Intelligent evaluation system and method for rehabilitation training exercises

Abstract

The application discloses a kind of intelligent evaluation system and method of rehabilitation training exercise.The method collects individual characteristic data of exerciser and its breathing, action signal in the exercise process of rehabilitation training exercise, the breathing, action signal collected is calculated, and the completion quality index of exerciser in rehabilitation training exercise is obtained, and machine learning model is established, and the rehabilitation utility score of rehabilitation training exercise is calculated according to individual characteristic data and completion quality index using model.The application combines the concept and decision of traditional health preserving exercise method, integrates traditional Chinese medicine knowledge into modern artificial intelligence technology, so that the completion quality and rehabilitation utility of rehabilitation training exercise are scientifically quantified and evaluated, and the exerciser can be effectively guided to adjust without the presence of Chinese medicine experts, to promote patient rehabilitation.

Description

Technical Field

[0001] This invention relates to the application of artificial intelligence technology in the field of health and wellness, specifically to the intelligent evaluation of the completion quality and rehabilitation effectiveness of rehabilitation exercises, and in particular to a system and method for intelligently evaluating the completion quality and rehabilitation effectiveness of rehabilitation exercises based on the fluency of movement, the smoothness of breathing, and the coordination of movement and breathing. Background Technology

[0002] Traditional health-preserving exercises are often used in rehabilitation training. Traditional health-preserving exercises such as the Five Animal Frolics, Eight Pieces of Brocade, and Tai Chi Chuan mostly revolve around "guiding techniques," combining breathing, movement, and mental focus to regulate the body and mind. For example, the Eight Pieces of Brocade is an exercise consisting of eight graceful and flowing movements. It integrates physical movement, breathing, and mental focus. Although it primarily focuses on physical movement, if the movements are fluid and coordinated with breathing, its regulatory effect on the body's internal organs and blood circulation will be even better, which is beneficial to the patient's recovery.

[0003] However, the traditional evaluation system for rehabilitation exercises only assesses whether the movements are performed correctly. It cannot conduct a more scientific and comprehensive evaluation based on the key points of traditional health-preserving exercises. For example, it cannot evaluate whether the movements are smooth, whether the breath is "deep, long, even, and fine," or whether the movements and breath are coordinated. Therefore, it cannot accurately measure the completion of rehabilitation exercises.

[0004] For example, Chinese patent application No. 202510253970.0, "Eight-Section Brocade Training and Evaluation System Based on Monocular Video 3D Human Pose Estimation," proposes extracting 3D human key points from monocular videos, calculating the key point differences between the user and the standard movements, and realizing the identification, feedback, and evaluation of movement deviations. Chinese patent application No. 202411709763.3, "Eight-Section Brocade Movement Evaluation Method and System Based on Dynamic Programming and Deep Learning," collects Eight-Section Brocade movement video data, extracts skeletal points, calculates the distance between the user and the standard, and further derives the total score. Existing similar methods include Chinese patent application No. 202411604079.9, "Eight-Section Brocade Intelligent Scoring System Based on Machine Learning and Sum-and-Product Method," and Chinese patent application No. 202411452201.5, "An Eight-Section Brocade Virtual Training Method and System Based on an 8K Set-Top Box." These patent applications focus on using video acquisition and processing technology to extract skeletal points, compare them with standards, calculate errors, and thus provide an evaluation. However, video detection inevitably involves differences in perspective and human anatomy across different scenarios, resulting in skeletal points or key points that cannot reflect the smoothness of movement. Furthermore, the methods described above are completely incapable of assessing the smoothness of breathing.

[0005] In summary, current technologies primarily focus on assessing whether a person's rehabilitation exercises are performed correctly, but cannot comprehensively evaluate the overall quality of the exercise's execution. Furthermore, current technologies lack effective evaluation or calculation of the actual rehabilitation efficacy of rehabilitation exercises, thus failing to scientifically guide individuals on how to practice them, and even less able to adjust the exercise methods from the perspective of rehabilitation effectiveness. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] The technical problem that this invention aims to solve is that existing evaluation systems for rehabilitation exercises cannot absorb and utilize the essence of traditional health-preserving exercises, and cannot scientifically quantify and evaluate the quality of rehabilitation exercises.

[0008] The technical problem solved by this invention is that existing technologies cannot evaluate the actual rehabilitation efficacy based on the completion quality of quantified rehabilitation training exercises, nor can they evaluate the influence of various quantitative indicators on rehabilitation efficacy, thus failing to guide practitioners to improve training methods to promote rehabilitation.

[0009] (II) Technical Solution

[0010] To address the aforementioned problems, this invention proposes an intelligent evaluation system for rehabilitation exercises, used to evaluate the completion quality and rehabilitation effectiveness of exercises performed by practitioners. The intelligent evaluation system includes: a data acquisition module for collecting individual characteristic data of the practitioner and their respiratory and movement signals during the exercise; a data processing module for calculating the respiratory and movement signals collected by the data acquisition module to obtain a completion quality index for the exercise; and an intelligent evaluation module for establishing a machine learning model and using the model to estimate the rehabilitation effectiveness score of the exercise based on the individual characteristic data and the completion quality index. The machine learning model is trained using historical data, including historical individual characteristic data, the completion quality index of the exercise, and rehabilitation effectiveness scores determined by experts.

[0011] According to a preferred embodiment of the present invention, the completion quality indicators include at least one of the following: movement fluency, breath smoothness, and movement-breath coordination information.

[0012] According to a preferred embodiment of the present invention, the smoothness of the movement is represented by the rate of change of acceleration of a part of the human body.

[0013] According to a preferred embodiment of the present invention, the motion signal is a linear acceleration signal of a human body part obtained by analyzing the video signal of rehabilitation training exercises.

[0014] According to a preferred embodiment of the present invention, the smoothness of breathing is expressed by the coefficient of variation of respiratory rate or the rate of change of respiratory airflow velocity.

[0015] According to a preferred embodiment of the present invention, the coordination degree of movement and breath is represented by the phase difference between key nodes of the acceleration signal and the breathing signal of the human body parts.

[0016] According to a preferred embodiment of the present invention, the intelligent assessment module is further used to calculate the degree of influence affecting the rehabilitation utility score.

[0017] According to a preferred embodiment of the present invention, the influence degree is the SHAP value of each input feature.

[0018] According to a preferred embodiment of the present invention, it further includes an expert scoring module, which is used to provide the machine learning model with a rehabilitation utility score given by an expert, which serves as output label data in the training dataset.

[0019] A second aspect of this invention proposes an intelligent evaluation method for rehabilitation exercises, used to evaluate the completion quality and rehabilitation effectiveness of exercises performed by practitioners. The intelligent evaluation method includes the following steps: collecting individual characteristic data of the practitioner and their respiratory and movement signals during the exercise; calculating the respiratory and movement signals to obtain a completion quality index for the exercise; establishing a machine learning model and using the machine learning model to estimate the rehabilitation effectiveness score of the exercise based on the individual characteristic data and the completion quality index, wherein the machine learning model is trained using historical data, including historical individual characteristic data, the completion quality index of the exercise, and rehabilitation effectiveness scores determined by experts.

[0020] (III) Beneficial Effects

[0021] This invention learns and introduces the concepts and key points of traditional health-preserving exercises such as Five Animal Frolics, Eight Pieces of Brocade, and Tai Chi, and proposes indicators for smooth movements, smooth breathing, and coordinated movements and breathing, and scientifically quantifies these indicators.

[0022] This invention integrates the requirements of traditional health-preserving exercises, such as the unity of movement, breath, and intention, and the "deep, long, even, and fine" breath, with modern artificial intelligence calculations, thereby achieving a scientific evaluation of the completion quality and rehabilitation effect of rehabilitation training exercises.

[0023] This invention can scientifically evaluate the actual rehabilitation effect of rehabilitation exercises on practitioners, and can calculate the influence of various completion quality indicators on rehabilitation efficacy. Thus, it can provide targeted guidance to practitioners to make targeted improvements even without the presence of a traditional Chinese medicine expert, effectively promoting the patient's rehabilitation. Attached Figure Description

[0024] Figure 1 This is a flowchart of the intelligent evaluation method for rehabilitation training exercises of the present invention.

[0025] Figure 2 This is a functional module architecture diagram of the intelligent evaluation system for rehabilitation training exercises according to the first embodiment of the present invention.

[0026] Figure 3 This is a functional module architecture diagram of the intelligent evaluation system for rehabilitation training exercises according to the first embodiment of the present invention during model training.

[0027] Figure 4 This is a functional module architecture diagram of the expert scoring module of the intelligent evaluation system for rehabilitation training exercises according to the first embodiment of the present invention.

[0028] Figure 5 This is a functional module architecture diagram of the intelligent evaluation system for rehabilitation training exercises according to the second embodiment of the present invention. Detailed Implementation

[0029] To address the problem that existing rehabilitation exercise evaluation systems cannot comprehensively assess the quality of a practitioner's performance and the rehabilitation effectiveness, this invention proposes an intelligent evaluation system for rehabilitation exercises. This system not only evaluates whether the movements are performed correctly, but also assesses the overall quality of performance. Furthermore, it uses machine learning to correlate the quality of the rehabilitation exercises with the practitioner's rehabilitation effectiveness.

[0030] We know that although traditional Chinese medicine and Western medicine theories have pointed out that traditional health-preserving exercises such as Baduanjin (Eight Pieces of Brocade) can dredge meridians, harmonize qi and blood, and resolve pathological states of "qi stagnation and blood stasis," showing significant effects on postoperative rehabilitation, chronic disease management, and nervous system diseases, the exact relationship between the quality of completion of these exercises and their effects remains inconclusive. However, with the rise of artificial intelligence technology based on machine learning, this complex problem of multi-parameter correspondence has found a solution. Based on this, the inventors of this invention have innovatively proposed that, through training a machine learning model, data related to the rehabilitation efficacy of actual rehabilitation patients can be correlated with the quality of completion of rehabilitation exercises (especially traditional exercises). This allows the trained machine learning model to infer the rehabilitation efficacy based on the individual characteristics of the rehabilitation patient and the quality of completion of the rehabilitation exercises, and to guide the rehabilitation patient to adjust their practice based on this inferred rehabilitation efficacy data and its attribution, such as how to regulate their movements or breathing for better rehabilitation.

[0031] Figure 1 This is a flowchart of the intelligent assessment method for rehabilitation training exercises according to the present invention. Figure 1 As shown, the method of the present invention includes the following steps:

[0032] S1. Collect individual characteristic data of the trainees and their respiratory and movement signals during the rehabilitation training exercises;

[0033] S2. Calculate the collected respiratory and movement signals to obtain the completion quality index of the exerciser's rehabilitation training exercises;

[0034] S3. Establish a machine learning model and use the machine learning model to estimate the rehabilitation efficacy score of the rehabilitation training exercise based on the individual characteristic data and completion quality indicators.

[0035] The machine learning model is trained using historical data, including historical individual characteristic data, quality indicators of rehabilitation training exercises, and rehabilitation utility scores determined by experts.

[0036] Unlike existing assessment methods that focus solely on whether movements meet standards, this invention learns from and incorporates the key principles and health concepts of traditional exercises. It proposes to assess the quality of completion by combining breathing, movement, and the coordination between the two, rather than simply evaluating whether a single movement meets the standard. More specifically, this invention proposes to describe the quality of rehabilitation exercises based on the smoothness of movement, the smoothness of breathing, and the coordination between movement and breath. This makes the assessment of rehabilitation exercises in this invention more closely aligned with the guiding principles of traditional Chinese medicine for health preservation exercises.

[0037] As a specific implementation method, the "motion smoothness" proposed in this invention includes the smoothness of the body part's motion state during a single movement, as well as the smoothness of the transition between different movements. To describe this "motion smoothness," one implementation method proposed in this invention uses the rate of change of acceleration. The rate of change of acceleration (i.e., the derivative of linear acceleration) can be calculated by differentiating the linear acceleration signal over time and then calculating the mean or root mean square of the rate of change of acceleration. A lower rate of change of acceleration in smooth movements indicates gentle acceleration changes, while a higher rate may indicate stuttering or discontinuity in the movement.

[0038] To evaluate the rate of change of acceleration values ​​for the overall movement, this invention requires measuring the acceleration values ​​of different parts of the body during exercise. One approach is to wear accelerometers on various parts of the body to directly acquire acceleration signals. This method yields relatively accurate acceleration values ​​and is unaffected by the scene or lighting conditions, but the process of wearing the accelerometers is cumbersome. Another option is image detection, which involves recording the exercise process, detecting the position of different parts of the body over time from the video image, and calculating the velocity and acceleration values ​​of each part, thereby calculating the rate of change of acceleration. This method is easy to implement but is affected by the scene and lighting conditions, and can also be inaccurate due to changes in the shooting angle.

[0039] Combining the two methods of acquiring acceleration described above, this invention also proposes a method to improve the accuracy of acceleration values ​​detected in video images by using the first measurement method to correct the second measurement method. Specifically, this invention can obtain a correction coefficient through the comparison of signals measured simultaneously in a limited number of cases, and use the correction coefficient to correct the acceleration values ​​acquired through image detection. For example, in a specific measurement example, for the same body part (e.g., the left arm), both the accelerometer signal (which can be called the standard acceleration) and video recording are performed. The positional change of the left arm is detected from the recorded video, and the acceleration curve over time is calculated (which can be called the calculated acceleration). The standard acceleration and the calculated acceleration are then aligned and compared.

[0040] Since this invention primarily focuses on the smoothness of movement, it pays more attention to the amplitude of acceleration rather than its phase. Therefore, this invention can correct only the amplitude of acceleration. When correcting the amplitude of acceleration, a simple linear correction can be used. That is, for the standard and calculated amplitude of acceleration at each time point, we can assume A = kB + b, where A is the amplitude of the standard acceleration, B is the amplitude of the calculated acceleration, and k and b are correction coefficients. The least squares method can be used to fit k and b to minimize the mean square error (MSE) between the corrected data and the accelerometer data. This invention does not exclude other correction methods, such as nonlinear and polynomial regression, or even neural networks. Since error correction methods are well known in the art, they will not be described in detail here.

[0041] The present invention is not limited to other indicators representing "motion smoothness", such as angular acceleration or rate of change of angular acceleration.

[0042] Another quality evaluation indicator proposed in this invention is "breathing smoothness." For traditional health-preserving exercises, unobstructed and smooth breathing is also an important evaluation indicator. For example, the traditional exercise Baduanjin requires breathing to be "deep, long, even, and fine." Therefore, this invention, combining the requirements of traditional health-preserving exercises, defines "breathing smoothness" as the stability of the airflow, including the stability of the airflow during a single exhalation or inhalation phase, the stability of the rhythm of multiple exhalation or inhalation phases, and the stability of the pause between adjacent breathing phases, etc. In other words, this invention attempts to quantify "breathing smoothness," requiring not only stable breathing and rhythm but also smooth transitions between breaths.

[0043] One feasible implementation is to use the coefficient of variation of respiratory rate (CV_RR) as an indicator of "breathing smoothness". The coefficient of variation of respiratory rate (CV_RR) = standard deviation of respiratory cycle duration / average respiratory cycle duration × 100%. The lower the coefficient of variation, the more stable the respiratory rhythm. For example, CV_RR is close to 0 during uniform breathing. This indicator can be obtained by measuring respiratory signals using impedance breathing belts, piezoelectric sensors worn on the chest and abdomen, etc., eliminating the need for airflow sensors.

[0044] Another feasible implementation is to use an airflow sensor to detect the respiratory airflow velocity or pressure, obtain a respiratory signal reflecting the breath, and then differentiate this respiratory information to obtain the rate of change of respiratory airflow velocity. The root mean square of this rate of change of airflow velocity is used to represent an indicator of "breathing smoothness".

[0045] Furthermore, the length of the pause between two adjacent breathing phases detected in the respiratory signal can be statistically analyzed, and the average pause duration and the number of pauses shorter than a predetermined threshold duration can be calculated. A pause duration that is too short may indicate some degree of shortness of breath or insufficient respiratory amplitude, while an uneven distribution of pause duration indicates an unstable respiratory rate. Therefore, this invention also uses this as an indicator of "breathing smoothness."

[0046] In addition to the above indicators, other indicators can be determined according to the requirements of "deep, long, even, and fine" breathing. For example, "deep" can be quantified as the total airflow within a breathing cycle, which should be as high as possible; "long" can be quantified as the duration of the breathing cycle, which should be as long as possible; "even" refers to the rate of change of airflow, which should be as low as possible, and the coefficient of variation of respiratory rate (CV_RR) mentioned earlier is an indicator of "even"; "fine" means that the airflow itself should be as fine as possible, which can be represented by the average airflow within a cycle.

[0047] This invention also proposes "coordination of movement and breath" as another key indicator of performance quality. "Smoothness of movement" and "smoothness of breath" have already been mentioned, and combining these two to assess the coordination of movement and breath is a crucial principle of traditional qigong practice. To integrate this principle into the evaluation of performance quality, this invention proposes the concept of "coordination of movement and breath," which in this invention refers to the consistency of the rhythm (frequency) of movement and breath at key time points.

[0048] For example, Baduanjin, a traditional Chinese exercise, emphasizes the unity of "form, breath, and intention," with the coordination of breathing and movement being the core principle. Its basic breathing patterns include inhaling when opening and exhaling when closing, and inhaling when rising and exhaling when descending. In the inhaling-closing-exhaling pattern, inhale when expanding the limbs (e.g., raising the arms, expanding the chest) and exhale when retracting (e.g., lowering the arms, tucking the chest). In the rising-and-descending-exhaling pattern, inhale when rising (e.g., standing up, raising the arms) and exhale when descending (e.g., squatting, lowering the arms). It is evident that the transition points between inhalation and exhalation should be consistent with the transition points between movements, and one breathing cycle may correspond to the completion of one or two movements.

[0049] Considering this situation, this invention proposes using the phase difference between action signals and respiratory signals at key nodes to represent "action-breath coordination." As mentioned earlier, the action signal can be an acceleration signal measured by an accelerometer, an acceleration signal obtained from video detection, or a corrected acceleration signal. For the acceleration signal, the transition node between actions can be considered a key node, which can be obtained by detecting the starting point of the action, since the starting point of the action is often the location of the maximum value of the acceleration signal. Therefore, the time node where the maximum value of the acceleration signal is detected is used as the key node. For the respiratory signal, the key node can be obtained by detecting the transition time node between exhalation and inhalation. Since the airflow is zero and the chest and abdominal expansion and contraction changes are also zero during the transition, the zero-crossing point of the respiratory signal can be used as the key node.

[0050] Once the key nodes of the action and respiration signals are acquired, the phase difference, i.e., the time deviation, between each key node of the respiration signal and its corresponding key node of the action signal can be calculated. When multiple actions are required during an exhalation or inhalation phase, a key node of the respiration signal may not have a corresponding key node of the action signal; in such cases, the calculation of the key node of that respiration signal is excluded. For all respiration key nodes with corresponding key nodes of action, the root mean square of the phase difference between all nodes can be calculated as an indicator of "action-breath coordination". Alternatively, the number of times the phase difference exceeds a predetermined threshold (e.g., 0.5s) can also be counted as an indicator of "action-breath coordination". That is, as long as it reflects the consistency of the key nodes of action and respiration, it can be used as an indicator of "action-breath coordination" in this invention.

[0051] The rehabilitation efficacy scoring data of the present invention will be further explained below.

[0052] The calculation of the evaluation index for the completion quality of rehabilitation exercises, innovatively proposed in this invention, has already been mentioned. However, the goal of this invention is not merely to evaluate the completion quality of rehabilitation exercises, but to explore more deeply the relationship between the completion quality of rehabilitation exercises and rehabilitation efficacy, and to provide more scientific guidance for the practice of rehabilitation exercises. Therefore, this invention proposes the use of the concept of "rehabilitation efficacy." The "rehabilitation efficacy" of this invention refers to the degree of effectiveness that rehabilitation exercises ultimately achieve in rehabilitation.

[0053] The effectiveness of rehabilitation exercises in overall rehabilitation is a difficult indicator to quantify. Many factors influence "rehabilitation effectiveness," including the individual's physical condition and the type of rehabilitation (e.g., post-operative rehabilitation, chronic disease rehabilitation, etc.). Furthermore, the duration, frequency, and intensity of the exercises also play a role. Therefore, "rehabilitation effectiveness" cannot be calculated using conventional quantitative methods and must be assessed by experienced experts based on their expertise and the patient's specific situation.

[0054] To incorporate expert experience into the assessment of "rehabilitation effectiveness," this invention proposes using big data and machine learning. Specifically, this invention proposes establishing a machine learning model and a training dataset. This training dataset needs to include a large amount of sample data. Each sample should include individual characteristic data of the practitioner of the rehabilitation exercises, the quality indicators of the exercise completion, and rehabilitation effectiveness scores determined by experts. The individual characteristic data and the quality indicators of the exercise completion are used as input features to the machine learning model, while the rehabilitation effectiveness scores determined by experts are used as output features to train the model. Thus, the model can acquire expert experience and assess the potential rehabilitation effectiveness based on individual characteristics and the quality of exercise completion.

[0055] Because this invention takes into account individual differences and can accurately describe the quality of rehabilitation exercises, as long as there is sufficient sample data for training, the trained model can accurately and quickly evaluate the actual rehabilitation effect of the rehabilitation exercises.

[0056] It should be noted that the individual characteristic data here includes the patient's gender, age, illness information, and rehabilitation training plan information. The completion quality of the rehabilitation exercises is a completion quality index obtained by processing the movements and breathing signals detected during the practice of the rehabilitation exercises as described above.

[0057] Furthermore, since this invention employs a machine learning model, the influence of input indicators affecting "rehabilitation efficacy" can be calculated based on the model's output. For example, for classification models, feature importance indicators can be directly used for calculation. Alternatively, the SHAP (SHapley Additive Explanations) value of each input feature can be calculated to determine the marginal contribution of each feature to the prediction.

[0058] The present invention will be further illustrated by specific embodiments below.

[0059] Figure 2 This is a functional module architecture diagram of the intelligent evaluation system for rehabilitation training exercises according to the first embodiment of the present invention. Figure 2 As shown, the intelligent evaluation system of the present invention includes a data acquisition module 1, a data processing module 2, and an intelligent evaluation module 3. The data acquisition module 1 includes an action signal acquisition unit 11, a respiratory signal acquisition unit 12, and an individual characteristic data acquisition unit 13.

[0060] The motion signal acquisition unit 11 is used to acquire the motion signals of the practitioner performing rehabilitation exercises. Since rehabilitation exercises typically consist of a series of continuous body movements, the practice status can be characterized by detecting the signals generated by these body movements over a period of time. This invention does not limit the physical quantity represented by the motion signal; displacement, velocity, and acceleration can all be used as motion signals. However, considering the subsequent calculation of indicators representing the quality of movement completion, such as "movement smoothness," acceleration signals are preferred as motion signals. Therefore, the motion signal acquisition unit 11 receives motion sensing signals (source signals) representing human body movements and converts them into the desired motion signals. The motion sensing signals are source signals generated by sensors, and their signal type depends on the type of sensor.

[0061] For example, such as Figure 2 As shown, the motion sensing signals received by the motion signal acquisition unit 11 can be linear acceleration signals and angular velocity signals obtained from an accelerometer and a gyroscope, or video signals obtained from an image sensor. Linear acceleration signals can be used directly as motion signals, or after filtering and other preprocessing. Angular velocity signals can be used directly as motion signals, or after differentiation, they can be converted into azimuth, pitch, angular acceleration, etc., signals before being used as motion signals. For video signals, the velocity or acceleration of a specific part of the human body can be calculated by analyzing the position of that part in the video signal. The conversion of the above signals and the algorithm for analyzing video signals to obtain the velocity or acceleration of various parts of the human body can be directly adopted from existing technologies, and therefore will not be described further here. Of course, this invention does not exclude the use of other types of sensors to obtain motion signals characterizing motion.

[0062] As a specific example of linear acceleration signals, they can be acquired by wearing three-dimensional accelerometers on the head, wrists, elbows, shoulders, hips, knees, and ankles. As for angular velocity signals, they can be acquired by wearing gyroscopes on the head, wrists, elbows, shoulders, hips, knees, and ankles. Alternatively, integrated inertial sensors can be used to simultaneously obtain linear acceleration and angular velocity signals. As a specific example of acquiring motion signals through video signals, to facilitate signal acquisition, a mobile phone camera can be used to record practice videos of rehabilitation exercises. An app installed on the phone can then detect positional changes of key body parts, calculating the linear acceleration of multiple body parts based on this. These key parts include, for example: nose, left corner of eye, right corner of eye; left shoulder, left elbow, left wrist, right shoulder, right elbow, right wrist; left hip, left knee, left ankle, right hip, right knee, right ankle. Although this key positional data is two-dimensional data projected onto a plane, it can still be used to assess the practitioner's movements.

[0063] In this embodiment, the motion signal acquisition unit 11 is also used to correct the linear acceleration signal calculated from the video signal. The specific correction method has been described above and will not be repeated here.

[0064] The respiratory signal acquisition unit 12 is used to acquire the respiratory signals of practitioners of rehabilitation exercises. Since rehabilitation exercises typically consist of a series of continuous body movements, the practice status can be characterized by detecting signals generated by these movements over a period of time. When a person breathes, in addition to changes in breath volume, the chest and abdominal cavities also change. Therefore, there are two methods for detecting respiratory signals: one is by detecting the rise and fall of the chest and abdominal cavities, and the other is by detecting the airflow. For the former, pressure signals can be measured using impedance breathing belts or piezoelectric sensors worn on the chest and abdomen to represent the respiratory signal. These signals can roughly reflect the strength and periodicity of breathing, but with slightly lower accuracy. For the latter, since the airflow changes directly reflect breathing, such as by using an airflow sensor to detect the airflow velocity, the respiratory state can be reflected more accurately and directly. The function of the respiratory signal acquisition unit 12 is to receive respiratory sensor signals representing human respiration and use them directly as respiratory signals, or to preprocess them through filtering or other methods before using them as respiratory signals.

[0065] The individual characteristic data acquisition unit 13 is used to directly receive user input or imported individual characteristic information. It may include a storage subunit, an input / output interface, and related input / output interfaces. As an example, users can input relevant individual characteristic information using input devices such as a keyboard and mouse, or directly import batches of individual characteristic information via a USB flash drive connected through a USB interface. The individual characteristic information or individual characteristic data in this invention refers to the information of the practitioner of the rehabilitation training exercises, including age, gender, and medical condition.

[0066] Next, we will refer to Figure 2 Let's introduce the data processing module 2 of the first embodiment.

[0067] like Figure 2 As shown, the data processing unit 2 includes a motion smoothness calculation unit 21, a breath smoothness calculation unit 22, and a motion-breath coordination calculation unit 23. The motion smoothness calculation unit 21 and the breath smoothness calculation unit 22 receive data from the motion signal acquisition unit 11 and the breathing signal acquisition unit 12, respectively, while the motion-breath coordination calculation unit 23 receives data from both the motion signal acquisition unit 11 and the breathing signal acquisition unit 12 simultaneously.

[0068] As its name suggests, the motion smoothness calculation unit 21 is used to calculate "motion smoothness". In this embodiment, in order to reflect the completion of the overall motion, the motion signal acquisition unit 11 obtains motion signals from multiple different parts of the human body, and the motion smoothness calculation unit 21 can calculate the "motion smoothness" of the motions of multiple different parts of the human body based on these signals. As mentioned above, although the present invention is not limited to the method of calculating the "motion smoothness" index, it can be any index that reflects the smoothness of the motion state of a single motion or the smoothness of the transition between motion states of different motions. However, this embodiment preferably uses the rate of change of acceleration (rate of change of linear acceleration) as the index of "motion smoothness". Therefore, the motion signal acquisition unit 11 should at least obtain the linear acceleration signals of multiple parts of the human body as motion signals, and the motion smoothness calculation unit 21 differentiates the linear acceleration signals of these multiple parts of the human body to obtain the rate of change of acceleration. For example, after the motion signal acquisition unit 11 acquires the linear acceleration of the nose, left corner of the eye, right corner of the eye, left shoulder, left elbow, left wrist, right shoulder, right elbow, right wrist, left hip, left knee, left ankle, right hip, right knee, and right ankle, the motion smoothness calculation unit 21 calculates the rate of change of acceleration of the nose, left corner of the eye, right corner of the eye, left shoulder, left elbow, left wrist, right shoulder, right elbow, right wrist, left hip, left knee, left ankle, right hip, right knee, and right ankle.

[0069] The breath smoothness calculation unit 22 is used to calculate "breath smoothness". As mentioned earlier, this invention combines the traditional requirements of "deep, long, even, and fine" breath in qigong, defining "breath smoothness" as the stability of the breathing airflow. Therefore, any indicator that represents the stability of breath over time, the stability of breath rhythm, or the smoothness of transitions between breaths can be used as an indicator of "breath smoothness". In this first embodiment, we use either the coefficient of variation of respiratory rate (CV_RR) or the rate of change of airflow to represent "breath smoothness". When using a piezoelectric sensor to detect the breathing signal, the coefficient of variation of respiratory rate is more suitable as "breath smoothness", while when using an airflow sensor, the rate of change of airflow is more suitable. The calculation of the above two indicators has been explained above and will not be repeated here.

[0070] In addition, in other embodiments, the respiratory signal may be processed in other ways to reflect the smoothness of breathing. For example, indicators related to pauses between respiratory cycles, such as the average pause duration and the number of pauses less than a predetermined duration threshold.

[0071] Another key indicator in this embodiment of the invention is "action-breath coordination," which is calculated by the action-breath coordination calculation unit 23 based on the comparison of action signals and breathing signals. Specifically, the action-breath coordination calculation unit 23 calculates the phase difference between the key nodes of each breathing signal and the key nodes of its corresponding action signal to represent the "action-breath coordination." Since the action signal may represent multiple action signals from multiple parts of the human body, the "action-breath coordination" can be calculated by comparing any one, any number, or all of the multiple action signals with the breathing signal at key nodes to calculate the average value of the phase difference, and then averaging the average values ​​of the phase differences of the action signals to calculate the total average value of the phase difference, which is used as the "action-breath coordination." The types of action signals, the selection of key nodes, and the omission of invalid nodes have been explained above.

[0072] The following continues to refer to... Figure 2 The intelligent evaluation module 3 of the first embodiment of the present invention will be described below.

[0073] like Figure 2As shown, the intelligent evaluation module 3 mainly includes a machine learning model 31. The machine learning model is used to predict the rehabilitation efficacy of the exercise based on individual characteristic data, movement fluency, breathing smoothness, and movement-breath coordination, and to provide a rehabilitation efficacy score. In a typical scenario, when a practitioner of Baduanjin (Eight Pieces of Brocade) inputs their individual characteristics and video signals generated during a set of Baduanjin exercises into the system, the data acquisition module 1 and data processing module 2, along with the calculations by the machine learning model 31, can provide a rehabilitation efficacy score for the exercise. The rehabilitation efficacy score is, for example, out of 10, with a minimum of 1 and a maximum of 10.

[0074] Furthermore, the machine learning model 31 can also calculate the influence of each input indicator on the rehabilitation efficacy score.

[0075] For example, in this first embodiment, it is assumed that a gradient boosting tree (XGBoost) model is used, and the input of the model is, for example, individual characteristic information of a patient, a video of practicing Baduanjin (Eight Pieces of Brocade), and respiratory airflow velocity signal.

[0076] Individual characteristics include:

[0077] Age: 65 years old;

[0078] Disease type: Stroke;

[0079] Duration of illness: 6 months;

[0080] Rehabilitation training plan: Baduanjin (Eight Pieces of Brocade), twice a day;

[0081] Rehabilitation goal: Recovery of upper limb function.

[0082] The video signal and respiratory airflow velocity signal were processed and calculated to obtain:

[0083] Root mean square of the rate of change of linear acceleration of the left arm (movement smoothness): 45 m / s² 3

[0084] Root mean square of the rate of change of linear acceleration of the right arm (movement smoothness): 40 m / s² 3

[0085] Rate of change of respiratory airflow (breathing smoothness): 8652.2 mL / s 2

[0086] Mean phase difference between left arm movement and breath (movement-breath coordination): 0.85s

[0087] Mean phase difference of breath during right arm movement (coordination of breath): 0.76s

[0088] The model outputs a rehabilitation efficacy score of 72 points.

[0089] The SHAP values ​​of each feature calculated by the model are as follows:

[0090] feature SHAP value age 0.1 Disease type 0.25 Course of disease 0.15 Rehabilitation training plan (frequency) -0.2 Root mean square of the rate of change of linear acceleration of the left arm (movement smoothness) 0.26 Root mean square of the rate of change of linear acceleration of the right arm (movement smoothness) 0.18 Rate of change of respiratory airflow velocity (smoothness of breathing) -0.6 Mean phase difference of breath during left arm movement (coordination of breath during movement) 0.15 Mean phase difference of breath during right arm movement (coordination of breath during movement) 0.14

[0091] The rehabilitation utility score and SHAP values ​​of each input feature in this example show that the Baduanjin exercise had a moderately high effect on rehabilitation, but there is still room for improvement. Relatively speaking, the root mean square of the rate of change of linear acceleration in the left arm had the greatest impact on rehabilitation utility, which suggests that patients should pay more attention to the smoothness of left arm movements.

[0092] As mentioned above, machine learning model 31 is used in the implementation of this invention. This invention can select an appropriate model type based on the actual results, such as classification models like SVM, neural network models, etc. After establishing machine model 31, it is necessary to train it. The following refers to... Figure 2 This section describes the module architecture for training the machine learning model in the intelligent evaluation system of the present invention.

[0093] Figure 3 This is a functional module architecture diagram of the intelligent evaluation system for rehabilitation training exercises according to the first embodiment of the present invention during model training. (See diagram below.) Figure 3 As shown, during training, with Figure 2 Compared to the previous architecture, the system also needs to introduce an expert scoring module 4. Expert scoring module 4 is used to provide the machine learning model with rehabilitation utility scores given by experts, which serve as output label data in the training dataset. Furthermore, in Figure 3 In the architecture, the data acquisition module 1 collects historical data, while the data processing module 2 processes the historical data collected by the data acquisition module 1 to obtain the historical data's motion smoothness, breathing smoothness, and motion-breath coordination.

[0094] Figure 4 This is a functional module architecture diagram of the expert scoring module of the intelligent evaluation system for rehabilitation training exercises according to the first embodiment of the present invention. Figure 4As shown, the expert scoring module 4 includes a data collection and presentation unit 41 and an expert scoring unit 42. The expert scoring module 4 is primarily used to facilitate expert scoring. The data collection and presentation unit 41 presents the input data from the model training dataset for expert review. Typically, it displays individual characteristic data from the historical dataset, as well as indicators of movement fluency, breathing smoothness, and movement-breath coordination output by the data processing module 2, using lists or graphical representations. The expert scoring unit 42 receives input from experts to obtain their rehabilitation efficacy scores. Optionally, the expert scoring module can also accept expert interpretations of the scoring process for filtering training data during model training.

[0095] Figure 5 This is a functional module architecture diagram of the intelligent assessment system for rehabilitation training exercises according to the second embodiment of the present invention. The second embodiment is an improvement based on the first embodiment, such as... Figure 4 As shown, compared to the first embodiment, the data processing module 2 further includes an action standard calculation unit, which processes the action signal to obtain the action standard. In other words, this invention can expand upon the basic scheme with more data in other dimensions to make the model's input and output more complete.

[0096] In this second embodiment, "movement standardization" refers to the degree of matching between the movements performed by the practitioner in the rehabilitation exercises and the standard movements. "Movement standardization" can be achieved using existing technologies. For example, it can be represented by calculating the key point differences between the user and the standard movements from video signals, as described in Chinese patent application number 202510253970.0. Alternatively, it can employ the method described in Chinese patent application number 202411709763.3, "A Method and System for Evaluating Eight-Section Brocade Movements Based on Dynamic Programming and Deep Learning," which extracts skeletal points from video data and calculates the distance between the user and the standard, etc.

[0097] In addition, the present invention can be extended based on the first or second embodiment to add more dimensions of data indicators. For example, respiratory signals and motion signals themselves can also be used as indicator data input to the model, and so on.

[0098] In summary, this invention combines the key principles and requirements of traditional health-preserving exercises with modern artificial intelligence technology. It integrates the concepts and knowledge of Traditional Chinese Medicine and traditional health-preserving exercises, enabling the scientific quantification of the quality and effectiveness of practitioners' rehabilitation exercises. This allows for effective guidance for adjustments and promotes recovery. The key to traditional health-preserving exercises such as Five Animal Frolics, Eight Pieces of Brocade, and Tai Chi lies in the unity of breathing, movement, and intention; smooth movements; and deep, long, even, and fine breathing. This invention, based on a deep understanding of these concepts, scientifically quantifies the quality of rehabilitation exercises and uses artificial intelligence technology to intelligently evaluate the actual rehabilitation effects of the exercises and the influence of various quality indicators on the rehabilitation efficacy. Therefore, it can provide targeted guidance for practitioners to improve their rehabilitation exercises even without the presence of a Traditional Chinese Medicine expert, thereby enhancing the effectiveness of rehabilitation practice.

[0099] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent evaluation system for rehabilitation exercise, for evaluating the completion quality and rehabilitation effect of a practitioner performing a rehabilitation exercise, characterized in that, The method comprises the following steps: The data acquisition module is used for collecting individual characteristic data of the exerciser and the breathing signal and the action signal of the exerciser in the exercise process of the rehabilitation training exercise; The data processing module is used for calculating the breathing signal and the action signal collected by the data acquisition module to obtain the completion quality index of the rehabilitation training exercise of the exerciser; The intelligent evaluation module is used for establishing a machine learning model and using the machine learning model to estimate the rehabilitation utility score of the rehabilitation training exercise according to the individual characteristic data and the completion quality index, wherein the rehabilitation utility refers to the effective degree of the rehabilitation training exercise finally played for rehabilitation; The machine learning model is trained using historical data, wherein the historical data comprises historical individual characteristic data, a completion quality index of the rehabilitation training exercise and a rehabilitation utility score determined by an expert; the completion quality index comprises action fluency, breath fluency and action-breath coordination information; the action fluency comprises the smoothness of the motion state of a body part in a single action and the smoothness of the motion state conversion of a body part between different actions; the breath fluency refers to the smoothness of the breathing airflow, which comprises the smoothness of the airflow in a single exhalation phase or inhalation phase, the stability of the rhythm of multiple exhalation phases or inhalation phases, and the time stability of the pause gap between adjacent two breathing phases; the action-breath coordination degree refers to the consistency of the rhythm of the action and the breath at a key time node; the action fluency is represented by the acceleration change rate of the human body part, the breath fluency is represented by the breathing frequency variation coefficient or the breathing airflow flow rate change rate, and the action-breath coordination degree is represented by the phase difference of the key nodes of the acceleration signal and the breathing signal of the human body part.

2. The intelligent evaluation system for rehabilitation training exercises according to claim 1, characterized in that, The action signal is the linear acceleration signal of the human body part obtained by analyzing the video signal of the rehabilitation training exercise.

3. The intelligent evaluation system for rehabilitation training exercises according to claim 2, characterized in that, The intelligent evaluation module is further used for calculating the influence degree of each input feature affecting the rehabilitation utility score.

4. The intelligent evaluation system for rehabilitation training exercises according to claim 3, characterized in that, The influence degree is the SHAP value of each input feature.

5. The intelligent evaluation system for rehabilitation training according to claim 1, characterized in that, The method further comprises an expert score module, which is used for providing the rehabilitation utility score given by the expert to the machine learning model as output label data in the training data set.

6. An intelligent evaluation method of rehabilitation exercise, for evaluating the completion quality and rehabilitation effect of a practitioner performing a rehabilitation exercise, characterized in that, The method comprises the following steps: Collecting individual characteristic data of the exerciser and the breathing signal and the action signal of the exerciser in the exercise process of the rehabilitation training exercise; The collected respiratory signals and action signals are calculated to obtain a completion quality index of the exerciser performing the rehabilitation training exercise, the completion quality index including action fluency, breath fluency and action-breath coordination information: the action fluency includes the smoothness of the motion state of the body part in a single action, and also includes the smoothness of the motion state conversion of the body part between different actions; the breath fluency refers to the smoothness of the respiratory airflow, including the smoothness of the airflow in a single exhalation phase or inhalation phase, also including the stability of the rhythm of multiple exhalation phases or inhalation phases, and further including the time stability of the pause gap between adjacent two breathing phases; the action-breath coordination refers to the consistency of the rhythm of the action and the breath at the key time nodes; the action fluency is represented by the acceleration change rate of the human body part, the breath fluency is represented by the respiratory frequency variation coefficient or the respiratory airflow flow rate change rate, and the action-breath coordination is represented by the phase difference of the key nodes of the acceleration signal and the respiratory signal of the human body part; A machine learning model is established and used to estimate the rehabilitation utility score of the rehabilitation training exercise according to the individual characteristic data and the completion quality index, wherein the rehabilitation utility refers to the effective degree of the rehabilitation training exercise finally played for rehabilitation, and the machine learning model is trained using historical data, the historical data including historical individual characteristic data, completion quality index of the rehabilitation training exercise and rehabilitation utility score determined by an expert.

Images