Children scoliosis training correction rehabilitation machine and control system and control method thereof

Abstract

The invention discloses a children scoliosis training correction rehabilitation machine and a control system and a control method thereof. The children scoliosis training correction rehabilitation machine comprises a main framework capable of bending along the spine, and the main framework is provided with opposite first back faces and a first attaching face attached to the back of the human body; the two side skeletons can bend and move along the spine, and the side skeletons are provided with opposite second back faces and second attaching faces attached to the waist of the human body; and the pneumatic boosting mechanism can enable the treatment of the scoliosis of the children to be more accurate, safer and more efficient.

Description

Technical Field

[0001] The present invention relates to the technical field of rehabilitation equipment, and in particular to a children's scoliosis training and correction rehabilitation machine and a control system and a control method thereof. Background Art

[0002] The device patented in Chinese patent CN202310973704.6 relies solely on surface pressure sensor data and does not integrate the Cobb angle parameter 4 of X-ray images, resulting in a correction force line calculation error of >15% (data source: "Chinese Journal of Rehabilitation Medicine" 2024, Issue 3);

[0003] Commercially available products (such as the German Medi-Mouse system) use preset training trajectories and are unable to adjust the force angle according to the child's real-time electromyographic signals, resulting in a muscle compensation rate of 29.7% (Chinese Journal of Physical Medicine and Rehabilitation, Issue 12, 2023);

[0004] Traditional braces (such as Chinese patent CN202322347556U) lack the function of synchronous relaxation of contralateral muscle groups, violate the "contraction-relaxation" alternating training principle, and result in a 37% decrease in correction efficiency (from a clinical report from the Children's Medical Center affiliated to Shanghai Jiao Tong University). Summary of the Invention

[0005] The purpose of the present invention is to provide a pediatric scoliosis training, correction and rehabilitation machine and its control system and control method, which can make the treatment of pediatric scoliosis more accurate, safe and efficient.

[0006] According to the present invention, a pediatric scoliosis training, correction and rehabilitation device is provided, comprising:

[0007] A main frame capable of bending along the spine, the main frame having a first opposite back surface and a first fitting surface that fits against the back of a human body;

[0008] Two side frames capable of bending along the spine, the side frames having a second opposite back surface and a second fitting surface that fits the waist of the human body;

[0009] A pneumatic boosting mechanism, used to apply thrust to the main frame and the side frames;

[0010] The pneumatic boosting mechanism includes a first electric cylinder and a second electric cylinder respectively hinged to the first back side and the second back side, and a first push rod and a second push rod respectively reciprocating in the first electric cylinder and the second electric cylinder, and the free ends of the first push rod and the second push rod are respectively hinged to the first back side and the second back side.

[0011] Furthermore, the first electric cylinder and the second electric cylinder are correspondingly connected to a harmonic reducer and a six-dimensional force sensor. The reduction ratio of the harmonic reducer is 1:100, and the six-dimensional force sensor is used to detect the torque of the first electric cylinder and the second electric cylinder.

[0012] Furthermore, a pressure applying unit, an electromyographic stimulation unit and an electromagnetic braking unit are installed on the main frame and the side frame;

[0013] The pressure applying unit is provided with a pressure fluctuation reference value and a safety threshold;

[0014] The pressure applying unit is connected to an LED warning light and a buzzer alarm, and when the pressure fluctuation value of the pressure applying unit exceeds the reference value by 20%, the LED warning light and the buzzer alarm are triggered;

[0015] The myoelectric stimulation unit is connected to the pressure application unit, and when the myoelectric signal spectrum entropy mutation rate exceeds 35%, the pressure application unit is automatically reduced to 50% of the safety threshold;

[0016] The electromagnetic brake unit is connected to an airbag protector, and after the electromagnetic brake unit is started, the airbag protector pops out synchronously.

[0017] The present invention also provides a control system for the above-mentioned child scoliosis training, correction and rehabilitation machine, comprising:

[0018] A multimodal data fusion system comprising an integrated medical image analysis module for capturing full-spine X-ray DICOM images and a three-dimensional body modeling module for collecting and modeling three-dimensional data of the patient's body posture;

[0019] An intelligent control system, wherein the intelligent control system is used to run and fuse a dynamic parameter adjustment algorithm based on an LSTM neural network, and the intelligent control system outputs control instructions;

[0020] The safety protection system is used for early warning of pressure fluctuations, monitoring of sudden changes in electromyographic signals, and emergency braking for posture imbalances.

[0021] Furthermore, the three-dimensional body modeling module includes:

[0022] The improved Mask R-CNN vertebral segmentation module is used to perform pixel-level segmentation on the input full-spine X-ray DICOM images, accurately identifying and separating individual vertebrae;

[0023] A Cobb angle-muscle tension correlation database, wherein the Cobb angle-muscle tension correlation database is used to establish a nonlinear mapping model based on clinical data;

[0024] A data synchronization interface is used to be compatible with DICOM image format and 3D body scanning point cloud data.

[0025] Furthermore, the hidden layer of the LSTM neural network model contains 32 neurons, and the weight initialization adopts the He regularization method; the dynamic parameter calculation formula is:

[0026] Ft=0.7·CobbX+0.2·sEMGrms+0.1·ΔTinfrared;

[0027] Among them, CobbX is the spinal Cobb angle parameter, sEMGrms is the root mean square value of surface electromyography, and ΔTinfrared is the temperature difference value of infrared thermal imaging;

[0028] Adaptive training strategy dynamically adjusts the force direction and stimulation parameters according to the frequency domain characteristics of the electromyographic signal (MFCC coefficient).

[0029] Furthermore, the safety protection system includes:

[0030] The pressure applying unit triggers the LED warning light and buzzer alarm when the pressure fluctuation exceeds the reference value by 20%;

[0031] The electromyographic stimulation unit automatically reduces the force of the pressure application unit to 50% of the safety threshold when the mutation rate of the electromyographic signal spectrum entropy exceeds 35%;

[0032] Electromagnetic brake module: when the posture imbalance lasts for 3 seconds or more, the electromagnetic brake is activated and the airbag protector is ejected.

[0033] The present invention also provides a control method for a control system for intelligent rehabilitation of scoliosis in children, comprising the following steps:

[0034] Step 1: Collect the patient's full spine X-ray DICOM image, automatically segment the vertebrae and calculate the Cobb angle using the improved Mask R-CNN algorithm;

[0035] Step 2: Perform a 3D body scan to obtain point cloud data and build a personalized biomechanical model;

[0036] Step 3: Start the training program, and the intelligent control system dynamically adjusts the force parameters and electrical stimulation frequency of the robotic arm based on the LSTM neural network;

[0037] Step 4: Real-time monitoring of pressure, electromyography, and temperature data to trigger the three-level safety protection mechanism;

[0038] Step 5: After the training, an evaluation report including pressure distribution diagram, muscle group activation rate and correction angle change will be generated.

[0039] Furthermore, in step 3, the force parameters of the robotic arm are dynamically adjusted to:

[0040] Force adjustment ΔF = k × (sEMG_deviation) + m × (temperature gradient),

[0041] Where k = 0.5 N / μV, m = 2 N / °C, and sEMG_deviation is the percentage difference between the current EMG amplitude and the baseline value. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a schematic diagram of the combined structure of the main skeleton and the pneumatic mechanism according to an embodiment of the present invention.

[0043] In the figure, 1-main frame; 2-first back surface; 3-first fitting surface; 4-side frame; 5-second back surface; 6-second fitting surface; 7-pneumatic mechanism; 71-first electric cylinder; 72-second electric cylinder; 73-first push rod; 74-second push rod. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0045] The present invention discloses a child scoliosis training, correction and rehabilitation machine, comprising:

[0046] A main frame 1 capable of bending along the spine, the main frame 1 having a first opposite back surface 2 and a first fitting surface 3 fitted to the back of a human body;

[0047] Two side frames 4 capable of bending along the spine, the side frames 4 having opposite second back surfaces 5 and second fitting surfaces 6 for fitting against the waist of the human body;

[0048] A pneumatic boosting mechanism 7, which is used to apply thrust to the main frame 1 and the side frames 4;

[0049] The pneumatic boosting mechanism 7 includes a first electric cylinder 71 and a second electric cylinder 72 respectively hinged to the first back side 2 and the second back side 5, and a first push rod 73 and a second push rod 74 respectively reciprocating in the first electric cylinder 71 and the second electric cylinder 72. The free ends of the first push rod 71 and the second push rod 72 are respectively hinged to the first back side 2 and the second back side 5.

[0050] The first electric cylinder 71 and the second electric cylinder 72 are also connected to a harmonic reducer and a six-dimensional force sensor respectively. The reduction ratio of the harmonic reducer is 1:100, and the six-dimensional force sensor is used to detect the torque of the electric cylinder 31.

[0051] The main frame 1 and the side frame 4 are equipped with a pressure force unit, an electromyographic stimulation unit and an electromagnetic brake unit;

[0052] The pressure application unit is provided with a pressure fluctuation reference value and a safety threshold;

[0053] The pressure applying unit is connected to an LED warning light and a buzzer alarm. When the pressure fluctuation value of the pressure applying unit exceeds the reference value by 20%, the LED warning light and the buzzer alarm are triggered.

[0054] The electromyographic stimulation unit is connected to the pressure application unit. When the mutation rate of the electromyographic signal spectrum entropy exceeds 35%, the pressure application unit will be automatically reduced to 50% of the safety threshold.

[0055] The electromagnetic brake unit is connected to the airbag protector, and after the electromagnetic brake unit is activated, the airbag protector pops out synchronously.

[0056] It is worth noting that the present invention mainly uses the pneumatic booster mechanism 7 to drive the side frame 4 to train and correct children's scoliosis, and the main frame 1 assists the spine.

[0057] The present invention also provides a control system for the above-mentioned children's scoliosis training, correction and rehabilitation machine, comprising:

[0058] Multimodal data fusion system, which includes an integrated medical image analysis module and a 3D body modeling module. The integrated medical image analysis module is used to capture full-spine X-ray DICOM images, and the 3D body modeling module is used to establish 3D data collection and modeling of the patient's body posture.

[0059] Intelligent control system, which is used to run and integrate the dynamic parameter adjustment algorithm based on the LSTM neural network, and output control instructions;

[0060] Safety protection system: The safety protection system is used for pressure fluctuation warning, electromyographic signal mutation monitoring and posture imbalance emergency braking.

[0061] The 3D body modeling module includes:

[0062] The improved Mask R-CNN vertebral segmentation module is used to perform pixel-level segmentation on the input full-spine X-ray DICOM images, accurately identifying and separating individual vertebrae;

[0063] Cobb angle-muscle tension correlation database, which is used to establish a nonlinear mapping model based on clinical data;

[0064] Data synchronization interface, the data synchronization interface is used to be compatible with DICOM image format and 3D body scanning point cloud data.

[0065] It is worth noting that the improved Mask R-CNN architecture described in this paper has undergone a series of optimizations based on the original Mask R-CNN. In particular, it has proposed multiple innovative improvements for the complex anatomical structures and small object segmentation of spinal X-ray images. Compared with the original Mask R-CNN, the main improvements include the following aspects:

[0066] 1. Improvement of the backbone network: The original Mask R-CNN usually uses ResNet as the backbone network, while this model combines ResNet with a CB-Net composite structure. The introduction of CB-Net enhances cross-level feature interaction through the composite connection module (CompositeConnection), so that low-level features can effectively receive support from high-level semantic information. This dual-path feature extraction method can improve the network's ability to model complex spinal anatomical structures (such as pedicles, osteophytes, and other irregular shapes). In addition, based on the original ResNet, deformable convolution (Deformable Conv) is used to adaptively adjust the convolution kernel sampling position. This improvement enables the network to more accurately capture irregular shapes in the spinal area, while the original Mask R-CNN relies on fixed convolution kernels, which performs poorly when processing irregular targets.

[0067] 2. Feature fusion and pyramid network enhancement: In the original Mask R-CNN, the Feature Pyramid Network (FPN) is mainly used for the fusion of multi-scale features, but its feature pyramid levels are usually achieved through simple upsampling and convolution. The improved architecture deeply fuses the multi-level features from ResNet and CB-Net by introducing a composite feature pyramid construction method. It not only performs upsampling and cross-level feature fusion through 3×3 convolution, but also specially designs cross-layer information flow to further enhance the coordinated expression of global semantics and local details. In particular, when processing small targets (such as intervertebral discs), a high-resolution preservation branch is introduced to ensure the accurate capture of details, which is not involved in the original Mask R-CNN.

[0068] 3. Dynamic Optimization of ROIAlign: The ROIAlign method in the original Mask R-CNN fixed the size and position of the ROI region. In this improved model, ROIAlign has been improved to a dynamic selection mechanism that automatically matches the optimal pyramid level based on the target size. In addition, the area outside the original ROI window is expanded by 15% to capture the associated features of the soft tissue surrounding the vertebra, further improving segmentation accuracy and the context of the target. This optimization enables the model to more accurately extract target features when processing targets of different scales.

[0069] 4. Optimization of multi-task collaborative design: In the collaborative optimization of detection and segmentation tasks, the improved Mask R-CNN has added several new mechanisms on the basis of the original Mask R-CNN. First, the classification branch of the detection head uses FocalLoss to alleviate the category imbalance problem in vertebral classification (such as the imbalance between normal vertebrae and diseased vertebrae). Secondly, CIoU Loss is used in the regression branch to improve the positioning accuracy of the vertebral bounding box. In addition, in the segmentation head, in addition to the traditional mask prediction, an edge refinement module (Edge-Aware Refinement) is added. The edge attention map generated by the cascaded deformable convolution and the Sobel operator is used to optimize the segmentation results of the vertebral endplate. This improvement can significantly improve the accuracy of the details of the segmentation edge, while the original Mask R-CNN does not perform such detailed processing on the segmentation edge.

[0070] 5. Feature Recalibration Mechanism: In the original Mask R-CNN, the information after feature fusion may contain redundant background noise. To suppress these irrelevant features, this improved model introduces a channel attention mechanism (SE Block) that adaptively recalibrates the fused features, enhancing focus on important targets. This mechanism effectively improves the distinction between the spine and the background in spinal images, significantly enhancing segmentation accuracy, especially in noisy medical images.

[0071] The hidden layer of the LSTM neural network model contains 32 neurons, and the weight initialization uses the He regularization method; the dynamic parameter calculation formula is:

[0072] Ft=0.7·CobbX+0.2·sEMGrms+0.1·ΔTinfrared;

[0073] Wherein, CobbX is the spinal Cobb angle parameter, sEMGrms is the root mean square value of surface electromyography, and ΔTinfrared is the temperature difference value of infrared thermal imaging;

[0074] Adaptive training strategy dynamically adjusts the force direction and stimulation parameters according to the frequency domain characteristics of the electromyographic signal (MFCC coefficient).

[0075] The safety protection system includes:

[0076] The pressure applying unit triggers the LED warning light and buzzer alarm when the pressure fluctuation exceeds the reference value by 20%;

[0077] The electromyographic stimulation unit automatically reduces the force of the pressure application unit to 50% of the safety threshold when the mutation rate of the electromyographic signal spectrum entropy exceeds 35%;

[0078] Electromagnetic brake module: when the posture imbalance lasts for 3 seconds or more, the electromagnetic brake is activated and the airbag protector is ejected.

[0079] The present invention also provides a control method for a control system for intelligent rehabilitation of scoliosis in children, comprising the following steps:

[0080] Step 1: Collect the patient's full spine X-ray DICOM image, automatically segment the vertebrae and calculate the Cobb angle using the improved Mask R-CNN algorithm;

[0081] Step 2: Perform a 3D body scan to obtain point cloud data and build a personalized biomechanical model;

[0082] Step 3: Start the training program, and the intelligent control system dynamically adjusts the force parameters and electrical stimulation frequency of the robotic arm based on the LSTM neural network;

[0083] Step 4: Real-time monitoring of pressure, electromyography, and temperature data to trigger the three-level safety protection mechanism;

[0084] Step 5: After the training, an evaluation report including pressure distribution diagram, muscle group activation rate and correction angle change will be generated.

[0085] In step 3, the dynamic adjustment of the robot arm force parameters is:

[0086] The force adjustment amount ΔF = k × (sEMG_deviation) + m × (temperature gradient), where k = 0.5 N / μV, m = 2 N / °C, and sEMG_deviation is the percentage difference between the current EMG amplitude and the baseline value.

[0087] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. Children's scoliosis training and correction rehabilitation machine, characterized by: include: A main frame capable of bending along the spine, the main frame having a first opposite back surface and a first fitting surface that fits against the back of a human body; Two side frames capable of bending along the spine, the side frames having a second opposite back surface and a second fitting surface that fits the waist of the human body; A pneumatic boosting mechanism, used to apply thrust to the main frame and the side frames; The pneumatic boosting mechanism includes a first electric cylinder and a second electric cylinder respectively hinged to the first back side and the second back side, and a first push rod and a second push rod respectively reciprocating in the first electric cylinder and the second electric cylinder, and the free ends of the first push rod and the second push rod are respectively hinged to the first back side and the second back side.

2. The children's scoliosis training, correction and rehabilitation machine according to claim 1, characterized in that: The first electric cylinder and the second electric cylinder are also connected to a harmonic reducer and a six-dimensional force sensor respectively. The reduction ratio of the harmonic reducer is 1:100, and the six-dimensional force sensor is used to detect the torque of the first electric cylinder and the second electric cylinder.

3. The children's scoliosis training, correction and rehabilitation machine according to claim 2, characterized in that: The main frame and the side frame are equipped with a pressure force unit, an electromyographic stimulation unit and an electromagnetic brake unit; The pressure applying unit is connected to an LED warning light and a buzzer alarm, and when the pressure fluctuation value of the pressure applying unit exceeds the reference value by 20%, the LED warning light and the buzzer alarm are triggered; The myoelectric stimulation unit is connected to the pressure application unit, and when the myoelectric signal spectrum entropy mutation rate exceeds 35%, the pressure application unit is automatically reduced to 50% of the safety threshold; The electromagnetic brake unit is connected to an airbag protector, and after the electromagnetic brake unit is started, the airbag protector pops out synchronously.

4. A control system for a children's scoliosis training, correction and rehabilitation machine as claimed in claim 3, characterized in that: include: A multimodal data fusion system comprising an integrated medical image analysis module for capturing full-spine X-ray DICOM images and a three-dimensional body modeling module for collecting and modeling three-dimensional data of the patient's body posture; An intelligent control system, wherein the intelligent control system is used to run and fuse a dynamic parameter adjustment algorithm based on an LSTM neural network, and the intelligent control system outputs control instructions; The safety protection system is used for early warning of pressure fluctuations, monitoring of sudden changes in electromyographic signals, and emergency braking for posture imbalances.

5. The control system of the children's scoliosis training, correction and rehabilitation machine according to claim 4, characterized in that: The three-dimensional body modeling module includes: The improved Mask R-CNN vertebral segmentation module is used to perform pixel-level segmentation on the input full-spine X-ray DICOM images, accurately identifying and separating individual vertebrae; A Cobb angle-muscle tension correlation database, wherein the Cobb angle-muscle tension correlation database is used to establish a nonlinear mapping model based on clinical data; A data synchronization interface is used to be compatible with DICOM image format and 3D body scanning point cloud data.

6. The control system of the children's scoliosis training, correction and rehabilitation machine according to claim 5, characterized in that: The hidden layer of the LSTM neural network model contains 32 neurons, and the weight initialization adopts the He regularization method; the dynamic parameter calculation formula is: Ft=0.7·CobbX+0.2·sEMGrms+0.1·ΔTinfrared; Among them, CobbX is the spinal Cobb angle parameter, sEMGrms is the root mean square value of surface electromyography, and ΔTinfrared is the temperature difference value of infrared thermal imaging; Adaptive training strategy dynamically adjusts the force direction and stimulation parameters according to the frequency domain characteristics of the electromyographic signal (MFCC coefficient).

7. The control system of the children's scoliosis training, correction and rehabilitation machine according to claim 6, characterized in that: The safety protection system includes: The pressure applying unit triggers the LED warning light and buzzer alarm when the pressure fluctuation exceeds the reference value by 20%; The electromyographic stimulation unit automatically reduces the force of the pressure application unit to 50% of the safety threshold when the mutation rate of the electromyographic signal spectrum entropy exceeds 35%; Electromagnetic brake module: when the posture imbalance lasts for 3 seconds or more, the electromagnetic brake is activated and the airbag protector is ejected.

8. A control method for a control system of a pediatric scoliosis training, correction and rehabilitation machine as claimed in claim 7, characterized in that: The following steps are involved: Step 1: Collect the patient's full spine X-ray DICOM image, automatically segment the vertebrae and calculate the Cobb angle using the improved Mask R-CNN algorithm; Step 2: Perform a 3D body scan to obtain point cloud data and build a personalized biomechanical model; Step 3: Start the training program, and the intelligent control system dynamically adjusts the force parameters and electrical stimulation frequency of the robotic arm based on the LSTM neural network; Step 4: Real-time monitoring of pressure, electromyography, and temperature data to trigger the three-level safety protection mechanism; Step 5: After the training, an evaluation report including pressure distribution diagram, muscle group activation rate and correction angle change will be generated.

9. The control method of the control system of the children's scoliosis training, correction and rehabilitation machine according to claim 8, characterized in that: In step 3, the dynamic adjustment of the robot arm force parameters is: Force adjustment ΔF = k × (sEMG_deviation) + m × (temperature gradient), Where k = 0.5 N / μV, m = 2 N / °C, and sEMG_deviation is the percentage difference between the current EMG amplitude and the baseline value.

Images