infrared acquisition module.

[0038] In one embodiment, the inertial measurement module placed in the arc-shaped cavity may use a six-axis or nine-axis inertial sensor to output quaternion data of the electrode placement position.

[0039] In addition, combined Figures 2 to 5 As shown, the integrated electrode also includes a signal transmission module disposed inside, which is used to transmit the electromyographic signal, near infrared signal and inertial signal collected by the integrated electrode to the controller. In addition, the signal transmission part also includes a shielded power line for supplying power to the electromyographic, near infrared and inertial acquisition modules.

[0040] Step S120, laying integrated electrodes on both sides of the bisecting line of the human spine on the back to collect electromyographic signals, near-infrared signals and inertial signals, and constructing a training set.

[0041] During the application process, the integrated electrodes are laid on both sides of the bisecting line of the human spine on the back to collect multimodal signals such as electromyographic signals, near-infrared signals and inertial signals, and the collected multimodal signals are divided into training sets and test sets. Figure 6 and Figure 7 This is a schematic diagram of laying integrated electrodes on both sides of the bisecting line of the human back spine, where Figure 6 Corresponding to the normal curvature of the spine, Figure 7 Illustration of scoliosis.

[0042] In actual operation, Figure 8As shown, the middle part of the human neck and the coccyx are connected to generate a bisection line of the human back spine. Further, based on different populations, different pairs of integrated electrodes are used and distributed on both sides of the bisection line of the human back spine according to a predetermined ratio, and the spacing between each pair of integrated electrodes is in a fixed coefficient ratio with the spacing between the upper and lower integrated electrodes.

[0043] For example, for children under 12 years old, 4 pairs of integrated electrodes are used, and the coefficient ratio is 1. For people over 12 years old, 5 pairs of integrated electrodes are used, and the coefficient ratio is 1.3.

[0044] When the integrated electrode is working, electrogel is applied to the outer edge of the upper electromyographic acquisition module and the lower arc-shaped cavity. The electrogel of the upper electromyographic acquisition electrode can reduce the contact impedance between the skin and the electrocardiogram electrode to ensure the clarity of the electromyographic signal transmission. The near-infrared acquisition module in the lower part can seal the outer edge of the arc-shaped cavity through the electrogel, thereby achieving a dark environment inside the arc-shaped cavity to avoid light. The inertial measurement module placed together collects inertial signals during muscle activity to obtain accurate data on joint angles.

[0045] like Figure 6 or Figure 7 As shown, after multiple integrated electrodes are symmetrically laid on both sides of the bisecting line of the human back spine, electromyographic signals, near-infrared signals and inertial signals can be collected simultaneously.

[0046] After the electromyographic signal, near-infrared signal and inertial signal are collected, the signal can be preprocessed by a preprocessing circuit to eliminate basic bioelectric noise and environmental noise interference. For example, most of the noise signals aliased in the electromyographic signal are common-mode signals, so the use of differential signal input can effectively suppress common-mode interference. In one embodiment, the basic bioelectric noise and environmental noise interference are eliminated by collecting using a pre-placed front-end electrode, and the pre-placed front-end electrode is used as a reference electrode for the electromyographic signal input.

[0047] For example, the input end of the reference electrode is coupled to the output end of the electromyography acquisition electrode to invert and amplify the power frequency signal and output it to the bioelectric information of the object being measured, thereby achieving preprocessing of bioelectric noise and environmental noise interference.

[0048] Furthermore, the EMG signal is analyzed by wavelet transform to obtain multi-layer wavelet coefficients, and the multi-layer wavelet coefficients are subjected to denoising threshold processing to obtain the final EMG signal data after denoising. The denoising threshold can be set as:

[0049]

[0050] Among them, η i is the threshold of the wavelet coefficient of layer i, i is the number of decomposition layers, σ is the standard deviation of the noise, and N is the length of the signal.

[0051] The electromyographic signal, near infrared signal and inertial signal are matched one by one according to the channel number, and a training set and a test set are generated respectively. The training set and the test set each contain multiple sample data, and each sample data reflects the corresponding relationship between the electromyographic signal, the near infrared signal and the inertial signal and the end cone position of the spine. The training set and the test set are both historical user data, and in addition to the electromyographic, near infrared and inertial signals, the data also include the position of the end cone position of each spine relative to the integrated electrode.

[0052] Step S130, using the constructed training set to train the graph neural network model until the set loss standard is met.

[0053] In this step, a neural network model is established and pre-trained. The neural network model uses electromyographic signals, near infrared signals, and inertial signals as inputs to determine the end cone of each vertebra. In one embodiment, a graph neural network (GNN) is constructed to learn graph structure data, extract and explore features and patterns in the graph structure data, and predict the end cone of the spine.

[0054] For example, for graph neural networks, small-scale sequences can be organized into graphs, and the hidden information in the sequence can be obtained through more complex topological structures. This is especially true for multiple groups of electromyographic signals, near-infrared signals, and inertial signals, which have large modal changes and no uniform characteristic rules. Fuzzy prediction tasks can be better accomplished. The graph described below is a graph constructed from small-scale sequence data, see Fig. 9 The modeling method of the example neural network model is shown in FIG. The method includes the following steps:

[0055] Step S1: input the sequence relationship to be predicted (the sequence relationship is the electromyographic signals, near infrared signals and inertial signals collected at multiple moments), and perform data cleaning and feature extraction on the sequence relationship to obtain low-noise sequence relationship data.

[0056] Step S2: Using a dual adjacency matrix, graph representation modeling is performed on the low-noise sequence relationship data, and using a betweenness centrality algorithm to introduce some prior knowledge into the graph to obtain initial features of the sequence relationship.

[0057] Step S3: Use the gated graph neural network to process the sequence relationship, model the topological structure in the sequence relationship, and obtain the topological relationship features in the sequence relationship, that is, the global features of the sequence relationship.

[0058] Step S4: Use the attention mechanism to model the relationship between different nodes in the sequence relationship, obtain the features of different nodes in the sequence relationship, that is, the local features of the sequence relationship, splice the above global features and local features to obtain spliced ​​features, predict the global and local spliced ​​features, and obtain the prediction of the next node in the final sequence.

[0059] Step S5: Using the training set as input, the neural network model is trained, and the key values ​​in the loss function are calculated based on the test set to obtain a trained neural network model.

[0060] First, a dual adjacency matrix is ​​used to store the graph. An out-degree matrix and an in-degree matrix are constructed for each graph. In order to facilitate input into the neural network, the degree values ​​are normalized, one matrix represents the out-degree and the other matrix represents the in-degree.

[0061] By using the out-degree matrix and the in-degree matrix, a model containing rich bidirectional relationships is obtained instead of simple unidirectional transfer relationships.

[0062] The betweenness centrality algorithm is used to calculate the betweenness centrality of node relationships in the graph, which is defined as:

[0063]

[0064] Among them, C B (v) represents the node betweenness centrality, s represents the node starting point, v represents all nodes, t represents the node end point, σ st (v) represents the number of shortest paths from s to t passing through node v, σ st Represents the number of shortest paths from s to t.

[0065] The idea of ​​data augmentation is adopted to augment a sequence into multiple sequences to obtain more information. Data augmentation solves the problem that the order of the original sequence is confused when the graph is directly generated based on the sequence, and it is impossible to clearly indicate which node the second node points to first.

[0066] After completing the modeling of the graph, the gated graph neural network GGNN is used to learn the interaction relationship between different nodes from the modeled adjacency matrix with initial prior knowledge.

[0067] The gated graph neural network is used to complete its information aggregation method by summing the neighboring nodes to obtain the input vector h of the gated graph neural network. v :

[0068]

[0069] Among them, w urepresents the neighbor node vector, N(v) represents the set of all nodes, W represents the weight value, Represents the adjacency matrix generated in the first step, including A (in) The in-degree matrix and A (out) The out-degree matrix represents the input vector of the i-th node of the graph neural network obtained by adding the bias term in the neural network.

[0070]

[0071] Among them, A g,i: represents the target graph adjacency matrix of the nodes, represents the D|V|-dimensional vector formed by the concatenation of all node features at time t-1, g represents the target graph containing n nodes, and b is the bias term in the neural network.

[0072] The hidden vectors of different nodes in the graph are extracted from the above input vectors to obtain vectors containing the node's own information and the relationship information of neighboring nodes.

[0073] After obtaining the input of the above neural network, the input is passed into the forget gate of GGNN. The formula of the forget gate is as follows:

[0074]

[0075] in, represents the input vector of the i-th forget gate, W z Represents the first weight value, U z represents the second weight value, Represents the characteristics of the i-th node.

[0076] After the forget gate determines the information to be discarded by the current unit, the above vector is input into the memory gate to select the new information stored in the unit module. The formula of the memory gate is as follows:

[0077]

[0078] in, represents the memory gate input vector, U r Represents the third weight value.

[0079] Use the output gate to determine the output value. The vector is fed into the Sigmiod function to determine which part of the neural network state needs to be output.

[0080] The network state of the above output is passed to the tanh function for processing, and the two are multiplied to get the final output. The formula of the output layer is expressed as:

[0081]

[0082] Among them, W O Represents the fourth weight value, U O Represents the fifth weight value, tanh is the hyperbolic tangent function, and this activation function is used to process the output of the previous layer.

[0083] After the training of the above GGNN is completed, the embedding vectors of all nodes are obtained. For each small-scale graph g, the gated graph neural network GGNN processes all nodes at the same time.

[0084] The information propagation between different nodes is carried out using the supervision of the gated graph neural network relationship matrix. For each node, the relationship between its adjacent nodes is extracted to generate a latent vector which is input into the subsequent GNN.

[0085] Use the forget gate and update gate to obtain the information that needs to be discarded and retained.

[0086] The candidate state is calculated by combining the current state, the previous state and the forget gate. The final state is calculated based on the above candidate state, the previous state and the update gate.

[0087] Repeat the above process until convergence to obtain the embedding vector of each node.

[0088] The above graph node representation generation process obtains the representation of each node. The following uses the attention mechanism to obtain the embedded representation of the graph, that is, the global information.

[0089] The last node embedding vector in the above small-scale sequence is used as the local embedding vector, that is,

[0090] s1=v n (8)

[0091] Among them, s1 represents the last node embedding vector in the small-scale sequence, v n represents the local embedding vector.

[0092] Define the global variable as the vector representation obtained by combining all node embedding vectors and adding different attention priorities. Using the attention mechanism, first get the attention weights α of different nodes i , and then add different attention to different nodes to obtain the global vector embedding result. The formula is as follows:

[0093] α i =q T σ(W1v n +W2v i +c) (9)

[0094] Among them, W1 and W2 are trainable weights in the attention mechanism, c represents the bias, and v n represents the local embedding vector, vi Represents the global embedding vector, and then combines the local embedding vector in the graph with the global embedding vector. The formula is as follows

[0095] s h =W3[s l ;s g ](10)

[0096] Among them, s l is the local vector, s g is the global vector, W3 is the trainable matrix that fuses the two into the same space, and finally the fused embedding vector s is obtained h , that is, the representation of the graph.

[0097] Use the graph embedding vector s obtained for each sequence h Calculate the score for each candidate node Use Softmax to get the predicted output. The formula is as follows

[0098]

[0099] in, For the final prediction score, we can use the real label combined with the predicted output to get the loss value through the cross entropy loss function, and use the back propagation algorithm based on the training set and the test set to perform reverse training of the propagation and output process network, and repeat the above process.

[0100] Step S140, for the target spine to be tested, the corresponding electromyographic signal, near-infrared signal and inertial signal are used as input, and the trained graph neural network model is used to determine the end cone of each vertebra, and then the center point of the spine is inferred based on the end cone of each vertebra, and the Cobb angle of the scoliosis curvature segment is calculated.

[0101] By using the trained neural network model, the EMG signal, near infrared signal and inertial signal collected from the target spine are used as input to predict the correspondence between the end cone of each vertebra of the user and the position of each adjacent integrated electrode acquisition module. Furthermore, the center point of the spine is estimated based on the end cone of each vertebra, and the Cobb angle of the scoliotic curvature segment is calculated.

[0102] Combination Fig.10 As described above, the calculation of the Cobb angle includes the following process:

[0103] Based on the correspondence between the end cones of each spine of the user and the positions of each adjacent integrated electrode acquisition module, the position of the spine center point is inferred, and then based on the adjacent spine center points, the position of the upper and lower vertebrae of each spine is determined;

[0104] For each vertebra, draw a horizontal line along the upper and lower vertebrae of the spine at the upper edge of the vertebral body of the upper vertebra, and draw a horizontal line at the lower edge of the vertebral body of the lower vertebra, and then draw a vertical line on each of the two horizontal lines;

[0105] Calculate the angle of each vertebra relative to the vertical line of other vertebrae, which is the Cobb angle.

[0106] The cases where the Cobb angle is greater than the preset threshold are screened, the vertebral number corresponding to the Cobb angle is recorded, and the scoliosis detection is completed.

[0107] The size of the Cobb angle is usually related to the severity of scoliosis. For example, if it is less than 10°, it is usually considered a variation within the normal range and does not require treatment; in the range of 10° to 25°, observation and regular review may be required; in the range of 25° to 50°, bracing may be required for treatment; if it is greater than 50°, surgical treatment may need to be considered. Based on the determination of the Cobb angle, the scoliosis curvature segment can be automatically determined by sorting according to the angle size.

[0108] In summary, compared with the prior art, the present invention has the following advantages:

[0109] 1) Based on the existing electromyographic acquisition electrodes, the present invention designs a new integrated electrode acquisition module that can synchronously collect electromyographic, near-infrared and inertial signals. Compared with the prior art, the measurement of inertial signals in different postures of the human body is added, and the real-time and accurate acquisition of limb movements and their spatial posture information is realized, which can further improve the accuracy of the neural network model in predicting the scoliosis angle. At the same time, it is easy to carry and has no radiation to the human body, which is helpful for the early and low-cost screening of scoliosis.

[0110] 2) The present invention constructs a neural network model, using multi-mode signals of electromyographic signals, near-infrared signals and inertial signals as input to predict the position of the spinal end cone. The model can obtain more detailed features of the input data through the convolution layer, and accurately locate the end cone of each spinal segment on the back of the human body. Experimental verification shows that the present invention can achieve high-precision detection of scoliosis.

[0111] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0112] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0113] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0114] The computer program instructions for performing the operation of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, Python, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present invention.

[0115] Various aspects of the present invention are described herein with reference to the flow charts and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each box of the flow chart and / or block diagram and the combination of each box in the flow chart and / or block diagram can be implemented by computer-readable program instructions.

[0116] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0117] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0118] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a part of a module, a program segment or an instruction, and a part of the module, a program segment or an instruction contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that it is equivalent to implement it by hardware, implement it by software, and implement it by combining software and hardware.

[0119] Embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the marketplace, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A scoliosis detection method based on muscle movement and body posture information, comprising the following steps: For the target spine, collect the corresponding electromyographic signals, near infrared signals and inertial signals; Inputting the electromyographic signal, the near infrared signal and the inertial signal into a trained graph neural network model to obtain the end cone position of each vertebra; According to the position of the end cone of each vertebra, the center point of the spine is determined, and the Cobb angle of the scoliotic curvature segment is calculated; Based on the Cobb angle, scoliosis curve segment information of the target spine is determined.

2. The method according to claim 1, characterized in that The electromyographic signal, the near-infrared signal and the inertial signal are collected by using an integrated electrode acquisition module, the integrated electrode acquisition module is arranged on both sides of the target spine bisector, and the integrated electrode acquisition module integrates a distributed electromyographic acquisition module, a near-infrared acquisition module and an inertial measurement module, the distributed electromyographic acquisition module is arranged in the first part of the integrated electrode acquisition module, and an arc-shaped cavity is arranged in the second part of the integrated electrode acquisition module, and the near-infrared acquisition module and the inertial measurement module are integrated in the arc-shaped cavity.

3. The method according to claim 2, characterized in that The infrared acquisition module includes a semiconductor light source, a photodetector and a shading plate. The semiconductor light source integrates 12 light-emitting diodes, of which 4 are light-emitting diodes with peak wavelengths of 850nm, 805nm and 735nm respectively. The number of the photodetectors is set to three, which are arranged in a triangle. After receiving the photoelectric signal, the three photodetectors are superimposed and coupled to form the near-infrared signal. A shading plate is set between the semiconductor light source and the photodetector, and the height of the shading plate matches the arc cavity of the near-infrared acquisition module. The inertial measurement module adopts a six-axis or nine-axis inertial sensor to output quaternion data of the electrode placement position.

4. The method according to claim 1, characterized in that: The integrated electrode acquisition module also includes a signal transmission module arranged inside, which is used to transmit the collected electromyographic signal, the near infrared signal and the inertial signal to a controller.

5. The method according to claim 1, characterized in that The electromyographic signal is preprocessed according to the following steps: The pre-placed front electrode is used as the reference electrode for electromyographic signal input. The input end of the reference electrode is coupled to the output end of the collection electrode of the electromyographic collection module to invert and amplify the power frequency signal and then output it to the bioelectric information of the object being measured; The collected electromyographic signal is transformed by wavelet to obtain multi-layer wavelet coefficients, and the multi-layer wavelet coefficients are subjected to denoising threshold processing to obtain denoised electromyographic signal, wherein the denoising threshold is set as: Among them, η i is the threshold of the wavelet coefficient of layer i, i is the number of decomposition layers, σ is the standard deviation of the noise, and N is the length of the signal.

6. The method according to claim 1, characterized in that The graph neural network model is trained according to the following steps: Input the sequence relationship to be predicted and perform data cleaning and feature extraction to obtain low-noise sequence relationship data, wherein the sequence relationship is the electromyographic signal, near infrared signal and inertial signal collected at multiple moments; The low-noise sequence relationship data is modeled using a dual adjacency matrix, and the betweenness centrality algorithm is used to introduce prior knowledge into the graph to obtain the initial features, where the node betweenness centrality C in the graph B (v) is defined as: Among them, C B (v) represents the node betweenness centrality, s represents the node starting point, v represents all nodes, t represents the node end point, σ st (v) represents the number of shortest paths from s to t passing through node v, σ st Represents the number of shortest paths from s to t; The graph neural network model is used to model the topological structure in the low-noise sequence relationship to obtain topological relationship features as global features. Using an attention mechanism to model the relationship between different nodes in the low-noise sequence relationship, and obtaining features of different nodes in the sequence relationship as local features; The global feature and the local feature are concatenated to obtain a concatenated feature, and then the concatenated feature is predicted to obtain a prediction of the next node of the final sequence; The graph neural network model is trained using a training set with a set loss function as an optimization target, wherein the training set reflects the correspondence between electromyographic signals, near-infrared signals, inertial signals and the end cone positions of the spine.

7. The method according to claim 6, characterized in that The graph neural network model is a gated graph neural network model, and the output of the i-th node is expressed as in: Among them, A g,i: represents the target graph adjacency matrix of the nodes, represents the vector formed by concatenating all node features at time t-1, g represents the target graph containing all n nodes, b is the bias term, and W z , W and U r is the weight of the corresponding item, represents the memory gate input vector, is the input vector of the graph neural network model, Represents the i-th node feature at time t-1.

8. The method according to claim 6, characterized in that In the process of training the graph neural network, a data augmentation strategy is used to augment one sequence into multiple sequences.

9. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images