A method for recognizing lower limb rehabilitation postures to prevent VTE and a smart ankle bracelet

Abstract

This invention relates to the field of lower limb rehabilitation data recognition and monitoring technology, specifically disclosing a method for recognizing lower limb rehabilitation movement postures and an intelligent ankle bracelet for preventing VTE, comprising: Step 1, real-time acquisition of three-dimensional spatial motion data and recognition of lower limb motion data, extracting kinematic parameter features; Step 2, identification of movement types based on kinematic parameter features, acquisition of movement pattern consistency scores, and determination of whether rehabilitation training is qualified; Step 3, acquisition of surface electromyographic signals of the rehabilitation participant's leg-raising force exertion movement during rehabilitation training and simultaneous acquisition of lower limb motion data, extraction of force exertion data, evaluation and analysis to generate a muscle activation efficiency score; Step 4, establishment of a mapping model based on historical hemodynamic data, inputting the extracted force exertion data into the mapping model to output a VTE prevention efficiency score; Step 5, calculation of the movement pattern consistency score, muscle activation efficiency score, and VTE prevention efficiency score through movement evaluation analysis to generate a movement evaluation quality coefficient.

Description

Technical Field

[0001] This invention relates to the field of lower limb rehabilitation data recognition and monitoring technology, specifically to a method for recognizing lower limb rehabilitation movement postures to prevent VTE and a smart ankle bracelet. Background Technology

[0002] Venous thromboembolism (VTE) is one of the most common serious complications after major orthopedic surgery, mainly including deep vein thrombosis (DVT) and pulmonary thromboembolism (pulmonary thromboembolism). Major orthopedic surgery (such as hip and knee replacement) is a very high-risk factor for VTE. In clinical prevention, in addition to pharmacological and physical prevention (such as compression stockings and intermittent pneumatic compression devices), early and standardized ankle pump exercises and quadriceps strengthening exercises are very important basic preventive measures. These movements can effectively utilize muscle contraction and relaxation to compress veins and promote blood return, thereby preventing thrombosis.

[0003] However, in current clinical rehabilitation practice, many post-orthopedic patients fail to complete the prescribed movements according to the frequency and quality instructions due to pain, fatigue, or insufficient understanding. Furthermore, in rehabilitation centers, the assessment of rehabilitation movements currently relies mainly on manual observation and verbal correction by medical staff, lacking objective, continuous, or quantitative monitoring methods. This makes it impossible for medical staff to know whether patients have accurately and regularly completed the flexion and extension range and frequency of ankle pump exercises without supervision, and consequently, it is impossible to determine whether the quadriceps muscles have achieved effective muscle exertion intensity during exercise, resulting in poor rehabilitation supervision. In addition, existing wearable devices mostly focus on recording macroscopic exercise volume such as steps and activity time. They lack specific identification settings for detailed identification and effect evaluation of specific movements for VTE prevention, and are even less able to establish a quantitative correlation between movement commands and hemodynamic improvement effects. Summary of the Invention

[0004] The purpose of this invention is to provide a method for recognizing lower limb rehabilitation postures and a smart ankle bracelet to prevent VTE, thereby solving the following technical problems: How can we achieve objective and real-time, refined monitoring of postoperative rehabilitation posture and movement assessment in orthopedic patients to improve the level of intelligent VTE prevention care and the efficiency of patient rehabilitation?

[0005] The objective of this invention can be achieved through the following technical solutions: A method for recognizing lower limb rehabilitation postures to prevent VTE, comprising: Step 1: Real-time acquisition of three-dimensional spatial motion data, identification of lower limb motion data based on the three-dimensional spatial motion data, and extraction of kinematic parameter features, including ankle flexion and extension features and leg lifting force features; Step 2: Identify the movement type based on kinematic parameter features, obtain the movement pattern consistency score, and determine whether the rehabilitation training is qualified based on the movement pattern consistency score. If yes, proceed to Step 3; otherwise, adjust the unqualified movement pattern consistency score and then feed it back to Step 1. Step 3: Obtain surface electromyographic signals of the rehabilitation participant's leg-raising force exertion during rehabilitation training and simultaneously collect lower limb motion data, extract force exertion data, evaluate and analyze to generate a muscle activation efficiency score; Step 4: Establish a mapping model based on historical hemodynamic data, input the extracted force data into the mapping model, and output the VTE prevention effectiveness score. Step 5: Calculate the motion assessment quality coefficient by performing motion pattern consistency score, muscle activation efficiency score, and VTE prevention effectiveness score.

[0006] Preferably, the method for extracting kinematic parameter features based on three-dimensional spatial motion data to identify lower limb motion data in step one is as follows: S101. Real-time acquisition and preprocessing to determine the triaxial acceleration and triaxial angular velocity at the ankle; S102. The preprocessed gyroscope angular velocity signal is numerically integrated, and the accelerometer signal is fused and corrected using a complementary filtering algorithm to obtain the real-time three-dimensional dynamic angle change curve of the ankle. S103. Based on the real-time three-dimensional dynamic angle change curve, the gravitational acceleration component of the acceleration signal is separated to obtain the vertical linear acceleration, and the integral term obtained by the second numerical integration is the real-time height displacement curve of the leg lift. S104. Collect the peak points of the real-time three-dimensional dynamic angle change curve, segment the motion cycle, extract and determine the peak values ​​of dorsiflexion and plantarflexion angles, motion amplitude, flexion and extension frequency and maximum angular velocity of each motion cycle, and form an ankle flexion and extension feature set. S105. Identify the acceleration mutation point on the real-time height displacement curve of the leg lift to determine the starting time of the leg lift, and extract the maximum height of the leg lift, the duration of isometric contraction and the rapid start index of force exertion to form a feature set of leg lift force exertion.

[0007] Preferably, the method for obtaining the action pattern consistency score in step two specifically includes: S201. Retrieve the standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template from the pre-stored rehabilitation history movement template library. The standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template contain the standard kinematic parameter feature sequence of their respective movements. S202. The real-time extracted ankle flexion-extension feature set and leg lift force feature set are dynamically time-aligned with the standard ankle pump motion template and the standard quadriceps isometric contraction leg lift template, and the similarity distance of the feature sequences is calculated; the similarity includes amplitude similarity, motion trajectory matching similarity and motion frequency similarity; S203. Calculate the three indicators respectively: amplitude similarity, motion trajectory matching similarity and motion frequency similarity, and assign a preset weight to each indicator; S204. Calculate the consistency score of the action pattern based on the weighted sum of the three indicators.

[0008] Preferably, step two, which determines whether the rehabilitation training is qualified based on the consistency score of movement patterns, includes: The consistency score of the action pattern is compared with the system's preset grading thresholds, which include qualified thresholds and critical thresholds. If the consistency score of the action pattern is greater than or equal to the passing threshold, the action is deemed to be qualified and proceeds to step three. If the consistency score of the action pattern is less than the qualified threshold but greater than the critical threshold, it is judged as a deformed action, triggering a feedback reminder but not proceeding to step three. If the consistency score of movement pattern continues to be below the critical threshold, it is determined that the rehabilitation training compliance has been interrupted, and the anti-forgetting reminder is triggered.

[0009] Preferably, the method for adjusting the consistency score of non-compliant action patterns in step two is as follows: S205. Analyze the real-time kinematic parameter characteristics of non-compliant actions, perform item-by-item deviation analysis with the standard template, and automatically identify non-compliant items; S206. Match the non-conformities with the intervention strategy library to generate corrective action guidance instructions; S207. Receive corrective action guidance instructions and provide real-time corrective guidance feedback to rehabilitation personnel; S208. After the rehabilitation personnel adjust their movements based on the feedback from the correction guidance, repeat steps one to two until the consistency score of the movement pattern reaches the qualified threshold.

[0010] Preferably, the specific method for extracting force data in step three is as follows: S301. Preprocess the electromyography signal on the action surface to remove power frequency interference and motion artifacts, and calculate the time-domain characteristics of the root mean square value and the time-domain characteristics of the integral electromyography value. S302. Simultaneously calibrate the starting point of lower limb motion data and establish a precise correspondence between electromyographic characteristics and kinematic characteristics on the time axis; S303. A muscle force estimation model is constructed using a bidirectional long short-term memory network. The real-time joint torque of the quadriceps femoris is estimated by taking the temporal characteristics of the root mean square value and the temporal characteristics of the integral electromyography value, as well as the lower limb motion data at the corresponding time, as input. The real-time joint torque is then used as the force data.

[0011] Preferably, the method for evaluating and analyzing the generated muscle activation efficiency score is as follows: The reference value of the maximum isometric voluntary muscle strength of rehabilitation personnel was used as the benchmark. Calculate the ratio of real-time joint torque to the reference torque to obtain the real-time force percentage; Based on medical rehabilitation information, preset the target exertion range and the optimal exertion range; If the real-time force exertion percentage falls within the optimal force exertion range, then the percentage will receive full marks, which will be used as a muscle activation efficiency score. If the muscle falls within the target force exertion zone, a linear interpolation score is performed based on the percentage distance from the optimal force exertion zone, which is used as the muscle activation efficiency score. If the value is lower than the minimum value of the target force range, a percentage-based decay score is applied as the muscle activation efficiency score.

[0012] Preferably, the process of establishing the mapping model in step four is as follows: Peak torque, torque-time integral, and force exertion duration are extracted from the force exertion data as predictive factors. Using predictive factors as input and the percentage increase in peak blood flow velocity as output, a recurrent neural network model is trained. The output percentage increase in peak blood flow velocity is then converted into an effective VTE prevention score using an evidence-based curve of the probability of thrombosis prevention.

[0013] Preferably, the fusion calculation method for generating the motion evaluation quality coefficient in step five is as follows: The formula for calculating the motion evaluation quality coefficient is as follows:

[0014] in, The quality coefficient for motion evaluation. The score represents the consistency of action patterns. The score represents the muscle activation efficiency. To prevent VTE from being effectively evaluated; , , These are the weight coefficients determined by principal component analysis based on clinical sample data in medical experience, and... .

[0015] A smart ankle bracelet is used to implement the aforementioned method for recognizing lower limb rehabilitation movements to prevent VTE. The smart ankle bracelet includes: The sensing module is used to simultaneously acquire motion data and surface electromyography signals. The sensing module includes a multi-axis inertial measurement unit and a flexible electrode array. The main control module, with its built-in microprocessor and hardware accelerator, is used to execute the feature extraction, pattern recognition, and action evaluation quality coefficient fusion algorithms in steps one through five. The specific steps include: Step 1: Identify lower limb motion data based on three-dimensional spatial motion data and extract kinematic parameter features, including ankle flexion and extension features and leg lifting force features; Step 2: Identify the movement type based on kinematic parameter features, obtain the movement pattern consistency score, and determine whether the rehabilitation training is qualified based on the movement pattern consistency score. If yes, proceed to Step 3; otherwise, return to Step 1. Step 3: Obtain surface electromyographic signals of the rehabilitation participant's leg-raising force exertion during rehabilitation training and simultaneously collect lower limb motion data, extract force exertion data, and evaluate and generate a muscle activation efficiency score. Step 4: Establish a mapping model based on historical hemodynamic data, input the extracted force data into the mapping model, and output the VTE prevention effectiveness score. Step 5: Calculate and generate a motion assessment quality coefficient by performing motion pattern consistency score, muscle activation efficiency score, and VTE prevention effectiveness score on motion assessment analysis. The feedback module provides graded tactile feedback and visual cues. The feedback module includes a miniature linear vibration motor and multi-color LED indicators. The communication module is used for data synchronization with the mobile terminal; The power module is used to provide power to the communication module.

[0016] The beneficial effects of this invention are: This invention achieves accurate extraction of rehabilitation movement features from lower limb movement data in three-dimensional spatial motion data through step one, including simultaneous extraction of ankle flexion and extension feature sets and leg lifting force feature sets; through step two, the invention calculates the movement pattern consistency score to dynamically compare the movement amplitude, movement trajectory, and movement frequency with a standard template, realizing the objective quantification of movement quality; through graded threshold judgment and automatic correction adjustment, further feedback is provided based on the qualified movement pattern consistency score, effectively solving the problem of patients' rehabilitation movements not meeting the standards and lacking real-time guidance, significantly improving rehabilitation training compliance and movement achievement rate.

[0017] (2) This invention solves the problem that it is impossible to judge whether the muscle has reached the effective force intensity based solely on the movement posture by synchronously collecting and aligning the surface electromyography signal and kinematic data in step three, and by using a bidirectional long short-term memory network to estimate the real-time joint torque of the quadriceps femoris. This achieves an objective quantitative evaluation of muscle activation efficiency. In step four, the recurrent neural network mapping model is constructed to directly correlate the force data with the blood flow velocity improvement index, and converts it into a VTE prevention efficiency score through the evidence-based relationship curve. This is the first time that a quantitative mapping from the mechanical parameters of rehabilitation movements to the VTE prevention effect has been realized, extending the movement assessment to the clinical endpoint index, and making the assessment results more clinically instructive. (3) In step five, the present invention weights and integrates the scores of three dimensions: consistency of movement pattern, muscle activation efficiency and VTE prevention efficiency, to generate a single movement assessment quality coefficient. This comprehensively reflects the overall quality of rehabilitation movements in terms of standardization, force exertion effectiveness and thrombosis prevention effect, providing medical staff with a simple and intuitive quantitative assessment indicator, which is convenient for remote monitoring and individualized rehabilitation program adjustment.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a step diagram of a lower limb rehabilitation exercise posture recognition method for preventing VTE according to the present invention; Figure 2 This is a step diagram of the method for extracting kinematic parameter features in step one of the present invention; Figure 3 This is a step-by-step diagram illustrating the method for obtaining the action pattern consistency score in step two of this method. Figure 4 This is a step diagram illustrating the method of adjusting the consistency score of non-compliant action patterns in step two of this invention; Figure 5 This is a step diagram illustrating the method for extracting force data in step three of this invention; Figure 6 This is a block diagram of a smart ankle bracelet structure according to the present invention; Figure 7 This is a detailed diagram of the structure of a smart anklet according to the present invention. Detailed Implementation

[0021] 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 As shown, this invention is a method for recognizing lower limb rehabilitation postures to prevent VTE, comprising: Step 1: Real-time acquisition of three-dimensional spatial motion data, identification of lower limb motion data based on the three-dimensional spatial motion data, and extraction of kinematic parameter features, including ankle flexion and extension features and leg lifting force features; Step 2: Identify the movement type based on kinematic parameter features, obtain the movement pattern consistency score, and determine whether the rehabilitation training is qualified based on the movement pattern consistency score. If yes, proceed to Step 3; otherwise, adjust the unqualified movement pattern consistency score and then feed it back to Step 1. Step 3: Obtain surface electromyographic signals of the rehabilitation participant's leg-raising force exertion during rehabilitation training and simultaneously collect lower limb motion data, extract force exertion data, evaluate and analyze to generate a muscle activation efficiency score; Step 4: Establish a mapping model based on historical hemodynamic data, input the extracted force data into the mapping model, and output the VTE prevention effectiveness score. Step 5: Calculate the motion assessment quality coefficient by performing motion pattern consistency score, muscle activation efficiency score, and VTE prevention effectiveness score.

[0023] In the above technical solution, firstly, during the signal acquisition process in step one, the acquisition target is the rehabilitation personnel, and three-dimensional spatial motion data is collected in real time at the ankle where the ankle band is worn, by setting a sampling frequency. This three-dimensional spatial motion data includes three-axis acceleration signals and three-axis angular velocity signals as raw signals, and the raw signals are processed by ground-pass filtering to remove high-frequency noise and muscle tremor interference. At the same time, the gyroscope is statically zero-biased calibrated to eliminate drift error. The processed signals are then subjected to attitude fusion through a complementary filtering algorithm, and numerical integration is used to determine the real-time three-dimensional dynamic angle change curves of the ankle in the sagittal, coronal, and horizontal planes. The kinematic parameter characteristics of the lower limb motion data are obtained through gravity separation and linear motion extraction, and these kinematic parameter characteristics include ankle pump motion characteristics and leg lifting force characteristics. Then, step two identifies the movement type based on kinematic parameter features and obtains the movement pattern consistency score. This is achieved through movement type identification and pattern consistency assessment. Specifically, standard ankle pump exercise templates and standard quadriceps isometric contraction leg raise templates are retrieved from a pre-stored rehabilitation history movement template library. These templates are generated by integrating data from multiple rehabilitation participants who have achieved the rehabilitation goals, using the DTW centroid averaging algorithm, and include standard kinematic parameter feature sequences. The real-time extracted features and corresponding standard templates are dynamically time-normalized to solve for the shortest normalized path. The cumulative distance is calculated, and based on this cumulative distance, three indicators—amplitude similarity, movement trajectory matching similarity, and movement strategy similarity—are calculated. These three indicators are then weighted and summed to obtain the movement pattern consistency score. After determining the consistency score of the movement pattern, the further evaluation of the movement qualification is performed based on the qualification score. Whether the correction adjustment needs to be triggered is determined by the qualification score. For qualified movements, the data related to the leg raising movement is further determined. For unqualified information, correction is performed. The correction is performed by analyzing the deviation between the real-time kinematic features and the standard template and comparing each item. The corresponding correction voice command is matched in the preset intervention strategy library, and the visual movement guidance is provided through Bluetooth or mobile device. The ankle band emits corresponding regular vibrations as a reminder. After the rehabilitation personnel complete the correction adjustment according to the instructions, the data collection and evaluation in step one are repeated. Next, in step three, the surface electromyography (EMG) signals of the leg-raising force exertion of the rehabilitation personnel with qualified movement patterns are obtained. Unlike directly collecting the leg-raising force exertion information, the surface EMG signals of the leg-raising force exertion are obtained and the lower limb movement data is extracted from the preprocessed EMG signals. The baseline value of the rehabilitation personnel in the preoperative assessment is determined to calculate the real-time force exertion percentage and determine the muscle activation efficiency score. Furthermore, in step four, a recurrent neural network mapping model is constructed based on historical clinical databases. The model training process includes collecting clinical hemodynamic data, then training the model to ensure that the output information and the antithrombotic probability form a certain mathematical mapping relationship, and then constructing a mapping model. By taking the predictor as input and the percentage increase in the peak value of popliteal vein blood flow velocity measured by Doppler ultrasound as output, a recurrent neural network mapping model is constructed. The percentage increase in the peak value of popliteal vein blood flow velocity measured by Doppler ultrasound is substituted into the preset antithrombotic probability mapping function, and the value within the normalized range is the VTE prevention effectiveness score. Finally, in step five, the three scores obtained in steps two through four are weighted and fused together to generate a value that is set as the action assessment quality coefficient. The magnitude of this action assessment quality coefficient comprehensively reflects the overall quality of the rehabilitation training of the patient in the current session. Furthermore, by displaying this value on a mobile interface and simultaneously displaying it on the medical system monitoring terminal, the medical staff of the rehabilitation center can remotely monitor the patient's rehabilitation progress and the prevention effect of VTE based on the real-time situation.

[0024] Please see Figure 2 As shown, in one embodiment of the present invention, the method for identifying lower limb motion data and extracting kinematic parameter features based on three-dimensional spatial motion data in step one is as follows: S101. Real-time acquisition and preprocessing to determine the triaxial acceleration and triaxial angular velocity at the ankle; S102. The preprocessed gyroscope angular velocity signal is numerically integrated, and the accelerometer signal is fused and corrected using a complementary filtering algorithm to obtain the real-time three-dimensional dynamic angle change curve of the ankle. S103. Based on the real-time three-dimensional dynamic angle change curve, the gravitational acceleration component of the acceleration signal is separated to obtain the vertical linear acceleration, and the integral term obtained by the second numerical integration is the real-time height displacement curve of the leg lift. S104. Collect the peak points of the real-time three-dimensional dynamic angle change curve, segment the motion cycle, extract and determine the peak values ​​of dorsiflexion and plantarflexion angles, motion amplitude, flexion and extension frequency and maximum angular velocity of each motion cycle, and form an ankle flexion and extension feature set. S105. Identify the acceleration mutation point on the real-time height displacement curve of the leg lift to determine the starting time of the leg lift, and extract the maximum height of the leg lift, the duration of isometric contraction and the rapid start index of force exertion to form a feature set of leg lift force exertion.

[0025] In the above technical solution, firstly, in step S101, after the patient wears the smart ankle bracelet, the IMU collects the triaxial acceleration signal at the ankle in real time at a sampling rate of 100Hz. , , ; and triaxial angular velocity signals , , The original signal is then subjected to a fourth-order Butterworth low-pass filter (cutoff frequency 20Hz) to remove high-frequency noise and muscle tremor interference; simultaneously, the gyroscope undergoes static zero-bias calibration to eliminate drift error; then, in step S102, a complementary filtering algorithm is used for attitude fusion, calculated as follows:

[0026] in, Trust coefficient The reference attitude angle calculated by the accelerometer is used to obtain the real-time three-dimensional dynamic angle change curves of the ankle in the sagittal, coronal and horizontal planes through numerical integration; Next, in step S103, the acceleration signal is rotated from the sensor coordinate system to the geographic coordinate system using the real-time attitude angle, and the standard gravitational acceleration is subtracted from the vertical axis of the geographic coordinate system to obtain the linear acceleration in the vertical direction. The real-time height displacement curve of the raised leg is obtained by a second integration. ; Furthermore, in step S104, peak points and zero-crossing points are detected on the sagittal angle curve to segment a single dorsiflexion-plantarflexion movement cycle. Ankle pump motion features are extracted within each cycle. The ankle pump motion features extracted within each cycle include: Peak backflex angle Peak plantar flexion angle The amplitude of motion is calculated by the absolute value of the difference between the two. ; flexion and extension frequency , The duration of the exercise cycle; Maximum angular velocity ; Finally, step S105 extracts the leg-lifting force characteristics to monitor the slope of the vertical linear acceleration signal and determines the moment when the slope exceeds a preset threshold, marking it as the start of the leg lift. Extracting from the height displacement curve: Maximum leg lift height ; and maintain a high altitude Over 80% of the duration is In fact, the instantaneous acceleration mutation is quantified as the force-initiating rapid index, and the above parameters constitute the leg-lifting force-generating feature set.

[0027] Please see Figure 3 As shown, in one embodiment of the present invention, the method for obtaining the action pattern consistency score in step two specifically includes: S201. Retrieve the standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template from the pre-stored rehabilitation history movement template library. The standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template contain the standard kinematic parameter feature sequence of their respective movements. S202. The real-time extracted ankle flexion-extension feature set and leg lift force feature set are dynamically time-aligned with the standard ankle pump motion template and the standard quadriceps isometric contraction leg lift template, and the similarity distance of the feature sequences is calculated; the similarity includes amplitude similarity, motion trajectory matching similarity and motion frequency similarity; S203. Calculate the three indicators respectively: amplitude similarity, motion trajectory matching similarity and motion frequency similarity, and assign a preset weight to each indicator; S204. Calculate the consistency score of the action pattern based on the weighted sum of the three indicators.

[0028] In the above technical solution, the calculated motion pattern consistency score reflects the degree of consistency between the current motion and the standard motion at the kinematic level. Specifically, firstly, step S201 retrieves the standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template from the pre-stored rehabilitation history motion template library; the templates are generated by collecting data from multiple rehabilitation patients who have achieved the rehabilitation goals, and are generated using the DTW centroid averaging algorithm, containing standard kinematic parameter feature sequences; then, step S202 performs dynamic time warping on the real-time extracted feature set and the corresponding standard template, solves the shortest warping path, and calculates the cumulative distance. ,in For real-time sequences, For the template sequence, the three indicators calculated are as follows: Amplitude similarity : ,in The peak value of the motion amplitude. The peak value of the motion amplitude in the real-time sequence. The peak value of the motion amplitude of the template sequence; similarity of motion trajectory matching. : Similarity of movement frequency ,in Frequency of movement; Step S203 assigns preset weights to each of the three indicators. ; ; ; S204. Calculate the movement pattern consistency score based on the weighted sum of the three indicators. The formula is: .

[0029] As one embodiment of the present invention, step two, which determines whether the rehabilitation training is qualified based on the consistency score of movement patterns, includes: The consistency score of the action pattern is compared with the system's preset grading thresholds, which include qualified thresholds and critical thresholds. If the consistency score of the action pattern is greater than or equal to the passing threshold, the action is deemed to be qualified and proceeds to step three. If the consistency score of the action pattern is less than the qualified threshold but greater than the critical threshold, it is judged as a deformed action, triggering a feedback reminder but not proceeding to step three. If the consistency score of the motion pattern continuously falls below the critical threshold, it is determined that the compliance of the rehabilitation training is interrupted, and an anti-forgetting reminder is triggered.

[0030] In the above technical solution, combined with the above consistency score of the motion pattern to perform a qualification judgment, and compare with the grading threshold, where the grading threshold includes, for example, a set passing threshold of 0.8 and a critical threshold of 0.5; judge ≥0.8, then it is judged as qualified and proceed to step three; 0.5 ≤ <0.8, it is determined that the motion is deformed, a vibration feedback reminder is triggered, and it is marked as to be corrected; <0.5 and continuously exceeds the preset time, it is determined that the compliance is terminated, and a forgetting reminder is triggered.

[0031] Please refer to Figure 4 As shown, as an embodiment of the present invention, the method for adjusting the unqualified consistency score of the motion pattern in step two is: S205. Analyze the real-time kinematic parameter characteristics of the unqualified motion, perform a deviation analysis item by item with the standard template, and automatically identify the unqualified items; S206. Match the unqualified items with the intervention strategy library to generate a correction motion guidance instruction; S207. Receive the correction motion guidance instruction and provide real-time correction guidance feedback to the rehabilitation personnel; S208. After the rehabilitation personnel adjust the motion according to the correction guidance feedback, repeat steps one to two until the consistency score of the motion pattern reaches the passing threshold.

[0032] In the above technical solution, according to the interpretation and adjustment of the unqualified motion in step two, common unqualified items include insufficient angular amplitude, too fast or too slow rhythm, and insufficient leg-lifting height; the following adjustment process is performed on the unqualified motion: Analyze the item-by-item deviation between the real-time kinematic characteristics and the standard template, such as identifying that "the peak dorsiflexion angle is 30% smaller"; match the corresponding correction instruction in the intervention strategy library, such as "Please forcefully hook the instep towards yourself"; push it to the mobile APP via Bluetooth for visual animation guidance, and at the same time, the ankle bracelet emits regular vibrations; after the rehabilitation personnel complete the adjustment according to the guidance, re-collect and evaluate, and repeat steps one to two until the consistency score of the motion pattern reaches the passing threshold, improve the feedback adjustment process, and ensure the smooth collection of other data items next.

[0033] Please refer to Figure 5 As shown, as an embodiment of the present invention, the specific method for extracting the force application data in step three is: S301. Preprocess the electromyography signal on the action surface to remove power frequency interference and motion artifacts, and calculate the time-domain characteristics of the root mean square value and the time-domain characteristics of the integral electromyography value. S302. Simultaneously calibrate the starting point of lower limb motion data and establish a precise correspondence between electromyographic characteristics and kinematic characteristics on the time axis; S303. A muscle force estimation model is constructed using a bidirectional long short-term memory network. The real-time joint torque of the quadriceps femoris is estimated by taking the temporal characteristics of the root mean square value and the temporal characteristics of the integral electromyography value, as well as the lower limb motion data at the corresponding time, as input. The real-time joint torque is then used as the force data.

[0034] In the above technical solution, after the action is qualified, the first step S301 is to acquire sEMG signals of the tibialis anterior muscle group and quadriceps femoris muscle region through a flexible electrode array, and then perform 50Hz notch filtering (to remove power frequency interference) and 20-450Hz bandpass filtering on the signal to calculate the time domain characteristics; the time domain characteristics include the root mean square value. Also, integrated electromyography (EMG) values. ; Then, step S302 uses the leg-raising start time (extracted in step S105) as the time anchor point to obtain the root mean square value and integral electromyography value sequence of the electromyography signal and aligns them on the time axis to determine the correspondence; that is, aligning the RMS and iEMG sequences of the electromyography signal with the linear acceleration and height displacement data on the time axis to establish an accurate correspondence. Next, step S303 involves constructing and deploying a bidirectional long short-term memory network model. The model input is a concatenated vector of time-aligned electromyographic and kinematic features, and the output is an estimated real-time quadriceps joint torque. The model is pre-trained using clinically collected sEMG and synchronous force plate data; the estimated joint torque is the force data.

[0035] As one embodiment of the present invention, the method for evaluating and analyzing the generated muscle activation efficiency score is as follows: The reference value of the maximum isometric voluntary muscle strength of rehabilitation personnel was used as the benchmark. Calculate the ratio of real-time joint torque to the reference torque to obtain the real-time force percentage; Based on medical rehabilitation information, preset the target exertion range and the optimal exertion range; If the real-time force exertion percentage falls within the optimal force exertion range, then the percentage will receive full marks, which will be used as a muscle activation efficiency score. If the muscle falls within the target force exertion zone, a linear interpolation score is performed based on the percentage distance from the optimal force exertion zone, which is used as the muscle activation efficiency score. If the value is lower than the minimum value of the target force range, a percentage-based decay score is applied as the muscle activation efficiency score.

[0036] The aforementioned technical solution specifically involves obtaining the baseline value of the maximum isometric voluntary contraction muscle strength measured during the preoperative assessment of the rehabilitation patient. Calculate the real-time force percentage : ; Set the optimal exertion zone according to medical guidelines and target power range Calculate the muscle activation efficiency score The maximum score is 1 point: ,but ; ≤ < ,but ; < ≤ ,but ; < or > : or (Decrease-rate rating).

[0037] In one embodiment of the present invention, the process of establishing the mapping model in step four is as follows: Peak torque, torque-time integral, and force exertion duration are extracted from the force exertion data as predictive factors. Using predictive factors as input and the percentage increase in peak blood flow velocity as output, a recurrent neural network model is trained. The output percentage increase in peak blood flow velocity is then converted into an effective VTE prevention score using an evidence-based curve of the probability of thrombosis prevention.

[0038] In the above technical solution, a recurrent neural network mapping model is constructed based on a historical clinical database to collect predictive factor data (peak torque) synchronously collected during rehabilitation training of orthopedic postoperative patients. Torque-time integral Duration of exertion (and the percentage increase in peak popliteal vein blood flow velocity measured by Doppler ultrasound) Using predictors as input, For output, train the GRU network. The forward propagation formula for a GRU unit is:

[0039]

[0040]

[0041]

[0042] The model output Substitute into the thrombosis prevention probability mapping function:

[0043] in The coefficients are derived from clinical evidence-based relationship curves. This is the VTE prevention effectiveness score, with a value range of [0,1). During rehabilitation training, predictive factors for force exertion data are extracted in real time, input into a trained GRU model, and output... Points.

[0044] As one embodiment of the present invention, the fusion calculation method for generating the motion evaluation quality coefficient in step five is as follows: The formula for calculating the motion evaluation quality coefficient is as follows:

[0045] in, The quality coefficient for motion evaluation. The score represents the consistency of action patterns. The score represents the muscle activation efficiency. To prevent VTE from being effectively evaluated; , , These are the weight coefficients determined by principal component analysis based on clinical sample data in medical experience, and... .

[0046] In the above technical solution, the three scores obtained in steps two to four are combined and calculated using a weighted formula. The weighting coefficients were determined from clinical sample data using principal component analysis, and the recommended values ​​are... , , ,satisfy The generated motion evaluation quality coefficient ∈ This comprehensively reflects the overall quality of the patient's current rehabilitation training, and the coefficient is displayed in real time on the mobile interface and synchronized to the nurse station monitoring terminal, making it convenient for medical staff to remotely monitor the rehabilitation progress and VTE prevention effect of the patients.

[0047] A smart ankle bracelet has been designed; please refer to [link / reference]. Figure 6-7 As shown, the smart ankle bracelet used to implement the above-mentioned method for recognizing lower limb rehabilitation movements to prevent VTE includes: The sensing module is used to simultaneously acquire motion data and surface electromyography signals. The sensing module includes a multi-axis inertial measurement unit and a flexible electrode array. The main control module, with its built-in microprocessor and hardware accelerator, is used to execute the feature extraction, pattern recognition, and action evaluation quality coefficient fusion algorithms in steps one through five. The specific steps include: Step 1: Identify lower limb motion data based on three-dimensional spatial motion data and extract kinematic parameter features, including ankle flexion and extension features and leg lifting force features; Step 2: Identify the movement type based on kinematic parameter features, obtain the movement pattern consistency score, and determine whether the rehabilitation training is qualified based on the movement pattern consistency score. If yes, proceed to Step 3; otherwise, return to Step 1. Step 3: Obtain surface electromyographic signals of the rehabilitation participant's leg-raising force exertion during rehabilitation training and simultaneously collect lower limb motion data, extract force exertion data, and evaluate and generate a muscle activation efficiency score. Step 4: Establish a mapping model based on historical hemodynamic data, input the extracted force data into the mapping model, and output the VTE prevention effectiveness score. Step 5: Calculate and generate a motion assessment quality coefficient by performing motion pattern consistency score, muscle activation efficiency score, and VTE prevention effectiveness score on motion assessment analysis. The feedback module provides graded tactile feedback and visual cues. The feedback module includes a miniature linear vibration motor and multi-color LED indicators. The communication module is used for data synchronization with the mobile terminal; The power module is used to provide power to the communication module.

[0048] In the above technical solution, the smart ankle bracelet adopts a modular structure design, including: The sensing module integrates a multi-axis inertial measurement unit (IMU, including a triaxial accelerometer and a triaxial gyroscope) and a flexible electrode array, used to synchronously acquire three-dimensional spatial motion data and surface electromyography (sEMG) signals, respectively; the main control module has a built-in ARM Cortex-M series microprocessor and hardware accelerator, responsible for core algorithm operations such as signal preprocessing, feature extraction, pattern recognition, and quality assessment coefficient fusion; the feedback module includes a miniature linear vibration motor and multi-color LED indicators, providing patients with graded tactile feedback and visual reminders; the communication module uses a low-power Bluetooth chip to achieve data synchronization with a mobile APP or nurse station terminal; and the power module contains a rechargeable lithium battery and power management circuitry to power all modules.

[0049] The smart ankle bracelet is encapsulated in an ergonomically designed curved waterproof shell and secured to the patient's ankle with a flexible, skin-friendly strap. The electrode array is attached to the skin area corresponding to the anterior tibial muscle group. The main control module summarizes the quality coefficient of all rehabilitation training movements of the patient on a daily (frequency) basis and calculates the daily average quality coefficient and the trend of total exercise volume. When the daily average quality coefficient or effective exercise duration is continuously lower than the warning threshold, the system automatically pushes a warning message to the nurse station, indicating that more intervention is needed.

[0050] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, and therefore described more simply; relevant parts can be referred to the descriptions of the method embodiments.

[0051] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this application, they should all fall within the protection scope of the present invention.

Claims

1. A method for recognizing lower limb rehabilitation postures to prevent VTE, characterized in that, The method includes: Step 1: Real-time acquisition of three-dimensional spatial motion data, identification of lower limb motion data based on the three-dimensional spatial motion data, and extraction of kinematic parameter features, including ankle flexion and extension features and leg lifting force features; Step 2: Identify the movement type based on the kinematic parameter features, obtain the movement pattern consistency score, and determine whether the rehabilitation training is qualified based on the movement pattern consistency score. If yes, proceed to Step 3; otherwise, adjust the unqualified movement pattern consistency score and then feed it back to Step 1. Step 3: Obtain the surface electromyographic signals of the rehabilitation participant's leg-raising force exertion during rehabilitation training and simultaneously collect the lower limb movement data, extract the force exertion data, evaluate and analyze to generate a muscle activation efficiency score; Step 4: Establish a mapping model based on historical hemodynamic data, input the extracted force data into the mapping model, and output the VTE prevention effectiveness score. Step 5: Calculate the motion assessment quality coefficient by performing motion pattern consistency score, muscle activation efficiency score, and VTE prevention effectiveness score.

2. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 1, characterized in that, The method for identifying lower limb motion data and extracting kinematic parameter features based on three-dimensional spatial motion data in step one is as follows: S101. Real-time acquisition and preprocessing to determine the triaxial acceleration and triaxial angular velocity at the ankle; S102. The preprocessed gyroscope angular velocity signal is numerically integrated, and the accelerometer signal is fused and corrected using a complementary filtering algorithm to obtain the real-time three-dimensional dynamic angle change curve of the ankle. S103. Based on the real-time three-dimensional dynamic angle change curve, the gravitational acceleration component of the acceleration signal is separated to obtain the vertical linear acceleration, and the integral term obtained by the second numerical integration is the real-time height displacement curve of the leg lift. S104. Collect the peak points of the real-time three-dimensional dynamic angle change curve, segment the motion cycle, extract and determine the peak values ​​of dorsiflexion and plantarflexion angles, motion amplitude, flexion and extension frequency and maximum angular velocity of each motion cycle, and form an ankle flexion and extension feature set. S105. Identify the acceleration mutation point on the real-time height displacement curve of the leg lift to determine the starting time of the leg lift, and extract the maximum height of the leg lift, the duration of isometric contraction and the rapid start index of force exertion to form a feature set of leg lift force exertion.

3. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 1, characterized in that, The specific methods for obtaining the action pattern consistency score in step two include: S201. Retrieve the standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template from the pre-stored rehabilitation history movement template library. The standard ankle pump exercise template and the standard quadriceps isometric contraction leg raise template contain the standard kinematic parameter feature sequence of their respective corresponding movements. S202. The real-time extracted ankle flexion-extension feature set and leg lift force feature set are dynamically time-aligned with the standard ankle pump motion template and the standard quadriceps isometric contraction leg lift template, and the similarity distance of the feature sequences is calculated; the similarity includes amplitude similarity, motion trajectory matching similarity and motion frequency similarity; S203. Calculate the three indicators respectively: amplitude similarity, motion trajectory matching similarity and motion frequency similarity, and assign a preset weight to each indicator; S204. Calculate the consistency score of the action pattern based on the weighted sum of the three indicators.

4. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 3, characterized in that, Step two, which involves determining the adequacy of rehabilitation training based on movement pattern consistency scores, includes: The consistency score of the action pattern is compared with the system's preset grading threshold, which includes a qualified threshold and a critical threshold. If the consistency score of the action pattern is greater than or equal to the passing threshold, the action is deemed to be passing and proceeds to step three. If the consistency score of the action pattern is less than the qualified threshold but greater than the critical threshold, it is determined to be an action deformation, triggering a feedback reminder but not proceeding to step three. If the consistency score of the action pattern continues to be below the critical threshold, it is determined that the rehabilitation training compliance has been interrupted, and the anti-forgetting reminder is triggered.

5. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 4, characterized in that, The method for adjusting the consistency score of non-compliant action patterns as described in step two is as follows: S205. Analyze the real-time kinematic parameter characteristics of the non-compliant actions, perform item-by-item deviation analysis with the standard template, and automatically identify non-compliant items; S206. Match the non-conforming items with the intervention strategy library to generate corrective action guidance instructions; S207. Receive corrective action guidance instructions and provide real-time corrective guidance feedback to rehabilitation personnel; S208. After the rehabilitation personnel adjust their movements based on the feedback from the correction guidance, repeat steps one to two until the consistency score of the movement pattern reaches the qualified threshold.

6. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 1, characterized in that, The specific method for extracting force data in step three is as follows: S301. Preprocess the electromyography signal of the action surface to remove power frequency interference and motion artifacts, and calculate the time-domain characteristics of the root mean square value and the time-domain characteristics of the integral electromyography value. S302. Simultaneously calibrate the starting point of lower limb motion data and establish a precise correspondence between electromyographic characteristics and kinematic characteristics on the time axis; S303. A muscle force estimation model is constructed using a bidirectional long short-term memory network. The real-time joint torque of the quadriceps femoris is estimated by taking the time-domain characteristics of the root mean square value and the time-domain characteristics of the integral electromyography value, as well as the lower limb motion data at the corresponding time, as input. The real-time joint torque is then used as the force exertion data.

7. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 6, characterized in that, The method for generating the muscle activation efficiency score through the evaluation and analysis is as follows: The reference value of the maximum isometric voluntary muscle strength of the rehabilitation personnel was used as the benchmark; Calculate the ratio of the real-time joint torque to the reference torque to obtain the real-time force percentage; Based on medical rehabilitation information, preset the target exertion range and the optimal exertion range; If the real-time force exertion percentage falls within the optimal force exertion range, then the percentage is given full marks and used as a muscle activation efficiency score. If the muscle falls within the target force exertion zone, a linear interpolation score is performed based on the percentage distance from the optimal force exertion zone, which is used as the muscle activation efficiency score. If the value is lower than the minimum value of the target force range, a percentage-based decay score is applied as the muscle activation efficiency score.

8. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 1, characterized in that, The process of establishing the mapping model in step four is as follows: Peak torque, torque-time integral, and force exertion duration are extracted from the force exertion data as predictive factors. Using the predictive factor as input and the percentage increase in peak blood flow velocity as output, a recurrent neural network model is trained. The output percentage increase in peak blood flow velocity is then converted into an effective VTE prevention score using an evidence-based curve of the probability of thrombosis prevention.

9. The method for recognizing lower limb rehabilitation exercise postures to prevent VTE according to claim 1, characterized in that, The fusion calculation method for the quality coefficient of the generated action evaluation described in step five is as follows: The formula for calculating the quality coefficient of the motion evaluation is as follows: in, The quality coefficient for motion evaluation. The score represents the consistency of action patterns. The score represents the muscle activation efficiency. To prevent VTE from being effectively evaluated; , , These are the weight coefficients determined by principal component analysis based on clinical sample data in medical experience, and... .

10. A smart ankle bracelet for implementing the lower limb rehabilitation exercise posture recognition method for VTE prevention as described in any one of claims 1-9, characterized in that, The smart ankle bracelet includes: A sensing module is used to simultaneously acquire motion data and surface electromyography signals. The sensing module includes a multi-axis inertial measurement unit and a flexible electrode array. The main control module, with a built-in microprocessor and hardware accelerator, is used to execute the feature extraction, pattern recognition, and action evaluation quality coefficient fusion algorithms in steps one through five. The specific steps include: Step 1: Identify lower limb motion data based on three-dimensional spatial motion data and extract kinematic parameter features, including ankle flexion and extension features and leg lifting force features; Step 2: Identify the movement type based on the kinematic parameter features, obtain the movement pattern consistency score, and determine whether the rehabilitation training is qualified based on the movement pattern consistency score. If yes, proceed to Step 3; otherwise, return to Step 1. Step 3: Obtain the surface electromyographic signals of the rehabilitation participant's leg-raising force exertion during rehabilitation training and simultaneously collect the lower limb motion data, extract the force exertion data, and evaluate and generate a muscle activation efficiency score; Step 4: Establish a mapping model based on historical hemodynamic data, input the extracted force data into the mapping model, and output the VTE prevention effectiveness score. Step 5: Calculate and generate a motion assessment quality coefficient by performing motion pattern consistency score, muscle activation efficiency score, and VTE prevention effectiveness score on motion assessment analysis. A feedback module is used to provide graded tactile feedback and visual alerts. The feedback module includes a miniature linear vibration motor and a multi-color LED indicator. The communication module is used for data synchronization with the mobile terminal; The power module is used to provide power to the communication module.

Images