Multi-dimensional physiological data automatic intelligent screening system for adolescent idiopathic scoliosis

Abstract

The application discloses a kind of multidimensional physiological data automatic juvenile idiopathic scoliosis intelligent screening system, it is related to scoliosis field, the system includes camera module, double shoulder height measurer, waist stress sensing device, back curve stress sensing device, foot pressure distribution testing device and scoliosis intelligent analysis module;Camera module is photographed to the back and side and the back when body is forward bent when standing upright;Double shoulder height measurer measures the height of double shoulder;Waist stress sensing device applies force to two sides waist, measures the reaction force of side waist;Back curve stress sensing device is used to measure back curve;Foot pressure distribution testing device measures the pressure of double foot;Scoliosis intelligent analysis module judges scoliosis risk degree according to photo, height difference of double shoulder, reaction force of side waist, back curve and the pressure of double foot, the application inspection process is fast and no radiation, and the accuracy of inspection result is high.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of scoliosis, in particular to a multi-dimensional physiological data automatic juvenile idiopathic scoliosis intelligent screening system. BACKGROUND

[0002] The existing juvenile idiopathic scoliosis screening is mainly through artificial palpation by specialist doctors, scoliosis detection ruler and clinical X-ray film for diagnosis, the accuracy of which is affected by the personal experience and method of the doctor and may miss diagnosis, and multiple X-ray diagnosis exists radiation, and the disease needs to be screened for many years, and single person screening needs 3-5 minutes, which is not suitable for multiple screenings of large population. SUMMARY

[0003] The purpose of the present application is to provide a multi-dimensional physiological data automatic juvenile idiopathic scoliosis intelligent screening system, which is fast and radiation-free in the checking process, and has high accuracy in the checking result.

[0004] To achieve the above purpose, the present application provides the following scheme: the present application provides a multi-dimensional physiological data automatic juvenile idiopathic scoliosis intelligent screening system, which comprises a camera module, a double-shoulder height measurer, a waist stress sensing device, a back curve stress sensing device, a foot pressure distribution testing device and a scoliosis intelligent analysis module.

[0005] The camera module is used for photographing the back and side of the user when standing upright and the back of the user when performing forward bending; the double-shoulder height measurer is used for measuring the height of the user's shoulders; the waist stress sensing device is used for applying equal forces to the user's two sides of the waist and measuring the values of the reaction forces of the two sides of the waist; the back curve stress sensing device is used for measuring the curve of the two sides of the spine processus spinosus to obtain the back curve; and the foot pressure distribution testing device is used for measuring the pressure of the user's feet.

[0006] The scoliosis intelligent analysis module is used for obtaining the height difference of the user's shoulders according to the height of the user's shoulders, obtaining the key points of each vertebra processus spinosus of the user's spine according to the back photos of the user when standing upright and when performing forward bending, obtaining the key point connecting line of the spine processus spinosus according to the key points of each vertebra processus spinosus of the user's spine, determining the angle between the trunk central axis of the user and the horizontal line according to the side photo of the user, and obtaining the risk degree of scoliosis of the user's spine according to the angle between the key point connecting line of the spine processus spinosus of the user and the vertical line, the angle between the trunk central axis and the horizontal line, the height difference of the shoulders, the values of the reaction forces of the two sides of the waist, the back curve and the pressure of the feet.

[0007] According to the specific embodiments provided in the application, the application has the following technical effects: the application provides a multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system, all diagnosis processes are diagnosed by a scoliosis intelligent analysis module according to a back photo, a side photo, the height of both shoulders, the value of the reaction force of both sides of the waist, the back curve and the pressure of both feet, without manual palpation, and the checking process is fast and the checking result is accurate, without CT and radiation. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0009] Figure 1 A multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system structure schematic diagram is provided for an embodiment of the application.

[0010] Figure 2 An improved HRNet network structure diagram is provided for the application.

[0011] Figure 3 A convolution block attention module structure diagram.

[0012] The drawings are as follows: double shoulder height measurer-1, waist stress sensing device-2, back curve stress sensing device-3, foot pressure distribution testing device-4, display-5, first camera-6, second camera-7, door-8. DETAILED DESCRIPTION

[0013] The technical solutions in the embodiments of the application will be described clearly and completely in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0014] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail in the following with reference to the drawings and specific embodiments.

[0015] In an exemplary embodiment, as Figure 1As shown, a multi-dimensional physiological data automatic intelligent screening system for adolescent idiopathic scoliosis is provided, comprising: a camera module, a double-shoulder height measurer 1, a waist stress sensing device 2, a back curve stress sensing device 3, a foot pressure distribution testing device 4, and an intelligent analysis module for scoliosis.

[0016] The camera module is used to take photos of the back and side of the user when standing upright and the back of the user when performing forward bending; the double-shoulder height measurer 1 is used to measure the height of the user's shoulders; the waist stress sensing device 2 is used to apply equal forces to the user's two sides of the waist and measure the values of the reaction forces of the two sides of the waist; the back curve stress sensing device 3 is used to measure the curve of the two sides of the spine's spinous process to obtain the back curve; and the foot pressure distribution testing device 4 is used to measure the pressure of the user's feet.

[0017] The intelligent analysis module for scoliosis is used to obtain the height difference of the user's shoulders according to the height of the user's shoulders; obtain the key points of each vertebra of the user's spine according to the photos of the back of the user when standing upright and the photos of the back of the user when performing forward bending; obtain the key point line of the spine's spinous process according to the key points of each vertebra of the user's spine; determine the angle between the user's trunk central axis and the horizontal line according to the photo of the side of the user; and obtain the risk degree of scoliosis of the user's spine according to the angle between the key point line of the spine's spinous process and the vertical line (a line perpendicular to the horizontal line), the angle between the user's trunk central axis and the horizontal line, the height difference of the user's shoulders, the values of the reaction forces of the two sides of the waist, the back curve, and the pressure of the feet.

[0018] In actual application, the camera module comprises a first camera 6 and a second camera 7. The first camera 6 is used to take the photo of the side, and the second camera 7 is used to take the photo of the back.

[0019] In another exemplary embodiment of the present application, the key points of each vertebra of the user's spine are obtained according to the photos of the back of the user when standing upright and the photos of the back of the user when performing forward bending, specifically: a spine key point detection algorithm is used to process the photos of the back of the user when standing upright and the photos of the back of the user when performing forward bending to obtain a spinous process key point heat map, and the key points of each vertebra of the spine are obtained according to the spinous process key point heat map.

[0020] The spine key point detection algorithm proposed in the present application takes HRNet as the basic backbone network, retains the core advantage of "high-resolution feature flow throughout", and realizes targeted improvement through embedding attention mechanism. In another exemplary embodiment of the present application, the spine key point detection algorithm is an improved HRnet model, such as Figure 2As shown, the HRNet network, the first CBAM network, the second CBAM network and the third CBAM network are included, the second stage and the third stage of the HRNet network are connected through the first CBAM network, the third stage and the fourth stage of the HRNet network are connected through the second CBAM network, and the output end of the HRNet network is connected with the third CBAM network.

[0021] The spine key point detection algorithm provided in the application includes a feature extraction layer, a feature fusion layer and an output layer. The feature extraction layer: the multi-branch parallel structure of the HRNet network is used, and a high-resolution-to-low-resolution feature map branch is gradually constructed through four stages. Each stage includes multiple residual units, which extract image bottom layer features (such as gray changes and edge contours) and middle layer features (such as overall morphology of vertebral bodies and intervertebral space distribution) through convolution, batch normalization and ReLU activation function. Different from the original HRNet network, the application embeds a convolution block attention module (CBAM) after the residual unit of each resolution branch from the second stage, so as to realize dynamic screening and enhancement of features. Through CBAM, the channel and spatial dimensions of each branch feature are doubly weighted, the feature response of key areas such as vertebral body edges, pedicles and dentate processes is highlighted, and the interference of irrelevant information such as soft tissues, noises and artifacts is suppressed.

[0022] The feature fusion layer: the cross-resolution fusion strategy unique to the HRNet network is adopted, the feature maps of different resolution branches are spliced and fused through bilinear interpolation upsampling and downsampling, so that the high-resolution branch can not only retain detailed information (such as the slight protrusions of vertebral body edges), but also fuse global semantic information (such as the continuity of the spine sequence) of the low-resolution branch, and finally output a high-resolution feature map.

[0023] The output layer: the fused feature map is mapped to a spinous process key point heat map through 1x1 convolution, the size of the heat map is the same as the high-resolution feature Figure 1 Therefore, each channel corresponds to a spinous process key point, and the peak position in the heat map is the coordinate of the key point.

[0024] In another exemplary embodiment of the application, as Figure 2 shown, the first CBAM network includes a first CBAM, a second CBAM and a third CBAM.

[0025] The input end of the first CBAM is connected with the output end of the first convolution block of the second stage of the HRNet network; the output end of the first CBAM is connected with the input end of the first convolution block of the third stage of the HRNet network.

[0026] The input end of the second CBAM is connected with the output end of the down-sampling of the second stage of the HRNet network and the output end of the up-sampling respectively; the output end of the second CBAM is connected with the input end of the up-sampling of the third stage of the HRNet network and the input end of the down-sampling respectively.

[0027] The input end of the third CBAM is connected with the output end of the second convolution block of the second stage of the HRNet network; the output end of the third CBAM is connected with the input end of the second convolution block of the third stage of the HRNet network.

[0028] In another exemplary embodiment of the present application, as shown in Figure 2 The second CBAM network includes a fourth CBAM, a fifth CBAM and a sixth CBAM.

[0029] The input end of the fourth CBAM is connected with the output end of the third convolution block of the third stage of the HRNet network; the output end of the fourth CBAM is connected with the input end of the first convolution block of the fourth stage of the HRNet network.

[0030] The input end of the fifth CBAM is connected with the output end of the fourth convolution block of the third stage of the HRNet network; the output end of the fifth CBAM is connected with the input end of the second convolution block of the fourth stage of the HRNet network.

[0031] The input end of the sixth CBAM is connected with the output end of the fifth convolution block of the third stage of the HRNet network; the output end of the sixth CBAM is connected with the input end of the third convolution block of the fourth stage of the HRNet network.

[0032] In another exemplary embodiment of the present application, as shown in Figure 3 The third CBAM network includes a seventh CBAM and a sixth convolution block connected in sequence; the input end of the seventh CBAM is connected with the output end of the fourth convolution block of the fourth stage of the HRNet network.

[0033] In actual application, the convolution block attention module is a kind of lightweight attention mechanism, structure as shown in σ Through the series cooperation of channel attention and spatial attention, the "importance weighting" of feature map is realized, and after embedding the residual unit of each branch of the HRNet network, the specific action mechanism is as follows: channel attention: focus on which feature channel is more important, by learning the weight of channel dimension, the channel containing key anatomical information (such as the high contrast channel of vertebral edge) is strengthened, and the noise or background channel is weakened. The implementation process is: the input feature map FGlobal Average Pooling and Global Max Pooling are performed on the feature map (shape of CxHxW, C is the number of channels, H and W are height and width respectively) to obtain two channel descriptors of 1x1xC; both are sent to a shared multi-layer perceptron (with 1 hidden layer) to output a channel weight vector; the weight is normalized to the interval of 0-1 through the Sigmoid activation function, multiplied with the channel dimension of the original feature map to realize channel-level feature enhancement. The calculation process can be represented as: .

[0034] wherein, represents the result of Global Average Pooling, with a dimension of ; represents a real number; represents the result of Global Max Pooling, with a dimension of ; represents the channel attention weight vector, with a dimension of , C is the number of channels of the feature map; MLP represents a multi-layer perceptron; represents the Global Average Pooling operation on the feature map F, represents the Global Max Pooling operation on the feature map F, σ is the Sigmoid activation function, and is the element-wise multiplication, is the broadcast operation (expanding the channel weight to the spatial dimension), 1 H×W is a full 1 matrix with a size of H x W , F ca is the feature map after channel attention enhancement.

[0035] Spatial Attention: Focus on which spatial positions in the feature map are more important, strengthen the feature response of key areas such as the edge of the vertebral body and the intervertebral space by learning the weight of the spatial dimension, and suppress non-key areas (such as surrounding soft tissues). Implementation process: average pooling and max pooling are performed on the feature map processed by channel attention in the channel dimension to obtain two spatial descriptors of HxWx1; the two are spliced in the channel dimension (forming HxWx2), compressed to a spatial weight map of HxWx1 through 1x1 convolution; after normalization through the Sigmoid activation function, it is multiplied with the feature map output by the channel attention in the spatial dimension to realize spatial-level feature enhancement. The calculation process can be represented as: .

[0036] wherein, represents the result of channel dimension average pooling on F ca . The dimension is 1xHxW, reflecting the average response intensity of each spatial position of the feature map in all channels, which is used to capture global spatial features. represents the result of channel dimension max pooling onF ca The result after channel-dimensional max pooling. The dimension is 1×H×W, focusing on the maximum response value across all channels at each spatial location of the feature map, used to highlight key local spatial features. MSa represents the spatial attention weight vector. This indicates that feature concatenation has been performed. AP is channel-level average pooling, MP is channel-level max pooling, Concat is channel-level concatenation, and Conv1×1 is a 1×1 convolution (compressing the number of channels to 1). Figure 1 Here, ⊙ represents the Sigmoid activation function, and ⊙ represents element-wise multiplication. For broadcast operations (extending spatial weights to the channel dimension), 1 C For length C A vector of all 1s F sa This is the final feature map after spatial attention enhancement.

[0037] For vertebrae with unique shapes, such as C1 and C2, CBAM can enhance the feature channels of unique structures such as the odontoid process and the anterior arch of the atlas through channel attention, and focus on the spatial position of these structures through spatial attention to reduce morphological confusion with other vertebrae. For intervertebral spaces that are obscured by osteophytes or have blurred edges (such as the C5 and C6 vertebrae), CBAM can dynamically enhance the feature response of the unobscured areas and improve the model's ability to recognize incomplete structures.

[0038] In practical applications, when using the WING loss function for spinous process keypoint detection, the WING loss applies equal weight to all keypoints, failing to specifically optimize for difficult points and thus limiting model performance. Specifically, among the keypoints of the seven vertebrae of the spine, vertebrae like C1 and C2, due to their unique morphology (lacking typical vertebral structures and containing odontoid processes) and high overlap with adjacent vertebrae, present significantly greater difficulty in locating keypoints (such as the tip of the odontoid process and the lower edge of C2) than those of conventional vertebrae like C3 to C7. Furthermore, even within the same vertebra, points easily obscured by osteophytes, such as pedicles and intervertebral spaces, are more difficult to locate than clearly defined areas like the vertebral midpoint. If all points are assigned the same loss weight, the model may tend to optimize easily located points (where the loss decreases quickly) while neglecting difficult points (where the loss decreases slowly), ultimately resulting in low detection accuracy for difficult points and failing to meet clinical needs for locating key anatomical structures. Therefore, this application improves the loss function by automatically assigning higher weights to high-difficulty points, forcing the model to prioritize these difficult-to-localize regions during training, thus increasing their learning priority and balancing the detection accuracy of key points of varying difficulty, adapting to the challenges posed by the unique anatomy of the spine. The specific improved formula is as follows: .

[0039] in, , represents the improved loss function, K represents the total number of key points, e represents the base of natural logarithm, represents the original WING loss, which is used to assign difficulty weights for each key point k w k , the average prediction error of each key point is calculated in real time during training , the greater the error, w k , the greater the error α , β is a regulation parameter, which realizes that the points with larger errors are paid more attention to.

[0040] In another exemplary embodiment of the present application, the user's scoliosis risk degree determination process is specifically: according to the angle between the user's spine spinous process key point connecting line and the vertical line, the preset angle range, the angle between the trunk central axis and the horizontal line, the preset normal value angle threshold, the height difference between the two shoulders, the preset normal threshold of the height difference between the two shoulders, the value of the reaction force of the two sides of the waist, the preset normal threshold of the stress of the two sides of the waist, the back curve, the preset back reference curve, the pressure of the two feet, and the preset reference pressure of the two feet, the user's scoliosis risk degree is obtained.

[0041] In actual application, the back curve stress sensing device includes an automatic spine palpation device and a trajectory supervision motion capture device. The automatic spine palpation device is used to measure the curve of the concave on both sides of the spine spinous process, and is connected with the trajectory supervision motion capture device, so that the motion curve of the automatic spine palpation device is recorded as the back curve. Specifically, the automatic spine palpation device is similar to the pulley of a massage chair.

[0042] In actual application, if the angle between the user's spine spinous process key point connecting line and the vertical line is not 0, it is judged that there is a risk of scoliosis. If it is below 10 degrees, it is low risk, if it is above 40 degrees, it is high risk, and if it is between the two, it is moderate risk.

[0043] The angle between the trunk central axis and the horizontal line is compared with the preset normal value angle threshold. If it is less than or greater than the threshold, it is determined that there is a risk of scoliosis.

[0044] The height difference between the two shoulders is compared with the preset normal threshold of the height difference between the two shoulders. If it is greater than the threshold, it is determined that there is a risk of scoliosis.

[0045] The value of the reaction force of the two sides of the waist is compared with the preset normal threshold of the stress of the two sides of the waist respectively. If it is greater than or less than the normal threshold, and the stress on both sides is asymmetric, it is determined that there is a risk of scoliosis.

[0046] The back curve is compared with the preset back reference curve. If the curvature exceeds the normal curvature range, it is determined that there is a risk of scoliosis. ​​

[0047] If the pressure asymmetry of the two feet exceeds 30%, and the difference with the preset two feet reference pressure is greater than 30%, it is determined that there is a risk of scoliosis.

[0048] Finally, according to the above results, it is determined whether there is a risk of scoliosis, and in addition, if the pressure difference of the two feet exceeds 50%, it is determined that there is a high risk of scoliosis.

[0049] In another exemplary embodiment of the present application, the double shoulder height measurer 1 is a double shoulder height laser height gauge, similar to an electronic height scale.

[0050] In another exemplary embodiment of the present application, the multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system further comprises a communication module; the scoliosis intelligent analysis module sends the user's scoliosis risk degree to the user's mobile phone through the communication module. The user views through the corresponding app of the mobile phone terminal.

[0051] In another exemplary embodiment of the present application, the multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system further comprises a display 5; the display 5 is used to display the use method of the multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system.

[0052] As shown in ​ The multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system provided by the present application is arranged in a room, and the specific use method is as follows: after a person enters through the door 8, faces the display 5 to watch the use method, and uses the second camera 7 to perform face recognition, the second camera 7 and the display 5 are arranged on the BHDE surface, takes off the upper garment, and the back is towards the display 5, at this time, the first camera 6 shoots a side photo, and the second camera 7 shoots a standing back photo, the first camera 6 is arranged on the GFEH surface, the pressure of the two feet is measured on the foot pressure distribution test device 4, the foot pressure distribution test device 4 is arranged on the ground, that is, the ABCD surface, the shoulder height is measured by the double shoulder height measurer 1, the double shoulder height measurer 1 is arranged on the AGCF surface, the stress of the two sides of the waist is measured by the waist stress sensing device 2, the waist stress sensing device 2 has two and is arranged on the ABCD surface and the GFEH surface, after the measurement is completed, the body is bent forward, and the second camera 7 shoots a back photo; then the face is towards the display 5, and the back is pasted on the back curve stress sensing device 3 to measure the back curve, the back curve stress sensing device 3 is arranged on the AGCF surface, and the data is uploaded to the scoliosis intelligent analysis module.

[0053] The multi-dimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system provided by the application can be automatically screened by the screened person all day long, the accuracy can reach the level of senior doctors (clinical work for more than 10 years), the single detection time is 1-2 minutes, the data is remotely transmitted by a cloud platform for diagnosis, and the result can be sent to the mobile phone of the parent of the screened person, a specialized hospital or a local disease control department, thereby providing important screening results for further early rehabilitation training or further clinical treatment of high-risk groups.

[0054] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0055] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0056] The principles and implementation modes of the application are described by applying specific examples in this paper, and the above embodiment description is only used to help understand the method and its core idea; at the same time, for those skilled in the art, according to the idea of the application, the specific implementation mode and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the application.

Claims

1. A multidimensional physiological data-based automated intelligent screening system for adolescent idiopathic scoliosis, characterized in that, The multidimensional physiological data-based automated intelligent screening system for adolescent idiopathic scoliosis includes: Camera module, shoulder height measuring device, lumbar stress sensing device, back curve stress sensing device, foot pressure distribution testing device, and scoliosis intelligent analysis module. The camera module is used to take pictures of the user's back and sides when standing upright, as well as the user's back when bending forward; the shoulder height measuring device is used to measure the height of the user's shoulders; the lumbar stress sensing device is used to apply equal force to both sides of the user's waist and measure the value of the reaction force on both sides of the waist; the back curve stress sensing device is used to measure the curve of the concave area on both sides of the spinous process of the spine to obtain the back curve; the back curve stress sensing device includes an automatic spinal palpation device and a trajectory monitoring motion capture device. The automatic spinal palpation device measures the curve of the concave area on both sides of the spinous process of the spine, and is connected to the trajectory monitoring motion capture device to record the motion curve of the automatic spinal palpation device as the back curve; the plantar pressure distribution testing device is used to measure the pressure of the user's feet. The intelligent scoliosis analysis module is used to obtain the height difference between the user's shoulders based on the height of the user's shoulders; and to obtain the key points of the spinous processes of each vertebra of the user's spine based on photos of the user's back when standing and when bending forward. Specifically, the module uses a spinal key point detection algorithm to process the photos of the user's back when standing and when bending forward to obtain a heat map of the spinous process key points. Based on the heat map of the spinous process key points, the key points of the spinous processes of each vertebra of the spine are obtained. Based on the key points of the spinous processes of each vertebra of the user's spine, the key points of the spinous processes of the spine are connected. Based on the user's side view, the angle between the user's trunk midline and the horizontal line is determined. Based on the angle between the line connecting the key points of the spinous processes of the spine and the vertical line, the angle between the trunk midline and the horizontal line, the height difference between the shoulders, the values ​​of the reaction forces on both sides of the waist, the back curve, and the pressure on both feet, the module obtains the degree of scoliosis risk of the user. The spinal key point detection algorithm includes an HRNet network, a first CBAM network, a second CBAM network, and a third CBAM network. The second and third stages of the HRNet network are connected through the first CBAM network, the third and fourth stages of the HRNet network are connected through the second CBAM network, and the output of the HRNet network is connected to the third CBAM network. The first CBAM network includes the first CBAM, the second CBAM, and the third CBAM. The input of the first CBAM is connected to the output of the first convolutional block of the second stage of the HRNet network; the output of the first CBAM is connected to the input of the first convolutional block of the third stage of the HRNet network. The input of the second CBAM is connected to the downsampling output and the upsampling output of the second stage of the HRNet network, respectively; the output of the second CBAM is connected to the upsampling input and the downsampling input of the third stage of the HRNet network, respectively. The input of the third CBAM is connected to the output of the second convolutional block of the second stage of the HRNet network; the output of the third CBAM is connected to the input of the second convolutional block of the third stage of the HRNet network; the second CBAM network includes a fourth CBAM, a fifth CBAM, and a sixth CBAM. The input of the fourth CBAM is connected to the output of the third convolutional block of the third stage of the HRNet network; the output of the fourth CBAM is connected to the input of the first convolutional block of the fourth stage of the HRNet network. The input of the fifth CBAM is connected to the output of the fourth convolutional block of the third stage of the HRNet network; the output of the fifth CBAM is connected to the input of the second convolutional block of the fourth stage of the HRNet network. The input of the sixth CBAM is connected to the output of the fifth convolutional block of the third stage of the HRNet network; the output of the sixth CBAM is connected to the input of the third convolutional block of the fourth stage of the HRNet network; the third CBAM network includes a seventh CBAM and a sixth convolutional block connected in sequence; the input of the seventh CBAM is connected to the output of the fourth convolutional block of the fourth stage of the HRNet network.

2. The multidimensional physiological data-based automatic intelligent screening system for adolescent idiopathic scoliosis according to claim 1, characterized in that, The process of determining the risk level of scoliosis in a user is as follows: the risk level of scoliosis in a user is determined based on the angle between the line connecting the key points of the spinous processes of the user's spine and the vertical line, the preset angle range, the angle between the trunk midline and the horizontal line, the preset normal value angle threshold, the height difference between the shoulders, the preset normal threshold for the height difference between the shoulders, the value of the reaction force on both sides of the waist, the preset normal threshold for lateral waist stress, the back curve, the preset back baseline curve, the pressure on both feet, and the preset baseline pressure on both feet.

3. The multidimensional physiological data-based automatic intelligent screening system for adolescent idiopathic scoliosis according to claim 1, characterized in that, The shoulder height measuring device is a shoulder height laser height measuring instrument.

4. The multidimensional physiological data-based automatic intelligent screening system for adolescent idiopathic scoliosis according to claim 1, characterized in that, The multidimensional physiological data automatic adolescent idiopathic scoliosis intelligent screening system also includes: a communication module; the scoliosis intelligent analysis module sends the user's scoliosis risk level to the user's mobile phone through the communication module.

5. The multidimensional physiological data-based automatic intelligent screening system for adolescent idiopathic scoliosis according to claim 1, characterized in that, The automated intelligent screening system for adolescent idiopathic scoliosis based on multidimensional physiological data also includes: a display; the display is used to show the usage method of the automated intelligent screening system for adolescent idiopathic scoliosis based on multidimensional physiological data.

Images