Intelligent screening system for adolescent scoliosis based on AI posture analysis
The intelligent screening system based on AI posture analysis solves the problem of existing adolescent scoliosis screening equipment relying on dedicated hardware and ionizing radiation, achieving high-precision, convenient, and non-invasive screening. It is applicable to multiple scenarios and improves screening coverage and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUXI TAIHU UNIV
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, adolescent scoliosis screening equipment relies on specialized hardware, is costly, complex to operate, and poses a risk of ionizing radiation. It is difficult to achieve large-scale, convenient, and non-invasive screening, and the screening capacity at the grassroots level is insufficient, resulting in low coverage.
The system employs an AI-based posture analysis-based intelligent screening system, which includes a posture perception module, a multimodal data fusion module, and a dual-engine decision-making module. Through improved deep learning algorithms and graph neural networks, it achieves high-precision non-invasive 3D spinal reconstruction and risk assessment. Combined with SLAM technology, it provides convenient screening reports and data management while ensuring data privacy.
It achieves sub-millimeter-level key point localization and non-invasive three-dimensional spinal reconstruction with accuracy reaching clinical standards. It has high sensitivity and low false positive rate, making it suitable for families, schools, and medical institutions. It reduces screening costs and promotes the popularization and intelligentization of spinal health screening.
Smart Images
Figure CN122158073A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical and health technology, and more specifically, to an intelligent screening system for adolescent scoliosis based on AI posture analysis. Background Technology
[0002] Adolescent scoliosis is a prevalent health problem among primary and secondary school students in my country, with an average annual incidence rate increase of 3.14%. However, the county-level screening coverage rate is less than 35%, and traditional screening methods have three major bottlenecks: First, insufficient coverage, relying on professional medical resources, and weak screening capabilities at the grassroots level; second, radiation risks, as traditional screening relies on X-ray imaging, which poses a hazard of ionizing radiation and is not suitable for routine screening; and third, resource shortages, with a shortage of more than 120,000 specialist physicians at the grassroots level, making it difficult to meet the needs of large-scale screening.
[0003] In the current technology, domestic competitors such as the Fusolt 3D Electronic Spine Measurement Instrument rely on dedicated hardware equipment, require professional operation, and have limited portability; the Youxing Health AI Posture Monitor focuses on daily posture monitoring, and its scoliosis screening lacks specificity and accuracy. Foreign competitors, such as the German DIERS formetric 4D 3D Spine Assessment System, have hardware costs exceeding 2 million yuan, are complex to operate, require professional training, and have high costs per screening; while the Israeli Sunlight SpineScan™ SH-105 Electronic Spine Measurement Instrument has high accuracy, it also relies on dedicated equipment, making it difficult to promote on a large scale.
[0004] Therefore, there is an urgent need to develop an intelligent screening system that requires no special hardware, is non-invasive and radiation-free, easy to operate, and has clinical-standard accuracy, in order to solve the pain points of traditional screening and promote the popularization and intelligentization of spinal health screening.
[0005] In view of this, the present invention is proposed to solve the above-mentioned technical problems. Summary of the Invention
[0006] The purpose of this invention is to provide an intelligent screening system for adolescent scoliosis based on AI posture analysis, which can achieve sub-millimeter-level key point positioning, non-invasive three-dimensional spinal reconstruction, and multimodal data fusion decision-making, while taking into account accuracy, convenience and low cost, to meet the screening needs of families, schools and medical institutions in multiple scenarios, and promote early screening and early intervention for scoliosis.
[0007] To achieve the above objectives, the present invention provides the following technical solution: An AI-based posture analysis-based intelligent screening system for adolescent scoliosis includes: Posture perception module: used to collect adolescent body posture data and extract spinal-related anatomical landmarks to achieve high-precision positioning; 3D Reconstruction Module: Based on the positioning data output by the posture perception module, the 3D curvature of the spine is reconstructed non-invasively and key clinical indicators are calculated. Multimodal data fusion module: Integrates at least two types of heterogeneous data to generate a dynamic assessment matrix for risk assessment; Dual-engine decision module: Outputs scoliosis risk level and treatment recommendations based on dynamic assessment matrix; Core functional modules: Adapt to screening needs in multiple scenarios, providing screening report generation, health tracking and data sharing functions; Compliance and privacy protection module: De-identifies and encrypts user data to ensure data security and privacy compliance.
[0008] Furthermore, the posture perception module employs an improved deep learning algorithm, including at least one of the improved YOLOv8 DeepPose algorithm and the HRFormer++ hybrid architecture. The localization error of the anatomical landmarks is <2mm, and motion artifacts are eliminated through optical flow and temporal modeling.
[0009] Furthermore, the 3D reconstruction module achieves 3D curvature reconstruction of the spine through at least one of monocular depth estimation network and multi-view depth estimation network, and simultaneously analyzes the coronal, sagittal and transverse morphology. Key clinical indicators include Cobb angle and spinal rotation angle, with the Cobb angle calculation error being <3°, and an adaptive Kalman filter is introduced to compensate for the shooting angle offset.
[0010] Furthermore, the multimodal data fusion module constructs a cross-modal fusion framework based on graph neural networks (GNNs). The integrated modal data includes at least two of the following: RGB images, IMU sensor data, growth curves, previous health records, dynamic videos, body movement data, and environmental data related to spinal health, to achieve data complementarity and bias correction.
[0011] Furthermore, the dual-engine decision-making module includes an AI layer and an expert layer; The AI layer is at least one of the LightGBM model and neural network model, used to output a four-level risk level; The expert layer integrates no fewer than 1,000 clinical pathway knowledge graphs to verify and supplement the output of the AI layer, ultimately generating a diagnostic report containing at least three clinical indicators.
[0012] Furthermore, the core functional modules include: Self-service screening unit: Supports capturing body images or videos via smart terminals and automatically generating diagnostic reports; Batch screening unit: Supports importing group information and generating summary health reports; Family management unit: Supports sharing of health data among multiple accounts, enabling long-term dynamic tracking; Medical Collaboration Unit: Provides screening data export function for professional physicians to refer to.
[0013] Furthermore, the compliance and privacy protection module uses a federated learning framework to de-identify edge data, uploading only encrypted feature data. A guardian authorization mechanism is enabled for users under the age of 14, and biometric data is stored for ≤6 months, and automatically and irreversibly erased after the expiration period.
[0014] Furthermore, the core functional modules also include an AI posture coaching unit, which uses SLAM spatial positioning technology to overlay 3D skeletal lines in real time and uses color coding to dynamically mark abnormal areas of trunk rotation and scapular imbalance, with an abnormal positioning accuracy of ±3°.
[0015] Furthermore, the smart terminals include mobile phones and tablets, which support Bluetooth connection to external high frame rate lenses and wearable IMU devices, and can complete core screening functions in environments without a network.
[0016] Furthermore, the core functional modules also include a data docking unit, which supports exporting screening data in at least one protocol format, such as HL7 or DICOM, to connect with hospital HIS systems and hierarchical diagnosis and treatment networks, and can realize digital twin of spinal health, AR visualization, and linkage functions with rehabilitation courses.
[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Leading Precision: Millimeter-level key point localization (error <2mm) and three-dimensional spinal reconstruction (Cobb angle error <3°) meet clinical screening standards, with a sensitivity of 92.3% and a false positive rate of <7%, which is superior to existing mobile screening solutions.
[0018] 2. Non-invasive and convenient: No special hardware or professional training is required. Screening can be completed simply by taking a picture with a mobile phone. There is no ionizing radiation. It is suitable for multiple scenarios such as home and school, and the cost is low.
[0019] 3. Multimodal fusion: Integrating 7 types of multimodal data to build data barriers, the risk assessment accuracy rate reaches 92.7%, reducing the threshold for diagnosis at the grassroots level.
[0020] 4. Full-chain management: Achieve full coverage of the "screening-assessment-intervention-referral" process, support long-term dynamic tracking and multi-scenario data sharing, and promote the forward shift of health management. Attached Figure Description
[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation of the invention. Obviously, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without creative effort. In the drawings: Figure 1 A three-dimensional reconstruction model of an AI-based intelligent screening system for adolescent scoliosis, provided in an embodiment of this application; Figure 2 A schematic diagram of the multimodal data fusion architecture of an AI-based intelligent screening system for adolescent scoliosis, provided in an embodiment of this application; Figure 3 A schematic diagram of the LightGBM four-level risk assessment model of the intelligent screening system for adolescent scoliosis based on AI posture analysis provided in the embodiments of this application. Detailed Implementation
[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0023] See Figures 1 to 3 As shown, the AI-based posture analysis-based intelligent screening system for adolescent scoliosis includes: a posture perception module, a 3D reconstruction module, a multimodal data fusion module, a dual-engine decision-making module, a core functional module, and a compliance and privacy protection module. The posture perception module collects adolescent posture data and extracts anatomical landmarks related to the spine to achieve high-precision positioning. The 3D reconstruction module, based on the positioning data output by the posture perception module, non-invasively reconstructs the 3D curvature of the spine and calculates key clinical indicators. The multimodal data fusion module integrates at least two types of heterogeneous data to generate a dynamic assessment matrix for risk assessment. The dual-engine decision-making module outputs the scoliosis risk level and treatment suggestions based on the dynamic assessment matrix. The core functional module adapts to screening needs in multiple scenarios, such as families, schools, and medical institutions, providing screening report generation, health tracking, and data sharing functions. The compliance and privacy protection module desensitizes and encrypts user data to ensure data security and privacy compliance.
[0024] The posture perception module employs an improved deep learning algorithm, including at least one of the improved YOLOv8 DeepPose algorithm and the HRFormer++ hybrid architecture. This algorithm processes photos or videos of the user's back from the front / side view, extracts anatomical landmarks related to the spine, and achieves pixel-level localization with a localization error of <2mm. Furthermore, it eliminates motion artifacts through optical flow and temporal modeling, thereby improving the stability of key point extraction.
[0025] Improvements: Added a dedicated attention mechanism for spinal key points, optimized anchor box size (to suit the body shape of teenagers aged 10-18), and adjusted the neck C2f module to C3 module to improve detection speed; HRFormer and MobileNetV3 are combined, with HRFormer extracting global features and MobileNetV3 extracting local detail features.
[0026] The 3D reconstruction module reconstructs the 3D curvature of the spine using at least one of a monocular depth estimation network or a multi-view depth estimation network, simultaneously analyzing the coronal, sagittal, and transverse morphology. Key clinical indicators include the Cobb angle and spinal rotation angle, with the Cobb angle calculation error being <3°. An adaptive Kalman filter is introduced to compensate for the shooting angle offset, compensating for depth estimation distortion caused by the mobile phone shooting angle offset, so that the difference between the Cobb angle measured in the standing and prone positions is controlled within ±2°. Kalman filter parameters: state equation A=[1,0.1;0,1], observation equation H=[1,0;0,1], process noise variance 0.01, observation noise variance 0.02.
[0027] The multimodal data fusion module is based on graph neural networks (GNN) to build a cross-modal fusion framework. The integrated modal data includes at least two of the following: RGB images, IMU sensor data, growth curves, previous health records, dynamic videos, body movement data, and environmental data related to spinal health (such as daily sitting environment and sports environment). A unified semantic space is constructed, and cross-modal embedding and graph learning are used to achieve data complementarity and bias correction, reduce the influence of bias in single modal data, and improve the screening accuracy to over 92.7%.
[0028] The dual-engine decision-making module includes an AI layer and an expert layer. The AI layer is at least one of the LightGBM model and a neural network model, used to output a four-level risk level (based on multimodal fusion data, the risk level of scoliosis is low, medium, high, and very high). The expert layer integrates no less than 1,000 clinical pathway knowledge graphs from tertiary hospitals to verify and supplement the output results of the AI layer, and finally generates a diagnostic report containing at least three clinical indicators.
[0029] The core functional modules include: Self-service screening unit: It supports users to take body posture images or videos through smart terminals and automatically generate diagnostic reports. The AI posture coach function uses SLAM spatial positioning technology to overlay 3D skeleton lines in real time and marks abnormal parts with red / yellow / green color coding. Batch screening unit: Supports importing group information (student name, age, gender) and generating summary health reports, adapting to large-scale campus physical examination scenarios; Family management unit: Supports sharing of health data among multiple accounts, allowing parents to remotely view their child's spinal health status, generate long-term health records, and achieve long-term dynamic tracking; Medical Collaboration Unit: Provides screening data export function, screening data can be exported as PDF format, connected to the hospital HIS system, and provided to professional physicians for reference.
[0030] The compliance and privacy protection module uses a federated learning framework to de-identify edge data, uploading only encrypted feature data. A guardian authorization mechanism is enabled for users under the age of 14. Biometric data storage period is ≤6 months, and it is automatically and irreversibly erased after the expiration period.
[0031] The core functional modules also include an AI posture coaching unit, which uses SLAM spatial positioning technology to overlay 3D skeletal lines in real time and uses color coding to dynamically mark abnormal areas of trunk rotation and scapular imbalance. The abnormality positioning accuracy reaches ±3°. Green indicates no abnormality (trunk rotation <3°, scapular imbalance <5°), yellow indicates mild abnormality (3°≤trunk rotation <5°, 5°≤scapular imbalance <8°), and red indicates moderate or above abnormality (trunk rotation ≥5°, scapular imbalance ≥8°).
[0032] The smart terminals include mobile phones and tablets, and support Bluetooth connection to external high frame rate lenses and wearable IMU devices. They can complete core screening functions such as body posture data collection, anatomical landmark extraction, three-dimensional curvature reconstruction and risk level assessment even in environments without network access.
[0033] The core functional modules also include a data docking unit, which supports exporting screening data in at least one protocol format, such as HL7 or DICOM, to connect with hospital HIS systems and hierarchical diagnosis and treatment networks, and can realize digital twin of spinal health, AR visualization, and linkage of rehabilitation courses.
[0034] Example 1 User operation: Under the assistance of their guardians, teenage users should use their mobile phones to take one photo of their back from the front and one photo from the side (or a 10-second video). During the shooting process, they should stand naturally with their backs exposed and unobstructed. Pose perception: The pose perception module calls the improved YOLOv8 DeepPose algorithm to automatically identify 12 key anatomical landmarks such as the spinous process of the spine, scapula, and iliac crest. It corrects slight shaking during shooting using optical flow and outputs stable key point coordinates with a positioning error of <2mm. 3D Reconstruction: The 3D reconstruction module is based on the MonoDepthV3 monocular depth estimation network. Combined with human biomechanical data, it transforms the key point coordinates into a 3D model of the spine. It compensates for the shooting angle offset (such as within ±15° of the shooting tilt angle) through adaptive Kalman filtering, and calculates the coronal Cobb angle, sagittal physiological curvature, and transverse rotation angle. The error is <3°. Multimodal fusion: The system automatically calls the user-bound IMU sensor data (such as daily body movement data) and growth curve records, and combines them with the captured image data. Through the GNN cross-modal fusion framework, the system performs data preprocessing and feature aggregation to construct a dynamic evaluation matrix. Dual-engine decision-making: The LightGBM four-level risk assessment model outputs risk levels based on a dynamic assessment matrix (low risk: Cobb angle <10°; medium risk: 10°≤Cobb angle <20°; high risk: 20°≤Cobb angle <40°; very high risk: Cobb angle ≥40°). An expert knowledge graph validates the model results and supplements targeted intervention suggestions (such as posture correction exercises for low-risk users and referral hospital recommendations for high-risk users). Report generation and application: The system generates diagnostic reports that include a 3D spinal model, key clinical indicators, risk level, and intervention recommendations. Family users can view the reports through a shared account. In school screening scenarios, class summary reports can be exported in batches. Hospital users can import the reports into the HIS system to assist in diagnosis and treatment.
[0035] Example 2 Five hundred adolescent subjects aged 10-18 years without other spinal diseases were included (250 patients with scoliosis, Cobb angle 5°-60°; 250 healthy controls). The imaging environment was indoors with natural light, light intensity 500-800 Lux, and the imaging angles were anteroposterior and lateral views of the back. The X-ray results were used as the standard. Sensitivity: The detection rate for scoliosis patients with a Cobb angle ≥10° was 92.3%; Specificity: The correct judgment rate for healthy controls was 93.0%; False positive rate: 6.7%; False negative rate: 7.7%; Consistency with Cobb angle measurement by X-ray: kappa value = 0.89, meeting the clinical equivalence standard.
[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. An intelligent screening system for adolescent scoliosis based on AI posture analysis, characterized in that, include: Posture perception module: used to collect adolescent body posture data and extract spinal-related anatomical landmarks to achieve high-precision positioning; 3D Reconstruction Module: Based on the positioning data output by the posture perception module, the three-dimensional curvature of the spine is reconstructed non-invasively and key clinical indicators are calculated; Multimodal data fusion module: Integrates at least two types of heterogeneous data to generate a dynamic assessment matrix for risk assessment; Dual-engine decision module: Outputs scoliosis risk level and treatment recommendations based on the dynamic assessment matrix; Core functional modules: Adapt to screening needs in multiple scenarios, providing screening report generation, health tracking and data sharing functions; Compliance and privacy protection module: De-identifies and encrypts user data to ensure data security and privacy compliance.
2. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 1, characterized in that, The posture perception module employs an improved deep learning algorithm, which includes at least one of the improved YOLOv8 DeepPose algorithm and the HRFormer++ hybrid architecture. The localization error of the anatomical landmarks is <2mm, and motion artifacts are eliminated through optical flow and temporal modeling.
3. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 1, characterized in that, The three-dimensional reconstruction module reconstructs the three-dimensional curvature of the spine through at least one of a monocular depth estimation network and a multi-view depth estimation network, and simultaneously analyzes the morphology of the coronal, sagittal, and transverse planes. The key clinical indicators include the Cobb angle and the spinal rotation angle, wherein the calculation error of the Cobb angle is <3°, and an adaptive Kalman filter is introduced to compensate for the shooting angle offset.
4. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 1, characterized in that, The multimodal data fusion module is based on a graph neural network (GNN) to build a cross-modal fusion framework. The integrated modal data includes at least two of the following: RGB images, IMU sensor data, growth curves, previous health records, dynamic videos, body movement data, and environmental data related to spinal health, so as to achieve data complementarity and bias correction.
5. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 4, characterized in that, The dual-engine decision-making module includes an AI layer and an expert layer; The AI layer is at least one of the LightGBM model and the neural network model, and is used to output a four-level risk level. The expert layer integrates no fewer than 1,000 clinical pathway knowledge graphs to verify and supplement the output of the AI layer, ultimately generating a diagnostic report containing at least three clinical indicators.
6. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 5, characterized in that, The core functional modules include: Self-service screening unit: Supports capturing body images or videos via smart terminals and automatically generating diagnostic reports; Batch screening unit: Supports importing group information and generating summary health reports; Family management unit: Supports sharing of health data among multiple accounts, enabling long-term dynamic tracking; Medical Collaboration Unit: Provides screening data export function for professional physicians to refer to.
7. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 1, characterized in that, The compliance and privacy protection module uses a federated learning framework to de-identify edge data, only uploading encrypted feature data. A guardian authorization mechanism is enabled for users under the age of 14. Biometric data storage period is ≤6 months, and it is automatically and irreversibly erased after the expiration period.
8. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 6, characterized in that, The core functional module also includes an AI posture coaching unit, which uses SLAM spatial positioning technology to overlay 3D skeletal lines in real time and uses color coding to dynamically mark abnormal areas of torso rotation and scapular imbalance, with an abnormal positioning accuracy of ±3°.
9. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 1 or 6, characterized in that, The smart terminal includes mobile phones and tablets, supports Bluetooth connection to external high frame rate lenses and wearable IMU devices, and can complete core screening functions in environments without network access.
10. The intelligent screening system for adolescent scoliosis based on AI posture analysis according to claim 6, characterized in that, The core functional module also includes a data docking unit, which supports exporting screening data in at least one protocol format, such as HL7 or DICOM, to connect with hospital HIS systems and hierarchical diagnosis and treatment networks, and can realize digital twin of spinal health, AR visualization, and linkage of rehabilitation courses.