Scoliosis rehabilitation robot for children

Abstract

The invention discloses scoliosis correction equipment for children, and aims to solve the technical problems that existing correction equipment depends on manual operation and lacks accurate real-time feedback and personalized self-adaptive adjustment. The equipment comprises a base plate, a stand column, an upper support, a lower support, an electric telescopic rod, an upper correcting ring, a middle correcting ring and a lower correcting ring, wherein the upper support, the lower support and the electric telescopic rod are in sliding fit with the stand column; the equipment collects a posture image of a user through a camera and constructs a three-dimensional posture model; and the PC dynamically outputs an optimized control instruction through a built-in machine learning model based on real-time posture data and historical data represented by the model, and drives the electric telescopic rod and the middle correction ring to apply accurate and self-adaptive correction force to the spine bending part of the user. According to the system, machine vision and artificial intelligence technologies are combined, closed-loop, accurate, personalized and intelligent control over the scoliosis correction process is achieved, and the corrective force applying accuracy and adaptability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation medical device technology, and more specifically, to a scoliosis rehabilitation robot for children that integrates machine vision, 3D modeling and artificial intelligence control technology. Background Technology

[0002] Childhood scoliosis is a common skeletal deformity affecting the physical and mental health of adolescents, and its incidence is increasing year by year. Early intervention is key to controlling the progression of scoliosis and avoiding surgery. Currently, non-surgical interventions for mild to moderate scoliosis mainly include physical correction, posture training, and wearing orthotic braces.

[0003] However, existing technologies have many limitations. Traditional orthotic braces correct deformities by passively applying force, but their design is often based on static plaster impressions or scan data, failing to adapt to the dynamic changes users experience in daily life. Furthermore, braces are uncomfortable to wear, easily leading to complications such as pressure sores, and user compliance is generally poor. Physical traction and manual repositioning methods heavily rely on the operator's personal experience and skills, lacking unified, quantifiable objective standards. The intervention process is difficult to accurately replicate and standardize, and the applied corrective force is often instantaneous, making it difficult to achieve continuous and stable corrective effects. Some existing mechanical orthotic devices typically use preset, fixed correction modes, operating through simple timed or force-controlled programs. These devices cannot sense subtle changes in the user's posture in real time and make adaptive adjustments, lacking intelligent feedback mechanisms. Their correction process is somewhat arbitrary and cannot provide optimized, personalized correction solutions for each user's unique needs.

[0004] Therefore, existing technologies generally suffer from drawbacks such as reliance on subjective experience during the correction process, lack of precise quantitative feedback, inability to achieve personalized adaptive adjustment, and poor user compliance. There is an urgent need to develop an intelligent correction device capable of real-time, precise dynamic capture of user posture, and able to make intelligent decisions and adaptively apply corrective forces based on feedback information. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a scoliosis correction device for children, which aims to achieve accurate perception, intelligent decision-making and adaptive closed-loop correction of the user's scoliosis status, thereby improving the automation, accuracy and safety of the correction process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A scoliosis correction device for children, comprising:

[0008] Base plate;

[0009] The column is vertically mounted on the base plate;

[0010] The upper support, lower support, and electric telescopic rod are all independently slidably fitted onto the column to achieve adjustment of their respective vertical positions along the column.

[0011] The upper correction ring is detachably fixed to the upper bracket;

[0012] The lower correction ring is detachably fixed to the lower bracket;

[0013] The central correction ring is detachably fixed to the telescopic end of the electric telescopic rod;

[0014] A camera is used to capture posture images of a user standing within the correction loop;

[0015] And a PC, which is electrically connected to the camera and the electric telescopic pole;

[0016] The PC is configured to: receive the posture images captured by the camera and construct a three-dimensional posture model of the user's torso based on the posture images; incorporate a machine learning model that takes real-time posture data and historical data represented by the three-dimensional posture model as input and outputs control commands for correcting scoliosis; and control the extension and retraction of the electric telescopic rod according to the control commands to drive the central correction ring to apply corrective force to the user's spinal curvature.

[0017] Compared with the prior art, the present invention has the following significant advantages:

[0018] (1) This invention transforms the traditional assessment method, which relies on operator visual observation and experience judgment, into a real-time and accurate quantitative perception of user posture through machine vision and 3D modeling technology. Combined with machine learning models, it can dynamically generate and optimize correction strategies based on each user's unique physiological data and real-time posture feedback.

[0019] (2) This invention constitutes an intelligent closed-loop feedback control system of "perception-decision-execution-re-perception". It can respond in real time to the user's posture fine-tuning or muscle compensation during the correction process, and make millisecond-level adaptive fine-tuning of the magnitude, direction and timing of the corrective force, avoiding the blindness and potential risks of traditional fixed-mode correction, and ensuring that the corrective force always acts on the most effective position.

[0020] (3) Intelligent real-time monitoring and adaptive adjustment can effectively avoid secondary injuries caused by improper force application or sudden user movement. At the same time, by continuously optimizing the correction strategy, it ensures that the correction force is always in the most efficient range, which can shorten the correction cycle and improve correction efficiency to the greatest extent while ensuring safety.

[0021] (4) This invention can completely record the data of each correction process, forming a quantitative and traceable electronic archive. This not only provides an objective basis for professionals to evaluate the effect and adjust the plan, but also provides a valuable data foundation for the continuous learning and self-optimization of the equipment's own machine learning model, enabling it to continuously evolve and become more and more intelligent.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, embodiments of the present invention are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a first-view structural diagram of the scoliosis rehabilitation robot described in an embodiment of the present invention;

[0025] Figure 2 This is a second-view structural diagram of the scoliosis rehabilitation robot described in an embodiment of the present invention;

[0026] In the diagram: 1-Upper correction ring, 2-Plastic pad, 3-Middle correction ring, 4-Lower correction ring, 5-Electric telescopic rod, 6-Upper bracket, 7-Lower bracket, 8-Column, 9-Camera, 10-PC, 11-Base plate. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0028] Please refer to Figure 1 and Figure 2 This invention provides a scoliosis correction device for children.

[0029] In this embodiment, the device includes a base plate 11 for providing stable support for the entire device. At least two uprights 8 are vertically mounted on the base plate 11. The upper support 6, lower support 7, and electric telescopic rod 5 are all independently and slidably fitted onto the uprights 8 via sliding sleeves or similar structures. This design allows their height to be freely adjusted and locked along the axial direction of the uprights 8 to accommodate users of different heights.

[0030] The device also includes three correction rings: an upper correction ring 1, a middle correction ring 3, and a lower correction ring 4. The upper correction ring 1 is fixed to the upper support 6 via a detachable connector; the lower correction ring 4 is fixed to the lower support 7 via a detachable connector; and the crucial force-applying component—the middle correction ring 3—is fixed to the telescopic end of the electric telescopic rod 5 via a detachable connector. These three rings together form a space for accommodating and positioning the user's torso.

[0031] To achieve intelligent perception of user posture, the device is equipped with multiple cameras 9 arranged in a spatial array around the correction ring. These cameras 9 are used to capture posture images of the user standing within the correction ring from different angles. The core control unit of the device is a PC 10, which is electrically connected to all cameras 9 and the motorized telescopic pole 5 via data cable or wirelessly.

[0032] When the device is in operation, the PC 10 receives multiple posture images simultaneously captured by multiple cameras 9 and constructs a three-dimensional posture model of the user's torso based on these images. The PC 10 has a built-in machine learning model that can take real-time posture data and historical data represented by the three-dimensional posture model as input. Through calculation, the model outputs a set of optimized control commands. Based on these commands, the PC 10 precisely controls the extension and retraction length and speed of the electric telescopic rod 5, thereby driving the central correction ring 3 to apply a corrective force with precise calculation in magnitude, direction, and timing to the most prominent part of the user's spinal curvature.

[0033] Specifically, the upper support 6 and lower support 7 mainly serve to provide auxiliary support and limit compensatory movements, and their height can be adjusted according to the user's shoulder and pelvic position. The middle correction ring 3, as the active force-applying component, is precisely positioned at the apex of the user's scoliosis, and applies corrective force through the reciprocating motion of the electric telescopic rod 5.

[0034] To achieve safety monitoring and accurate feedback, in a preferred embodiment, the inner walls of the upper corrective ring 1, the middle corrective ring 3, and the lower corrective ring 4—the surfaces that directly contact the user's body—are all equipped with plastic pads 2. These plastic pads 2 not only enhance comfort but also integrate a flexible pressure sensor array. This sensor array is electrically connected to the PC 10. During the application of corrective force, the sensor array can collect and generate a real-time pressure distribution map that is visualized and covers all areas of body contact. The PC 10 uses this real-time pressure distribution map as a dynamic safety boundary condition for the machine learning model to calculate when generating control commands, thereby proactively avoiding any risks that may lead to excessive local pressure and cause discomfort or injury.

[0035] To achieve high-precision 3D perception of user posture, this embodiment uses multiple cameras 9, whose spatial array distribution ensures that the shape of the user's torso can be captured without blind spots. The PC 10 is configured to invoke a multi-view stereo vision algorithm to fuse and reconstruct 3D posture images simultaneously acquired by the multiple cameras 9. This process generates a dynamic 3D point cloud model containing tens of thousands of spatial coordinate points, which can reflect the fine morphology of the user's torso surface with high fidelity. Furthermore, based on this dynamic 3D point cloud model, the PC 10, through its built-in geometric calculation algorithm, can extract and continuously track a series of key posture feature parameters in real time, such as the estimated value of the 3D Cobb angle, the rotation degree of the vertex vertebra, and the 3D spatial tilt matrix of the scapula and pelvis used to characterize the body's balance.

[0036] To ensure the accuracy of the corrective force application, the device in this embodiment also includes a fully automated calibration program before use. This program first guides the operator to import the user's DICOM format medical image data. Then, the program performs 3D-to-3D rigid registration between the dynamic 3D point cloud model constructed in real-time by camera 9 and the imported DICOM data. After registration, the system can accurately map and lock the 3D spatial coordinates of the vertebra at the apex of the scoliosis on the user's real-time body surface point cloud model. Subsequently, PC 10 automatically calculates the target vertical position required to align the geometric center of the central corrective ring 3 with the mapped point on the body surface and automatically generates drive commands to complete the movement.

[0037] In this embodiment, the machine learning model can specifically employ a deep reinforcement learning model trained on historical data. Its input data includes not only real-time posture data and historical data, but also a comprehensive feature vector containing information such as the user's age, weight, scoliosis type, and Risker rating. Therefore, its output control commands are highly personalized, specifically manifested as a time series defining the corrective force loading pattern throughout the complete correction cycle. This sequence is meticulously planned, incorporating multiple stages including micro-stress adaptation, logarithmic progressive loading, plateau period constant force maintenance, and linear unloading, forming a non-linear force-time curve.

[0038] To further enhance the intelligence and effectiveness of control, PC 10 is configured to execute an advanced predictive closed-loop feedback control method. In each control cycle, the machine learning model, based on the user's biomechanical model, predicts potential undesirable compensatory movement trends that the body might exhibit under the current main control command. Based on this prediction, the system generates compensatory fine-tuning commands in advance and superimposes them onto the main control command. These fine-tuning commands simultaneously control the upper support 6 and lower support 7 to make minor vertical position adjustments, thereby actively suppressing compensatory movements.

[0039] In addition, to ensure absolute safety during equipment operation, PC 10 executes a multi-level redundant safety monitoring program. The first level is physical monitoring based on data from a flexible pressure sensor array, which immediately triggers protection when the pressure value or its rate of change exceeds the limit. The second level is electrical monitoring based on the encoder and current sensor built into the electric telescopic rod 5, which detects mechanical jamming or abnormal resistance by comparing commands and feedback in real time. If any level detects an abnormality, it will trigger the system to enter a safe shutdown mode.

[0040] In terms of data management, the PC10 also features data archiving and report generation capabilities. It can generate multi-dimensional assessment reports of the correction process, including comparisons of 3D posture models before and after correction, changes in key posture characteristic parameters, and in-depth analysis indicators such as user posture stability index and task cooperation score. This report can be exported and integrated with hospital information systems.

[0041] To continuously improve the intelligence level of the device, the PC 10 in this embodiment is also used to continuously optimize the machine learning model through federated learning or transfer learning mechanisms. While protecting user data privacy, the system can use anonymized data accumulated from multiple devices to train the global base model and update the optimized model parameters locally; simultaneously, it can use individual user's personalized data to fine-tune the model online.

[0042] It is worth noting that the fully automatic calibration program further includes a calibration module for determining the optimal direction of the corrective force. After the corrective ring 3 is positioned, the PC 10 controls the electric telescopic rod 5 to apply a probe force and guides the user to perform a slight axial rotation. During this process, the PC 10 analyzes the deformation field distribution on the three-dimensional posture model to calculate the optimal force vector direction that most effectively transmits the corrective force to the apex vertebra while minimizing compensatory deformation of the body, and uses this direction as the primary direction of subsequent correction.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A scoliosis rehabilitation robot for children, characterized in that, include: A scoliosis correction device for children, characterized in that it comprises: Base plate (11); The column (8) is vertically installed on the base plate (11); The upper bracket (6), the lower bracket (7), and the electric telescopic rod (5) are all independently slidably fitted to the column (8) to achieve vertical position adjustment along the column (8); The upper correction ring (1) is detachably fixed to the upper bracket (6); The lower correction ring (4) is detachably fixed to the lower bracket (7); The middle correction ring (3) is detachably fixed to the telescopic end of the electric telescopic rod (5); Camera (9) is used to capture posture images of a user standing within the correction ring; And a PC (10), which is electrically connected to the camera (9) and the electric telescopic rod (5); The PC (10) is used to: receive the posture image and construct a three-dimensional posture model of the user's torso based on the posture image; and, through its built-in machine learning model, take the real-time posture data and historical data represented by the three-dimensional posture model as input, output control commands for controlling the electric telescopic rod (5) to drive the middle correction ring (3) to apply corrective force to the user's spinal curvature.

2. The scoliosis correction device of claim 1, wherein, Plastic pads (2) are provided on the inner walls of the upper correction ring (1), middle correction ring (3) and lower correction ring (4), and a flexible pressure sensor array is integrated inside the plastic pads (2); the flexible pressure sensor array is electrically connected to the PC (10) and is used to collect and form a real-time pressure distribution map covering the body contact area in real time during the application of the corrective force; the PC (10) uses this real-time pressure distribution map as a dynamic safety boundary condition for the machine learning model to calculate when generating the control command, so as to actively avoid outputting a corrective force that may cause the local pressure to exceed the preset safety threshold.

3. The scoliosis correction device of claim 2, wherein, There are multiple cameras (9), which are arranged in a spatial array around the correction ring; the PC (10) is used to: call a multi-view stereo vision algorithm to fuse and reconstruct the multi-channel two-dimensional posture images synchronously acquired by the multiple cameras (9) to generate a dynamic three-dimensional point cloud model containing tens of thousands of spatial coordinate points; and, based on the dynamic three-dimensional point cloud model, extract and continuously track key posture feature parameters in real time through geometric calculations. The key posture feature parameters include the estimated value of three-dimensional parameters used to characterize the curvature of the spine, the rotation of the vertex vertebra, and the three-dimensional spatial tilt matrix of the scapula and pelvis.

4. The scoliosis correction device of claim 3, wherein, The PC (10) is also used to execute a fully automatic calibration program before use; the fully automatic calibration program is used to: perform three-dimensional rigid registration between the real-time constructed dynamic three-dimensional point cloud model and the pre-imported user DICOM format medical image data, so as to accurately map and lock the three-dimensional spatial coordinates of the vertebral apex of the spinal curvature on the point cloud model. In addition, the target vertical sliding position of the electric telescopic rod (5) on the column (8) is automatically calculated, and a corresponding driving command is generated, thereby accurately moving the geometric center of the middle correction ring (3) to the user's body surface action point corresponding to the vertex.

5. The scoliosis correction device of claim 4, wherein, The machine learning model is a deep reinforcement learning model trained on historical data. Its input data includes real-time posture data, historical data, and a comprehensive feature vector containing user age, weight, scoliosis type, and Risser rating information. The control command is specifically a time series that defines the corrective force loading mode within the complete correction cycle. This sequence plans a multi-stage, nonlinear force-time curve including a micro-stress adaptation stage, a logarithmic progressive loading stage, a plateau constant force maintenance stage, and a linear unloading stage to simulate the preset corrective force application strategy.

6. The scoliosis correction device of claim 5, wherein, The PC (10) is executed by predictive closed-loop feedback control. In each control cycle, the actual posture change is compared with the expected change. The machine learning model is based on the user's biomechanical model to predict the undesirable compensatory movement trend that the body may produce under the current control command. Based on this prediction, a compensatory fine-tuning command is generated in advance and superimposed on the main control command. The fine-tuning command can simultaneously control the upper support (6) and the lower support (7) to adjust their vertical positions to actively suppress the occurrence of compensatory movements, thereby ensuring the targeting of the corrective force.

7. The scoliosis correction device of claim 6, wherein, The PC (10) executes a multi-level redundant safety monitoring program; the first level of the redundant safety monitoring program is physical layer monitoring based on the data of the flexible pressure sensor array, and protection is immediately triggered when any pressure value or its rate of change exceeds the safety threshold. The second level of the redundant safety monitoring program is electrical layer monitoring based on the encoder and current sensor built into the electric telescopic pole (5). By comparing the command displacement with the actual displacement in real time and monitoring the motor load current, it can detect potential mechanical jamming or external abnormal resistance. If any level detects an abnormality, it will trigger the system to enter the safe shutdown mode and lock all actuators.

8. The scoliosis correction device of claim 7, wherein, The PC (10) is also used for data archiving and generating a multi-dimensional correction process report. The multi-dimensional correction process report includes not only a comparison of the three-dimensional posture model before and after correction and the change values ​​of key posture feature parameters, but also in-depth analysis indicators such as user posture stability index, correction task cooperation score, and short-term correction effect prediction curve based on the current data update. The multi-dimensional correction process report can be exported and integrated with information systems to realize the electronic management of correction files.

9. The scoliosis correction device of claim 8, wherein, The PC (10) is also used to continuously optimize the machine learning model through federated learning or transfer learning mechanisms, including: periodically training the global basic model using anonymized data accumulated from multiple devices while protecting data privacy, and sending the optimized model parameters to the local machine; at the same time, using personalized data continuously generated by a single user to fine-tune the updated model online, so that the device can learn universal laws from wide-area data and provide highly personalized and adaptive control decisions for each user.

10. The scoliosis correction apparatus for children according to claim 9, wherein The fully automatic calibration program also includes a calibration module for the optimal direction of the corrective force. In this calibration module, when the middle correction ring (3) is positioned at the body surface position corresponding to the vertex, the PC (10) will control the electric telescopic rod (5) to apply a detection force and guide the user to rotate axially. During this process, by analyzing the deformation field distribution caused by the detection force on the three-dimensional posture model, the optimal force vector direction that enables the corrective force to be transmitted to the vertex most effectively and minimizes the ineffective compensatory deformation of the body is calculated, and this direction is used as the main direction of the corrective force in the subsequent correction process.

Images