Parkinson patient rehabilitation training method based on artificial intelligence

Abstract

The invention relates to the technical field of rehabilitation training, and discloses a Parkinson's disease patient rehabilitation training method based on artificial intelligence. A wearable sensor and an environment sensor are adopted to collect motion data, heart rate data, gait data and home environment data of a patient, and noise removal, filtering, standardization and normalization processing are carried out on the collected original data; step S2, on the basis of the data in the step S1, performing weighted fusion on the data acquired by different sensors to form a comprehensive index reflecting the overall health state of the patient; and S3, making a training scheme on the basis of the comprehensive indexes obtained in the step S2. By carrying out safety risk assessment on the home environment of the patient and combining environment data and a principal component analysis method, targeted environment improvement suggestions are automatically provided, the rehabilitation environment of the patient can be optimized through environment safety assessment and improvement, potential safety risks can be effectively avoided, and a safe home rehabilitation environment is provided for the patient.

Description

Technical Field

[0001] The present invention relates to the technical field of rehabilitation training, and specifically to a rehabilitation training method for Parkinson's patients based on artificial intelligence. Background Art

[0002] Parkinson's disease is a neurodegenerative disease of the nervous system, mainly manifested as symptoms such as movement disorders, tremors, stiffness, and gait instability, seriously affecting the daily living ability of patients. Due to the progression of the patient's condition, Parkinson's patients need long-term support from family members or professional caregivers. How to improve the quality of life of Parkinson's patients and reduce the burden of home care has become an important topic in the combination of medicine, rehabilitation, and intelligent technology.

[0003] Most current Parkinson's rehabilitation trainings adopt standardized training methods without fully considering the differences in the individual motor abilities, cognitive states, and living environments of patients. The lack of personalized programs will lead to insignificant rehabilitation effects and the inability to make real-time adjustments according to the actual needs of patients. Existing technologies generally cannot provide accurate training content and dynamic adjustment mechanisms based on the comprehensive health status of patients.

[0004] Many existing rehabilitation training methods have a certain function of collecting motion data, but they are often lagging in real-time feedback and data processing, and cannot accurately adjust the training effect in a short time. The lack of support from adaptive algorithms makes the rehabilitation training unable to effectively respond in a timely manner according to the physiological and motion feedback of patients, resulting in limitations in the personalization and accuracy of the training effect.

[0005] Home rehabilitation is very important for Parkinson's patients, but existing technologies will ignore the safety assessment of the living environment, and the environmental factors of patients at home have not been given enough attention. Safety hazards in the environment, such as slips and obstacles, may cause accidents to patients, further affecting the rehabilitation effect. Existing technologies also lack intelligent environmental improvement suggestions and cannot effectively avoid safety risks.

[0006] Existing rehabilitation treatment technologies lack an effective remote medical management platform, and the health data of patients cannot be transmitted to doctors in real time for tracking and intervention. Especially Parkinson's patients usually require long-term rehabilitation training and monitoring. If doctors cannot obtain the rehabilitation data of patients in real time, they will not be able to make timely treatment adjustments, thus affecting the treatment effect. The limitations of existing technologies in remote medical support lead to blind spots in treatment management.

[0007] Current rehabilitation programs often focus on short-term rehabilitation effects and lack dynamic optimization for the long-term rehabilitation process. The rehabilitation effect fluctuates over time, and existing technologies fail to provide a comprehensive evaluation and adjustment mechanism based on long-term data. There is a lack of the ability to comprehensively evaluate multiple health indicators, which cannot ensure the continuous optimization of the rehabilitation program during the treatment cycle, resulting in stagnant rehabilitation effects for patients.

[0008] Many existing technologies lack prognosis prediction models and risk intervention mechanisms. The rehabilitation process of patients is affected by various factors, and existing technologies fail to conduct comprehensive evaluations through big data analysis and integrated learning methods. The lack of means to predict rehabilitation effects leads to patients being unable to obtain personalized intervention programs in a timely manner, increasing medical costs and risks.

[0009] Therefore, those skilled in the art provide a rehabilitation training method for Parkinson's patients based on artificial intelligence to solve the above-mentioned problems. Summary of the Invention

[0010] In view of the deficiencies of the prior art, the present invention provides a rehabilitation training method for Parkinson's patients based on artificial intelligence to solve the problems raised in the above background technology.

[0011] To achieve the above objectives, the present invention is realized through the following technical solutions: A rehabilitation training method for Parkinson's patients based on artificial intelligence, comprising:

[0012] Step S1, using wearable sensors and environmental sensors to collect patients' motion data, heart rate data, gait data, and home environment data, and performing noise removal, filtering, standardization, and normalization processing on the collected raw data;

[0013] Step S2, based on the data in Step S1, performing weighted fusion on the data collected by different sensors to form a comprehensive index reflecting the overall health status of the patient;

[0014] Step S3, based on the comprehensive index obtained in Step S2, formulating a training plan according to the patient's motor ability, cognitive ability, and home environment safety;

[0015] Step S4, during the implementation of the training plan formulated in Step S3, adjusting the training plan according to the real-time collected training feedback data to make the training content conform to the actual state of the patient;

[0016] Step S5, based on the environmental data collected in Step S1, performing a safety risk assessment on the home environment and providing environmental improvement suggestions;

[0017] Step S6, based on the implementation results of the training plan adjusted in Step S4 and the evaluation results in Step S5, comprehensively calculating the patient's comprehensive health index, and transmitting the evaluation results to a remote medical platform for medical staff to refer to;

[0018] Step S7: Integrate the data and information obtained in the above steps through a standardized interface and adopt encryption measures to ensure the security of data transmission and storage.

[0019] Preferably, step S1 further includes:

[0020] Step 1.1: Use the adaptive Kalman filtering algorithm to eliminate signal noise based on the collected raw data, and use the prediction and update recurrence method for state estimation and noise correction. The prediction step uses the formula: Y k|k-1 = F k Y k-1|k-1 + B k U k ,

[0021] The prediction error covariance formula:

[0022] The Kalman gain formula:

[0023] The state update formula: Y k|k = Y k|k-1 + K k (Z k - H k Y k|k-1 ),

[0024] where Y k|k-1 is the predicted signal state estimate at time k, F k is the state transition matrix describing the signal change law, Y k-1|k-1 is the state estimate at the previous time, B k is the control influence matrix, U k is the control input vector reflecting external disturbances, E k|k-1 is the prediction error covariance matrix, Q k is the process noise covariance matrix reflecting the inherent uncertainty of the signal, H k is the observation matrix mapping the state to the observation space, Z k is the actual observation value at time k, R k is the observation noise covariance matrix reflecting the sensor measurement error, K k is the Kalman gain used to adjust the weight between prediction and observation, Y k|k is the state estimate at the current time, is the transpose matrix of H k , is the transpose matrix of F k ;

[0025] Step 1.2: Based on the signal processed by Kalman filtering in Step 1.1, use the multi-scale wavelet decomposition and reconstruction algorithm to filter, standardize, and normalize the signal. The formula is as follows:

[0026]

[0027] where Θ is the signal after filtering and standardization processing, Π is the reconstruction scale factor, J is the number of wavelet decomposition layers, Ω j is the detail coefficient obtained from the j-th layer decomposition, is the result of processing the j-th layer detail coefficient using the threshold function, and Σ j is the j-th layer adaptive threshold value.

[0028] Preferably, the step S2 further includes:

[0029] Step 2.1: Based on the output signals ω i of each sensor obtained through filtering, standardization, and normalization processing in Step S1, use the calibration model to achieve multi-sensor data alignment. The calibrated data is expressed using the formula: Φ i = α i ·ω i + β i ,

[0030] where Φ i is the output signal of the i-th sensor after calibration, α i is the calibration gain of the i-th sensor, β i is the bias correction value of the i-th sensor, and ω i is the original signal of the i-th sensor obtained from the standardization processing in Step S1;

[0031] Step 2.2: Based on the calibrated data Φ i obtained in Step 2.1, use the weight assignment algorithm based on exponential smoothing to determine the contribution ratio of each sensor in data fusion. The weights of each sensor are expressed using the formula:

[0032] where w i is the weight of the i-th sensor, γ i is the measurement uncertainty of the i-th sensor, and j is the index of all sensors;

[0033] Step 2.3: Based on the calibrated data Φ i obtained in Step 2.1 and the weights w i determined in Step 2.2, use the weighted linear fusion algorithm to achieve the integration of multi-sensor data. The integrated health status index generated after fusion is expressed using the formula:

[0034] Among them, Δ is the final fusion result, w i is the weight of the i-th sensor, and Φ i is the output signal of the i-th sensor after calibration.

[0035] Preferably, the step S3 further includes:

[0036] Step 3.1, based on the fusion comprehensive index Δ obtained in step S2, extract the patient's motor ability, cognitive ability, and home environment safety data, and perform normalization processing on each data. Use a weighted linear normalization model to achieve data integration, and use a formula to express the patient ability data integration result:

[0037]

[0038] Among them, is the training capacity factor, χ m is the motor ability weight coefficient, ρ m is the normalized patient motor ability score, χ c is the cognitive ability weight coefficient, ρ c is the normalized patient cognitive ability score, χ h is the home environment safety weight coefficient, ρ h is the normalized home environment safety score;

[0039] Step 3.2, based on the training capacity factor ρ h obtained in step 3.1, fuse the comprehensive health status index Δ generated in step S2, and use a multi-factor linear combination model to calculate the patient's training readiness index. Use a formula to express the training readiness index:

[0040] Among them, κ is the patient's overall training readiness index, μ is the fusion comprehensive index weight coefficient, Δ is the final fusion result, ν is the training capacity factor weight coefficient, is the training capacity factor;

[0041] Step 3.3, based on the training readiness index κ obtained in step 3.2, use a non-linear mapping model to generate personalized rehabilitation training plan parameters. Use a formula to express the training intensity parameters:

[0042] I = ξ·ln(1 + κ) + ζ,

[0043] Among them, I is the generated training intensity parameter, which reflects the training load level in the rehabilitation training plan, ξ is the training intensity scale coefficient to adjust the index amplification factor, κ is the patient's overall training readiness index, and ζ is the training benchmark offset.

[0044] Preferably, the step S4 further includes:

[0045] Step 4.1, based on the personalized training plan generated in step S3, collect the motion data and physiological feedback data generated by the patient during the training in real time. Use an adaptive Kalman filter to monitor and feedback-adjust the training process in real time based on the data, and use the formula to perform state estimation on the training feedback signal: Y k|k-1 = F k Y k-1|k-1 + B k U k ,

[0046] where Y k|k-1 is the signal state estimate predicted at time k, F k is the state transition matrix, B k is the control influence matrix, U k is the control input vector, and Y k-1|k-1 is the state estimate at the previous moment;

[0047] Step 4.2, according to the real-time feedback data obtained in sub-step 4.1, use a dynamic adjustment algorithm based on fuzzy logic to correct the training plan in real time. The adjustment process uses the fuzzy logic control formula:

[0048] Δu = K f · sigm(L(∈)),

[0049] where Δu is the adjustment amount of the training parameter, K f is the fuzzy control gain, L is the training adjustment rule set, and ∈ is the current training feedback error;

[0050] Step 4.3, based on the training plan adjusted in step 4.2, use the new feedback data of the patient to verify the effect of the training plan, and use an adaptive learning algorithm to correct the optimization of the plan. The optimization process uses the reinforcement learning algorithm formula:

[0051]

[0052] where Q(s,a) is the expected reward in the current state s and action a, a is the learning rate, R(s,a) is the immediate reward, γ is the discount factor, s ′ is the new state, and a ′ is the new action;

[0053] Step 4.4, according to the optimized training plan in step 4.3, carry out the actual training implementation, and collect and analyze the training effect data during the implementation process. The evaluation of the training effect uses a system evaluation algorithm, and a comprehensive evaluation is carried out based on real-time feedback and target setting. Use the following formula:

[0054]

[0055] Among them, E is the comprehensive evaluation result, w i is the weight of each evaluation index, Δ i is the change in the training effect of each index, v j is the weight of external influencing factors, η j is the corresponding influence amount, and λ1 and λ2 are weight adjustment coefficients.

[0056] Preferably, the step S5 further includes:

[0057] Step 5.1, based on the home environment data collected in step S1, use the principal component analysis method to extract features and reduce the dimension of the environment data, and use the formula to express the extracted environmental features:

[0058]

[0059] Among them, X e is the original home environment data after standardization and normalization, μ e is the mean vector, W e is the eigenvector matrix, F e is the environmental feature vector;

[0060] Step 5.2, based on the environmental feature vector F e extracted in step 5.1, use the logistic regression model to evaluate the home environment safety risk, and use the formula to express the environmental safety risk score:

[0061]

[0062] Among them, R e is the home environment safety risk score, f e,i is the i-th feature component in the environmental feature vector F e α i is the weight coefficient corresponding to the i-th feature, n is the total number of environmental features, β is the bias term of the logistic regression model, and exp(·) is the natural exponential function;

[0063] Step 5.3, according to the safety risk score R e obtained in step 5.2, use the improved suggestion generation algorithm based on fuzzy inference, and use the membership function to perform defuzzification mapping on the risk score. The mapping process is expressed as: S e =μ F (R e ; θ1, θ2),

[0064] Among them, μ F is the membership function, defined as:

[0065]

[0066] Among them, S e is the generated environmental improvement suggestion index, R e is the home environment safety risk score, μ F (R e ; θ1, θ2) is the membership function of the risk score under different risk levels,

[0067] θ1 is the low-risk threshold of the risk score. When it is lower than the threshold, it indicates a low environmental risk;

[0068] θ2 is the high-risk threshold of the risk score. When it is higher than the threshold, it indicates a high environmental risk.

[0069] Preferably, the step S6 further includes:

[0070] Step 6.1, based on the real-time health status index θ 1,t obtained in step S2, the training preparation index θ 2,t obtained in step S3, and the environmental improvement suggestion index θ 3,t generated in step S5, a weighted average fusion method is used to comprehensively evaluate the long-term rehabilitation effect of the patient, and the long-term effect evaluation result is expressed by the formula as follows:

[0071] Among them, E t is the comprehensive rehabilitation effect evaluation value at time t, θ 1,t is the real-time health status index obtained in step S2 at time t, θ 2,t is the training preparation index obtained in step S3 at time t, θ 3,t is the environmental improvement suggestion index obtained in step S5 at time t, and π1, π2, and π3 are the reliability weights of the corresponding indexes;

[0072] Step 6.2, based on the prediction error between the long-term effect evaluation value E t calculated in step 6.1 and the actually observed rehabilitation effect, the reliability weights of each index are adaptively updated through an online learning algorithm, and the update formula is used:

[0073] π i (t + 1) = π i (t) + η[ε t ·θ i,t - λ·π i (t)],

[0074] Among them, π i (t + 1) is the weight of the i-th index updated at time t + 1, π i (t) is the current weight of the i-th index at time t, η is the learning rate, ε tis the prediction error at time t, θ i,t is the value of the i-th index at time t, and λ is the regularization parameter;

[0075] Step 6.3, after completing Steps 6.1 and 6.2, based on the latest comprehensive rehabilitation effect evaluation value E t use the non-linear mapping model to determine whether the rehabilitation training plan needs to be adjusted, and adopt the decision function: D t = σ(ζ · (E t - τ)),

[0076] where, D t is the plan adjustment decision variable at time t, σ is the sigmoid function, E t is the comprehensive rehabilitation effect evaluation value at time t, τ is the preset rehabilitation effect target threshold, and ζ is the decision sensitivity coefficient.

[0077] Preferably, the step S7 further includes:

[0078] Step 7.1, based on the real-time health status indicators obtained in Step S2, the training preparation index generated in Step S3, the real-time feedback data collected in Step S4, the environmental improvement suggestion indicators extracted in Step S5, and the long-term rehabilitation effect value calculated in Step S6, fuse the data into a comprehensive feature vector z, and use the prediction model based on ensemble learning to construct a prognosis prediction model, using the formula:

[0079] to express the prognosis prediction output,

[0080] where, is the predicted future rehabilitation effect value, M is the number of base learners, θ j is the weight coefficient of the j-th base learner, g j (z) is the prediction output of the j-th base learner for the input feature vector z, and z is the comprehensive feature vector;

[0081] Step 7.2, based on the prognosis prediction model constructed in Step 7.1, train and optimize the model parameter set ΘΘ through the mean square error loss function with a regularization term, using the formula:

[0082] to express the model loss,

[0083] where, L(Θ) is the total loss value of the model parameter set Θ, N is the number of training samples, is the rehabilitation effect value obtained by the i-th sample through model prediction, Y i is the actually observed rehabilitation effect value of the i-th sample, Θ is the set of all trainable parameters of the model, and λ is the regularization coefficient;

[0084] Step 7.3, based on the predicted rehabilitation effect obtained after training and optimization in Step 7.2 Combine with the preset rehabilitation effect target threshold Y th , use the non-linear mapping decision model to generate risk warning signals and personalized intervention strategy adjustment suggestions, using the formula:

[0085] Express the decision output,

[0086] where W is the risk warning weight or intervention strategy adjustment signal, σ represents the sigmoid function, κ is the decision sensitivity coefficient, is the predicted value of the rehabilitation effect output by the prognosis prediction model, and Y th is the preset rehabilitation effect target threshold.

[0087] A terminal device, the device includes:

[0088] A processor for executing the steps of the Parkinson's patient rehabilitation training method;

[0089] A memory for storing the relevant data of the method;

[0090] A sensor interface for connecting to and receiving data from wearable sensors and environmental sensors;

[0091] A user interface for presenting the feedback results and adjustment suggestions of the patient's rehabilitation training to the user;

[0092] A communication module for data transmission with a remote medical platform.

[0093] A storage medium for storing computer-executable program code, and the program code implements the Parkinson's patient rehabilitation training method when executed, specifically including:

[0094] A data acquisition module for collecting the patient's movement data, heart rate data, gait data and home environment data through sensors;

[0095] A data processing module for performing noise removal, filtering, standardization and normalization processing on the collected data, and performing multi-sensor data fusion;

[0096] A training plan generation module for formulating a personalized rehabilitation training plan based on health status indicators, motor ability, cognitive ability and home environment safety;

[0097] A feedback adjustment module for adjusting the training plan according to real-time training feedback data;

[0098] An evaluation module for evaluating the patient's rehabilitation effect and generating a personalized intervention strategy according to the evaluation result;

[0099] A data storage module for storing the training data, evaluation results, and all relevant data of patients.

[0100] The present invention provides a rehabilitation training method for Parkinson's patients based on artificial intelligence. It has the following beneficial effects:

[0101] 1. The present invention adopts a method based on multi-sensor data fusion to comprehensively obtain the patient's motion data, heart rate data, gait data, and home environment data. After processing, the data generates a comprehensive health status index, and a personalized rehabilitation training plan is formulated according to the patient's motor ability, cognitive ability, and home environment safety. At the same time, by dynamically adjusting the training content, it ensures that the rehabilitation plan highly matches the patient's actual health condition, so as to effectively improve the rehabilitation effect.

[0102] 2. The present invention adopts an adaptive Kalman filter and a fuzzy logic control algorithm, which can collect and analyze the motion data and physiological feedback data generated by the patient during the training process in real time, and through systematic real-time adjustment, it can quickly respond to the changes during the training process to ensure the personalization and effectiveness of the training.

[0103] 3. The present invention conducts a safety risk assessment on the patient's home environment, combines the environmental data and the principal component analysis method, and automatically provides targeted environmental improvement suggestions. The environmental safety assessment and improvement can help optimize the patient's rehabilitation environment, and can effectively avoid potential safety risks, providing a safe home rehabilitation environment for the patient.

[0104] 4. The present invention can comprehensively calculate the patient's comprehensive health index, transmit the evaluation results to the remote medical platform in real time, provide accurate patient health data support for medical staff, promote remote medical management and tracking, help medical staff timely understand the patient's rehabilitation situation and conduct remote intervention, and reduce the risks during the patient's treatment process.

[0105] 5. The present invention constructs a prognosis prediction model based on multiple health data, combines the ensemble learning method, can predict the future rehabilitation effect according to the patient's rehabilitation data, and at the same time, generates a personalized intervention strategy through a non-linear mapping decision model, warns of potential risks in advance, further improves the patient's rehabilitation effect and reduces medical costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0107] To enable those skilled in the art to understand the solution of the present invention, the following will clearly and completely describe the technical solution in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are partial embodiments of the present invention, rather than all embodiments. Based on the embodiments in the present invention, other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0108] The following will describe the present invention in detail with reference to the accompanying drawings:

[0109] Embodiment:

[0110] Please refer to the attached Figure 1 , the embodiment of the present invention provides a rehabilitation training method for Parkinson's patients based on artificial intelligence, including:

[0111] Step S1, using wearable sensors and environmental sensors to collect the patient's motion data, heart rate data, gait data, and home environment data, and performing noise removal, filtering, standardization, and normalization processing on the collected original data;

[0112] Step 1.1, based on the collected original data, using the adaptive Kalman filtering algorithm to eliminate signal noise, and using the prediction and update recurrence method for state estimation and noise correction. The prediction step uses the formula: Y k|k-1 =F k Y k-1|k-1 +B k U k ,

[0113] The prediction error covariance formula:

[0114] The Kalman gain formula:

[0115] The state update formula: Y k|k =Y k|k-1 +K k (Z k -H k Y k|k-1 ),

[0116] where Y k|k-1 is the signal state estimate predicted at time k, F k is the state transition matrix describing the signal change law, Y k-1|k-1 is the state estimate of the previous moment, B k is the control influence matrix, U k is the control input vector reflecting external disturbances, E k|k-1 is the prediction error covariance matrix, Qk The process noise covariance matrix reflects the inherent uncertainty of the signal, and H k is the observation matrix that maps the state to the observation space, and Z k is the actual observation value at time k, and R k is the observation noise covariance matrix that reflects the sensor measurement error, and K k is the Kalman gain used to adjust the weight between prediction and observation, and Y k|k is the state estimate at the current time, is H k 's transpose matrix, is F k 's transpose matrix;

[0117] Step 1.2, Based on the signal processed by Kalman filtering in Step 1.1, use the multi-scale wavelet decomposition and reconstruction algorithm to filter, standardize, and normalize the signal. Use the formula:

[0118]

[0119] where, Θ is the signal after filtering and standardization, Π is the reconstruction scale factor, J is the number of layers of wavelet decomposition, and Ω j is the detail coefficient obtained from the j-th layer decomposition, is the result of processing the j-th layer detail coefficient using the threshold function, and Σ j is the j-th layer adaptive threshold value;

[0120] Step S2, Based on the data in Step S1, perform weighted fusion on the data collected by different sensors to form a comprehensive index reflecting the overall health status of the patient;

[0121] Step 2.1, On the basis of the output signals ω i of each sensor obtained by filtering, standardizing, and normalizing in Step S1, use the calibration model to achieve multi-sensor data alignment. Express the calibrated data using the formula: Φ i = α i ·ω i + β i ,

[0122] where, Φ i is the output signal of the i-th sensor after calibration, α i is the calibration gain of the i-th sensor, β i is the offset calibration value of the i-th sensor, and ω i is the original signal of the i-th sensor obtained by standardization in Step S1;

[0123] Step 2.2, On the calibrated data Φ iBased on this, the contribution ratio of each sensor in data fusion is determined by using a weight allocation algorithm based on exponential smoothing, and the weights of each sensor are expressed by formulas:

[0124] Among them, w i is the weight of the i-th sensor, and γ i is the measurement uncertainty of the i-th sensor, and j is the index of all sensors;

[0125] Step 2.3, on the basis of the calibration data Φ i obtained in Step 2.1 and the weights w i determined in Step 2.2, a weighted linear fusion algorithm is adopted to realize the integration of multi-sensor data, and the comprehensive health status index generated after fusion is expressed by a formula:

[0126] Among them, Δ is the final fusion result, w i is the weight of the i-th sensor, and Φ i is the output signal of the i-th sensor after calibration;

[0127] Step S3, based on the comprehensive index obtained in Step S2, a training plan is formulated according to the patient's motor ability, cognitive ability, and home environment safety;

[0128] Step 3.1, based on the fusion comprehensive index Δ obtained in Step S2, extract the patient's motor ability, cognitive ability, and home environment safety data, and perform normalization processing on each data. A weighted linear normalization model is adopted to realize data integration, and the patient ability data integration result is expressed by a formula:

[0129]

[0130] Among them, is the training capacity factor, χ m is the motor ability weight coefficient, ρ m is the normalized patient motor ability score, χ c is the cognitive ability weight coefficient, ρ c is the normalized patient cognitive ability score, χ h is the home environment safety weight coefficient, ρ h is the normalized home environment safety score;

[0131] Step 3.2, based on the training capacity factor ρ h obtained in Step 3.1, fuse the comprehensive health status index Δ generated in Step S2, and calculate the patient training readiness index by using a multi-factor linear combination model. The training readiness index is expressed by a formula:

[0132] Where κ is the overall training preparation index of the patient, μ is the weight coefficient of the fusion comprehensive index, Δ is the final fusion result, and ν is the weight coefficient of the training capacity factor. is the training capacity factor;

[0133] Step 3.3: Based on the training preparation index κ obtained in Step 3.2, use a non-linear mapping model to generate personalized rehabilitation training plan parameters. Express the training intensity parameter using the formula:

[0134] I = ξ·ln(1 + κ) + ζ,

[0135] where I is the generated training intensity parameter, reflecting the training load level in the rehabilitation training plan; ξ is the training intensity scale coefficient, adjusting the index magnification; κ is the overall training preparation index of the patient; and ζ is the training reference offset.

[0136] Step S4: During the implementation of the training plan formulated in Step S3, adjust the training plan according to the real-time collected training feedback data to make the training content conform to the actual state of the patient.

[0137] Step 4.1: Based on the personalized training plan generated in Step S3, collect the motion data and physiological feedback data generated by the patient during the training in real time. Use an adaptive Kalman filter to monitor and feedback-adjust the training process in real time based on the data. Use the formula to perform state estimation on the training feedback signal: Y k|k-1 = F k Y k-1|k-1 + B k U k ,

[0138] where Y k|k-1 is the signal state estimation predicted at time k, F k is the state transition matrix, B k is the control influence matrix, U k is the control input vector, and Y k-1|k-1 is the state estimation at the previous moment;

[0139] Step 4.2: According to the real-time feedback data obtained in Sub-step 4.1, use a dynamic adjustment algorithm based on fuzzy logic to correct the training plan in real time. The adjustment process uses the fuzzy logic control formula:

[0140] Δu = K f ·sigm(L(∈)),

[0141] where Δu is the adjustment amount of the training parameter, K f is the fuzzy control gain, L is the training adjustment rule set, and ∈ is the current training feedback error;

[0142] Step 4.3, based on the training plan adjusted in Step 4.2, verify the effectiveness of the training plan using the new feedback data of the patient, and correct the optimization of the plan using an adaptive learning algorithm. The optimization process uses the formula of the reinforcement learning algorithm:

[0143]

[0144] Among them, Q(s,a) is the expected reward in the current state s and action a, a is the learning rate, R(s,a) is the immediate reward, γ is the discount factor, s ′ is the new state, a ′ is the new action;

[0145] Step 4.4, according to the optimized training plan in Step 4.3, conduct actual training implementation, and collect and analyze the training effect data during the implementation process. The evaluation of the training effect uses a system evaluation algorithm, and a comprehensive evaluation is conducted based on real-time feedback and target setting, using the following formula:

[0146]

[0147] Among them, E is the comprehensive evaluation result, w i is the weight of each evaluation index, Δ i is the change in the training effect of each index, v j is the weight of external influencing factors, η j is the corresponding influence amount, and λ1 and λ2 are weight adjustment coefficients;

[0148] Step S5, conduct a safety risk assessment of the home environment based on the environmental data collected in Step S1 and provide environmental improvement suggestions;

[0149] Step 5.1, based on the home environment data collected in Step S1, use the principal component analysis method to extract features and reduce the dimension of the environmental data, and express the extracted environmental features using the formula:

[0150]

[0151] Among them, X e is the original home environment data after standardization and normalization, μ e is the mean vector, W e is the feature vector matrix, and F e is the environmental feature vector;

[0152] Step 5.2, based on the environmental feature vector F e extracted in Step 5.1, use a logistic regression model to evaluate the safety risk of the home environment, and express the environmental safety risk score using the formula:

[0153]

[0154] Among them, R e is the safety risk score of the home environment, and f e,i is the i-th feature component in the environmental feature vector F e , α i is the weight coefficient corresponding to the i-th feature, n is the total number of environmental features, β is the bias term of the logistic regression model, and exp(·) is the natural exponential function;

[0155] Step 5.3, based on the safety risk score R e obtained in Step 5.2, an improved recommendation generation algorithm based on fuzzy inference is adopted, and the membership function is used to perform defuzzification mapping on the risk score. The mapping process is expressed as: S e = μ F (R e ; θ1, θ2),

[0156] where μ F is the membership function, defined as:

[0157]

[0158] where S e is the generated environmental improvement recommendation index, R e is the safety risk score of the home environment, and μ F (R e ; θ1, θ2) is the membership function of the risk score at different risk levels,

[0159] θ1 is the low-risk threshold of the risk score. When it is lower than the threshold, it indicates a low environmental risk;

[0160] θ2 is the high-risk threshold of the risk score. When it is higher than the threshold, it indicates a high environmental risk;

[0161] Step S6, based on the implementation results of the training plan adjusted in Step S4 and the evaluation results in Step S5, comprehensively calculate the patient's comprehensive health index, and transmit the evaluation results to the remote medical platform for medical staff to refer to;

[0162] Step 6.1, on the basis of the real-time health status index θ 1,t obtained in Step S2, the training preparation index θ 2,t obtained in Step S3, and the environmental improvement recommendation index θ 3,t generated in Step S5, a weighted average fusion method is used to comprehensively evaluate the long-term rehabilitation effect of the patient, and the long-term effect evaluation results are expressed using the formula as follows:

[0163] Among them, E t is the comprehensive rehabilitation effect evaluation value at time t, θ 1,t is the real-time health status index obtained through step S2 at time t, θ 2,t is the training preparation index obtained through step S3 at time t, θ 3,t is the environmental improvement suggestion index obtained through step S5 at time t, and π1, π2, and π3 are the reliability weights of the corresponding indexes;

[0164] Step 6.2, based on the long-term effect evaluation value E calculated in step 6.1 t and the prediction error between the actually observed rehabilitation effect, adaptively update the reliability weights of each index through an online learning algorithm, and adopt the update formula:

[0165] π i (t + 1) = π i (t) + η[ε t ·θ i,t - λ·π i (t)],

[0166] Among them, π i (t + 1) is the updated weight of the i-th index at time t + 1, π i (t) is the current weight of the i-th index at time t, η is the learning rate, ε t is the prediction error at time t, θ i,t is the value of the i-th index at time t, and λ is the regularization parameter;

[0167] Step 6.3, after completing steps 6.1 and 6.2, based on the latest comprehensive rehabilitation effect evaluation value E t use a non-linear mapping model to determine whether it is necessary to adjust the rehabilitation training plan, and adopt the decision function: D t = σ(ζ·(E t - τ)),

[0168] Among them, D t is the plan adjustment decision variable at time t, σ is the sigmoid function, E t is the comprehensive rehabilitation effect evaluation value at time t, τ is the preset rehabilitation effect target threshold, and ζ is the decision sensitivity coefficient;

[0169] Step S7, integrate the data and information obtained in the above steps through a standardized interface and adopt encryption measures to ensure the security of data transmission and storage;

[0170] Step 7.1: Based on the real-time health status indicators obtained in Step S2, the training preparation index generated in Step S3, the real-time feedback data collected in Step S4, the environmental improvement suggestion indicators extracted in Step S5, and the long-term rehabilitation effect value calculated in Step S6, fuse each data into a comprehensive feature vector z, and construct a prognosis prediction model using a prediction model based on ensemble learning. Use the formula:

[0171] to express the prognosis prediction output,

[0172] where, is the predicted future rehabilitation effect value, M is the number of base learners, and θ j is the weight coefficient of the j-th base learner, and g j (z) is the prediction output of the j-th base learner for the input feature vector z, and z is the comprehensive feature vector;

[0173] Step 7.2: Based on the prognosis prediction model constructed in Step 7.1, train and optimize the model parameter set Θ using the mean squared error loss function with a regularization term. Use the formula:

[0174] to express the model loss,

[0175] where, L(Θ) is the total loss value of the model parameter set Θ, N is the number of training samples, is the rehabilitation effect value obtained by the model prediction for the i-th sample, and Y i is the actually observed rehabilitation effect value for the i-th sample, Θ is the set of all trainable parameters of the model, and λ is the regularization coefficient;

[0176] Step 7.3: Based on the predicted rehabilitation effect obtained after training and optimization in Step 7.2, combine it with the preset rehabilitation effect target threshold Y th , and use a non-linear mapping decision model to generate a risk warning signal and a personalized intervention strategy adjustment suggestion. Use the formula:

[0177] to express the decision output,

[0178] where, W is the risk warning weight or intervention strategy adjustment signal, σ represents the sigmoid function, κ is the decision sensitivity coefficient, is the rehabilitation effect prediction value output by the prognosis prediction model, and Y th is the preset rehabilitation effect target threshold.

[0179] Step S1: Multi-sensor data collection and preprocessing. Use wearable sensors and environmental sensors to simultaneously obtain patient movement, heart rate, gait, and home environment data. After collection, the raw data is subjected to noise removal and filtering.

[0180] In step 1.1, the adaptive Kalman filtering algorithm is used. The system first makes a prediction and then updates the state estimate. The prediction formula, prediction error covariance, and Kalman gain are all calculated according to the set formulas.

[0181] Next, in step 1.2, the multi-scale wavelet decomposition and reconstruction algorithm is applied to the filtered signal. Threshold processing is performed on the detail coefficients of each layer and the signal is reconstructed to complete signal standardization and normalization. The algorithm makes full use of the local characteristics of wavelet decomposition to ensure fine data processing.

[0182] Step S2: Multi-sensor data fusion. The preprocessed multi-channel data are aligned and weighted fused to form a comprehensive index reflecting the overall health status of the patient.

[0183] In step 2.1, a calibration model is used to achieve data alignment. The signals of each sensor are multiplied by their respective calibration gains and the biases are subtracted.

[0184] Then, in step 2.2, the weights of each sensor are determined based on the exponential smoothing algorithm. The measurement uncertainties of each sensor affect the weight distribution to ensure the rationality of data fusion.

[0185] In step 2.3, the weighted linear fusion algorithm is used to add the calibrated data according to the preset weights to generate a comprehensive health status index. The comprehensive index can reflect the real-time status of the patient.

[0186] Step S3: Generate a personalized rehabilitation training plan. Based on the fusion index in step S2, data on the patient's movement, cognition, and home environment safety are extracted.

[0187] In step 3.1, each item of data is normalized and integrated into a training capacity factor using a weighted linear normalization model. Each index has its own weight, and the result is intuitive and easy for subsequent calculations.

[0188] In step 3.2, by combining the comprehensive health index and the training capacity factor, the patient's training readiness index is obtained through a multi-factor linear combination.

[0189] Furthermore, in step 3.3, specific training intensity parameters are generated through a non-linear mapping model. A regulation factor is used in the mapping process, taking into account the patient's existing physical fitness and also considering the rehabilitation needs to achieve the goal of personalized plan formulation.

[0190] Step S4: Real-time feedback and dynamic adjustment of the plan. During the training process, the system continuously collects the patient's movement and physiological feedback data.

[0191] First, in step 4.1, the real-time data are processed again through an adaptive Kalman filter to obtain an accurate state estimate.

[0192] Step 4.2 introduces a fuzzy logic control algorithm to dynamically adjust the training plan according to the current feedback error. The fuzzy control rules are not restricted to fixed formulas, but combine experience and real-time data to ensure that the training load always matches the patient's condition.

[0193] In step 4.3, a reinforcement learning algorithm is used to perform online learning on the plan. The new data continuously updates the training model, making the plan tend to be optimal.

[0194] Finally, in step 4.4, through the system evaluation algorithm, various training effect indicators are integrated to verify whether the adjusted training plan achieves the expected effect. The evaluation results provide a basis for subsequent adjustments.

[0195] Step S5: Home environment safety assessment and improvement suggestions, considering the safety of the patient during home training.

[0196] In step 5.1, the principal component analysis method is used to extract the features of the home environment data. After the data is standardized and dimension-reduced, the key features are retained.

[0197] Then in step 5.2, a logistic regression model is used to evaluate the environmental safety risk and give a risk score. The score is intuitive and quantifiable.

[0198] In step 5.3, through fuzzy inference, a defuzzification mapping is performed on the risk score to generate targeted environmental improvement suggestions. Here, it is no longer restricted to fixed rules, but is flexibly adjusted to ensure the practicality of the suggestions.

[0199] Step S6: Comprehensive rehabilitation effect evaluation and remote data transmission. During and after the training process, the system comprehensively evaluates the long-term rehabilitation effect of the patient.

[0200] In step 6.1, by combining real-time health indicators, training preparation index and environmental improvement suggestions, the weighted average method is used to calculate the rehabilitation effect evaluation value. The weight distribution fully considers the reliability of each indicator.

[0201] In step 6.2, the index weights are adaptively updated through an online learning algorithm to reduce the prediction error.

[0202] In step 6.3, it is determined whether further adjustment of the plan is needed according to the non-linear mapping decision model. This decision function is simple and direct, and can timely feedback potential risks.

[0203] The evaluation results are transmitted to the remote medical platform in real time, providing timely data support for doctors and realizing remote monitoring and intervention.

[0204] Step S7: Data integration and secure transmission. Finally, the data generated in each step is integrated through a standardized interface.

[0205] Step 7.1 Aggregate each data into a comprehensive feature vector as the input of the prognosis prediction model.

[0206] Step 7.2 Train the prediction model based on the mean square error loss function with a regularization term, and continuously optimize the model parameters.

[0207] Step 7.3 Use the non-linear mapping decision model to generate risk warning signals and suggestions for adjusting intervention strategies. Finally, the system ensures the security of all data transmission and storage through encryption measures to ensure that patients' privacy and data security are not threatened.

[0208] A terminal device, which includes:

[0209] A processor, which is used to execute the steps of the rehabilitation training method for Parkinson's patients;

[0210] A memory, which is used to store the relevant data of the method;

[0211] A sensor interface, which is used to connect and receive data from wearable sensors and environmental sensors;

[0212] A user interface, which is used to display the feedback results and adjustment suggestions of the patient's rehabilitation training to the user;

[0213] A communication module, which is used to transmit data with a remote medical platform.

[0214] A storage medium, which is used to store computer-executable program codes. When the program codes are executed, they implement the rehabilitation training method for Parkinson's patients, specifically including:

[0215] A data acquisition module, which is used to collect the patient's movement data, heart rate data, gait data, and home environment data through sensors;

[0216] A data processing module, which is used to perform noise removal, filtering, standardization, and normalization processing on the collected data, and perform multi-sensor data fusion;

[0217] A training plan generation module, which is used to develop a personalized rehabilitation training plan based on health status indicators, motor ability, cognitive ability, and home environment safety;

[0218] A feedback adjustment module, which is used to adjust the training plan according to real-time training feedback data;

[0219] An evaluation module, which is used to evaluate the rehabilitation effect of the patient and generate a personalized intervention strategy according to the evaluation results;

[0220] A data storage module, which is used to store the patient's training data, evaluation results, and all relevant data.

[0221] The present invention integrates multi-sensor data acquisition, filtering processing, data fusion, personalized solution generation, real-time dynamic adjustment, environmental safety assessment, and remote data transmission into a complete rehabilitation system. The method is based on multiple health data of patients and realizes the real-time generation and adjustment of personalized training solutions through algorithms such as adaptive Kalman filtering, wavelet decomposition, multi-factor fusion, and non-linear mapping. The system can monitor the movement and physiological parameters of patients in real time at home, and at the same time conduct a safety risk assessment of the living environment, providing a basis for doctors to carry out remote intervention. The implementation process of this technology has clear steps, a compact and efficient processing flow.

[0222] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A rehabilitation training method for Parkinson's disease patients based on artificial intelligence, characterized in that: include: Step S1, using wearable sensors and environmental sensors to collect patient motion data, heart rate data, gait data and home environment data, and performing noise removal, filtering, standardization and normalization processing on the collected raw data; Step S2, based on the data in step S1, weighted fusion is performed on the data collected by different sensors to form a comprehensive index reflecting the overall health status of the patient; Step S3, formulating a training plan based on the comprehensive indicators obtained in step S2 and the patient's motor ability, cognitive ability and home environment safety; Step S4, during the implementation of the training program formulated in step S3, adjusting the training program according to the training feedback data collected in real time, so that the training content is consistent with the actual condition of the patient; Step S5, based on the environmental data collected in step S1, a safety risk assessment is performed on the home environment and environmental improvement suggestions are provided; Step S6, comprehensively calculating the patient's comprehensive health index based on the implementation results of the training program adjusted in step S4 and the evaluation results in step S5, and transmitting the evaluation results to the telemedicine platform for reference by medical personnel; Step S7, integrating the data and information obtained in the above steps through a standardized interface and using encryption measures to ensure data transmission and storage security.

2. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S1 further comprises: Step 1.1, based on the collected raw data, the adaptive Kalman filter algorithm is used to eliminate the signal noise, and the prediction and update recursive method is used for state estimation and noise correction. The prediction step uses the formula: Y k|k-1 =F k Y k-1|k-1 +B k U k , Forecast error covariance formula: Kalman gain formula: Status update formula: Y k|k =Y k|k-1 +K k (Z k -H k Y k|k-1 ), Among them, Y k|k-1 is the signal state estimate predicted at time k, F k is the state transfer matrix describing the signal change law, Y k-1|k-1 is the state estimate at the previous moment, B k is the control influence matrix, U k is the control input vector reflecting the external disturbance, E k|k-1 is the forecast error covariance matrix, Q k is the process noise covariance matrix reflecting the intrinsic uncertainty of the signal, H k is the observation matrix that maps the state to the observation space, Z k is the actual observed value at time k, R k is the observation noise covariance matrix reflecting the sensor measurement error, K k is the Kalman gain used to adjust the weight between prediction and observation, Y k|k is the current state estimate, Yes H k The transposed matrix of Yes F k The transposed matrix of Step 1.2, based on the signal processed by Kalman filtering in step 1.1, a multi-scale wavelet decomposition and reconstruction algorithm is used to filter, standardize and normalize the signal, using the formula: Where Θ is the signal after filtering and normalization, Π is the reconstruction scale factor, J is the number of layers of wavelet decomposition, and Ω j is the detail coefficient obtained by decomposition at the jth layer, is the result of threshold function processing on the j-th layer detail coefficient, Σ j is the adaptive threshold value of the jth layer.

3. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S2 further comprises: Step 2.1, the output signal ω of each sensor obtained after filtering, standardization and normalization in step S1 i Based on this, a correction model is used to realize multi-sensor data alignment, and the corrected data is expressed by the formula: Φ i =α i ·ω i +β i , Among them, Φ i is the output signal of the ith sensor after correction, α i is the sensor correction gain, β i is the bias correction value of the i-th sensor, ω i is the original signal of the i-th sensor obtained by the standardization process in step S1; Step 2.2, the correction data Φ obtained in step 2.1 i On this basis, a weight allocation algorithm based on exponential smoothing is used to determine the contribution ratio of each sensor in data fusion, and the weight of each sensor is expressed by the formula: Among them, w i is the weight of the ith sensor, γ i is the measurement uncertainty of the i-th sensor, the index of all sensors in j; Step 2.3, the correction data Φ obtained in step 2.1 i The weight w determined in step 2.2 i Based on this, a weighted linear fusion algorithm is used to realize the integration of multi-sensor data, and the comprehensive health status index generated after fusion is expressed by the formula: Among them, Δ is the final fusion result, w i is the weight of the ith sensor, Φ i is the output signal of the ith sensor after correction.

4. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S3 further comprises: Step 3.1, based on the fusion comprehensive index Δ obtained in step S2, extract the patient's motor ability, cognitive ability and home environment safety data and normalize the data. Use a weighted linear normalization model to achieve data integration and use the formula to express the patient's ability data integration results: in, is the training capacity factor, χ m is the athletic ability weight coefficient, ρ m is the normalized patient motor ability score, χ c is the cognitive ability weight coefficient, ρ c is the normalized patient cognitive ability score, χ h is the weight coefficient of home environment safety, ρ h is the normalized home environment safety score; Step 3.2, the training capacity factor ρ obtained in step 3.1 h Based on the fusion step S2, the comprehensive health status index Δ is generated. The training readiness index of the patient is calculated based on the multi-factor linear combination model. The training readiness index is expressed by the formula: Among them, κ is the patient's overall training readiness index, μ is the weight coefficient of the fusion comprehensive index, Δ is the final fusion result, ν is the weight coefficient of the training capacity factor, is the training capacity factor; Step 3.3, based on the training readiness index κ obtained in step 3.2, a nonlinear mapping model is used to generate the parameters of the personalized rehabilitation training program and the training intensity parameters are expressed using the formula: I=ξ·ln(1+κ)+ζ, Among them, I is the generated training intensity parameter reflecting the training load level in the rehabilitation training program, ξ is the training intensity scale coefficient adjustment index magnification, κ is the patient's overall training readiness index, and ζ is the training baseline offset.

5. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S4 further comprises: Step 4.1, based on the personalized training program generated in step S3, the motion data and physiological feedback data generated by the patient during the training process are collected in real time, and the training process is monitored and feedback adjusted in real time using an adaptive Kalman filter based on the data, and the training feedback signal is estimated using the formula: Y k|k-1 =F k Y k-1|k-1 +B k U k , Among them, Y k|k-1 is the signal state estimate predicted at time k, F k is the state transfer matrix, B k is the control influence matrix, U k is the control input vector, Y k-1|k-1 is the state estimate at the previous moment; Step 4.2, according to the real-time feedback data obtained in sub-step 4.1, the training plan is corrected in real time using a dynamic adjustment algorithm based on fuzzy logic. The adjustment process uses the fuzzy logic control formula: Δu=K f ·sigm(L(∈)), Among them, Δu is the adjustment amount of the training parameters, K f is the fuzzy control gain, L is the training adjustment rule set, ∈ is the current training feedback error; Step 4.3: Based on the training program adjusted in step 4.2, the new feedback data from patients is used to verify the effect of the training program, and the adaptive learning algorithm is used to modify the program optimization. The optimization process uses the reinforcement learning algorithm formula: Among them, Q(s,a) is the expected reward under the current state s and action a, a is the learning rate, R(s,a) is the immediate reward, γ is the discount factor, s ′ is the new state, a ′ It’s a new move; Step 4.4, according to the training plan optimized in step 4.3, actual training is implemented, and training effect data is collected and analyzed during the implementation process. The evaluation of training effect adopts a system evaluation algorithm, and a comprehensive evaluation is performed based on real-time feedback and goal setting, using the following formula: Among them, E is the comprehensive evaluation result, w i is the weight of each evaluation indicator, Δ i is the change in the training effect of each indicator, v j is the weight of external influencing factors, η j is the corresponding influence, λ1 and λ2 are weight adjustment coefficients.

6. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S5 further comprises: Step 5.1: Based on the home environment data collected in step S1, the principal component analysis method is used to extract features and reduce the dimension of the environment data, and the extracted environment features are expressed using the formula: Among them, X e is the original data of the home environment after standardization and normalization, μ e is the mean vector, W e is the eigenvector matrix, F e is the environmental feature vector; Step 5.2, based on the environmental feature vector F extracted in step 5.1 e , a logistic regression model was used to evaluate the home environmental safety risk, and the environmental safety risk score was expressed using the formula: Among them, R e is the home environment safety risk score, f e,i is the environmental feature vector F e The i-th eigencomponent, α i is the weight coefficient corresponding to the i-th feature, n is the total number of environmental features, β is the bias term of the logistic regression model, and exp(·) is the natural exponential function; Step 5.3: Based on the security risk score R obtained in step 5.2 e , an improved suggestion generation algorithm based on fuzzy reasoning is adopted, and the membership function is used to defuzzify the risk score. The mapping process is expressed as: S e =μ F (R e ; θ1,θ2), Among them, μ F is the membership function, defined as: Among them, S e is the generated environmental improvement suggestion index, R e is the home environment safety risk score, μ F (R e ; θ1, θ2) are the membership functions of risk scores at different risk levels, θ1 is the low risk threshold of the risk score, and when it is below the threshold, it indicates that the environmental risk is low; θ2 is the high risk threshold of the risk score. When it is higher than the threshold, it indicates that the environmental risk is high.

7. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S6 further comprises: Step 6.1, the real-time health status indicator θ obtained in step S2 1,t , the training preparation index θ obtained in step S3 2,t and the environmental improvement suggestion index θ generated in step S5 3,t On this basis, the weighted average fusion method is used to comprehensively evaluate the long-term rehabilitation effect of patients, and the long-term effect evaluation results are expressed by the formula as follows: Among them, E t is the comprehensive rehabilitation effect evaluation value at time t, θ 1,t is the real-time health status indicator obtained by step S2 at time t, θ 2,t is the training readiness index obtained at time t by step S3, θ 3,t is the environmental improvement suggestion index obtained by step S5 at time t, π1, π2, π3 are the reliability weights of the corresponding index; Step 6.2: Based on the long-term effect evaluation value E calculated in step 6.1 t The prediction error between the actual observed rehabilitation effect is adaptively updated by the online learning algorithm for the reliability weight of each indicator, using the update formula: p i (t+1)=π i (t)+η[ε t ·i i,t -l·p i (t)], Among them, π i (t+1) is the i-th index weight after updating at time t+1, π i (t) is the current weight of the i-th indicator at time t, η is the learning rate, ε t is the prediction error at time t, θ i,t is the value of the i-th index at time t, and λ is the regularization parameter; Step 6.3: After completing steps 6.1 and 6.2, based on the latest comprehensive rehabilitation effect evaluation value E t The nonlinear mapping model is used to determine whether the rehabilitation training program needs to be adjusted, and the decision function is adopted: D t =σ(ζ·(E t -τ)), Among them, D t is the decision variable for adjusting the solution at time t, σ is the sigmoid function, E t is the comprehensive rehabilitation effect evaluation value at time t, τ is the preset rehabilitation effect target threshold, and ζ is the decision sensitivity coefficient.

8. The artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to claim 1, characterized in that: The step S7 further comprises: Step 7.1, based on the real-time health status index obtained in step S2, the training readiness index generated in step S3, the real-time feedback data collected in step S4, the environmental improvement suggestion index extracted in step S5, and the long-term rehabilitation effect value calculated in step S6, each data is integrated into a comprehensive feature vector z, and a prognosis prediction model based on ensemble learning is constructed, using the formula: Express prognostic prediction output, in, is the predicted future rehabilitation effect value, M is the number of base learners, θ j is the weight coefficient of the j-th basis learner, g j (z) is the predicted output of the j-th base learner for the input feature vector z, where z is the comprehensive feature vector; Step 7.2, based on the prognosis prediction model constructed in step 7.1, the model parameter set ΘΘ is trained and optimized through the mean square error loss function with regularization term, using the formula: Expression model loss, Where L(Θ) is the total loss value of the model parameter set Θ, N is the number of training samples, is the rehabilitation effect value of the i-th sample predicted by the model, Y i is the actual observed rehabilitation effect value of the i-th sample, Θ is the set of all trainable parameters of the model, and λ is the regularization coefficient; Step 7.3, based on the predicted rehabilitation effect obtained after training optimization in step 7.2 Combined with the preset rehabilitation effect target threshold Y th , a nonlinear mapping decision model is used to generate risk warning signals and personalized intervention strategy adjustment suggestions, using the formula: Express decision output, Where W is the risk warning weight or intervention strategy adjustment signal, σ represents the sigmoid function, κ is the decision sensitivity coefficient, is the predicted value of rehabilitation effect output by the prognostic prediction model, Y th It is the preset rehabilitation effect target threshold.

9. A terminal device, characterized in that: The equipment includes: A processor, configured to execute the steps of the artificial intelligence-based rehabilitation training method for Parkinson's disease patients according to any one of claims 1 to 8; A memory for storing data related to the method; A sensor interface for connecting to and receiving data from wearable sensors and environmental sensors; A user interface, used to show users the feedback results and adjustment suggestions of the patient's rehabilitation training; Communication module, used for data transmission with the telemedicine platform.

10. A storage medium, characterized in that: The medium is used to store computer executable program codes, and when the program codes are executed, the method for rehabilitation training of Parkinson's patients based on artificial intelligence according to any one of claims 1 to 8 is implemented, specifically comprising: A data acquisition module, which is used to collect patient movement data, heart rate data, gait data and home environment data through sensors; The data processing module is used to remove noise, filter, standardize and normalize the collected data, and perform multi-sensor data fusion; A training program generation module is used to develop personalized rehabilitation training programs based on health status indicators, motor ability, cognitive ability, and home environment safety; A feedback adjustment module is used to adjust the training plan according to real-time training feedback data; The evaluation module is used to evaluate the patient's rehabilitation effect and generate personalized intervention strategies based on the evaluation results; The data storage module is used to store the patient's training data, evaluation results, and all related data.

Images