Intelligent training method for preventing lower limb thrombus and related device

Abstract

The invention discloses an intelligent training method for preventing lower limb thrombus and a related device, and the method comprises the steps: inputting the physiological parameters of a patient, the initial exercise ability evaluation information and the personalized rehabilitation target information into a DQN multi-mode fusion model, and generating an ankle pump exercise training scheme meeting the rehabilitation demands of the patient; monitoring ankle pump motion execution information of the patient in real time through wearable equipment; according to the ankle pump motion training scheme and the ankle pump motion execution information of the patient, a lower limb thrombus training effect evaluation result of the patient is calculated, and the lower limb thrombus training effect evaluation result comprises the blood circulation improvement degree, the joint motion range and the gait characteristics; according to a lower limb thrombus training effect evaluation result of the patient, predicting lower limb thrombus potential recovery disorder and a falling risk level of the patient, and starting a grading alarm program according to the potential recovery disorder and the falling risk level. The training effect is comprehensively evaluated through the multi-modal data fusion training method, and the lower limb thrombus rehabilitation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical rehabilitation assistance technology, and in particular to an intelligent training method and device for preventing lower limb thrombosis, and a computing device. Background Art

[0002] Due to the decline in physical function, the elderly are more likely to have problems such as poor blood circulation in the lower limbs, especially surgical patients (patients undergoing major surgery such as orthopedics and gynecology), who are at high risk of deep vein thrombosis in the lower limbs. These patients need to stay in bed for a long time or reduce their activities after surgery and require auxiliary function training.

[0003] To solve the above problems, the present invention proposes an intelligent training method for preventing lower limb thrombosis, which comprehensively evaluates the training effectiveness through a multimodal data fusion training method and improves the efficiency of lower limb thrombosis rehabilitation. Summary of the invention

[0004] In view of the above problems, the present invention provides an intelligent training method and device, and a computing device for preventing lower limb thrombosis.

[0005] According to one aspect of the present invention, there is provided an intelligent training method for preventing lower limb thrombosis, comprising:

[0006] Inputting the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, exercise pattern, exercise rhythm and rest time of the ankle pump exercise;

[0007] The patient's ankle pump exercise execution information is monitored in real time by a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump exercise execution information includes ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change and plantar pressure distribution;

[0008] Calculating the patient's lower limb thrombosis training effect evaluation result according to the patient's ankle pump exercise training program and ankle pump exercise execution information, wherein the lower limb thrombosis training effect evaluation result includes the degree of improvement in blood circulation, the index of reducing the risk of venous thrombosis, muscle fatigue, joint range of motion, pain level, exercise endurance and gait characteristics;

[0009] Based on the evaluation results of the patient's lower limb thrombosis training effect, the patient's potential recovery obstacles and fall risk level of lower limb thrombosis are predicted, and a graded alarm program is initiated based on the potential recovery obstacles and fall risk level.

[0010] In an alternative embodiment, the physiological parameters include age, weight, height, medical history, type of surgery, number of days after surgery, allergy history, and heart rate variability.

[0011] In an alternative embodiment, the multi-axis inertial sensor is disposed at the distal tibia above the ankle joint, near the medial or lateral malleolus, and is configured to measure the movement angle, movement speed, movement amplitude, and posture change of the ankle joint.

[0012] The electromyography sensor is attached to the surface of the main muscle groups of the calf and is configured to measure the contraction strength and fatigue degree of the muscles during ankle pump exercise.

[0013] The biofeedback sensor is disposed at the fingertip or earlobe of the patient and is configured to monitor the heart rate and heart rate variability index of the patient to evaluate the exercise endurance and physiological response of the patient.

[0014] The pressure sensor is embedded in the insole and located in the forefoot and heel regions of the sole of the foot, and is configured to measure the plantar pressure distribution to evaluate the exercise rhythm and gait characteristics.

[0015] In an alternative embodiment, the DQN multi-modal fusion model includes a multi-modal input layer, a multi-modal feature extraction layer, a cross-modal interaction layer, a multi-modal feature fusion layer, a DQN hidden layer, and a Dueling Q-value output layer.

[0016] In an alternative embodiment, the loss function of the DQN multi-modal fusion model is:

[0017]

[0018] where δ(r, s, a, θ, θ - ) is the Huber loss function, r is the immediate reward, s is the current state, a is the current action, θ is the parameter of the current Q-network, and θ - is the parameter of the target Q-network;

[0019]

[0020] where is the target Q-value, s ′ is the new state transferred after executing the action a, a ′ is the action that may be taken in the future, A is the action space; λ is the regularization coefficient; is the L2 regularization term.

[0021] In an alternative embodiment, the calculation formula for the evaluation result of the lower limb thrombosis training effect is:

[0022]

[0023] where, w i is the weight of each evaluation index; P i is the expected training parameter value; A i is the actual training execution value; n is the number of evaluation indexes.

[0024] In an alternative manner, the multi-modal feature extraction layer includes a GRU network layer, a 1DCNN network layer, and a 2DCNN network layer;

[0025] wherein, the GRU network layer is used to process multi-axis inertial sensor data; the 1DCNN network layer is used to process electromyogram data; the 2DCNN network layer is used to process plantar pressure data.

[0026] In an alternative manner, the multi-modal feature fusion layer includes a modal alignment algorithm to calculate the similarity and correlation between different modal features, and align and fuse the similar features.

[0027] According to another aspect of the present invention, there is provided an intelligent training device for preventing lower limb thrombosis, including:

[0028] A personalized training plan generation module, configured to input the physiological parameters, initial motor ability evaluation information, cognitive ability, and personalized rehabilitation goal information of a patient into the DQN multi-modal fusion model to generate an ankle pump exercise training plan that meets the rehabilitation needs of the patient; wherein, the ankle pump exercise training plan includes the intensity, frequency, duration, exercise mode, exercise rhythm, and rest time of the ankle pump exercise;

[0029] An ankle pump exercise monitoring module, configured to monitor the ankle pump exercise execution information of the patient in real time through a wearable device; wherein, the wearable device includes a multi-axis inertial sensor, an electromyogram sensor, a biofeedback sensor, and a pressure sensor; the ankle pump exercise execution information includes the ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change, and plantar pressure distribution;

[0030] An ankle pump exercise evaluation module, configured to calculate the evaluation result of the lower limb thrombosis training effect of the patient according to the ankle pump exercise training plan and the ankle pump exercise execution information of the patient, wherein, the evaluation result of the lower limb thrombosis training effect includes the degree of improvement in blood circulation, the venous thrombosis risk reduction index, muscle fatigue, joint range of motion, pain degree, exercise endurance, and gait characteristics;

[0031] A risk warning module, configured to predict the potential recovery obstacles and fall risk level of the patient's lower limb thrombosis according to the evaluation result of the patient's lower limb thrombosis training effect, and start a hierarchical alarm program according to the potential recovery obstacles and fall risk level.

[0032] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0033] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned intelligent training method for preventing lower limb thrombosis.

[0034] According to the solution provided by the present invention, the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information are input into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, movement pattern, movement rhythm and rest time of the ankle pump exercise; the patient's ankle pump exercise execution information is monitored in real time by a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump exercise execution information includes the ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change and plantar pressure distribution; according to the patient's ankle pump exercise training program and ankle pump exercise execution information, calculate the patient's lower limb thrombosis training effect evaluation results, wherein the lower limb thrombosis training effect evaluation results include blood circulation improvement degree, venous thrombosis risk reduction index, muscle fatigue, joint range of motion, pain degree, exercise endurance and gait characteristics; according to the patient's lower limb thrombosis training effect evaluation results, predict the patient's lower limb thrombosis potential recovery obstacles and fall risk level, and start the graded alarm program according to the potential recovery obstacles and fall risk level. The present invention comprehensively evaluates the training effect through a multimodal data fusion training method and improves the rehabilitation efficiency of lower limb thrombosis. Specifically, the patient's comprehensive physiological parameters (heart rate, blood pressure, etc.), initial exercise ability evaluation information (muscle strength, flexibility, etc.), cognitive ability (attention, memory, etc.) and personalized rehabilitation goal information (expected recovery time, expected activity level, etc.) are input into the DQN (deep Q network) multimodal fusion model to generate a highly personalized ankle pump exercise training program to ensure that the training is both safe and effective. According to the patient's ankle pump exercise training program and actual execution information, the patient's lower limb thrombosis training effect evaluation results are calculated to fully reflect the patient's rehabilitation progress and physical condition. Through real-time monitoring, potential risks are discovered and handled in a timely manner to ensure the safety of patients; and effective management of different risk levels is achieved through graded alarms.

[0035] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. Brief Description of the Drawings

[0036] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0037] Figure 1 A flowchart showing the intelligent training method for preventing lower limb thrombosis according to an embodiment of the present invention is shown;

[0038] Figure 2 A schematic diagram of the DQN multimodal fusion model according to an embodiment of the present invention is shown;

[0039] Figure 3 A schematic diagram of the framework of the intelligent training device for preventing lower limb thrombosis according to an embodiment of the present invention is shown;

[0040] Figure 4 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. Detailed Embodiments

[0041] The exemplary embodiments of the present invention will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present invention can be more thoroughly understood and the scope of the present invention can be completely conveyed to those skilled in the art.

[0042] Figure 1 A flowchart showing the intelligent training method for preventing lower limb thrombosis according to an embodiment of the present invention is shown.

[0043] Specifically, as Figure 1 shown, the following steps are included:

[0044] Step S101, input the physiological parameters, initial motor ability assessment information, cognitive ability, and personalized rehabilitation goal information of the patient into the DQN multimodal fusion model to generate an ankle pump exercise training plan that meets the rehabilitation needs of the patient; wherein, the ankle pump exercise training plan includes the intensity, frequency, duration, exercise mode, exercise rhythm, and rest time of the ankle pump exercise.

[0045] In this embodiment, the DQN (Deep Q-Network) learns how to select the best ankle pump exercise training plan according to the patient's status (physiological parameters, motor ability, cognition, goals). Integrate data from different sources (physiological parameters, assessment information, cognition, goals) to comprehensively understand the patient's condition. Specifically, construct an action space. For example, the intensity of ankle pump exercise (high, medium, low), frequency (10 times / minute, 15 times / minute), duration (5 minutes, 10 minutes), exercise mode (both feet at the same time, single foot alternating), exercise rhythm (uniform, variable speed), rest time (30 seconds, 60 seconds). Define the "good or bad" criteria for training the DQN multimodal fusion model. For example, the range of motion of the ankle joint increases after training, the patient feels comfortable, and the motor ability is improved, etc. Input the patient's real-time data into the trained DQN model to generate a personalized ankle pump exercise training plan.

[0046] In an alternative way, the physiological parameters include age, weight, height, medical history, type of surgery, number of days after surgery, allergy history, and heart rate variability.

[0047] In this embodiment, through physiological parameters such as age, weight, height, medical history, type of surgery, number of days after surgery, allergy history, and heart rate variability, the DQN model can generate a more personalized, safer, and more effective ankle pump exercise training plan, which not only better adapts to the individual differences of patients, but also can monitor the patient's physical status in real time to avoid adverse reactions during the training process.

[0048] In an alternative way, the DQN multimodal fusion model includes a multimodal input layer, a multimodal feature extraction layer, a cross-modal interaction layer, a multimodal feature fusion layer, a DQN hidden layer, and a Dueling Q-value output layer.

[0049] In this embodiment, there may be a complementary relationship between different modality information. Through the cross-modal interaction layer, the deep associations between different modalities can be mined, so as to more comprehensively understand the patient's rehabilitation status. The Dueling Q-value output layer decomposes the Q-value into a value function and an advantage function, which can more stably estimate the Q-value and avoid overestimation. Among them, the multi-modal input layer receives the raw data from different modalities (physiological parameters, motion assessment, cognitive assessment, rehabilitation goals, etc.). The multi-modal feature extraction layer extracts features from the raw data of different modalities and converts them into a more expressive vector representation. The feature extraction layer of each modality works independently without interference. The cross-modal interaction layer realizes the information interaction between different modality features and captures the correlations between modalities. The multi-modal feature fusion layer fuses the different modality features output by the cross-modal interaction layer to generate a unified representation vector. The DQN hidden layer performs a non-linear transformation on the fused multi-modal features to learn a more abstract representation. The Dueling Q-value output layer converts the output of the DQN hidden layer into the Q-value of an action, which is used to guide the policy selection. The Dueling network decomposes the Q-value into two parts: a value function (describing the value of the current state) and an advantage function (describing the advantage of taking a certain action in the current state). By combining these two functions, the Q-value can be learned more stably. The Dueling output layer includes two branches, one outputs the state value, and the other outputs the advantage value of each action. Finally, the two are combined to obtain the Q-value.

[0050] In an alternative manner, the multi-modal feature extraction layer includes a GRU network layer, a 1DCNN network layer, and a 2DCNN network layer;

[0051] Among them, the GRU network layer is used to process multi-axis inertial sensor data; the 1DCNN network layer is used to process electromyography data; the 2DCNN network layer is used to process plantar pressure data.

[0052] In this embodiment, as Figure 2 shown, the GRU network layer inputs the inertial sensor data into the GRU network to extract the features in the time series. The 1DCNN network layer inputs the preprocessed electromyography signal into the 1DCNN to extract the spatial features in the signal. The 2DCNN network layer converts the plantar pressure map data into a 2D format and inputs it into the 2DCNN to extract the spatial distribution features.

[0053] In an alternative manner, the multi-modal feature fusion layer includes a modality alignment algorithm to calculate the similarity and correlation between different modality features, and align and fuse the similar features.

[0054] In this embodiment, the modal alignment algorithm identifies and aligns similar or relevant features in different modalities, thereby effectively fusing the complementary information of each modality. For example, feature extraction is performed on the data of each modality to obtain their respective feature vectors. A similarity matrix between modal features is calculated using a similarity metric method (such as cosine similarity, Euclidean distance). Based on the similarity matrix, methods such as selecting a threshold or using the Hungarian algorithm are used to align the features with higher similarity. The aligned features are fused.

[0055] In an alternative approach, the loss function of the DQN multi-modal fusion model is:

[0056]

[0057] where δ(r, s, a, θ, θ - ) is the Huber loss function, r is the immediate reward, s is the current state, a is the current action, θ is the parameter of the current Q-network, and θ - is the parameter of the target Q-network;

[0058]

[0059] where, is the target Q-value, s ′ is the new state transferred after performing the action a, a ′ is the action that may be taken in the future, A is the action space; λ is the regularization coefficient; is the L2 regularization term.

[0060] In this embodiment, the Huber loss function combines the advantages of the mean squared error (MSE) and the mean absolute error (MAE). When the error is small, MSE is used, which is sensitive to small errors and can converge quickly; when the error is large, MAE is used, which is not sensitive to outliers and can improve the stability of training. In DQN, there may be large errors in the estimation of Q-values, especially in the early training stage, and the Huber loss function can alleviate the unstable effects brought by these large errors. The target Q-network is used to calculate the target Q-value, and its parameters are periodically copied from the current Q-network, which can avoid the oscillation of the target value during the training process and ensure the stability of the learning process. Using the target Q-network in DQN can reduce the instability problem of the target value for Q-value update and make the training process more stable. The loss function itself does not directly depend on the nature of the modality, so whether the input state is from a single modality or the result of fusing multiple modalities, it can effectively optimize the Q-network.

[0061] Step S102, real-time monitor the execution information of the ankle pump exercise of the patient through a wearable device; wherein, the wearable device includes a multi-axis inertial sensor, an electromyography sensor, a biofeedback sensor, and a pressure sensor; the execution information of the ankle pump exercise includes the ankle joint movement angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change, and plantar pressure distribution.

[0062] In this embodiment, compared with the traditional visual observation or subjective report, the data collected by the wearable device is quantified into numerical indicators, such as the ankle joint movement angle, movement speed, muscle contraction strength, etc., which is convenient for quantitatively evaluating the exercise effect and rehabilitation progress. The wearable device is lightweight and easy to wear, and the patient can monitor the ankle pump exercise anytime and anywhere, without being restricted by time and location.

[0063] In an optional manner, the multi-axis inertial sensor is arranged at the distal tibia above the ankle joint, near the medial malleolus or lateral malleolus, and is used to measure the movement angle, movement speed, movement amplitude, and posture change of the ankle joint;

[0064] The electromyography sensor is attached to the surface of the main muscle groups of the calf and is used to measure the muscle contraction strength and fatigue degree during the ankle pump exercise;

[0065] The biofeedback sensor is arranged at the fingertips or earlobes of the patient and is used to monitor the heart rate and heart rate variability index of the patient to evaluate the exercise endurance and physiological response of the patient;

[0066] The pressure sensor is embedded in the insole and located in the forefoot and heel areas of the sole of the foot, and is used to measure the plantar pressure distribution to evaluate the movement rhythm and gait characteristics.

[0067] In this embodiment, the multi-axis inertial sensor measures the movement angle, movement speed, movement amplitude, and posture change of the ankle joint, so as to understand the movement range, flexibility, and control ability of the ankle joint. The electromyography sensor measures the muscle contraction strength and fatigue degree of the main muscle groups of the calf during the ankle pump exercise, so as to understand the muscle activation pattern, force output, and endurance level. The pressure sensor combines the plantar pressure distribution data to analyze the interaction force between the foot and the ground during the movement and understand the stability of the ankle joint and the gait pattern. The cooperation of multiple sensors can comprehensively evaluate the function of the ankle joint from the perspectives of kinematics, dynamics, and physiology. For example, the multi-axis inertial sensor is fixed to the distal tibia with a wearable strap, near the medial malleolus or lateral malleolus. The electromyography sensor uses a medical adhesive patch to attach the electrodes to the surface of the main muscle groups of the calf (such as the gastrocnemius muscle, tibialis anterior muscle, etc.). The biofeedback sensor includes a finger clip sensor or an ear clip sensor, which is worn on the fingertips or earlobes of the patient. The pressure sensor is embedded in a special insole, and the insole is placed in the patient's shoe to ensure that the pressure sensors in the forefoot and heel areas can accurately contact the sole of the foot.

[0068] Step S103: Calculate the evaluation result of the patient's lower limb thrombosis training according to the patient's ankle pump exercise training plan and ankle pump exercise execution information. The evaluation result of the lower limb thrombosis training includes the degree of improvement in blood circulation, the index of reduction in venous thrombosis risk, muscle fatigue, joint range of motion, pain level, exercise endurance, and gait characteristics.

[0069] In an optional manner, the calculation formula for the evaluation result of the lower limb thrombosis training is:

[0070]

[0071] where w i is the weight of each evaluation index; P i is the expected training parameter value; A i is the actual training execution value; and n is the number of evaluation indexes.

[0072] For example, the evaluation indexes and parameters of a patient with lower limb venous thrombosis rehabilitation are shown in Table 1:

[0073] Table 1

[0074]

[0075]

[0076] Substitute the above formula to calculate the evaluation result E, which is approximately 0.789, indicating that the patient's training effect is good.

[0077] Step S104: Predict the potential recovery obstacles and fall risk level of the patient's lower limb thrombosis according to the evaluation result of the patient's lower limb thrombosis training, and start a hierarchical alarm program according to the potential recovery obstacles and fall risk level.

[0078] In this embodiment, by evaluating the training effect of lower limb thrombosis, potential recovery disorders and fall risks in patients can be detected early, thus providing opportunities for timely intervention. Hierarchical alarm can allocate medical resources according to the risk level, ensuring that high-risk patients receive priority attention and improving the utilization rate of medical resources. For example, based on the evaluation data, statistical methods (such as regression analysis, machine learning) are used to establish a prediction model to predict the potential recovery disorders and fall risks of patients. According to the risk scores output by the prediction model, patients are divided into three risk levels: high, medium, and low. Among them, high-risk factors include: motor function disorders (significant decrease in muscle strength, poor balance ability, abnormal gait), severe pain (persistent severe pain, affecting activities), obvious edema (increase in the circumference of the affected limb), low quality of life (significant decrease in daily living ability), and neurological function disorders (poor motor control ability). Medium-risk factors include: mild limitation of motor function, tolerable pain, and mild edema. Low-risk factors include: normal or slightly limited motor function, mild or no pain, and no edema.

[0079] According to the solution provided by the present invention, the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information are input into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, movement pattern, movement rhythm and rest time of the ankle pump exercise; the patient's ankle pump exercise execution information is monitored in real time by a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump exercise execution information includes the ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change and plantar pressure distribution; according to the patient's ankle pump exercise training program and ankle pump exercise execution information, calculate the patient's lower limb thrombosis training effect evaluation results, wherein the lower limb thrombosis training effect evaluation results include blood circulation improvement degree, venous thrombosis risk reduction index, muscle fatigue, joint range of motion, pain degree, exercise endurance and gait characteristics; according to the patient's lower limb thrombosis training effect evaluation results, predict the patient's lower limb thrombosis potential recovery obstacles and fall risk level, and start the graded alarm program according to the potential recovery obstacles and fall risk level. The present invention comprehensively evaluates the training effect through a multimodal data fusion training method and improves the rehabilitation efficiency of lower limb thrombosis. Specifically, the patient's comprehensive physiological parameters (heart rate, blood pressure, etc.), initial exercise ability evaluation information (muscle strength, flexibility, etc.), cognitive ability (attention, memory, etc.) and personalized rehabilitation goal information (expected recovery time, expected activity level, etc.) are input into the DQN (deep Q network) multimodal fusion model to generate a highly personalized ankle pump exercise training program to ensure that the training is both safe and effective. According to the patient's ankle pump exercise training program and actual execution information, the patient's lower limb thrombosis training effect evaluation results are calculated to fully reflect the patient's rehabilitation progress and physical condition. Through real-time monitoring, potential risks are discovered and handled in a timely manner to ensure the safety of patients; and effective management of different risk levels is achieved through graded alarms.

[0080] Figure 3 The schematic diagram of the framework of the intelligent training device for preventing lower limb thrombosis according to the embodiment of the present invention is shown. The intelligent training device for preventing lower limb thrombosis comprises:

[0081] The personalized training program generation module 310 is used to input the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, exercise mode, exercise rhythm and rest time of the ankle pump exercise;

[0082] Ankle pump motion monitoring module 320, used for real-time monitoring of the patient's ankle pump motion execution information through a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump motion execution information includes ankle joint activity angle, motion speed, muscle contraction strength, motion amplitude, motion rhythm, posture change and plantar pressure distribution;

[0083] An ankle pump exercise evaluation module 330 is used to calculate the patient's lower limb thrombosis training effect evaluation result according to the patient's ankle pump exercise training program and ankle pump exercise execution information, wherein the lower limb thrombosis training effect evaluation result includes the degree of improvement of blood circulation, the index of reducing the risk of venous thrombosis, muscle fatigue, joint range of motion, pain level, exercise endurance and gait characteristics;

[0084] The risk warning module 340 is used to predict the potential recovery obstacles and fall risk levels of the patient's lower limb thrombosis according to the patient's lower limb thrombosis training effect evaluation results, and to initiate a graded alarm program according to the potential recovery obstacles and fall risk levels.

[0085] Figure 4 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0086] like Figure 4 As shown, the computing device may include: a processor (processor) 402 , a communications interface (Communications Interface) 404 , a memory (memory) 406 , and a communication bus 408 .

[0087] The processor 402, the communication interface 404, and the memory 406 communicate with each other via the communication bus 408. The communication interface 404 is used to communicate with other devices such as a client or other server network elements. The processor 402 is used to execute the program 410, which can specifically execute the relevant steps in the above-mentioned embodiment of the intelligent training method for preventing lower limb thrombosis.

[0088] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0089] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. One or more processors included in the computing device may be of the same type of processor, such as one or more CPUs; or may be of different types of processors, such as one or more CPUs and one or more ASICs.

[0090] A memory 406 for storing a program 410. The memory 406 may include high-speed RAM memory and may also include non-volatile memory, such as at least one magnetic disk memory.

[0091] According to the solution provided by the present invention, the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information are input into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, movement pattern, movement rhythm and rest time of the ankle pump exercise; the patient's ankle pump exercise execution information is monitored in real time by a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump exercise execution information includes the ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change and plantar pressure distribution; according to the patient's ankle pump exercise training program and ankle pump exercise execution information, calculate the patient's lower limb thrombosis training effect evaluation results, wherein the lower limb thrombosis training effect evaluation results include blood circulation improvement degree, venous thrombosis risk reduction index, muscle fatigue, joint range of motion, pain degree, exercise endurance and gait characteristics; according to the patient's lower limb thrombosis training effect evaluation results, predict the patient's lower limb thrombosis potential recovery obstacles and fall risk level, and start the graded alarm program according to the potential recovery obstacles and fall risk level. The present invention comprehensively evaluates the training effect through a multimodal data fusion training method and improves the rehabilitation efficiency of lower limb thrombosis. Specifically, the patient's comprehensive physiological parameters (heart rate, blood pressure, etc.), initial exercise ability evaluation information (muscle strength, flexibility, etc.), cognitive ability (attention, memory, etc.) and personalized rehabilitation goal information (expected recovery time, expected activity level, etc.) are input into the DQN (deep Q network) multimodal fusion model to generate a highly personalized ankle pump exercise training program to ensure that the training is both safe and effective. According to the patient's ankle pump exercise training program and actual execution information, the patient's lower limb thrombosis training effect evaluation results are calculated to fully reflect the patient's rehabilitation progress and physical condition. Through real-time monitoring, potential risks are discovered and handled in a timely manner to ensure the safety of patients; and effective management of different risk levels is achieved through graded alarms.

[0092] Those skilled in the art can understand that the modules in the devices in the embodiments can be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise explicitly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature that provides the same, equivalent, or similar purpose. In addition, those skilled in the art can understand that although some of the embodiments herein include certain features included in other embodiments rather than other features, the combination of the features of different embodiments means that it is within the scope of the present invention and forms different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination. The present invention can be implemented by means of hardware including several different elements and by means of a properly programmed computer. In the unit claims listing several devices, several of these devices can be embodied by the same hardware item. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the execution order.

Claims

1. An intelligent training method for preventing lower limb thrombosis, characterized in that: include: Inputting the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, exercise pattern, exercise rhythm and rest time of the ankle pump exercise; The patient's ankle pump exercise execution information is monitored in real time by a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump exercise execution information includes ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change and plantar pressure distribution; Calculating the patient's lower limb thrombosis training effect evaluation result according to the patient's ankle pump exercise training program and ankle pump exercise execution information, wherein the lower limb thrombosis training effect evaluation result includes the degree of improvement in blood circulation, the index of reducing the risk of venous thrombosis, muscle fatigue, joint range of motion, pain level, exercise endurance and gait characteristics; Based on the evaluation results of the patient's lower limb thrombosis training effect, the patient's potential recovery obstacles and fall risk level of lower limb thrombosis are predicted, and a graded alarm program is initiated based on the potential recovery obstacles and fall risk level.

2. The intelligent training method for preventing lower limb thrombosis according to claim 1, characterized in that: The physiological parameters include age, weight, height, medical history, type of surgery, days after surgery, allergy history, and heart rate variability.

3. The intelligent training method for preventing lower limb thrombosis according to claim 1 is characterized in that: The multi-axis inertial sensor is arranged at the distal end of the tibia above the ankle joint, close to the medial or lateral malleolus, and is used to measure the activity angle, movement speed, movement amplitude and posture change of the ankle joint; The electromyographic sensor is attached to the surface of the main muscle group of the calf and is used to measure the contraction strength and fatigue degree of the muscle during ankle pump exercise; The biofeedback sensor is arranged at the fingertips or earlobes of the patient and is used to monitor the patient's heart rate and heart rate variability index to evaluate the patient's exercise endurance and physiological response; The pressure sensor, embedded in the insole and located in the forefoot and heel areas of the sole, is used to measure the plantar pressure distribution to evaluate the movement rhythm and gait characteristics.

4. The intelligent training method for preventing lower limb thrombosis according to claim 1, characterized in that: The DQN multimodal fusion model includes a multimodal input layer, a multimodal feature extraction layer, a cross-modal interaction layer, a multimodal feature fusion layer, a DQN hidden layer and a Dueling Q value output layer.

5. The intelligent training method for preventing lower limb thrombosis according to claim 1, characterized in that: The loss function of the DQN multimodal fusion model is: Among them, δ(r,s,a,θ,θ - ) is the Huber loss function, r is the immediate reward, s is the current state, a is the current action, θ is the parameter of the current Q network, θ - are the parameters of the target Q network; in, is the target Q value, s ′ is the new state after executing action a, a ′ is the possible action to be taken in the future, A is the action space; λ is the regularization coefficient; is the L2 regularization term.

6. The intelligent training method for preventing lower limb thrombosis according to claim 1, characterized in that: The calculation formula for the evaluation result of the lower limb thrombosis training effect is: Among them, w i is the weight of each evaluation indicator; P i is the expected training parameter value; A i is the actual training execution value; n is the number of evaluation indicators.

7. The intelligent training method for preventing lower limb thrombosis according to claim 4, characterized in that: The multimodal feature extraction layer includes a GRU network layer, a 1DCNN network layer and a 2DCNN network layer; Among them, the GRU network layer is used to process multi-axis inertial sensor data; the 1DCNN network layer is used to process electromyography data; and the 2DCNN network layer is used to process plantar pressure data.

8. The intelligent training method for preventing lower limb thrombosis according to claim 4, characterized in that: The multimodal feature fusion layer includes a modality alignment algorithm to calculate the similarity and correlation between different modality features and align and fuse similar features.

9. An intelligent training device for preventing lower limb thrombosis, characterized in that: include: A personalized training program generation module is used to input the patient's physiological parameters, initial motor ability assessment information, cognitive ability and personalized rehabilitation goal information into the DQN multimodal fusion model to generate an ankle pump exercise training program that meets the patient's rehabilitation needs; wherein the ankle pump exercise training program includes the intensity, frequency, duration, exercise mode, exercise rhythm and rest time of the ankle pump exercise; Ankle pump exercise monitoring module, used for real-time monitoring of the patient's ankle pump exercise execution information through a wearable device; wherein the wearable device includes a multi-axis inertial sensor, an electromyographic sensor, a biofeedback sensor and a pressure sensor; the ankle pump exercise execution information includes ankle joint activity angle, movement speed, muscle contraction strength, movement amplitude, movement rhythm, posture change and plantar pressure distribution; An ankle pump exercise evaluation module is used to calculate the patient's lower limb thrombosis training effect evaluation results according to the patient's ankle pump exercise training program and ankle pump exercise execution information, wherein the lower limb thrombosis training effect evaluation results include blood circulation improvement degree, venous thrombosis risk reduction index, muscle fatigue, joint range of motion, pain degree, exercise endurance and gait characteristics; The risk warning module is used to predict the potential recovery obstacles and fall risk levels of patients' lower limb thrombosis based on the evaluation results of the patients' lower limb thrombosis training effects, and to initiate a graded alarm program based on the potential recovery obstacles and fall risk levels.

10. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned intelligent training method for preventing lower limb thrombosis.

Images